Distributed Situation Awareness for the Civil Cockpit: Theory and Operationalisation Abstract This thesis combines the theoretical frameworks of internal human cognition with the framework of distributed cognition for the civil cockpit, to create a combined situation awareness theory of distributed situation awareness. The theory is operationalised by being related to measurement methods. An experiment was conducted to test the theory and the distribution of attention in the cockpit as well as the adequacy of the situation awareness system in an altitude bust anomaly. The theory was found to be adequate for the operationalisation of the situation awareness system of the experiment. Distributed attention was found to exist for the landing phase of one flight, and the situation awareness system was found to be adequate for safe handling of the altitude bust event, but inadequate for the safe handling of the map shift event. Reference notice: Lundberg, J. (1999). Distributed Situation Awareness for the Civil Cockpit: Theory and operationalisation. LIU-KOGVET-D-0013-SE, Linköpings Universitet, Linköping. i Table of contents 1. Introduction.......................................................................................................... 1 2. Theory .................................................................................................................. 3 2.1. Theoretical grounding of the SA system ....................................................... 5 2.1.1. SA as a product....................................................................................... 5 2.1.2. Updating SA ........................................................................................... 7 2.1.3. Examples of distributed SA.................................................................... 9 2.1.4. Compatibility with SA definitions of the literature .............................. 10 2.2. The proposed SA theory.............................................................................. 13 2.2.1. Situation models, mediations and function........................................... 14 2.2.2. Updating and sampling processes......................................................... 15 2.2.3. The presentation process....................................................................... 16 2.2.4. The monitoring process ........................................................................ 17 2.2.5. The relevance process........................................................................... 18 2.2.6. Properties of distributed and localised systems .................................... 19 2.2.7. SA Error................................................................................................ 20 2.3. Measurement techniques for SA systems.................................................... 22 2.3.1. Eye point of Gaze ................................................................................. 22 2.3.2. Questionnaires ...................................................................................... 23 2.3.3. Physiological measures......................................................................... 23 2.3.4. Event-related potentials ........................................................................ 24 2.3.5. Transcripts of verbal messages ............................................................. 24 2.3.6. Performance measures .......................................................................... 25 2.4. Theory summary and discussion ................................................................. 26 2.4.1. Concluding operationalisation model for SA ....................................... 26 2.4.2. Relevant SA.......................................................................................... 27 2.4.3. Safe distributions of labour between actors and artefacts..................... 28 2.4.4. Safe practices, tradition and training .................................................... 28 2.4.5. Optimal SA distribution and component failure................................... 29 2.4.6. Consequences for design ...................................................................... 30 3. The experiment .................................................................................................. 34 3.5. Method ........................................................................................................ 34 3.5.1. Test site................................................................................................. 34 3.5.2. Participants ........................................................................................... 34 3.5.3. Procedure .............................................................................................. 34 3.5.4. Design................................................................................................... 37 3.5.5. Equipment............................................................................................. 37 3.5.6. Research questions ............................................................................... 38 3.5.7. Operationalisation of the research questions ........................................ 39 3.6. Measurement results.................................................................................... 42 3.7. Conclusions ................................................................................................. 44 4. General discussion ............................................................................................. 47 5. References.......................................................................................................... 49 6. Appendix A........................................................................................................ 52 7. Appendix B ........................................................................................................ 55 ii 1. INTRODUCTION To have situational awareness (SA) has traditionally been to know what is going on, and to predict what will happen. Recent developments of the study of SA has presented a more complex, yet still simple and straightforward view of what it is to have good SA. It is now widely accepted to view SA as a dynamic process rather than a product. The next step has been to view SA as something localised to the individual, yet co-ordinated between, built in concert with, and maintained in, a team of actors and tools. Tool free environments are very scarce, and so are environments where there is just one actor, making these issues essential to a theory of SA. To have good SA, then, is now to know what is going on, how that came to be, and to predict what will happen in the near and more distant future. For a system of actors and tools, to have good SA is to have a sufficiently redundant SA divided over the actors, to have accurate SA, and to have a good knowledge of the quality of the SA in the system. To have an environment for good SA, is to have a safe distribution of labour, with safe practices for maintaining SA, between actors and tools. The aim of this thesis The aim of this thesis is to operationalise the concept of situation awareness (SA) for individuals in co-operation with each other in a dynamic environment, the civil cockpit. In the first part of this thesis I will present a theory and a descriptive framework for SA systems with actors and artefacts. It is common to discuss how to operationalise the concept of SA, how to relate theory to measurement tools useable in real and experimental situations. This will be done in this thesis as well, to anchor the theory in practice. The model will finally be applied to an experimental flight situation, to evaluate the SA in the situation, and to evaluate the suitability of the theory to a practical problem. VINTHEC This thesis is made in co-operation with the VINTHEC (Visual interaction in the cockpit) project, sponsored by the European Commission. The VINTHEC project is a joint research effort between companies1 in several countries. The project aim is to clarify the concept of situational awareness and to test and evaluate measurement tools for 1 National Aerospace Laboratory (NL) British Aerospace (UK), Defence Evaluation & Research Agency (UK), National Defence Research Establishment (SE), Luleå University of Technology (SE), Saab AB (SE), H.A. Mooij Holding B.V (NL) 1 situational awareness, pilot workload and performance in simulation environments. 2 2. THEORY After this short introduction to SA, there is a chapter reviewing SA theory from the literature. In the next chapter the combined theory of localised and distributed SA, which emerged from the review, will be described. Then I will present a review of measurement methods for SA. The theory part of this thesis will end with a summary and a discussion of the proposed SA system. Situation awareness (SA) SA is in short knowledge of what is going on. It is conscious knowledge of what is happening, what has happened, what will happen if, and what will happen if not. But SA is not the ability or knowledge of how to act. Since SA incorporates awareness it is not automatic, unconscious use of data. As part of SA an estimation of the quality of the awareness can be included, an estimation of its correctness and completeness. But since this can also be excluded form the concept of SA, it is simply termed meta-SA, to distinguish quality from content. Device models, to know how some device works, or how a human will act in a given situation, is not part of SA. However, to have an accurate device model is essential to get SA, since if one knows how some artefact or human will act in a situation, then it is possible to predict what will happen, in short, to get SA of future states. SA consists of, first, unstructured aspects of the dynamic situation, such as for an aircraft the current airspeed and the current weight. Second, SA is made of what gives the aspects meaning, situation models, such as the landing situation model. Active situation knowledge (ASK) For devices, it is pointless to discuss awareness. Instead devices can actively process data about a dynamic situation, thus devices can have active situation knowledge. Since the quality and number of situations a device recognises and handles may vary, the weaker term active situation knowledge (ASK) will be used for devices taking part in the SA system. By using a different term for mechanical information processors executing SA processes, simple mechanical agents can be recognised as contributing to the SA of the system without attributing full fledged awareness, equivalent to human awareness, to them. Mediation 3 SA or ASK is always mediated, either in a mental model, such as in the human working memory, or in a physical media, such as a display of some kind Situation assessment Furthermore, the maintenance of SA relies on several processes. Together these are called situation assessment processes. First, there is an updating process, responsible for the maintenance of the model, to keep it up to date. Second, there may be a presentation process, which presents the model for some other agent to perceive. Third, there may be a monitoring process, which evaluates whether the two processes work, and whether the SA is adequate. Scope of this thesis The model of this thesis can be used to model the distribution of processes between agents, by what physical medium the situation models and aspects are mediated. But the physical possibilities for relevant action, the exact properties of the physical medium used, and the physical possibilities for manipulation of the situation models, mental or physical, are beyond the scope of this thesis. These properties can and should be discussed with the resulting situation model for which they are relevant, but the properties are not described in this thesis, but have to be found elsewhere. 4 2.1. Theoretical grounding of the SA system The objective of this chapter is to ground the theoretical model presented in the subsequent chapters in the models and observations of SA related research. 2.1.1. SA as a product Endsley (1988, 1990; Jones & Endsley, 1996) proposes three components, or levels, of SA, the first level being knowledge of aspects of the environment, the second being comprehension of the meaning of the aspects, and the third being projection of future events. Related to the model of this thesis, the first level aspects can be found as non interpreted, that is not as parts of any situation model, or as interpreted, that is as parts of one or more situation model. The second level of comprehended models, can be the result of comprehending elements of the situation, or comprehending a represented situation model. The level three, future events can be the result of an internal projection process or the result of comprehending an external representation of future situations. Artman and Garbis (1998) propose three functional types of SA models, contemporary (current states) SA, prospective (future states) and retrospective (past states) SA. The current states and future states SA corresponds with the level two and level three SA of Endsley’s (1988) model, whereas the past states SA is added. The focus of Artman and Garbis study is on the usage and mediation of SA in artefacts. Device models The importance of a correct device model for pilots of passenger jet aircraft is illustrated by an auto pilot failure incident (Tenney, Adams, Pew, Huggins, Rogers, 1992). As one engine of the aircraft failed, and the aircraft started to turn, the auto pilot started to compensate for this by turning the control wheel of the aircraft to keep the programmed flight path. Oblivious to this tug-of-war between the aircraft and the autopilot, after four minutes, the pilot took manual command of the wheel and turned of the autopilot. Instantly the aircraft won the tug of war, and was only stabilised after a fall of 10 000 feet. This illustrates how an inaccurate or missing device model of what was going on, contributed to 5 an incident when the control was transferred from automated equipment to a crew member. Hutchins (1990) describes how the crew of the ship form a chain of information processing, where the simple perception is done at one end, and the highest level interpretation is done at the other end. The human components of this processing chain, as individuals start in the perception end of the chain, and may move up the chain of processing only when the current task is mastered. This gradual ascent of the chain of processing causes the individual to know what has been done with the information received, and how it has been processed. Thus, the individual has a device model of the team navigation chain covering all steps before him or her. Meta-SA In the VINTHEC report (forthcoming) SA is described as consisting of first, all the information that concerns the dynamic situation. Then there is the subset of information about the situation that a person is aware of. This can be compared with the subset of information that is required to be fully aware of what is relevant in the current situation. The area covered both by the pilot awareness and required awareness constitutes the actual SA. From the VINTHEC model, the relative awareness can be seen as the quotient between the required awareness and the actual awareness. Then, the estimation of the relative awareness by the pilot is called the awareness of relative awareness. It may have four outcomes, a high SA with a correct estimation, a low SA with a correct estimation, a high SA with an underestimation of SA, or a low SA with an overestimation of SA. This awareness is what is in this thesis called meta-SA, and the result of some monitoring process. Summary SA consists of six kinds of components, models of present events, models of future events, models of past events, and knowledge of aspects of the environment. Device models are knowledge about how actors and tools function. This is not SA, but essential to achieve SA, especially when the functionality of actors or tools change. A sixth component of SA is made of meta-SA, awareness about the quality of the SA. 6 2.1.2. Updating SA In its simplest form, the perceptual cycle (Neisser, 1976) consists of three entities, an object, a schema2 and perceptual exploration. The object is in this case the world, the schema is in most cases a mental situation model, and perceptual exploration is the sampling of sensory data from the world. In Neisser’s model, the schema directs perceptual exploration, causing the components of the situation model to change. Endsley (1988, 1990; Jones & Endsley, 1996) presents SA as emerging in three levels, the first level being perception of elements in the environment, the second being comprehension of the situation, and the third being projection of future status. This process is guided by expectation, by preconceptions, objectives, experience and training and restricted by the ongoing task workload. In this thesis, the level three projection does not have to be mental projection, it can be mechanical projection as in the case with the GPWS (Bateman, 1997). The resulting situation model can then be presented for another actor to perceive. The model of this thesis is partly analogous to the Wellens (1993) extended dynamic decision-making model for group decision making. In the model for the single actor, all raw data, including current and past events makes an information space. Integration and assessment creates a situation space. Decision rules then select action from an action space consisting of all behavioural options, which can effect the environment. A feedback loop is then connecting the action space and the information space. Several decision-making units can be combined via an information bridge. Compared to the model of this thesis, the information space most closely resembles the unordered situation aspects. Wellens also includes events among the aspects; in this thesis events consist of groups of aspects, and are not separate entities. The integration and assessment component, is in this thesis the updating processes. The situation space, is a collection of all situation models. The decision rules as well as the action space is part of the relevance process. In this thesis, the limitations of mental and 2 In the present thesis, the schema (Ashcraft, 1994, Sarter & Woods, 1991), is a situation model skeleton that has to be filled with relevant aspect values to become a situation model. 7 physical manipulation are mentioned as safe or unsafe combinations of mental, physical and device systems. The exact limitations and properties of mental, device and physical manipulation is beyond the scope of this thesis. The communication bridge is in this thesis specified as co-ordination of presentation processes and relevant attention. This corresponds to the Clark (1996) first level of symbol based communication, execution of behaviour by one agent co-ordinated with attention to that behaviour by another agent. Figure 1 (adapted from Ashcraft, 1994) shows the medium for human situation models. The three components, sensory memory, working memory and long term memory, and their interconnections have different properties. The sensory memory data are short lived. Sensory memory contains the rawest form of information that has currently been sampled from the environment. This information is available to working memory and to long term memory for identification, categorisation and further processing. Its content is therefore highly dependent on the processes of attention focusing. What is in sensory memory is not automatically attended to by working memory. The working memory, where most of the processing of the situation models is done has limited resources. If too many tasks are given, there will not be enough capacity available in working memory to perform them all and some tasks will have to be ignored. The long-term episodic memory contains situation schemes, memories of past events used to interpret the sensory data to build a situation model. It also contains a history of events. For example, it contains the events of the current flight, and how acted upon. It should be noted that what is stored in long term memory may be distorted and difficult to retrieve. 8 PHYSICAL MEDIUM FOR MENTAL MODELS WORKING MEMORY mental model & mental manipulation perceived physical model SENSORY MEMORY Currently attended environment LONG TERM MEMORY Model structure (schemes) History of past events Figure 1. Physical medium for mental models Summary SA models in a dynamic environment must constantly be updated. Three kinds of updating processes can be found in the literature; aspect sampling processes, structure deciding processes and projective processes. In the human cognitive system, attention focusing, directed by the situation schemes from long term memory, decides the contents of the sensory memories. What is in the sensory memories does not automatically become part of SA. 2.1.3. Examples of distributed SA For SA to be distributed, it may be co-ordinated using shared representations. Private representations may also be actively presented to some other agent. Ballard & Rippy (1994) describes a knowledge based decision aid, a mechanical tool, sampling and interpreting data, thus constituting a localised mechanical ASK system. This example illustrates that not only humans may contain situation models, and that ASK systems can potentially do many things a human SA system can do. 9 The research of Mogford (1997) on air traffic control (ATC), suggests how aspects of a situation model may be distributed between the environment, and the human working memory. It is suggested that it may be sufficient to maintain directions and altitudes of aircraft in the human working memory, and to sample further information from the environment on demand. In this way the limited human working-memory capacity is optimally used. Mogford suggest that three kinds of information are present in the ATC environment, relevant information that has to be maintained in working memory, relevant information that can be sampled when needed and then forgotten, and third irrelevant information that should be ignored. Artman and Garbis (1998) has documented presentation and sharing of representations in Rescue Agency Centres. Sharing of model aspects allowing reconstruction of the situation model by co-agents has been documented in London underground control rooms (Heath & Luff, 1992). Ground proximity warning systems present their internal model, a map to the pilots (Bateman, 1997) and Heads up display systems (HUDS) present aspects of the environment replacing information usually perceivable from the outside view during bad weather (Hartman & Moylan, 1994). Together these examples shows that the aspects and situation models of the SA system are distributed between different agents, and that the models and aspects are presented for other agents to perceive. 2.1.4. Compatibility with SA definitions of the literature Several definitions of SA, or as it is sometimes termed, situational awareness, can be found in the literature. These will be related to the model of this thesis, to show if they are consistent with it, if the model of this thesis can indeed be claimed to concern what has previously been named situation(al) awareness. The definitions of Endsley (1988, 1990) of Jones and Endsley (1996) and of Artman & Garbis (1998) has been discussed earlier in this chapter and will therefore not be discussed further. Metalis (1993) describes SA as the mental processes involved in the development, maintenance and use of the incomplete and constantly changing variables of a complex system. In this thesis, the mentioned mental process has been categorised, and their importance emphasised. 10 Fracker (1988) reviews several SA definitions from the literature. First, SA can be; “the ability to envision the current and future disposition of both red and blue aircraft and surface threats”. In the model of this thesis, this would be an example of situation aspects of a current and a future situation model, specified for relevance in some relevance process. Second, SA can be; “Where: the pilot’s knowledge of the spatial relationships among aircraft and other objects What: the pilot’s knowledge of the presence of threats and their objectives, and of his own aircraft system state variables When: The pilots knowledge of the evolution of events over time.” The ‘where’ and ‘what’ would in the model of this thesis be current situation models, unspecified for usage by any relevance process, and the ‘when’ projections of the models in the future. Since the what and where parts of the above SA definition can not be translated into two concepts, but to one concept, the situation model, this is a difference between the above definition and the definition suggested in this thesis. Third, SA can be; “the pilot’s knowledge about his surroundings in light of his mission goals” Mission goals would have to be translated to the reasons why the relevance processes are considered relevant. In this thesis, the surroundings must be specified as the surroundings at the present, future or at some past point in time. However, these definitions have only concerned individual SA, definitions of group SA has been made by Wellens (1993), Artman & Garbis (1998) and Endsley (1988). Wellens defines group SA as; “The sharing of a common perspective between two or more individuals regarding current environmental events, their meaning and projected future status.” Artman & Garbis defines group SA as; “The active construction of a model of a situation partly shared and partly distributed between two or more agents, from which one can anticipate important future states in the near future.” 11 Endsley defines group SA as the; “degree to which every crew member posses the SA required for his position.” The SA model of the present thesis suggests that the sharing of a common perspective is defined as the degree to which the participants are aware of the same situation aspects, and the degree to which use them as parts in equivalent situation models. The meaning of the aspects is thus defined as the situation models they are part of. Their projected future status is the future situation model they are related to. In a dynamic environment, every situation model has to be constantly reconstructed. For future situations, in this thesis every future situation model can be included as relevant SA, not only SA models for the near future. Finally, in this thesis the degree to which every crew member posses the SA required for his position, decides how much SA is missing, how much is redundant, and how much SA is misdistributed. 12 2.2. The proposed SA theory From the perspective of distributed cognition, SA becomes a system of agents, humans or artefacts, mediating, creating, transforming and presenting situation models, using the situation models, making them relevant to the destiny of the aircraft (Figure 1). The situation model is what is commonly referred to as the SA whereas the processes to maintain it are termed situation assessment processes. The processes related to the models, are the updating, presentation, monitoring and relevance processes. The updating processes place the unstructured aspects of the situation in structures to give them meaning, update the aspect values, and can project the model into the past or the future. Presentation processes can make the models available to other agents of the system and monitoring process evaluate whether there is something wrong with the models, creating meta-SA. Finally, the relevance process uses the model to execute some task. With an analysis of distribution, it becomes clear to what extent the processes are performed by the same or by different agents and which agents are co-operatively performing the same process. It can also be seen what or who mediates the models, which will influence whom, can access the models, and how easily these can be distorted. This gives insight into how reliable the system is and where measurements must be made clarify how well the system works. ASPECTS (unstructured) UPDATING • of structure • of aspects • of projection SITUATION MODEL (structured aspects) PRESENTATION MONITORING 13 USAGE IN RELEVANCE PROCESS Figure 2. Components of the SA system 2.2.1. Situation models, mediations and function All situation models are physically mediated, by for example a computer memory, a human working memory, or a verbal utterance. The physical form constrains who can perceive and alter the model, how easily it can be perceived, comprehended, and altered, and how the model can be disrupted. The physical form of the model and the location of the agents also allow or disallow monitoring of attention to the model, by some agent for some other agent. Physical situation models Physical models accessible to several agents allow distribution of labour, since different agents can update the model, and several agents can use the same model to achieve SA. In the cockpit, if one pilot could program the flight path to be visible on the screen of the other pilot, then the shared representation might save some work. The physical model can be modelling the current situation, for example the current position of the aircraft. Even the outside view of the cockpit can be seen as a situation model of the current environment. When it is blocked the relevant contents has to be modelled in some other way. It can also model some previous situation, and in this way work as a memory. In aviation, models of future situations are of importance. A model could, for example, combine a presentation of the current position of the aircraft with a possible future situation. The ideal is to model the relevant aspects of the situation, and their values at the relevant time, which may be any combination of the current state, and possible future and past states. Mental situation models For human cognition it may be interesting to distinguish between present models and mental3 models, since they rely on different mental resources. A mental model resides in the human working memory, and can be 3 Here, the term ‘mental’, in ‘mental model’ does not refer to the functional type of the model, but to the physical medium of the model. In the literature, the term ‘mental model’ has been used to denote situation awareness, as well as device models, and what is in this model denoted situation schemes. 14 transformed by mental manipulation, whereas a present model relies on the perception of models present in the environment, therefore the model may be physically manipulated, by the perceiver or by someone else. 2.2.2. Updating and sampling processes There are three kinds of updating, of structure, of aspect values and of relative time. Updating of structure changes the meaning of the situation. Updating of aspects provides the contents in the structure. The entire situation model, or parts of it, can be projected either forwards in time, for a prospective SA, or backwards in time, for a retrospective SA. These three things can be distributed between several agents, several agents may provide structures and contents, and projection may be done by the same or different agents. Sampling of the environment is the basis for updating SA. Situation elements can be sampled and comprehended into a situation model. Alternatively, a mediated situation model can be sampled, this model can be of future, present, past states, or a combination of the three. There can be a combination of physical manipulation of a physical model, and mental manipulation of a mental model. A mental model that is a perceived physical model is then co-represented by the world and the perceiver. In this way, if the mental model of the world model is not perceived, it is not lost, but can easily be reconstructed by attending to the model in the world. On the other hand, if the perceived model is manipulated mentally, for example projected into the future, this projected model will be lost if it is not maintained in the working memory of the agent, or externalised by manipulation of the physical model. Updating by ASK devices As an example of ASK updating, consider the ground proximity warning system (GPWS). It collects the aspects it needs, the current position of the aircraft and the current surrounding terrain. It also provides a structure for the aspects where it can project the future position of the aircraft in relation to the current terrain. Human updating The updating process in the human cognitive system relies on the environment and on the situation schemes of long-term memory. For sampling of the environment, it relies on attention attraction of the environment, such as light and sound, and by expectation created by the 15 situation scheme, for example the pilot expects to see information about the landing gear on the relevant indicator, and is assumed to know when to attend to the landing gear indicator. The reliance on situation schemes to interpret sensory data can be illustrated by the example of a crew warned of possible windshear at the takeoff from the runway (Tenney, Adams, Pew, Huggins, Rogers, 1992). This made them interpret the information using the windshear scheme, which resulted in a crash, because the symptoms were caused by the flap settings of the aircraft. In this case the symptoms could also be interpreted using the scheme of inappropriate flap settings. As Tenney et al. interprets this event, the crew used the wrong scheme because it had been primed by the windshear alert. The situation scheme thus has two functions, it directs attention, and it is used to make sense of what is perceived. In this case, the windshear situation model was not consistent with all possible information in the environment, since the flap settings were wrong. Apparently, the windshear scheme does not direct attention to the flap settings. In the accident described above, the cause of the inappropriate flap settings was an interruption during the pre flight checklist, which caused the checklist not to be completed. At the time of the interruption, this can be viewed as a problem with the current situation model. In this case the SA is incomplete. This in turn caused a prospective SA error, since the problems of the inappropriate flap settings were not foreseen. At the time of the accident, it is transformed to a retrospective SA problem, since the pilots’ model of how the current state came to be was flawed at the time of the accident. 2.2.3. The presentation process A presentation process can do two things. First it can present a situation model for some agent to perceive. Second, it can direct the attention of agents to the situation model. The task of presenting the model and the task of directing attention to it may be distributed between several agents. Presentation of a situation model may be done for optimal own cognition, since then the model does not have to be maintained in memory. It may be done for other agents, to reduce their workload, and it may be done to co-ordinate situation models of different agents, when it is of importance 16 that they agree on one model of the situation for co-ordinated action. Furthermore, it may be presented to allow some other agent monitor the situation model. The other agent can then evaluate the accuracy, completeness and relevance of the model, and also use it to understand the action of the other agent. Presentation by ASK devices In the case of the GPWS, a correct projection of the future aircraft relative the terrain does not automatically mean that a crash can be avoided. If the projection is not presented to the process that controls the navigation of the aircraft, the modelling is done for nothing. To make sure that it is attended to when needed, the GPWS evaluates the consequences of the future position, and, if needed, directs the attention of the pilot to the result of the evaluation. If the GPWS presents the map with the projection, the pilot can always decide to look at it to evaluate what can be a future problem, or if there already is an identified future problem, evaluate how to avoid it. Human presentation One disadvantage of mental models is that other agents cannot directly perceive them. They can, however, be implicitly or explicitly communicated. By acting, a pilot may reveal how the situation is perceived. What information a pilot attends to, can reveal what the pilot currently finds important. If a co-pilot perceives something important, and also perceives the pilot flying does not attend to it or acts upon it, he / she can assume that the pilot flying is unaware of it. Of course the pilots may also use explicit means of communication, such as discussing the events with each other. If a third party overhears such discussion, this third party may use it to infer what the pilots are aware of. Presentation is subject to distortion, and the perceiver is subject to possible misperception, misunderstanding, or distraction. Therefore, it is a good idea to test the presentation, such as a digital map, in a realistic situation, if the presentation must work. 2.2.4. The monitoring process In a multi-agent system with human and machine agents, the agents can monitor each other, to evaluate how well the system as a whole, or some specific agent, is functioning. For this to work, an agent can either 17 monitor the actions, or results of actions, from other agents. Alternatively, the internal processes of the agent can be monitored, if these are presented. Monitoring by ASK devices In the modern cockpit a good example of a non-human agent is the autopilot. It can for example be programmed to hold a specified altitude. This task may be monitored by checking the current altitude, to see if it deviates from the selected altitude. In this case, what is monitored is the effects of the actions of the agent. Monitoring by humans An agent may monitor and evaluate its own SA, to decide how complete and accurate it is. This meta-SA can then have four outcomes. Either the awareness is both complete and accurate, and the pilots believes that it is, or it can be complete and accurate when the pilot believes that it is not. Furthermore, the awareness can be incomplete or inaccurate when the pilot also believes that it is incomplete and inaccurate. The least desirable outcome is that the SA is incomplete or inaccurate, when the pilot believes it to be complete (VINTHEC WP 1). The pilots may monitor the load on the own cognitive system, this subjective assessment of workload can be used to decide when to delegate tasks or when to ignore some tasks in favour of the most important ones. 2.2.5. The relevance process The relevance process are not part of SA, but makes use of the situation model. For example, a situation model containing the future and current position of the own aircraft, an enemy aircraft, and the ground, may be used in two processes. One process may need the own aircraft position and the enemy aircraft position to lock some weapon at the enemy. Another process may need the position of the own aircraft and the position of the ground, to avoid crashing into the ground. This way the military awareness as well as the navigation awareness is based upon the same model. Thus, different processes make use of aspects of the same model. Sometimes, as Endsley (1988) observes, one of the processes, but not the other is, performed correctly. If hunting the enemy is done at the cost of maintaining awareness of the ground, a crash may result. This may according to this model be either because the ground position of the model is not maintained, or because the ground position relative the 18 aircraft is not correctly projected into the future, or simply because the consequences of the model, a crash, is not evaluated. A situation model is only relevant if the information is needed to perform or to evaluate the need for some current or future action. Action can, however, be done without complete SA. For example, one crew taxied to the wrong runway, by habit, instead of the assigned runway (Jones & Endsley, 1996). Awareness relevant for future actions is for example to know the weather conditions of the destination airport, before arrival. If there is a risk of weather closing, this awareness can be used to plan ahead, to decide in advance what to do if there should be weather closing. An experiment where some crews obtained such awareness, whereas some pilots failed to obtain it, showed that such awareness reduces the mental workload of the pilots during the weather closing. (Tenney et. al. 1992) 2.2.6. Properties of distributed and localised systems The properties of distributed systems The processes and models have to be found and located to some agent or physical representation. When this is done, it is possible to see what processes each agent is responsible for. Then it can be examined whether the processes are done with conditional or continuos updating. With continuos processing, a process is performed continuously, or with regular intervals, unconditionally, it is not possible for the agent to leave the process undone to execute some other process instead. Agents with continuous processing, such as the GPWS, can dramatically increase aircraft security, as seen in the reduction of accidents with the introduction of GPWS systems (Bateman, 1997). Here, ASK, although weaker than human SA, is sufficient to increase the SA of the entire SA system. With conditional processing, the process can be left undone. Then care has to be taken to ensure that vital processes are still maintained even during high workload. Human agents, in particular, have to switch between different processes. This may mean that some process is performed at the cost of another process not being performed. It may happen that some process is not performed, even though the resources are present. Such an action, or lack of action, depends on traditions and 19 training. Since monitoring processes are only vital when something goes wrong, these are especially vulnerable to the problems of processes being left undone during high workload. The properties of localised systems The main properties of the internal human cognitive system are first that it relies on situation schemes to direct what aspects of the environment to attend to, and how to interpret these aspects. The second main property is the limited capacity of the working memory to maintain the mental focus on the situation models, or on the focus on the task. If the human working memory is overloaded some situation model has to be neglected, or the execution of some task has to be abandoned. These sometimes conflicting tasks of maintaining awareness, may therefore result in a maintenance of SA on cost of maintenance of relevant tasks. For example, if a model of a future situation which is not immediately relevant (but will be relevant in a later stage of the flight) is maintained in a high workload situation, and the execution of a relevant task therefore has to be abandoned, high SA is then achieved at the expense of performance, instead of providing the basis for good performance (Wellens, 1993). If compared with the situation of distributed cognition, all the aspects of SA cognition may be localised to one individual. The individual then represents the model internally, updates and manipulates it internally, and can also monitor its quality and present the model when needed. Localised cognition is not only a possible property of a human agent. A mechanical system may also work independently of other agents. For example it is possible to imagine an ASK system such as the GPWS, which instead of directing the attention of the pilot to a possible future danger, is designed to assume control of the aircraft and to perform the evasive manoeuvres without communicating this to the pilots. 2.2.7. SA Error In a system of distributed SA, when some agent has the relevant SA, but not the agent that should have it, then the SA is misdistributed, as compared to missing SA. There can also be redundant SA, that no agent needs, or that is shared between an agent that needs it and one that doesn’t need it. When SA is 20 shared between two or more agents that need it, then the SA is not redundant. Redundant SA is necessary as a precaution against device failure, or in the case of a human agent for some reason not being able to act as desired. In these cases, the redundant SA becomes vital for the remaining agents to be able to take over the processes performed by the missing agents. The SA for the individual may be incomplete, when some aspect of the relevant situations, or some relevant situation, is missing. It may be inaccurate when some aspect is interpreted using the wrong situation scheme, or when some aspect is inaccurately perceived. 21 2.3. Measurement techniques for SA systems For the operationalisation of SA, several measurement tools are available. In this chapter eye-point of gaze equipment, voice transcripts, physiological measures, event-related potentials, and questionnaires as well as performance measures will be discussed. 2.3.1. Eye point of Gaze Eye point of gaze (EPOG) equipment (VINTHEC WP 4) measures what a human is looking at. It measures a time history of fixations, where the environment is divided into areas of interest. The equipment records the time, duration and position of fixations on these areas. For each fixation, pupil diameter and blink rate may be measured as well, by some EPOG measurement equipment. Translated into the model of this thesis, the fixated areas correspond to the contents of the visual sensory memory of the human participants. The time history corresponds to the changing contents of the visual sensory memory. The LTM scripts together with other attention guiding devices guide head- and eye movement. The time history may be transformed into dwell time, or fixation number, percentages for some period of time on the defined areas. It may also be transformed into a transition matrix, where the numbers of transitions from area X to area Y, from area X to area Z, e t c, are described. Given one GazeTracker and one agent, the equipment can still be seen not only as a measure of internal cognition, but of usage of external information sources. In a real situation some source of information supposedly contains the relevant information. The EPOG measure can be used to evaluate to what extent the areas with relevant information are attended to. This is particularly important when information is only visible a short time, or has to be reacted to as soon as it is presented. Assuming that a pilot knows what to do when the information is comprehended, in the case of a pilot not acting, it can be inferred if the information was attended to at all. If so, it was not comprehensible. If it was not attended to, then it was not salient enough. Several EPOG equipped pilots in the same environment give rise to parallel time histories. These can be compared, to find differences in the time distribution of the attended areas, and to examine whether simultaneous dwells on the same area occurs. 22 The pupil diameter and blink rates are usually used as measures of workload. The workload measures of the EPOG equipment are mostly a measure of internal cognition, but even this is important from a holistic perspective, since a too high workload means that the work distribution has to be altered. 2.3.2. Questionnaires Questionnaires can be used to measure meta-SA, subjective workload and subjective performance, as well as more objective data such as working memory contents. The meta-SA questionnaires measure the result of some monitoring process. (VINTECH WP 3) Consequently, one can measure the meta- SA of ones own SA, assess the SA of some other agent(s) or the SA of the system as a whole. However, it has been argued that an observer rating of SA quality and completeness must be incomplete, since the observer can only observe behaviour and does not have access to what other agents are thinking (Berggren, 1998). There are two approaches to questionnaire measures of SA. The subjective SA method asks the human agent for a SA score on a scale. The other method queries the agent for specific aspects of the environment, when the relevant aspects in the environment are removed. An example of such a method is the SAGAT technique (Endsley, 1988). These two methods measure very different things. While the SA score method attempts to measure the result of a monitoring process, subjective meta-SA, the other method measures the components of sensory memory, working memory or long term memory. The subjective workload measures the experienced workload of the agent. This can be compared with physiological measures such as the mentioned blink rates, and measures such as heart rate variability. 2.3.3. Physiological measures Physiological measures such as heart rate, heart rate variability, perspiration, and blink rates can be used as indicators of mental workload. Unfortunately, they can be indicators of physical work as well, which means that physiological measures has to be interpreted with care (VINTHEC WP 1,3). 23 2.3.4. Event-related potentials EEG Event-related potentials (ERP) can be used to shed light on the processes of understanding of stimuli (Derks & Gillkins, 1993). In short, using electrodes attached to the human skull, the magnetic fields of the brain are measured. The fields are then analysed after the presentation of stimuli to examine the changing electrical field in different parts of the brain as recognition of stimuli proceeds. The comprehension processes are assumed to occur in parallel, with a negative peak at 100 ms for “orientation”, and at 200 ms for “labelling”. These peaks are different for each modality. The positive peak at 300 ms is associated with categorisation, where unexpected but relevant stimuli give a high peak. There is a highly interesting 400 ms-depolarisation peak associated with categorisation failure, it is reported to be specifically sensitive to violation of semantic expectancies. At 600 ms, there may appear a positive polarisation that corresponds to incongruity resolution, however this is not well documented. In the model of this thesis, what is used as the known variable is the content of sensory memory. This has to be known to be able to discuss what is being labelled and categorised. A categorisation failure would indicate that the pilot is unable to find a situation model in which the sampled sensory data would make sense, or fails to decide which conflicting model the data would fit. A high categorisation “surprise” peak would indicate that the pilot finds the event surprising but comprehensible. The 600 ms resolution peak, could, if this peak is correctly understood, indicate that the subject has managed to make sense of the stimulus after all. Thus, an ERP analysis will not, even after the appearance of a resolution peak, tell us which model the pilot is using. Using ERP alone does not tell us whether the pilot has an accurate situation model, or if the model is inaccurate, but rather if the pilot has at all got a model, whether the pilot has been considering alternative models and how surprising the situation is. 2.3.5. Transcripts of verbal messages Transcripts of communications between aircraft and air traffic control can illuminate the number of errors occurring due to misperception or 24 miscomprehension of voice based messages (Jones & Endsley, 1996). It has furthermore been used to analyse the interactions between coordinators of the London underground (Heath & Luff, 1992) and of the interactions in an emergency co-ordination centre (Artman & Garbis, 1998). 2.3.6. Performance measures High performance is not always accompanied by high SA. System performance with for example an active autopilot may be excellent, whereas pilot SA may be quite low (Tenney et. al, 1992; Shively & Goodman, 1994). As discussed, when performance of cognitive functions and performance of relevance processes are distributed from human to machine, performance may improve, whereas pilot SA may decrease (Shively & Goodman, 1994). Care must therefore be taken to ensure that, for the measured situation, SA is required for performance. 25 2.4. Theory summary and discussion In this chapter I will summarise the SA model, and discuss the consequences for designing SA critical systems. 2.4.1. Concluding operationalisation model for SA Scheme directed sampling & externally directed sampling AGENT A MENTAL MODEL & SCRIPTS Mental manipulation AGENT B MENTAL MODEL & SCRIPTS Mental manipulation Physical manipulation PHYSICAL MODEL AVAILIABLE TO A AND B Physical manipulation Scheme directed sampling & externally directed sampling Figure 3. Combined SA system To model an SA system, first, the agents are added, then the physical representations of interest. In this distributed system there can be different physical models, modelling the same information, such as in cockpits with double instrumentation, one for each pilot. Sampling is, as illustrated in Figure 3, either guided by mental situation schemes or by external attention allocation mechanisms. As we see, there can also be a combination of physical manipulation of a physical model, which may be shared, and mental manipulation of a mental model. The mental model can be a perceived physical model, or a comprehended set of situation aspects. When we want to predict the SA of a modelled situation, first we can examine if the relevant dynamic aspects are present in the environment. If not, then we can stop here, because then there can be no SA. If they are present then we want to know if they will be attended to at the right time, 26 and if so, whether they will be comprehended. For this we may design an experimental setting and measure the SA system. If SA is required for performance, and the pilots perform well, then we know the dynamic aspects have been comprehended. Comprehension of situations and situation elements can also be measured with the SAGAT method. The process of comprehension can be investigated with ERP methods. To measure comprehension in verbal interaction, the review suggests transcripts of verbal messages. If comprehension is not achieved or if we want to know where in the environment the dynamic aspects were sampled, we can measure attention. To measure visual attention, the review suggests the usage of Eye Point of Gaze tools. To measure Meta-SA the review suggests the usage of a subjective SA quality rating scale. To measure the pilot workload affecting SA, the review suggests physiological measures or a subjective workload rating scale. 2.4.2. Relevant SA Having defined a model for SA, relevant SA can be defined as those models and aspects present in a team that are relevant for some relevance process. For the SA to be correctly distributed, it has to be present for the agent executing the relevance process. For the SA to be accurate, the aspects must be interpreted using accurate situation schemes. For the SA to be consistent there can be no situation models that gives the same aspect conflicting meanings or values. For the SA to be sufficiently redundant, if some agent should be unable to function, the remaining agent(s) should have sufficient SA to maintain the relevant tasks. To say that this abstract description leaves the concept of SA too general to be useful is merely a play of words, since the model framework of this thesis shows that SA is a label for any system containing situation models. This means that anyone interested in some specific SA system, be it a distributed or localised system, can compare the system of interest to already examined systems to seek similarities that may make measurement redundant, or that may show how measurement can be performed. It allows sharing of measurement tools, and of modelling 27 frameworks, as well as sharing of observations about the properties of some physical situation representation. 2.4.3. Safe distributions of labour between actors and artefacts Device usage in the cockpit transforms the task of remembering cockpit speeds (Hutchins, 1995) from one which heavily relies on the unreliable human working memory to one that relies on simple manipulation of simple devices, and a far more reliable visual comparison task. Basically, a list of landing speeds given an aircraft weight and flap settings, serve as a long term memory of landing speeds. Given the current weight of the aircraft read from a display, a sheet with relevant numbers can be selected and put in front of the pilots. The sheet is functioning as a working memory, and a means for both pilots to monitor the task, whenever they want they can compare the weight specification on the card, with that of the weight indicator of the aircraft. Next, the pilots set the speed bugs on their airspeed indicators, which show the airspeed where the flap settings have to be changed. As the settings are cross checked by both pilots, the reliance on human working memory is removed, and the task is successfully transformed into a simple visual task of observing the airspeed needle and compare it to the position of the speed bugs. In this way, the cognition is distributed over several artefacts and a reliable part of the human cognitive system. The point of this example was to show a maintenance of a physical future situation device, the speed bugs. They do not predict the future situation, they are used to compare a physical model of the current speed, with the desired future speeds at which the flap settings are to be changed. In this way, the model is maintained by physical manipulation, sampled and used for mental evaluation. The external usage of speed bugs at the same time allows the pilots to monitor each others representations of the future situation, and the appropriate monitoring process was observed in the above example. 2.4.4. Safe practices, tradition and training That a set of components and practices exists that can constitute a safe SA system, does not mean that the practices are followed or that the human agents use the components properly. 28 Sollenbergher & Stein (1995) discuss the problem with unused memory aids in air traffic control environments. The concern is that usages of memory aids were resisted due to controller culture. Weick & Roberts (1993) discusses the phenomenon as it is manifested on aircraft carriers. They name it heedful interrelating, and includes in it not only acting with high SA, but also the individual usage of practices leading to a high SA. On air carrier decks heedful interrelating is preserved as experienced personnel introduce newcomers to the tradition of heedful interrelating through stories of the consequences of heedless interrelating. How training can theoretically increase SA trough training the pilots how to recognise signs of low SA, is described by several authors (Schwartz, 1997; Caretta & Zelinsku, 1996; Regal, Rogers & Boucek, 1988). Related to the SA system model of this thesis, such training would, if successful, increase the accuracy of the human monitoring processes. This would lead to a more accurate meta-SA and consequently to safer flight. 2.4.5. Optimal SA distribution and component failure Distributing actions between man and machine may have very subtle, yet potentially very grave consequences, for the pilot’s ability to handle automation failures. Shively and Goodman (1994) used an arcade game like simulation of a star trade cruiser to examine these subtle effects. In the simulation, the calculation needed for a successful landing was to evaluate the distance from the sun to the own position. The pilots either had to calculate this themselves, or the correct position was calculated by the game computer and presented as an X on the navigation display. The consequence of having a computer presenting the correct position, was an increase in performance, but a decreased awareness of how the correct position should be calculated. In the framework this corresponds to decreased device awareness for the process. It also corresponds to a potentially decreased awareness of the correct distance to the sun, a critical aspect of the situation. Shively and Goodman theorise that automatic systems producing this effect may leave the pilot without the necessary device knowledge of the process, to perform it during a device failure. In the model of this thesis, this corresponds to an optimal distribution of cognitive processing for the fully functional system, since the performance with this distribution is improved compared to a system without it. Then it corresponds to a system where one agent is suddenly removed. The result may then be that 29 the pilot is unable to perform the cognitive process of the device, or that performance is poor. This means that optimal distribution of cognitive processing may lead to poor device models of the human agents. How to perform some action, such as a calculation is not part of SA, but if the result, awareness of the correct distance to the sun is considered part of SA, then this decreased device knowledge results in decreased SA. This result is to be contrasted to display enhancements where cognitive processing is not delegated, but where the action space is made more salient, such as making the storage gauge more comprehensible. With this kind of enhancement SA was enhanced as well as performance (Shively & Goodman, 1994). The SA system may consist of combination of agents using conditional and continuos processes. Then the properties of the combined system will depend on the reliability of this combination. The mechanical agents most often responsible for the continuos updating may of course fail, but they may none the less be less likely to fail than if the task is delegated to a human agent, especially if the human agent simultaneously has to attend to other tasks. If action is not delegated to the mechanical agent with continuos processing, then the agent should direct the attention of the pilot to the relevant process, when needed. The advantage of mechanical agents is therefore that they can often reliably maintain continuos processes, whereas the advantage of a human processor is that they often can handle unexpected processes, or processes that are yet too complex to be automated. 2.4.6. Consequences for design The measured and modelled system will show to what extent the system is optimised. Design efforts can then be directed at the weakest spots of the system. Furthermore it may give insight to the distributed and localised effects of the introduction of new technology. The design of situation models for mechanical agents to achieve ASK, and the design of situation models to present for the human agents, reflects the understanding of the SA needs of the designer. This is true especially for non-human agents since these normally do not have the option to decide what aspects to sample, or how to combine the aspects to a situation model. They normally do not decide how to evaluate the resulting situation model either. Thus, the performance of the cognitive 30 processes is delegated to a mechanical device, whereas the design of the process resides in the hands of the designer. Thus, all situation models delegated to a non-human agent contain only what is put there by the designer. Even for self-learning systems, the designer controls what can be learned. A weaker dependence on the designer is present for the human actor, which has to rely on what is presented to him or her by aircraft equipment. The dependence is weaker only since the human agent can decide to sample what a designer did not intend to be sampled, and put it into a situation model the designer does not know of, and evaluate the resulting model in a way unknown by the designer. Still, however, the availability and comprehensibility of situation aspects and models is largely in the hands of the designer. The applied model together with appropriate measurement may show the reliance of the system on internal attention management, what aspects and models are maintained internally, and whether they are projected internally in time. Different design solutions may have to be used to make world aspects more comprehensible, attended to at the appropriate time, to reduce the need for human internal projection of aspects and models, and to reduce the effort needed to either perceive a model, to perceive the relevant aspects of that model, or to construct a model from presented aspects. The theory does not deal with all design problems, of course, but only those associated with SA. A major problem for cockpit design is the limited space. There can’t be too many big displays, and decisions may be time critical. That means, if some critical information proves to be incomprehensible, and there is no space to present a comprehensible picture, or if some information will be unattended with no means of attention attraction, this model suggest three options. First, routines may be changed, training may direct attention and increase the comprehensibility of messages. Second, the monitoring task may be distributed, either to a new human agent, or to a mechanical agent. Third, if neither of these options are possible, then if the experiment was valid, as it should, then the problem will remain, and if it is considered too dangerous to fly with this problem, then there is always the option to stay on the ground. But the limited space and the time pressure is exactly what makes an analysis of the SA system vital, since it may suggest what the limited space should be used to present, and how to manage the distribution of labour in the SA system. 31 This model, thus, is to be seen as complementary to models for the design of comprehensible and space economical representations, and to models addressing other design issues for the cockpit. This model is complementary, since it is of no importance that a representation is comprehensible, if it is either irrelevant or unattended. It is furthermore of no help to have an attended but incomprehensible representation of a situation. 32 33 3. THE EXPERIMENT The experiment was made in the framework of the VINTHEC (Visual interaction and human effectiveness in the cockpit) project. The experiment was designed to evaluate the accuracy and usefulness of eye point of gaze (EPOG) equipment, and for measuring SA, and workload during the experimental conditions. Due to the fact that there were two pilots in each crew, the experimental data could be used for the analysis of this thesis as well. Relevant measures from the experiment were selected for inclusion in the data analysis of this thesis. 3.5. Method 3.5.1. Test site The experiment was performed at the NLR (National Aerospace Laboratories) in Amsterdam, the Netherlands. 3.5.2. Participants The participants were 20 active pilots normally flying civil passenger jets, most of them familiar with Boeing 737 procedures. Their flight experience ranged from 2000 to 20000 flight hours. In the preexperiment questionnaire, no pilot reported any activities within twentyfour hours, judged as likely to interfere with experiment behaviour. The questionnaires are described in appendix B. 3.5.3. Procedure The pilots were divided into ten crews of pilot and co-pilot. Each crew flew two simulated flights, the first from Amsterdam to London, the second from London to Amsterdam. The entire flights, except landing and takeoff were simulated to be above thick clouds, to avoid relevant information to appear on the outside view. Each flight crew was briefed and made familiar with the simulator during the morning. In the afternoon the two experimental flights were made. Before the experiment the EPOG equipment was calibrated, to ensure that dwells on areas of interest were also recorded as such. Without 34 calibration, the EPOG equipment does not work up to standard. The calibration was checked twice during each flight, one during the cruise and once on the ground after the taxi phase of each flight, with a possibility of recalibration between the two flights. The calibration sequences were removed from the experimental data and analysed separately. Before each flight, the briefing questionnaire was handed out. During each flight the four subjective questionnaires were handed out. After each flight a debriefing questionnaire was handed out. Before each flight, the in flight observer gave the pilots the flight log. The log described nothing of consequence for the functionality of the aircraft for the first, Amsterdam-London, flight. For the second flight, the log described an autopilot failure where the autopilot failed to capture the selected altitude. In addition to the physiological measures made by the EPOG equipment, heart rate was recorded, and heart rate variability was derived from the heart rate. Each flight may be divided into six flight segments, the takeoff, climb, cruise, descent, approach and landing. The flight was segmented because the workload, the actions and goals of the pilots and their looking behaviour are assumed to differ between the segments, but stay the same within the segments, if nothing unusual happens. The takeoff segment was excluded from the present study, leaving five segments for the analysis. There were four abnormal events in the experiment, two occurring in each flight. They will here be presented in the order they occurred, the same order for each crew. The map shift event occurred during the cruise phase in the Amsterdam London flight. It was manifested by a short period of strong wind. It took place in a situation with almost no wind, and returned to almost no wind after a short time. The strength of the wind was displayed on the map display, and was a simulated computer error, which made the computer project the aircraft on the map at the location it would have been if there had been a strong wind. This means that the projection error remained after the disappearance of the wind. Such a strong wind is extremely 35 unlikely, if not impossible, if it occurs and disappears as was indicated on the map display. A gear unsafe event occurred during the landing in the Amsterdam London flight. This was shown as a missing green light, of the three lights for the landing gear. The missing green light indicates that the landing gear has not been safely extended. An altitude bust occurred at the end of the climb in the second flight. This was constructed as a failure of the autopilot to catch the selected altitude. This means that the aircraft continued the climb beyond the selected altitude. It could be seen as a continued rising altitude on the altitude indicators. The fourth and final event occurred during the London-Amsterdam landing. The pilots were given a flap asymmetry warning, consisting of a warning light, a sound, and a flap asymmetry warning text on the engine display. A flap asymmetry causes a situation where the flap settings can not be changed during the landing without risk of an increased asymmetry. The function of the flaps is to change the weight/speed ratio with which the aircraft can fly. The events were grouped as follows Low workload High workload Low SA Map Shift Flap Asymmetry High SA Altitude Bust Gear Unsafe The map shift and the altitude bust were designed to be low workload events, whereas the gear unsafe and flap asymmetry occurred during the landing, and were therefore considered high workload events. Regarding SA, the pilots were warned in the techlog, that the autopilot had failed to capture the desired altitude during a previous flight, and the gear unsafe was practised during the practice flights, making these two events high SA events because of the preparedness. The map shift and flap asymmetry were neither practised nor prepared for, making them low SA events. For this thesis the altitude bust event and the map shift events were selected for analysis. The gear unsafe event was ill suited for the analysis of the present thesis since the gear unsafe warning lights were not included as an area of interest for the gazeTracker recordings. The 36 flap asymmetry event was not analysed statistically but will be mentioned briefly in the discussion section. 3.5.4. Design A 2*2*2*5*20 mixed design was used. The first variable refers to Role, as pilot flying or as co-pilot, the second refers to Workload (high/low), the third refers to SA (high/low), the fourth refers to the five Flight phases (climb, cruise, descent, approach and landing) and the fifth to the number of Subjects (20). 3.5.5. Equipment The Fokker 100 simulator was based on a full-scale multipurpose flight simulator with motion. The simulator had seats for two pilots and one in flight observer. Communications using headsets was recorded, and a small stationary video camera was used to record the events in the cockpit. On the same video tape also the primary flight display (PFD) , the navigation display (NAV) and engine display was recorded, together with all audio that went via the cockpit headphones. For the EPOG data collection, the cockpit was divided into ten areas of interest. The primary flight display (PFD)(*2), the navigation display (NAV)(*2), the outside view (OUT)(*2), the engine display (ENG), the CDU (*2) and the autopilot (A/P). Two headsets were used for the EPOG data collection, one for each pilot. The EPOG equipment used at the NLR consists of a head mounted tracking device. Eye Point-Of-Gaze (EPOG) is the point on a predefined surface were a line coming straight from the centre of the eye crosses that surface via the lens of the eye. As such this is the central point in the pilot’s field of vision. This point was measured by means of the EPOGrecorder called GazeTracker (VINTHEC WP4). The co-ordinates on the surface, together with the duration the pilot looked at a particular area of interest, is called a fixation and has been recorded in a computer file. Besides the scanning pattern, the pupil diameter and eye blink rate were recorded. The marker signals for the events and flight phases of the experiment were attached to the timeline of the co-pilot. For this experiment, the synchronisation accuracy was +/- one second due to the time accuracy of the ethernet connection. It should be noted that a much higher synchronisation accuracy can be obtained with other now available 37 methods. This accuracy is only of importance for the simultaneous dwell analysis. The scanning behaviour was considered to be an indicator of the pilot’s mental preoccupation. Pupil diameter was used as an indicator of mental workload while eye blink rate indicated the visual workload. The accuracy used to discriminate between fixations was 1,5 degree visual angle from the eye and a minimum fixation duration of 150 ms. This means, for the analysis it was assumed, only if the subject looked at a point more than 1,5 degrees from the previous fixation and for longer than 150 ms. a new, cognitive meaningful, fixation was made. For the determination of the eye blink rate the blink had to be longer then 40 ms. Shorter blinks are impossible for human beings to make and as such are considered to be artefacts. Blinks are all further situations during which the pupil diameter was zero (closed eyelids). It has to be noted that pupil diameter is dependent on workload as well as other variables amongst which the amount of light falling on the eye. In order to compensate for this illumination effect care has been taken to use, for the analyses, only the pupil diameter measured when the pilot looked at one display only for each analysis. This way, knowing that the illumination within the cockpit didn’t change, the pupil diameter can be used as an indicator of mental workload. Questionnaires for SA, workload and performance were used as subjective measurement techniques before, after and during flights. Questionnaires for sleep quality, and other individual variables were used as well as briefing and debriefing questionnaires before and after each flight respectively. All questionnaires were handed out on sheets of paper. 3.5.6. Research questions The first research question which can be partly answered by an experiment, is whether the operationalisation model for distributed situation awareness presented in this thesis, is sufficiently clear, complete, and correct to explain found results. That is, to explain what components of the SA system are responsible for the achieved SA. If the achieved SA is too low, it should shed light on the reasons for this, as well as the reasons for a sufficient SA. In the case of no results regarding 38 SA, it should explain whether this is due to quality problems with the equipment or the number of subjects, the scenario, or the choice of measurement techniques. This question will be answered as an evaluation of whether the following research questions can be answered. The second main question the experiment should answer is for this thesis whether relevant attention is a factor for achieving relevant SA. If the second question, as expected, is confirmed, an additional question is posed. The question, then, is whether this particular attention management system is sufficient for achieving or for not achieving the relevant SA. In particular, if the system is sufficient to achieve relevant SA in the altitude bust and map shift events. The third main question is whether there exists divided attention in the cockpit. If the pilots look at different things, this can mean that they update different situation models, for themselves, or to be shared with each other. If they look at the same things but at different times, in a dynamic environment this means that the team as a whole has a more complete coverage of what has been present in the environment. If the perceiver can also comprehend what has been attended to can, as noted before, not be evaluated using attention measures alone. A side question, which will be answered, is whether the measurement techniques were sufficient for the desired measurements, or if some question demands some alterations to the equipment or entirely different techniques. A part of this side question is whether the scenario was appropriate for the questions in this thesis. When this side question is answered, the reader should keep in mind that the experiment was not made exclusively for this thesis, but as part of a larger research project. 3.5.7. Operationalisation of the research questions To operationalise the questions, they were reformulated as questions related to measurement data and interpreted using the SA system model presented in this thesis, applied to the experimental situations. The first step is to relate the measurements to the applied SA system model. Below is a model (Figure 4) of the two human agents, the pilot flying and the pilot non-flying and of the physical areas of interest, a selection of the physical representation surfaces, in the cockpit. The cockpit has duplicated Primary Flight Displays (PFD), two Navigation 39 Displays (NAV), two front windows (OUT), two CDUs, a shared Autopilot (A/P) and one Engine display. The content of sensory memory were estimated by the EPOG measures. The arrows represent the EOPG measures of visual attention. The attention of one pilot on the areas of interest of the other pilot could not be measured, as indicated in the illustration. This could not be done due to low precision of the EPOG equipment with such an angle. As indicated in the illustration the EPOG measures tell us what raw display information is present to the visual sensory memory. It should be noted that mere presence to visual sensory memory does not mean that the information is attended to by working memory, or that the information is comprehended. The working memory situation models were not measured directly. Instead the subjective quality of the situation models, the meta-SA, as estimated by the pilots, was operationalised using the subjective SA questionnaire. The load on working memory was recorded using a subjective workload measures questionnaire. In the case of the altitude bust abnormality, by observed attention to the aircraft techlog, the altitude bust scheme is assumed to be present, allowing an altitude bust situation model to be formed. The maps shift scheme had, instead, to be evoked only with the help of the cues of a map shift visible on the map shift display. The quality of the long term memory schemes were estimated as the hours of flight experience of each pilot. The aircraft log was, as noted above, handed out on each flight, and attended to by the pilots. The log represents the physical medium for the retrospective SA of the aircrew. The team performance was measured, for the altitude bust, as the time elapsed from the onset of the altitude bust, to the time of altitude capture. For the map shift event, time was measured from the onset of the event to the moment the map was reset, either the crew discovered it or it was revealed to not interfere with the next event. For the individual pilots, pilot performance was estimated by the pilots using a subjective performance questionnaire. Performance is not a part of the situation awareness system, but a control variable, which under specific circumstances can be used as an estimation of the quality of the situation model, or the quality of the SA system. 40 Sampled sensory data of the team can be measured as the total percentage of time an area was attended to by at least one pilot. For each flight segment or event, the time the pilots spend simultaneously on each display can be calculated. Together with the times they each spend fixating on each display, the total time they monitored each area together can be calculated. AIRCRAFT LOG Observed usage PILOT FLYING SAMPLED SENSORY DATA PHYSICAL AREAS OF INTEREST PFD NAV WORKING MEMORY Comprehension: sSA sWL sP Primed scheme Performance (event resolution time) LONG TERM MEMORY Flight experience OUT CDU A/P Engine SAMPLED SENSORY DATA PFD NAV OUT CDU PILOT NON FLYING WORKING MEMORY Comprehension: sSA sWL sP Primed scheme Performance (Event resolution time) LONG TERM MEMORY Flight experience Figure 4. Applied SA system model to the experiment. sSA stands for subjective SA, sWL stands for subjective workload, and sP stands for subjective performance. The first research question was operationalised as a discussion over the results of the other research questions. 41 The second research question will be operationalised first as an analysis of the possible dwell time percentage differences on the various areas of interest between the flight phases. If there are no differences then attention can not be a good measure for situation models, because clearly the landing demands a different situation model than the cruise phase, since the situations are very different. If no differences are found, then the differences will still be assumed to exist. The conclusion will instead be that the differences can not be caught using this experiment. Second, to get a more clear demonstration of the importance or lack of importance of relevant attention, the altitude bust and map shift events will be examined. For these events, are there any differences in dwell time on the relevant display (PFD) between pilots flying and co-pilots? Is there a mean difference between the groups, when divided into groups of fast and slow action: subjective workload, dwell time, subjective SA, subjective performance and flight hours? If there are any results regarding the altitude bust and the map shift, these results, or lack of results, will be discussed to answer the additional question regarding the attention management system. If there are any differences in the attention between the flight phases, the third question will be operationalised as two subquestions. Are there differences on the time spent on the same display, for the pilots of the same crew, when analysed for simultaneous dwells, that can not be accounted for by assuming that the co-ordination is completely random? Are there differences in the percentage of time spent on the displays when the crews are divided into pilots flying and co-pilots? The side question, regarding the validity of the experiment will be addressed in the general discussion part of this thesis. 3.6. Measurement results The evaluation of the EPOG calibration data resulted in the exclusion of two crews from the analysis. Furthermore, the analysis of the remaining pilots showed the desired high between area point of gaze accuracy. The within area precision varied between pilots, therefore precautions must be taken when selecting pilots if a high within area precision is desired. No within area analyses are performed in the current thesis. Between the main flight segments there were significant mean dwell time differences, for both the pilots flying and the co-pilots. For example, in 42 the Amsterdam-London flight for the co-pilots there was a significant dwell time difference on the outside view for the climb-cruise-descent phase compared to the approach-landing phase (Dunnet C: df = 9, F = 123.456, p < 0.05). For the pilot flying, there is an additional dwell time difference between the approach and the landing (Dunnet C: df = 9, F = 123.456, p < 0.05). A list of significant differences can be found in appendix A. Mean attention on OUT (AL) 80 60 40 Value 20 MEAN_PNF MEAN_PF 0 24.00 25.00 26.00 27.00 28.00 Segment Figure 5. Division of attention on the outside view. (climb - 24) ( cruise - 25) (descent - 26) (approach - 27) (landing - 28) (PF - Pilot flying) (PNF - Co-pilot) There was a significant mean dwell time difference between the crew members for the landing flight segment on the Amsterdam-London flight (Dunnett C: df = 9, F = 123.456, p < 0.05) (see Figure 5). (A nonsignificant trend to the same effect was visible for the LondonAmsterdam flight.) This difference was expected and reflects the assumed division of attention between the pilot flying and the pilot nonflying. In Figure 5, the attention during the climb (24), the cruise (25), the descent (26) the approach (27) and the landing (28) can be seen on the Y axis, as the percentage of time the eyes of each pilot were focused on the outside view. There were no significant mean differences between the measured simultaneous dwell time and the expected simultaneous dwell time based on random co-ordination of attention, for the main flight segments. Since the simultaneous dwells measures method was not valid on data with the 43 time line synchronisation accuracy obtained in this experiment, these results should be viewed as a demonstration of the simultaneous dwell time analysis procedure only. For the altitude bust event -> +2 min, the groups of pilots were divided using the performance measure, time to correct the altitude bust. There was a significant difference between the groups regarding the percentage of time spent on the primary flight display, the relevant area of interest (ANOVA: df = 1, F = 7.105, p < 0.05). When further divided into groups of pilots flying and co-pilots, there was a significant dwell time percentage difference for the co-pilots (ANOVA: df = 1, F = 8.038, p < 0.05). No significant dwell time percentage differences (Dunett C: p < 0.05) were found when groups were divided into groups of pilot flying / copilot, into groups with high / low subjective SA, subjective workload or subjective performance, or into groups with high/low flight hours. However, the co-pilot flying had significantly longer dwells on the relevant display (PFD) than the pilot flying (df = 1, F = 5.636, p < 0.05). For the map shift event, no significant correlation was found. However, the pilots did spend time looking at the display during the event. 3.7. Conclusions The answer to the first research question, is that the operationalisation model is sufficient to explain the found results, as we shall se below. It will be discussed in more detail in the concluding general discussion chapter. The second research question was as expected confirmed. As there were differences between the dwell time percentages on the relevant displays for the flights segments, the null hypothesis of no differences was disproved. Furthermore, the significant dwell time differences on the PFD for the altitude bust event, which showed the fast agents to spend more time on the PFD, confirm this further, by disproving that the time spent on the relevant display does not matter. The answer to the additional question then, whether the SA system was sufficient to handle the altitude bust event, is that even for the slowest agents, the attention was sufficient to ensure safe handling of the event. For the map shift event, despite the time spent looking at the display, the 44 event was either not discovered at all, or after several minutes of flight with inaccurate SA regarding the position of the aircraft. The third research question could not be completely answered since the analysis of simultaneous dwells could not be carried out on the available data. There was however a significant division of attention during the Amsterdam-London landing. The answer to the additional question, regarding the attention management system is that, for the altitude bust event that the long-term memory schemes were sufficiently good. There was no need for further training. Furthermore, local attention distribution was sufficient. This implicates that there was no need for attention direction by some aircraft system. There was also no need for any mechanical system to be installed to monitor the altitude capture, in case the one in the autopilot would fail. It can also be concluded that the relevant aspects on the display were comprehensible and salient enough. The aircraft log, finally, was in this case functioning adequately as a mediator of retrospective SA, since it was in all cases properly attended to and comprehended. For the map shift event, the SA system did not work properly. The relevant display was looked at and attended to by the crews during the event and despite this the instrument malfunction was not discovered. The conlusion then is that the malfunction was neither comprehensible nor salient enough to be recognised by an unprepared crew. Here either an additional mechanical monitoring system has to be installed, or the display has to be redesigned, or the pilots has to receive more training to discover the event. The flap asymmetry event was not statistically analysed in the present thesis, however, all crews discovered the event swiftly. The SA question raised by the flap asymmetry event is somewhat different from the other events. Whereas there are not many alternative ways of reacting to a gear unsafe event, than to try and extend the landing gears again, for the flap asymmetry there are at least two possible actions, to make a go-around and land later, and to land immediately. The correct choice depends on if the flaps will be increasingly asymmetrical with time or if the safest thing to do is to delay the landing. This relates to the to the knowledge of how the aircraft works. This knowledge will influence the projected future situation, either to give the pilots an accurate prospective SA or an inaccurate prospective SA. This illustrates that device models are necessary means to reach accurate SA. In this report, the focus is not on the impact of device models for accurate SA. Therefore this was not 45 examined more closely. As a remark, it can be mentioned that the procedure for a flap asymmetry for this experiment was available and used by the pilots, as mediated by a written instruction in the cockpit. 46 4. GENERAL DISCUSSION The present thesis was conducted in order to scrutinise SA in the civil cockpit. The model proposed in this thesis is intended to be sufficient to model SA systems with varying degrees of localisation and distribution of SA processes and models. The model has been fully applied to one experimental situation, the one of this thesis. Further research is needed to establish the general usefulness of the model. The model could however be successfully applied to the experimental situation described in this thesis. The research questions could be operationalised by relating the measurement techniques to the applied SA model and the questions could be answered as well. This does not prove that the model will be successful for every experimental situation regarding distributed SA, but this has to be examined with further research. A careful reader may have noticed that the data collection in the experiment section is based on a very small group of pilots and crews. Of course, a larger group would have been desirable. For economical reasons, however, a larger group was not an option. Instead, a simpler simulation could have been made, with more crews. Then, however, it would have been uncertain how the results would have transferred to a real world situation, since the actual layout, the design of the instrumentation, and the feeling of real flight may influence both workload and the use of normal procedures and eye scan patterns. The last reason being the stronger for this thesis, since the assumption of realistic eye scan patterns is what grounds the applied SA system model in reality. If the eye scan patterns were not realistic, this model can’t tell us this, and it will only show the properties of a system with the measured eye scan patterns and the measured performance. For further research, it is recommended to make more detailed artefact studies of the instrumentation mediating the cues for the events. This would give a more detailed description of what in the presentation was, or was not, sufficiently good. Furthermore, since statistical analyses on small groups are of questionable value, the emphasis on these can be reduced. Not removed, since it is of importance to know whether the instrumentation was at all attended. Instead a qualitative interpretation could be of value, questioning how much time before onset to recognition to action, would be acceptable, and of how much training is dedicated to the discovery of the anomalies. Furthermore it would be 47 possible to analyse whether the discovered attention patterns and behaviours are expected by pilots and by pilot trainers. As this experiment was made within the VINTHEC project, the experiment could not be optimally arranged to answer the questions of this thesis. A compromise had to be made to get acceptable data for all involved in the project. This meant that not all aspects of the SA system could be measured. This could not have been done in any case, since too many measures would have been too intrusive, and made the experiment too unrealistic. To reach the local situation model of the agent, for example, a probing technique such as Endsley’s SAGAT could have been used, but to use this measure on top of the other measures would have made the entire mass of measures too intrusive. Considering the accuracy of the EPOG equipment in a real simulator environment, this was in itself a research question for the VINTHEC group, and was therefore not completely known in advance. The accuracy, as expected, turned out to be sufficient both to answer the VINTHEC theoretical questions, and to answer the general questions of this thesis. To answer more detailed questions of the role of attention for the updating process of situation models, higher measurement accuracy than needed for the general conclusions, will have to be achieved. As a result, parts of the SA definition of this thesis has not been grounded in the empirical work of this thesis, but has been grounded in SA theory and in the results of previously made SA research. 48 5. REFERENCES Artman, H., & Garbis, C. (1998). SA as Distributed Cognition. Proceedings of ECCE’98, Limerick. Ashcraft, M. H. (1994). Human Memory and Cognition. New York: Harper Collins Ballard, D., & Rippy, L. (1994). A knowledge-based decision aid for enhanced SA. IEEE-94CH-3573-0. Bateman, D. (1997) Enhanced GPWS Improves Situational Awareness and Provides More Effective Protection Near Terrain. ICAO Journal, 52(4), 7-8 Berggren, P. (1998). Situational Awareness, mental workload, and pilot performance. Magisteruppsats, Institutionen för pedagogik och psykologi, Linköpings Universitet. Caretta, T. R., Zelenski, M. W. (1996) Situational Awareness: There is no substitute for experience. Flying safety, 52(3), 18-20. Derks, P.L., & Gillkin L. S. (1993). Incongruity, Incongruity Resolution, and Mental States: The measure and Modification of Situational Awareness and Control. NASA-CR-194568 Clark, H. C. (1996). Using Language. Cambridge: Cambridge University Press. Endsley, M. R. (1988). SA Global Assessment Technique (SAGAT). Proceedings of the IEEE National Aerospace and Electronics Conference. Endsley, M. R. (1990). Predictive Utility of an Objective Measure of SA. Proceedings of the Human Factors Society - 34th Annual Meeting, p 4145. Fracker, M. L. (1988). A Theory of Situation Assessment: Implications for Measuring SA. Proceedings of the Human Factors Society -32 nd Annual Meeting, p 102-106. 49 Hartman, B. K., Moylan, P. (1994) Head-Up Guidance Systems: The Foundation Of SA Systems. IEEE 0-7803-2425-0, p88-92 Heath, C. & Luff, P. (1992). Collaboration and Control - Crisis management and multimedia technology in London underground line control rooms. CSCW - An international journal, 1, 69-94. Hutchins, E. (1990). The technology of team navigation. In J. Galegher, R. E. Kraut & C. Egido (Eds.) Intellectual Teamwork - Social and Technological Foundations of Co-operative Work. (pp 22-51). Hillsdae, NJ:Erlbaum. Hutchins, E. (1995) How a cockpit remembers its speeds. Cognitive Science, 19, 265-288. Jones, D. G. & Endsley, M. R. (1996). Sources of SA in Aviation. Aviation, Space and Environmental Medicine, 67, p 507-512. Metalis, S. A. (1993). Assessment of Pilot Situational Awareness: Measurement Via Simulation. Proceedings of the Human Factors Society - 37th annual meeting. Mogford, R. H. (1997). Mental Models and SA in Air Traffic Control. The Internaltional Journal of Aviation Psychology, 6(4), 331-341. Neisser, U. (1976). Cognition and Reality: Principles and Implications of Cognitive Psychology. San Fransisco: W H Freeman and Company. Regal, D. M., Rogers, W.H., Boucek, G. P. (1988). Situational Awareness in the Commercial Flight Deck: Definition, Measurement and Enhancement. Aerospace Technology Conference and Exposition. Sarter, N. B. & Woods, D. D. (1991). SA: A Critical But Ill-defined Phenomenon. The International Journal of Aviation Psychology, 1(1), 45-57. Schwartz, D. (1994). Training for SA. Air Line Pilot, 20-23, May edition. Shively, R. J., Goodman, A. D. (1994). Effects of Perceptual Augmentation of Visual Displays: Dissociation of Performance and Situational Awareness. Proceedings of the Human Factors and Ergonomics Society 38th Annual Meeting. 50 Sollenbergher, R.I., & Stein, E. S. (1995). The Effects of Structured Arrival and Departure Procedures on TRACON Air Traffic Controller Memory. DOT/FAA/CT-TN95/27 Tenney, Y.J., Adams, M. J., Pew, R. W. Huggins, A. W. F, & Rogers, W. H. (1992). A Principled Approach to the Measurement of SA in Commercial Aviation. NASA-CR-4451 VINTHEC Consortium. (1996). Technical Report: Human Awareness Criteria. VINTHEC-WP1-TR 01, NLR, Amsterdam VINTHEC Consortium. (1997). Technical Report: Review of Eye-pointof-gaze Equipment and Data Analysis. VINTHEC-WP2-TR 01, NLR, Amsterdam VINTHEC Consortium. (1997). Technical Report: Dynamic Measures of Pilot Mental Workload (PMWL), Pilot Performance (PP), and Situational Awareness (SA). VINTHEC-WP3-TR 01, NLR, Amsterdam VINTHEC Consortium. (1998). Technical Report: Testplans for Pilot Experiments. VINTHEC-WP4-TR 01, NLR, Amsterdam VINTHEC Consortium. (1998). Technical Report: Technical Implementation. VINTHEC-WP5-TR 01, NLR, Amsterdam VINTHEC Consortium. (1998). Technical Report: Pilot Studies. VINTHEC-WP6-TR 01, NLR, Amsterdam Weick, K. E. & Roberts, K. H. (1993). Collective minds in organisations: Heedful interrelating on flight decks. Administrative Science Quarterly, 38, 357-381. Wellens, A. R. (1993). Group SA and distributed decision making: From military to civilian applications. In N. J. Castellan (ed. ). Individual and Group Decision Making: Current issues. (pp. 267-87). Erlbaum. 51 6. APPENDIX A Significant dwell time percentage differences (Dunnet C: p < 0.05) between the flight phases, for the pilot flying (PF) and the pilot nonflying (Where PF is not indicated). AL PFD Cruise <> Approach PF Climb <> PF Approach & PF Landing PF Cruise <> PF Approach & PF Descent PF Descent <> PF Cruise & PF Approach & PF Landing PF Approach <> PF Climb & PF Cruise & PF Descent & PF Landing AL NAV Cruise <> Approach Descent <> Approach Approach <> Cruise & Descent PF Climb <> PF Approach & PF Landing PF Cruise <> PF Approach & PF Landing PF Descent <> PF Approach & PF Landing PF Approach <> PF Climb & PF Cruise & PF Descent PF Landing <> PF Climb & PF Cruise & PF Descent AL OUT Climb <> Approach & Landing Cruise <> Approach & Landing Descent <> Approach & Landing Approach <> Climb & Cruise & Descent Landing <> Climb & Cruise & Descent PF Climb <> PF Approach & PF Landing PF Cruise <> PF Approach & PF Landing PF Descent <> PF Approach & PF Landing PF Approach <> PF Climb & PF Cruise & PF Descent & PF Landing PF Landing <> PF Climb & PF Cruise & PF Descent & PF Approach AL ENG AL CDU 52 Climb <> Approach & Landing Descent <> Landing Approach <> Climb Landing <> Climb & Descent AL A/P Descent <> Landing Landing <> Descent PF Climb <> PF Landing PF Cruise <> PF Landing PF Descent <> PF Approach & PF Landing PF Approach <> PF Descent PF Landing <> PF Climb & PF Cruise & PF Descent LA PFD Climb <> Approach Cruise <> Approach Approach <> Climb & Cruise PF Climb <> PF Approach & PF Landing PF Cruise <> PF Descent & PF Approach & PF Landing PF Descent <> PF Cruise & PF Landing PF Approach <> PF Climb & PF Cruise & PF Landing PF Landing <> PF Climb & PF Cruise & PF Descent & PF Approach LA NAV Climb <> Approach Cruise <> Approach Descent <> Approach Approach <> Climb & Cruise & Descent PF Climb <> PF Approach & PF Landing PF Cruise <> PF Approach & PF Landing PF Descent <> PF Approach & PF Landing PF Approach <> PF Climb & PF Cruise & PF Descent PF Landing <> PF Climb & PF Cruise & PF Descent LA OUT Climb <> Approach & Landing Cruise <> Approach & Landing Descent <> Approach & Landing 53 Approach <> Climb & Cruise & Descent Landing <> Climb & Cruise & Descent PF Climb <> PF Landing PF Cruise <> PF Landing PF Descent <> PF Landing PF Landing <> PF Climb & PF Cruise & PF Descent LA ENG LA CDU Cruise <> Approach & Landing Approach <> Cruise Landing <> Cruise PF Cruise <> PF Approach & PF Landing PF Approach <> PF Cruise PF Landing <> PF Cruise LA A/P Climb <> Landing Cruise <> Landing Descent <> Landing Landing <> Climb & Cruise & Descent PF Cruise <> PF Approach & PF Landing PF Approach <> PF Cruise PF Landing <> PF Cruise 54 7. APPENDIX B Nationaal Lucht- en Ruimtevaartlaboratorium National Aerospace Laboratory (NLR) VINTHEC Questionnaire Beoordelings Schaal Mentale Inspanning (BSMI) Wilt u door middel van het zetten van een streepje op de verticale lijn aangeven hoeveel inspanning het u heeft gekost om de zojuist door u verrichtte taak te voltooien. 150 140 130 120 Ontzettend inspannend 110 Heel erg inspannend 100 90 Erg inspannend 80 Behoorlijk inspannend 70 60 Tamelijk inspannend 50 40 Enigzins inspannend 30 Een beetje inspannend 20 Nauwelijks inspannend 10 Helemaal niet inspannend 0 Volgorde number: ……. Vlieger: flying / non-flying Datum: ….... april 1998 Crew nummer: ….... Vlucht: AMS - LHR / LHR - AMS Vlieger nr.: ……. (Links / Rechts) Tijd ….... : ….... 55 Nationaal Lucht- en Ruimtevaartlaboratorium National Aerospace Laboratory (NLR) VINTHEC Questionnaire Situational Awareness (SA) Pilot Rating Scale Yes Was your SA on a satisfactory level? No Yes Was your SA on an acceptable level? No My SA included all aspects of importance as well as a lot of other aspects of the dynamic situation. 10 My SA included all aspects of importance as well as some other aspects of the dynamic situation. 9 My SA included all aspects of importance (neither more nor less) of the dynamic situation. 8 My SA was insufficient. A few important aspects were out of my control. 7 My SA was reduced. Some important aspects were out of my control. 6 My SA was low. A lot of important aspects were out of my control. 5 My SA was very low. About half of the important aspects were out of my control. 4 My SA was very, very low. Most important aspects were out of my control. 3 My SA was extremely low. Almost all important aspects were out of my control. 2 Yes Was it possible to perform the task with respect to your SA? No No SA at all. All important aspects were out of my control. Start here! Pilot: flying / non-flying Date: April ….... 1998 Crew number: ….... Pilot number: ……. (Left / Right) 1 Order number: ……. Flight: AMS - LHR / LHR - AMS Time: ….... : ……. Remark: The actual size of this questionnaire is bigger. The size has been adjusted so that it fits on a report page together with the standard headers that are used throughout the entire report. 56 Sleep questionnaire Length of sleep • At what time did you go to bed last night? • At what time did you get up this morning? ……………….. ……………….. Scale for quality of sleep The following 14 questions refer to your sleep during the past night. The questions can provide a good impression of one's quality of sleep. Please answer every question by ticking 'yes' or 'no', according to the alternative that best describes your rest of last night. Scale for Quality of Sleep (SQS) 1. I did not sleep at all last night 2. Last night, I woke up and had trouble getting back to sleep 3. I got out of bed last night 4. I find that I slept badly last night 5. Last night, I fell asleep very easily 6. I slept less than five hours last night 7. Last night, before I fell asleep, I was awake for at least half an hour 8. Last night, I woke up several times 9. Last night, I tossed and turned all night 10 I find that I slept well last night . 11 I feel as if I only slept a couple of hours . 12 After getting up this morning, I felt tired . 13 I feel that I did not get enough sleep last night . 14 After getting up this morning, I felt well rested . Please make sure that you’ve answered all questions! 57 ye s no Pre flight questions 1. How well do you think you will succeed with this flight? Not at all Average Very 2. How motivated are you to perform this flight? Not at all Average Very 3. How difficult do you think this flight will be? Very simple Average Very difficult 4. How would you characterise this flight, what are typical aspects of this flight that you can expect? (You can think of the holding, the workload that ATC might generate, and so on.) ………………………………………………………………………………………………………… ………………………………………………………………………………………………………… ………………………………………………………………………………………………………… ………………………………………………………………………………………………………… 58 During flight ratings 1. I would describe the mission performance during this period as: HopelessPoor Faulty Mediocre Fair Good Excellent 2. Did something peculiar occur during this period? No Go to question 2A Yes Stop 2A. If yes, at what amount of relevant displays had you been looking immediately before the odd situation occurred? None Some Most All 2B. If yes, how long did it take you to recognise that something odd has been going on? < 5 sec.½ min.> 1 min. > 2 min.> 3 min.> 4 min.> 5 min. Post flight questions 1. How difficult did you find the flight? Very simple Average Very difficult 2. How well would you say you succeeded with the flight? Not very well Average Very well 3. How would you assess the level of realism of the flight? Very poor Average Very good 4. How much ‘spare time’ did you have during the flight? None at all Average A great amount 5. How much physical effort was required for the flight? 59 None at all Average Excessive 6. How much mental effort was required for the flight? None at all Average Excessive 7. What was the overall level of physical workload? None at all Average Excessive 8. What was the overall level of mental workload? None at all Average Excessive 9. Did you find it difficult to evaluate any of the available information? Very simple Average Very difficult 10. To what extend were you disturbed by information other than that required for the task? Not at all Average A great deal 11. Concerning the flight control task… I need to consciously think what to do Average I acted without conscious consideration 12. Concerning the navigation task… I need to consciously think what to do Average I acted without conscious consideration 13. Considering the entire flight the helmet is comfortable - uncomfortable / heavy - light 60 / steady - sliding a bit Post Amsterdam - London questions 1. Have you experienced a map shift before? Yes No 1A. If yes, can you give an estimation about how often you experienced a map shift? …………………………………………………………………………………………………. 1B. If yes, do you remember the last time you experienced a map shift? …………… …………………………………………………………………………………………………. 2. Did you ever experience a flap asymmetry before? Yes No 2A. If yes, can you give an estimation about how often you experienced a flap asymmetry? …………………………………………………………………………………………………. 2B. If yes, do you remember the last time you experienced a flap asymmetry? …………… …………………………………………………………………………………………………. 61 Post London - Amsterdam Questions 1. Have you experienced an altitude bust before? Yes No 1A. If yes, can you give an estimation about how often you experienced a altitude bust? …………………………………………………………………………………………………. 1B. If yes, do you remember the last time you experienced a altitude bust? …………… …………………………………………………………………………………………………. 2. Have you experienced a gear unsafe before? Yes No 2A. If yes, can you give an estimation about how often you experienced a gear unsafe? …………………………………………………………………………………………………. 2B. If yes, do you remember the last time you experienced a gear unsafe? …………… …………………………………………………………………………………………………. 62 Questionnaires instructions Before, after and during the simulator questionnaires will be handed to you. experiment some The responses to the questions as well as the psychophysiological measurements will be stored separately from your personal information such as your name or employer. Besides this separation no one except for the experiment operators is allowed and will be able to access this information. All the information will be used for the purpose of this experiment only and will not be used for medical purposes or become available to third parties. Besides the measurements and recordings that will be made in the simulator we are interested in your background, and how you perceive the experiment. This information will be compared with the experimental results and psychophysiological measurements. Therefore we kindly ask you to take your time to fill in these questionnaires. Thanking you in advance, The experiment management. 63 Flight experience Please answer the following questions about your flight experience: Aircraft type In the function of Hour s Period (MM/YY MM/YY) ….. / ….. - ….. / 1. ….. ….. / ….. - ….. / 2. ….. ….. / ….. - ….. / 3. ….. ….. / ….. - ….. / 4. ….. ….. / ….. - ….. / 5. ….. ….. / ….. - ….. / 6. ….. ….. / ….. - ….. / 7. ….. ….. / ….. - ….. / 8. ….. ….. / ….. - ….. / 9. ….. ….. / ….. - ….. / 10 ….. . Please make sure that you’ve answered all questions! If you have any experience with PC-based computer games that simulate aircraft like Microsoft flight simulator, flight unlimited, and so on, please give a brief description below of this experience and the period you used those simulations regularly. …………………………………………………………………………… …………………………………………………………………………… …………………………………………………………………………… …………………………………………………………………………… …………………………………………………………………………… …………………………………………………………………………… …………………………………………………………………………… … 64 Use of substances Did you use any of the following substances in the past 24 hours: Substance Ye s No If yes, what and how much 1. Alcohol 2. Caffeine 3. Medication 4. Nicotine 5. Other • • • • coffee: ……………….. cups tea: ……………….. cups cola: ……………….. glasses other:……………….. Please make sure that you’ve answered all questions! 65