RISKMAN – Subproject proposal 1.1. Subproject full title: New Concept of Diagnostic Dynamic Expert Systems 1.2. Subproject Acronym: DIAGES 1.3. RISKMAN Research Area: RA2: Systems Performance Monitoring and Diagnostics 2.1. 2.2. Proposing Organisation: Institute of Fluid Flow Machinery of the Polish Academy of Sciences, Gdansk, Poland Contact Person Name: Prof. Jan Kicinski 2.3. Address: 2.4. Tel.: (+48) 58 341 12 71 2.5. Fax: (+48) 58 341 61 44 2.6. e-mail: kic@imp.gda.pl 2.7. Web site: www.imp.gda.pl 2.8. Participating companies, name and country: Fiszera 14, PL – 80-952 Gdansk, Poland 1. (IMP-1) Institute of Fluid Flow Machinery – Department of Thermomechanics of Fluids, (O2),Gdansk, Poland, Leader: Prof. P. Doerffer e-mail: doerffer@imp.gda.pl 2. (IMP-2) Institute of Fluid Flow Machinery - Department of Machine Mechanics, Gdansk, Poland, Leader: Prof. J. Kicinski e-mail: kic@imp.gda.pl 3. (SUT-1) Silesian University of Technology- Department of Machinery Design, Gliwice, Poland, Leader: Prof. W. Cholewa e-mail: wch@polsl.pl 4. (SUT-2) Silesian University of Technology- Department of Power Systems, Gliwice, Poland, Leader: Prof. T. Chmielniak e-mail: imiue@rie5.ise.polsl.gliwice.pl 5. (UMM) University of Mining and Metallurgy, Department of Vibroacustics and Robotics, Krakow, Poland, Leader: Prof. T. Uhl e-mail: tuhl@rob.wibro.agh.edu.pl 6. (TUG) Technical University of Gdansk, Department of Ship Automation and Turbine Propulsion, Gdansk, Poland, Leader: Dr J. Gluch, e-mail: jgluch@imp.gda.pl 7. (MD) Machinery Diagnostics (SME), Gdansk, Poland, Leader: Prof. A. Gardzilewicz e-mail: gar@imp.gda.pl 8. (EC) Energocontrol sp. Z.o.o., Krakow, Poland, Leader: Dr. T. Barszcz, e-mail: info@energocontrol.com.pl 3. Proposal summary The main goal of the proposed subproject is the formulation and application of knowledge- and model-based artificial intelligence methods such as: heuristic modelling, multi-models approach, model inversion techniques, adaptive systems, fuzzy logic methods, neural networks methods, genetic algorithms methods and their combination to the monitoring and control of industrial systems. 1 A new concept here consists in construction of a system in which generated conclusions depend on the time available. The longer the available time, the more reliable are the conclusions. This is the main idea of, so-called, dynamic expert systems. The model reversing technique as one of the methods of knowledge acquisition plays an important role in this approach. Other new concept of intelligent sensor will be formulated and developed. The intelligent smart sensor (ISS) will have several flexible inputs from primary sensors as accelerometers, pressure sensors, and force sensors. The sensor is designed with the use of artificial neural networks or fuzzy logic intelligent algorithms depending on application and implemented hardware. DIAGES Diagnostic Dynamic Expert System RTD Tasks Dynamic Diagnostics Thermal Diagnostics Intelligent Smart Sensor (ISS) DEM Tasks Prototype of DIAGES – System Demonstration on virtual plant Possible implementation in real power plant TRA Tasks Training of personnel involved in operation of DIAGES Fig.1. General DIAGES subproject structure The application of these tools on industry level is the main breakthrough idea of the proposal presented here. Development of these tools needs sound "knowledge base" concerning the test object, which is a thermal power plant. Present power stations should be converted into energy factories of the future making use of such diagnostic systems leading to appropriate organisation models. The general structure of the DIAGES subproject is presented in Fig.1. 2 4. Objectives Design and implementation of a low cost, intelligent data storage diagnostic units for risk management algorithms, running on higher levels of distributed diagnostic networks, is the general objective of the subproject. To achieve such a goal the following group of objectives has to be realised: - formulation and implementation of modern knowledge- and model-based approach to secure and predictive maintenance of power industry machinery, - measurement data reliability determination, - development of new concepts of knowledge acquisition for supervisory systems, - development and implementation in industry of the procedures for better performance and safety of machinery taking into account the mechatronic and micro-devices aspects, - modeling methodology of selected objects (e.g. model reversing using artificial intelligence, modeling of couplings, heat exchangers, etc), - creation of knowledge basis and expert systems for industry, - optimisation of the decision making by genetic algorithms, - development of communication tools according to future standards (UMTS), - development of first level diagnostic algorithms for ISS subsystems, - development of data compression algorithms according to risk management task, - development of self diagnostic actions algorithms, - development of hardware solution of ISS coupler according reliability and data safety, - determination of relations between reasons and symptoms using new methods of fuzzy logic, genetic algorithms and artificial neural networks. These objectives are aimed at improvement of our products according to risk management related tasks. The module of this diagnostics should be a part of an open dynamic and hybrid diagnostic system. Access to necessary results of measurements is possible in cooperation with industrial partners and SME. 5. Deliverables (new products, new processes and services, radical innovations; prime deliverable is expected to be a breakthrough in applicable knowledge to be transferred to industry and society) D1) New dynamic expert system ready for operation in the plant (full documentation of the system, prototype hardware and software environment) D2) Prototype modules of knowledge base and fusion software. Complete set of routines for dynamic and thermal diagnostics based on artificial intelligence and new updatable knowledge base (documentation). D3) Prototype modules of system models and diagnostic multi-models (documentation, models of dynamical states, model of heat exchangers degradation, computer codes) D4) New algorithms for first level diagnostic tasks implemented in ISS (Intelligent Smart Sensor) and/or ISS coupler D5) Prototype of communication module for ISS coupler using new communication standards D6) New on-line and off-line learning algorithms for ISS. 3 D7) Prototype of monitoring system using developed ISS (several different units) and ISS coupler installed on industrial or infrastructural object D8) www nano-server embedded in ISS coupler. Host computer-supervising installation of all ISS components. D9) Procedures for thermal diagnostics based on entropy methods. D10) Trained neural networks for all new modelling methods D11) Satisfactory tests of system operation on virtual heat cycle D12) The set of algorithms for evaluation of DIAGES impact on environment D13) Final report, conclusion and operation 6. Justification and potential impact (economic impact, direct and indirect economic benefits, European dimension, training and education, conformity with EC societal objectives: quality of life, health, safety, working conditions, employment and environment) Main types of energy necessary for human society consist of electricity and heat (or chill). Sufficient amount of energy is produced by continuously operating power plants. District heating power plants are much more efficient sources providing energy than individual heating systems. Besides, the excess energy in power plants is used to generate electricity, increasing the economic quality of the system. Usually they are placed in the neighbourhood of towns and other centres of civilisation. This close neighbourhood causes many problems, not only technical but also social. Technical problems concern efficient energy production, safety for the crew and other people, lifetime of the components etc. while social problems concern enterprise’s organisation, environment protection, wastes utilisation, communication, etc. Nowadays each of these problems is solved individually and in many cases improvement at one place causes worsening at the others. The new procedures for such enterprise combining all the problems, applying more sophisticated and knowledge-based concepts are needed in order to fulfil requirements not only known nowadays but also to answer new demands of the future. That should be solved not only during the design process but also during exploitation under continuously changing demands and conditions. Diagnostic systems based on modern methods, making use of artificial intelligence, provide the best solution for new concept of cheap (efficient), flexible, integrated, safe and clean energy production. Broad definition of the diagnostics covers different fields of application. Vibration diagnostics, foundation diagnostics, material and life-time diagnostics, control systems diagnostics help to assure safety of production. Thermal (or performance) diagnostics let to decrease internal and external production costs mainly by reduction in consumption of primary resources and by improvement of environmental protection against pollutants. Diagnostics of measuring systems helps to provide reliable experimental data. Economical and organisational diagnostics assures overall low cost production. All these diagnostic systems applying artificial intelligence methods, heuristic modelling, multi-models approach, model inversion technique, training of adaptive systems, fuzzy logic methods, neural networks methods, genetic algorithms methods and new concept of intelligent sensor should cooperate. Distributed systems theory known from computer technology and from systems theory let to combine them efficiently and to create open dynamic and hybrid supervising system. Those system lets to take into account not only known but also new unknown production conditions and technologies and thus it is flexible and easy to develop. However it is a very challenging task taking into account its complexity. Present power stations should be converted into energy factories of the future making use of such diagnostic systems leading to appropriate organisation models. 4 7. Description of the work (technological approaches and methods, work tasks, their description, deliverables and work effort in person month) marked according to the following components (RTD = Research, technological development and innovation-related activities, DEM = Demonstration activities, TRA = training) Dynamic diagnostics. Expert system integration All the developed modules are to be integrated into a dynamic diagnostic expert system. The expert system will take advantage of data, knowledge and information fusion from different sources. As data sources both the internal and external (distributed) databases, model-based data sources as simulation units, and measurements from multiple sensors will be used. This task addresses development of an expert system able to operate in full dynamic regime with taking into account dynamics of the system being diagnosed or monitored. By the development of the system under consideration the broad experience of the consortium in implementing of such a class of expert systems operating in industrial conditions will be of great advantage. The main principle of the system structure consists in its open character since it will be possible to include into the system other modules, which were developed for the sake of other systems and by other RTD groups. Heuristic modelling is applied to identify models from data that describe behaviour of dynamic objects and processes operated by humans, mainly in open control loop. The models are especially useful in such cases where both continuous and discrete, event-type controls and uncontrolled inputs to the object/process occur. Regardless heuristic nature, they make possible the accurate predictions of system/process parameters. The objective of the models may be manifold. History of w1...wn values Input supervisor model 1 w1 model 2 w2 Output ... wn model n Multi-models are aggregates of several different models whose inputs are fed to the common multi-model input, whereas the multi-model output is combined from outputs of constituting models –Fig.2. There are many solutions to a way of the outputs combination. One of them being their ordinary, linear combination. The diagnostic multi-model corresponds to a negotiating team of experts, where each component model corresponds to an expert, a model’s output may be interpreted as an expert’s opinion. The multi-model output corresponds to the opinion of the team of experts, and weights may be interpreted as experts’ competences. Such multi-models are going to be applied in the expert system. The essence of data and information fusion consists in gathering information at the same time or at some intervals from many data sources. This information enables better 5 identifying technical state of the examined object. Precision and reliability that may be obtained using the Data Fusion methods could not be achieved using standard manners (with single sensors). Methods of data and information fusion are to be implemented among dataand knowledge sources, yielding required integration between units of the expert system. Diagnostic expert systems use different knowledge representations and reasoning algorithms. An important branch of research dealing with modelling diagnostic rules is connected with Bayesian networks, introduced (or reintroduced) by J. Pearl, called also belief networks (BN). The most important step in the designing of the BN is connected with acquisition of conditional probabilities. They may result from different sources, e.g. such as known physical or numerical models, results of experiments or passive observations as well as opinions of domain experts. An expert opinion is often the only source of information on the discussed probability. The quality of the BN depends on the quality of network structure and parameters introduced by experts. It should be explained that the quality measure of such data is unknown and validation of the complex BN, which use such data is very difficult or even impossible. Diagnostic model of an object describes the relations between observed symptoms and their causes, i.e. technical states of the object. It exemplifies a class of inverse models i.e. models determining the matching results to causes. Direct specification of such models is (quite) impossible due to complex nature of these relations. We propose to deal with the general idea of the inverted diagnostic models, i.e. models obtained as the result of a kind of inverting transformation applied to known cause-effect models. Development and demonstration of new methods for the efficient definition, identification and updating of inverted models is one of important goals of the subproject. Thermal performance diagnostics The need of thermal performance diagnostics emerges from the new and rational tendency in power unit operation consisted in elongation of periods between repairs. This tendency formulates serious task for performance diagnostics whose role is to keep power station at the highest possible efficiency and economic level. That diagnostics assures low consumption of primary resources which in turn influence economics, safety and eventually decreases environmental impact of energy conversion by thermal power plant. It helps also to make optimisation of energy conversion by whole energy production system. In this new approach risk based analysis is necessary, leading to optimal decision-making for maintenance planning as long as off line aspects are concerned, but also on-line decisions connected with safety issues. Analysis of heat-flow cycle makes use of the diagnostic relations. These relations are to be created with the use of new knowledge-based methods aiming at application of the artificial intelligence methods. They help to recognise the faulty (inefficient) components of the system as well as scale of their degradation. Such a diagnostic system gives data for economically founded innovative, reliable, smart and cost-effective exploitation procedures. This system on the one hand is based on expert knowledge and intelligence and on the other hand it develops this knowledge, verifies it and transfers new experience into these procedures. Moreover it supplies knowledge from the exploitation to the design, helping in a creation of more and more efficient life-cycle systems. Application of knowledge-based artificial intelligence methods consists in: Fuzzy logic methods, Neural networks methods, Genetic algorithms methods, Application of combination of these three. 6 It is believed, that smart and efficient system should apply all of this methods for particular purpose. For example fuzzy logic methods could supplement diagnostics of measurement lines and devices. Neural networks make calculations easier and they are efficient in searching of faulty components and in determination of degradation scale for objects operating under known conditions. Genetic algorithm method helps to find inefficient components for objects operating under unknown or newly established conditions. All these actions are the base for technical, economical and environmental analysis. Thus, this system could be treated as hybrid system and it needs strong cooperation between industry and researchers in order to be reliable. This helps them also to exchange knowledge. The system developed in this way is an open system which could include new ideas and procedures as well as be a part of a supervising distributed system creating appropriate organisation models. It could be used as well for operation as for training purposes. It improves knowledge based management and it could be a part of industrial systems of the future. Intelligent Smart Sensor (ISS) The intelligent smart sensor (ISS) will have several flexible inputs from primary sensors e.g, accelerometers, pressure sensors, force sensors, voltage or current. The sensor is designed with the use of artificial neural networks or fuzzy logic. Intelligent algorithms depend on application and implemented in hardware. For hardware implementation FPGA technology and new design procedure will be formulated. Currently many diagnostic systems use vision as primary input. Proposed solution of ISS is also capable to handle such inputs. Local ISS coupler is a kind of hub for several ISS units. Together it creates real time monitoring network for monitoring and first level diagnostics. The coupler is highly sophisticated embedded computer. It performs the following tasks: - data acquisition from ISS, - data recording (in range of 1GB memory) - management of ISS group e.g. Update of artificial neural network parameters inside ISS using chosen learning algorithms - sharing data on local www nano-server - execution of the first level diagnostics algorithms Notification about detected damage can be made available using wireless and remote communication tools like; GPRS, SMS, radio. A local autonomous diagnostic centre (unit) described above can be integrated into diagnostic networks and can be applied for any devices (vehicles, power plants, ect.) or infrastructure elements (tunnels, bridges, dams, buldings, ect.) Proposed system is fully modular and can be installed as autonomous or fully integrated with diagnostic network Implementation of dynamic diagnostic expert system The goal of this task is to put the dynamic diagnostic expert system into operation by the consortium. The intended expert system is going to be implemented in the Kozienice Power Plant at turbine generator sets of 230 or 500 MW. To put the system into operation some investment work is needed, that covers complete hardware including measuring lines, transmission of signals and data, signal processing hardware and software, required computer systems etc. Finally, the system will undergo extensive validation with the use of especially elaborated validation software and techniques. The possibility of realisation of this task depends upon the interest and the resources of the IP. 7 Training of personnel involved in operation of dynamic diagnostic expert system To enable the industrial partner to operate the system into the plant, an extensive training is intended to carry out. This training will take part with the application of the expert system operating on data collected from the real-existing object. Moreover, training on the system operating in several simulation modes will be applied. To make it possible, special data and knowledge sources will be included into the system, and respective training facilities and materials prepared.. Work split Task No. 1 2 2.1 2.2 3 3.1 3.2 4 4.1 4.2 4.3 4.4 4.5 Task description Partner Deliverable Expert system integration (RTD) SUT-1 D1 Heuristic modelling of systems (RTD) SUT-1 D2 IMP-2 Development of new models and computer codes IMP-2 describing the dynamical state of selected objects (RTD) Verification of new models and codes both in laboratory IMP-2 scale and on real objects (RTD) Diagnostic multi-models (RTD) SUT-1 D3 IMP-2 IMP-1 Methodology of description of coupled aerodynamic, SUT-1 mechanic and electric interactions in large power industry IMP-2 objects (RTD) IMP-1 Identification procedure of coupled multi-models (RTD) SUT-1 IMP-2 Thermal performance diagnostics (RTD) IMP-1 D9 TUG D10 SUT-2 D11 MD Measurement data reliability determination. IMP-1 Analysis of large sets of measurement. Determination of TUG the possible reasons for their standard deviations. Code preparation for artificial neural-networks to determine reliability of measurements. (RTD) Creation of knowledge base containing all elements of the IMP-1 whole heat-flow cycle. Preparation of codes allowing the MD up-dating of knowledge base. (RTD) Formation of a complete set of symptom – reason IMP-1 correlations for all elements included in the analysis. IMP MD PAN, MD (RTD) Procedures for thermal diagnostics based on entropy SUT-2 methods (RTD) Preparation and verification of on-line diagnostic TUG procedures (RTD) Effort MM 15 20 30 87 8 Task No. 4.6 4.7 4.8 4.9 5 5.1 5.2 5.3 5.4 5.5 6 7 8 9 9.1 9.2 9.3 9.4 10 11 Task description Development of new modelling of heat exchangers degradation (RTD) Training of neural networks for all new modelling methods (RTD) Demonstration of system operation on a virtual plant, (DEM) Evaluation of new system impact on environment by means of genetic algorithms (RTD) Intelligent Smart Sensor (ISS) – (RTD) Partner Delive- Effort rable MM IMP-1 IMP-1 IMP-1 TUG IMP-1 UMM EC Specification of complete system (RTD) UMM Design of product (RTD) UMM Prototype of system components (RTD) UMM EC Integration of diagnostic system (RTD) UMM Demo installation (DEM) EC Data and information fusion for diagnostics and SUT-1 monitoring (RTD) Acquisition of diagnostic knowledge (RTD) SUT-1 Acquisition and improvement of diagnostic belief SUT-1 networks (RTD) Identification and updating of inverted models(RTD) SUT-1 IMP-2 Generation of training data (RTD) IMP-2 Investigation of model sensitivity on selected defects IMP-2 (RTD) Building and training of adaptive systems (RTD) SUT-1 Building of diagnostic relations (RTD) SUT-1 IMP-2 Implementation of dynamic diagnostic expert system SUT-1 (DEM) IMP-2 UMM IMP-1 Training of personnel involved in operation of dynamic SUT-1 diagnostic expert system (TRA) UMM D4D8 58 D3 10 D3 D3 10 10 D2 40 D1 D12 D13 60 10 9 8. Partners involved, partner profiles (business idea, size, competence) and the role of each partner (IMP-2) Institute of Fluid Flow Machinery – Established as European Centre of Excellence for Clean and Safe Technologies in Power Engineering Department of Machine Mechanics- coordination of the subproject, research on dynamic diagnostics, building of large object models, analyses of coupled non-linear vibration in rotor- bearing-foundation systems, generation of training data for adaptive systems. High expertise in model-based diagnostics, structural mechanics, crack identification, FEM-Methods, industrial applications (e.g. new journal bearings for turbo-sets). Size : 19 employers (4 Professors, 3 Assistant Professors). (IMP-1) Institute of Fluid Flow Machinery - Department of Thermomechanics of Fluids, thermal diagnostics, diagnostics of turbine flow path, monitoring of heat cycle and intelligent analysis of its condition, searching faulty components of heat cycle, heat cycle components degradation, research on flow phenomena, artificial intelligence methods. 30 employees (1 professor, 4 assistant professors) (SUT-1) Silesian University of Technology- Department of Machinery Design, Research on dynamic diagnostics and expert systems (2 professors, 8 assistant professors, 10 PhD students) High expertise in: diagnostics, development of diagnostic databases and their implementation in diagnostics and monitoring diagnosing with hidden models, identification and application of approximate and/or uncertain models for technical diagnostic development of real-time dynamic expert systems, diagnostics of complex machinery in varying conditions of operations, model-based prediction of state propagation, industrial applications of diagnostic and intelligent monitoring systems (eg. DT200 employed in the 200 MW Turbine Generator in a power station in Poland – together with IMP2, SUT2, UMM, EC), (SUT-2) Silesian University of Technology- Department of Power Systems, (UMM) University of Mining and Metallurgy, Department of Vibroacustics and Robotics. Research on Intelligent Smart Sensor, monitoring and diagnostics systems, modal analyses. Size: 16 employers (3 Professors, 5 Assistants) and 12 PhD students. (TUG ) Technical University of Gdansk, Department of Ship automation and Turbine propulsion, Gdansk, Poland, Leader: Dr J. Gluch, e-mail: jgluch@imp.gda.pl (MD ) Machinery Diagnostics (SME), Gdansk, Poland, Leader: Prof. A. Gardzilewicz e-mail: gar@imp.gda.pl (EC ) Energocontrol sp. Z.o.o., Krakow, Poland. SME- high tech company, 40 employees, manufacturer of monitoring and diagnostic systems 10 including algorithms hardware and software, RTD performer, mechanical design, mechanical integrity calculation, diagnostic service performer. 9. Resources for total subproject and for each partner (resources needed: personnel, equipment etc; costs, work effort in person month) Partner Total IMP-1 IMP-2 SUT-1 UMM TUG SUT-2 EC (SME) MD (SME) Total 10. 170.000 200.000 230.000 100.000 80.000 50.000 70.000 Direct Overhead Travel Equipment Materials Personal MM costs 136.000 34.000 4.000 10.000 3.000 119.000 61 160.000 40.000 5.000 15.000 3.000 137.000 70 184.000 46.000 5.000 10.000 5.000 164.000 83 80.000 20.000 4.000 10.000 3.000 63.000 33 64.000 16.000 1.000 3.000 1.000 59.000 30 40.000 10.000 1.000 2.000 1.000 36.000 18 56.000 14.000 5.000 2.000 49.000 25 80.000 64.000 16.000 - 5.000 2.000 57.000 30 980.000 784.000 196.000 20.000 60.000 20.000 684.000 350 Duration (starting date, duration in month) 1.01.2004, 36 months, 10. Financial plan Years DIAGES 2004 Total 12. 2005 330 000,- 330 000,- 2006 320 000,- Other issues (e.g. ethical, gender, EC policy related issues) Introduction of higher standards and new methods of plant control and monitoring will increase the safety and efficiency of energy supply and the education standard of the crew, which has an impact on society. 11