Flight Control Systems and Autopilots Brian Jewell Department of Engineering, Calvin College Engineering 315 Final Paper Professor Ribeiro Abstract: Autopilot systems have been crucial to flight control for decades and have been making flight easier, safer, and more efficient. However, these autopilot systems are complex devices that require precise control and stability. These systems usually include a form of digital control systems to allow for easier implementation. One example, the Beaver Autopilot system, uses an inner, outer loop system to maintain control while simpler systems often only require something as simple as a PID controller to keep the aircraft stable. 1. Introduction Since the creation of the first aircraft, the ability for people to travel large distances in relatively short periods of time has drastically increased. However, in the beginning of the airplane age, traveling on these craft was difficult as there are many components of air travel and controlling these components can be an extremely difficult task for a pilot. In modern aircraft, there are simply too many things for the pilot to control them all so some form of automation must be done. Also, long air trips can cause problems for the pilots. While the plane is traveling along the same trajectory, flying can and does become a rather monotonous job for the pilot and the pilot runs a higher risk of falling asleep or suffering from a reduced response time. For these reasons, autopilot systems have become a blessing to the aerospace industry. However, these autopilot systems are not simple systems. They require complex control systems with robust measuring equipment. The scope of this paper is to give an overview of autopilot systems (with a description of digital controls), to examine the Beaver Autopilot System, and to explore a simple flight control example. 2. History The earliest form of autopilots have been in existence for decades. The first systems were created an implemented about ten years after the Wright Brothers flew the very first airplane: the Kitty Hawk. These earlier systems were simply a gyroscope that gave the plane a smoother flight path. The only control that these systems gave was in the altitude and pitch. They kept the plane from flying with its nose pointed down and from flying crooked. Around the time of Second World War, the very first fully functional autopilot systems were designed and tested. These earlier systems had the capability to keep the flight path level as well as launch and land the aircraft. This revolutionary system, however, was not a simple task and it was prone to failure. The systems would often break down and crash the plane (by making the plane point straight down as opposed to straight forward. To keep the systems operating properly, the plane required a crew that was more than twice as large as the original autopilot-less planes. This caused problems because it took far more work to keep the autopilot working than it did to actually fly the plane without such a device. For this reason, the autopilot was not used very often for many years. It wasn’t until the 1960’s that the device really took off (to use an atrocious pun). At this point, more sophisticated systems were being introduced and a new form of flight control was in the works: a computer controlled flight. This technique called “fly by wire” didn’t become a standard until closer to the 80s but it had its beginnings in the late 60s. The basic concept of the “fly by wire” systems is digital control systems. To understand this form of autopilot a brief description of digital control systems must be discussed. 3. Digital Control Systems In a typical control system, the equipment to measure and control signals can be extremely complicated and can require enormous levels of sophistication. However, if it were possible to use a computer to control the systems, a simple “off the counter” processor could be able to handle complex control systems. For example, a system that requires complicated mathematical algorithms for control could simply digitize the signal and send it to a processor to take care of the calculations. The processor (usually called a “minicomputer”) is usually inexpensive and relatively easy to implement. The control system using these sorts of devices is shown below in Figure 1: Digital Control System Block Diagram. Figure 1: Digital Control System Block Diagram In this system, the digital computer reads in the digital signals from the feedback loop and the input and it sends them to the D/A converter. This converter takes the digital values from the computer and converts the signals into a usable value for the actuator and process. The signals are then sent via the feedback loop through the measurement sensors until it reaches the analog to digital convert. This portion takes the measured value and converts it into a binary digital signal that can be read by the computer. This digitizing is where the real challenge enters into the picture. Since computers cannot read the same kind of signals that an analog device can, some method of transforming the analog data to readable digital data must be introduced. This is done through a method called sampling. A simple circuit is set up with a switch (shown below in Figure 2: Switch Digitizer). The switch samples the value of the signal at regular intervals. like function. An example of this is shown below in Figure 3: Discrete Signal. Figure 3: Discrete Signal As can be seen, the resulting function is no longer a smooth signal but it is a series of step functions that represent the original signal. To deal with these functions, the zTransform must be used. The z-Transform is an extension of the LaPlace transform and is of the following form: Z = esT where s is the value from the LaPlace transform and T is the period of the sampler (how many seconds between samples). In the z-Plane, there are several methods of stability analysis. For example, a sampled system is stable if all of the poles of the closed loop transfer function T(z) lie within the unit circuit of the zplane. These systems can readily be constructed in MATLAB. The function can be defined as it would in any transfer function however the sampling time is also included (so the definition is as follows: sys=tf(num,den,Ts) where Ts is the sampling time). The systems can be converted from continuous to digital and back again using the c2d and d2c commands. Below, in Figure 4: Discrete Step Response the following MATLAB code was entered: >> num=[1];den=[1 1 0]; >> sysc=tf(num,den); >> sysd=c2d(sysc,1,'zoh'); >> sys=feedback(sysd,[1]); >> T=[0:1:20];step(sys,T) This code yields the following graph: Figure 2: Switch Digitizer These samples must be at a high enough frequency to accurately represent the input signal. This process takes the signal and breaks it up into a step Figure 4: Discrete Step Response As can be seen by the above plot, the digital signal is no longer a smooth curve. Depending on the sampling rate (in this case, once every second) the value is measured and then held constant until the next value is measured. For a digital control system, this value is converted into a binary number that is then fed into the minicomputer for analysis. This sort of system is extremely beneficial to the aviation industry. With a digital control system, the autopilots can have a much more precise control of the flight path of the aircraft. Rather than feed the controlled signal (such as altitude or yaw) through a complex process (which would be extremely difficult to design), the signal can be converted to a digital signal and sent to the minicomputer. In the minicomputer, the signal can be operated on with a higher level of ease because the computer can directly run the algorithms on the digital numbers. This method can yield a higher quality control system for the aircraft. Although it does require more circuitry (the minicomputer) it is usually smaller and easier to implement backup systems. With this general understanding of digital control systems, the Beaver Autopilot System is the focus of the rest of this paper. This digital control system overview will help understand some of the mystery of the proprietary (and therefore undisclosed) portions of the Beaver system. guidance and control. The guidance function of an auto-pilot determines the speed and the course to be followed by the craft. This is done by measuring the current actual values and comparing them to some reference system. The control function is the function that takes the data from the guidance system and applies the proper corrections. For example, if the guidance says that the altitude of the aircraft is 200m too high, the control function would move the wing flaps to bring the craft back down to the appropriate level. The control does not usually contain any sort of measuring devices as this function is delegated entirely to the guidance function. The guidance loop acts as a commander to the control loop and the control loop commands the physical movement and response of the aircraft. As is evident, it is desirable for these control systems to have a fast and stable response. They must also be able to withstand any disturbances from the surrounding environment. This is extremely important because, if a disturbance caused a critical control system to become unstable, the autopilot would cease to function and possibly put the lives of the passengers at risk. The best way to describe the two functions is to think of them as two interrelated loops. The control loop is the inner loop as it is controlled by the outer loop guidance system. This is best shown in Figure 5: Basic Block Diagram of Autopilot. 4. Beaver Autopilot System The Beaver autopilot document that is used for this report entitled “A Simulink Toolbox for Flight Dynamics and Control Analysis” and is written by Marc Rauw. This document describes the Beaver system in great detail and discusses its implementation in the Flight Dynamic Control toolbox for SIMULINK. However, since this toolbox was not written by the SIMULINK people, the standard SIMULINK package does not contain this toolbox. Therefore, for any attempted simulations, regular SIMULINK will be utilized and any unknown functions will be estimated as best as is capable by this author. 4.1 Functions The functions of an autopilot system can be broken down into two major categories: Figure 1: Basic Block Diagram of Beaver Autopilot These two controllers (guidance and control) control two major areas of the flight path of the aircraft: Longitudinal and Latitudinal direction. The Longitudinal Mode is discussed below. The Latitudinal mode is of a similar format with slightly different constants and different control blocks. Figure 6: Block Diagram of Longitudinal Mode 4.1.1 Longitudinal Mode Overview This portion of the auto-pilot controls the pitch angle and the altitude of the aircraft. The complete block diagram for this portion of the autopilot is shown above in Figure 6: Block Diagram of Longitudinal Mode. As can be seen by the above diagram, there are several components that make up this portion. The main three blocks of the system are the controllers of this function. They take the values from the outer loops which pass through constants and integrators and output the appropriate controls for the aircraft. The input signal Href is the current altitude as measured from the guidance systems. This value then is taken with the control’s new altitude as well as the pitch angle (θ) and fed into the control blocks of the diagram. It should be noted that the gains shown in the feedback loops of the system are all variable and depend on the velocity of the aircraft. This happens because the control to the aircraft will change as the speed does. Wind resistance and other factors contribute to this. The final portion of the longitudinal mode of the aircraft is the Approach mode. It should be noted that the Glideslope device is a unique device to the Beaver autopilot mode. This portion of the control system is a feedback loop from the Hdot output (from Figure 6) to the input Hdotref signal. The glideslope receiver is an on-board measurement device that interacts with a transmitter on the airport runway. This system is an extra feedback loop that has more control over the descent of the aircraft. To properly operate, the distance to the runway is calculated using Distance Measurement Equipment (DME). However, this equipment doesn’t often work well with autopilot modules because of hardware limitations. Therefore, a different approach must be used. The three dimensional distance to the runway is calculated using the following equation. In this equation, R is the three dimensional distance to the runway, Href is the height above the runway, and γgs is the reference flight path angle or, the angle the plane makes when flying along the nominal path. Generally, a radio altimeter is used to determine the value of Href. During a glideslope approach, there are two different modes of operation. The first is the “glideslope armed mode. As the authors of the Beaver document say: “This phase is engaged as the approach mode is selected by the pilot. The longitudinal autopilot mode in which the aircraft flew before selecting the approach mode, usually ALH, will be maintained until the aircraft has reached the glideslope reference plane” (Rauw, 177). This mode simply tells the aircraft that it is going to be landing soon and that it needs to get ready for landing. It does not affect any of the current flight paths. The second mode is the glideslope coupled mode. The author describes this mode by saying that “This phase is initiated as soon as the aircraft passes the glideslope reference plane for the first time. In this phase the control laws of the GS mode take over the longitudinal guidance task of the autopilot” (Rauw, 177). In this mode, the GS actually takes over the rest of the autopilot. It does this by adding its own signals with the Hdotref input signal. When the GS is not engaged, the signal that is the output of the GS is zero allowing the aircraft to operate as it normally would. The timing of the coupling of the GS is extremely important. If the GS is coupled too earlier, the aircraft will follow the path as shown below in Figure 7: Result of Early GS Coupling block. This block is assumed to be a simple signal delay and is set to delay the signal by one second. The next block is the Actuator and Cable Dynamics block. This block, for the sake of the simulation is assumed to be a simple transfer function (a second order is used). At the heart of the block diagram is the Beaver Dynamics. This block is the portion of the control system that reads in the measured values and operates on them. For the sake of simulation, the block was assumed to be three transfer functions: one of which is a simple constant value, the second is an integrator function, and the third is a derivative function. After the system was constructed, the outputs were examined. The θ plot is shown below in Figure 8: θ Simulink Plot. Figure 7: Result of Early GS Coupling As can be seen by the above figure, the timing is important because, if the GS is coupled too early, the aircraft tries to approach the reference line (the slanted dotted line in the figure) before it is supposed to. The result of this is that the plane will rise to the reference line and then have to suddenly shift down after the line is crossed. To properly couple the GS, the mode controller is constantly examining the state of the aircraft. When the aircraft reaches the correct point, the mode controller immediately switches to coupled mode and the aircraft can land. This system is not perfect and it does yield a slight overshoot but the resulting overshoot is far more desirable than the overshoot shown in Figure 7. 4.1.2 Longitudinal Mode Simulation To get a better picture of how exactly this autopilot system works, it became desirable to simulate the system in MATLAB. All of the K values (as seen in Figure 6) are given in the Beaver document and can be constructed in Simulink. However, there are still a few different blocks that are not explained in the Beaver document. This is due to the fact that the autopilot system is proprietary information (or if it isn’t, the authors of the document don’t disclose the information). Because of this, the following assumptions are made. The first is in the Computational Delay Figure 8: θ Simulink Plot As can be seen, this is invalid data. The plot shows nothing and every one of the outputs looks about the same. Changes made in the functions of the control system yielded no results either. The system would either be zero and then drop off to negative infinity or the signal would be zero and rise to positive infinity. The system did not compensate for anything. The reason that these simulations did not yield valid results is because of the fact that the setup of the Beaver Dynamics is unavailable for research. The actual system for the Beaver Dynamics I probably not actually a simple transfer function. If it was a simple transfer function, changes to the simulated system would have had a larger effect. The Beaver Dynamics is probably a control system in and of itself. Since the system is a moderately complex system, it can probably be assumed that the Beaver Dynamics is a digital control system. There is probably a minicomputer (as discussed earlier) at the heart of the system that does all of the actual control for the aircraft. For this reason, it cannot be adequately simulated for this paper. For that to happen, more information about the Beaver Dynamic would have to be available for study. To make up for this lack of simulation, a simple aircraft flight control system is examined in the final portion of this paper. 4.2 Beaver Conclusion The Beaver Autopilot system is a complex system that cannot be readily simulated due to its complexity and a lack of information concerning its primary controllers (the Beaver Dynamics to be precise). However, that aside, it is still a fascinating system. The document by Marc Rauw gives a solid and detailed description of the system and makes it simple for people with basic experience in control systems to understand the peripherals. While it does not go into detail on the heart of the system, it at least gives a general overview of how the system works. The Beaver autopilot system is a unique system that does not follow all of the same rules that its brethren follow. For example, the GS as discussed is a unique feature to the Beaver Autopilot System in that it is not how the landing mode is implemented. Most other systems use the distance measurement equipment (DME). It should be noted that the description given in this paper only gives a brief overview of the system. Many of the details of the system (as well as the lateral and turn compensation modes) were not discussed. To better understand those modes, or to understand the system in greater detail, see the paper by Marc Rauw. 5. MATLAB Simulation 5.1 Purpose The purpose of the following simulation is the better understand a particular example of a flight control system. Since the Beaver Autopilot could not be simulated in this paper, a simpler example is necessary. The following sections give a description of a particular simple system. They then run through some simulations to examine how the system responds. 5.2 The Problem This example comes from page 624 of Modern Control Systems by Richard Dorf and Robert Bishop. The control system of the problem is that of a bi-wing aircraft. The control system is shown below in Figure 9: Bi-Wing Flight Controller Figure 9: Bi-Wing Flight Controller As can be seen in Figure 9, the aircraft consists of an engine, a controller and Aircraft Dynamics. There is also a little disturbance in the system caused by wind. The goal of this exploration is to determine the best way to control a step input signal (simulating some variable of flight that needs to be rectified). Because this control system is controlling the flight of an aircraft, the percent overshoot and steady-state errors must be as small as possible (less than 5% for steady state error). 5.3 Analysis For this example, the Aircraft Dynamics and the Engine blocks are considered unchangeable and that it is the controller than must be modified to meet the appropriate requirements. The simplest approach is to simply use a constant gain function to control the system. The gain largely has control over the stability of the system. For certain gains, the system is stable, and for others, it is not. After setting the system up in SIMULINK, the value of the gain (K) was varied and the results examined. It was found that values of K greater than approximately 1.5 caused the system to become unstable. However, gains of les than 1.5 caused stability but the steady state error was extremely high. As is shown in Figure 10: Steady State Error of Pure Gain System, the steady state error for the system ranges from 20% for gains closer to 1.0 and up to 50% for smaller gains. As can be seen, this response is already better than any of the constant gain system responses. The steady state error is eliminated and the percent overshoot is less (although it is still about 30%). However, as can be seen, the response of the system is relatively slow. This would not be acceptable for an aircraft as it is imperative that the system respond quickly to changes in its state. To speed up the response time, the proportional portion of the PID controller is increased until the desired response time is met. If the response time is defined to be the time needed to rise from 10% to 90%, a good response time would be 0.5 seconds. This is achieved when the proportional component is set to 5. It yields the result shown below in Figure 12: PID Controlled System (P=5, I=1, D=1). Figure 10: Steady State Error of Pure Gain System As can be seen by the above plots, the system that has the faster and smoother response also has a completely unacceptable steady state error. The system that has a slower response and a large amount of overshoot has a better stead state error but it is still around 15% which is unacceptable. If a better result is to be attained, a different approach becomes necessary. To eliminate this error and overshoot, implementing some other kind of controller will be beneficial. The usual approach is to implement a PID controller. The components in this kind of controller can take sensitive control systems and give them a more stable, smoother, faster, and less oscillatory response. A PID controller was added and the system was tested in SIMULINK. The first step in setting up a PID controller is to set all of the values to 1. The results of this test are shown below in Figure 11: PID Controlled System (P=1, I=1, D=1). Figure 11: PID Controlled System (P=1, I=1, D=1) Figure 12: PID Controlled System (P=5, I=1, D=1) As can be seen, this is a fairly good response. The steady state error is very close to zero, the percent overshoot is 2% and the response time is less than 0.5 seconds. However, this system is not perfect. Although it may not be initially evident by the above plot, there is oscillation in the signal and the oscillations last for more than 10 seconds. While some systems may not have serious problems with these oscillations, flight control systems may have complications. For example, if this system was the altitude controller, the aircraft may see these oscillations as vibrations of the aircraft. These vibrations would add to the vibrations that are already present and could increase the natural wear and tear of the aircraft devices. Also, if the vibrations occurred at the exact right (or wrong) frequency, some systems that use this signal could become unstable. Therefore, the system must be modified again to accommodate for these oscillations. The best results that could be attained in the time span allowed for the experiment ended up not yielding a much better response. The oscillations were eliminated but the percent overshoot is substantially higher as is shown below in Figure 13: PID Controlled System (P=1, I=3, D=0.2) example of a flight control system was explored to see the effects of various controllers on the system. Autopilot systems and, more broadly, flight control systems are complex systems that generally are sensitive systems. However, in order for flight to be smooth, the systems must be robust and able to handle disturbances. If a disturbance could potentially cause instability, the flight would become dangerous. However, to combat these unexpected instabilities, most flight systems are 2 and 3 times redundant: they have several backup systems that can step in and take over in the event of a failure. These emergency systems can mean the difference between flight and a crash in the event of a sudden disturbance. However, it is still up the control systems on the aircraft to insure that the plane will make it from point A to point B and back again. Figure 13: PID Controlled System (P=1, I=3, D=0.2) As can be seen by the above plot, if the system can handle a 25% overshoot, this response is better because of the fact that the oscillations are not present here. However, if the system can handle oscillations, the first PID controller is much more desirable because of its high response time, low overshoot, and low steady state error. Bibliography Rauw, Marc. A Simulink Toolbox for Flight Dynamics and Control Analysis. Published by Marc Rauw, 1994 – 2001 Richard Dorf and Robert Bishop. Modern Control Systems. New Jersey: Prentice Hall, 2001 5.4 Summary The system presented in the previous example, shows how an aircraft responds to changes in the system. For all of the explorations, the system was given a step input which could be equated to a change in altitude ordered by the pilot. The pilot would change the altitude and the system would have to respond quickly and smoothly. Because the system presented in the problem is inherently sensitive to changes, a PID controller is necessary to aid the system in responding effectively. It was found that a larger proportional controller as well as a unit integrator and derivator were necessary to keep the system stable and responding quickly. However, as were discovered, there were tiny oscillations present that could potentially cause problems for the aircraft. 6. Conclusion Through this paper, a brief overview of Digital Control systems was given to show how they relate to autopilot systems, the Beaver Autopilot system was analyzed and studied, and an Flight Dynamic and Control Toolbox page. < http://home.wanadoo.nl/dutchroll/> Hughes, Arthur. History of Air Navigation. Great Britain: Unwin Bothers Limited Working, 1946 Shevell, Richard. Fundamentals of Flight. NJ: Prentice Hall, 1983