Baseline UAV Controller Initial Thesis Report

Size: px

Start display at page:

Download "Baseline UAV Controller Initial Thesis Report"

Juniper Pitts
5 years ago
Views:

1 Baseline UAV Controller Initial Thesis Report Callum J. Newton 1 The development of a baseline Unmanned Aerial Vehicle (UAV) controller for use on the Australian Defence Force Academy s Telemaster UAV will provide redundancy to the flight control system and assist in the certification of the UAV. Via the application of control methods which are common in industrial applications, this baseline controller will be designed using gain scheduling of Proportional, Integral, Derivative (PID) controllers. Key Work Index Gain Scheduling, Unmanned Aerial Vehicle (UAV), Telemaster, nonlinear systems, flight control, Aerosim, PID controller. k p k i k d! : Proportional gain : Integral gain : Derivative gain Nomenclature : Angular displacement about Y axis, measured clockwise from the horixon I. History of the UNSW UAV Programme The development of autonomous flight for commercially available remote controlled aircraft is an emerging niche capability with interesting commercial applications. Large aircraft developers, such as Boeing or Northrop Grumman, have largely ignored this cheap alternative as they focus on sophisticated military Unmanned Aerial Vehicles (UAV), allowing the domain to be investigated and dominated by research conducted at universities around the globe. The University of New South Wales (UNSW) Australian Defence Force Academy (ADFA) Aeronautical Engineering department is currently in the process of developing its UAV expertise, with programmes in both fixed and rotary wing flight. The fixed wing programme began in 1999 as a practical way for students to test and implement different stability augmentation systems and control strategies for model aircraft [1]. It has since evolved into a project seeking to develop a cheap UAV that would be available to organisations - such as the Australian Customs Service - on the commercial market. The programme currently employs a team of PhD students under the guidance of Dr A.G. Sreenatha and other lecturers including control theory, electrical and aerodynamic specialists to accelerate the development of a working prototype. While the UNSW@ADFA UAV programme is not as mature as competing universities, such as Cornell Universities operationally certified UAV or University of Sydney s self manufactured UAV aircraft, the work conducted has nontheless contributed to the global push forward for economical UAV production. Fully autonomous flight has not been achieved by ADFA as yet. The school currently operates ten platforms for testing and developing various ideas in the rotary and fixed wing fields. This thesis project will solely be focused on the fixed wing variant, Telemaster (figures 11 & 12, appendix B). Development of an automated Telemaster aircraft has thus far included augmentation of sensors and devices required to provide data such as angle of attack, aircraft dynamic conditions and GPS to track the aircraft. The primary controller has been developed using modern control theory, including fuzzy logic and neural network controllers. 1 Bachelor of Engineering (Aeronautical) ; ZACM

2 Although good progress has been made to date, no successful development of a baseline controller has been achieved to check the validity of the results gained via the modern control methods or act as a redundancy, ensuring that automated flight could be maintained should the primary control system fail. II. Generic UAV Flight Control Processes Most of these attempts at autonomous flight focus on the interaction between three broad components, navigation and guidance, controller and the interface with the actuators to combine and achieve successful UAV flight. The schematic identifying the relationship between these components and external influences, such as weather conditions and overriding commands from a ground based pilot, is shown in figure 1. The selector linking the primary controller and the radio pilot input ensures that any input from the human pilot overrides the flight inputs from the automated controller. Figure 1: Typical UAV Flight Control Flowchart With some exceptions, alteration of the physical aircraft itself is not generally conducted, as the aircraft is usually available in kit form. This thesis focuses on development and enhancement of the mathematics governing the flight controller. III. Extension of the UNSW UAV Programme This project will enhance the work done on the school UAV by contributing to the development of the automated flight system. Therefore, the aim of this thesis is to develop a baseline flight control system for the UNSW Telemster Unmanned Aerial Vehicle (UAV) via the application of classical control theory. This work is important, as the development of a baseline flight control system will enhance the UAV by; Providing redundancy should the primary control system fail, Acting as a check to verify and compare the results with the modern control theory methods, Assisting in the certification process for the final UAV package. As the Telemaster weight remains under 150kg, it is classed as a small UAS under Civil Aviation Safety Authority guidelines [2]. Thus although Telemaster requires no airworthiness certificate to operate, the development of a baseline controller would be beneficial to ADFA as the enhanced academic knowledge could be utilised should development of a UAV of this larger weight be pursued. Aside from the official certification requirements, the redundancy provided by a baseline controller would enhance the failsafe ability of the Telemaster and potentially avoid costing ADFA and future commercial operators another crashed UAV 2. 2 ADFA suffered a destroyed Mega Soar UAV platform in early Although this loss was attributed to pilot radio range issues and not a failure in the modern control flight control system, it illustrates the money and inconvenience a destroyed UAV can cause for an organisation. 2

3 The baseline controller will operate in parallel to the primary controller as indicated in figure 2. An additional selector operating between the primary and baseline controller is included to monitor the quality of the outputted signals and judge if the primary controller is outputting inappropriate values for the UAV to respond to. Development of the navigation and guidance system is undertaken by a different team and for the purposes of this thesis, simply provides the input data for the control programmes to use. The baseline controller generated in this thesis will operate the longitudinal motion of the aircraft. Time permitting, the baseline controller will also be determined for the more complex lateral motion of the aircraft. Figure 2: Proposed UAV Flight Control Flowchart IV. Baseline Controller Options 1. Transfer Function An appropriate method to model the aircraft (system) using classical control theory must be established to enable production of the baseline controller. The method of selection is complicated by the coupled, nonlinerarity of the aircraft dynamics, which prevents the easy determination of a transfer function and plotting of response via a method such as root locus analysis. The only way the transfer function could be determined for such a complex system would be to ascertain the aerodynamic and control derivatives of the aircraft via wind tunnel testing. Unfortunately, the wind tunnel facilities of ADFA are too small to hold our aircraft. The time required to determine these coefficients would be large and the effective range of results small to be of any real use for our aircraft. Therefore, generating the baseline controller using standard transfer function methods will not work. 2. Extended Linearisation Determination of the feedback control of nonlinear systems is possible by extended linearisation. Several factors limit its use however including the single input requirement [3] which is useless for our aircraft with multiple input sources. Indeed, the quantitative estimates of the improvement in accuracy afforded by extended linearisation are likely to be so conservative and restrictive as to be uninteresting from an engineering viewpoint [3]. and thus this method can be discounted for our purposes. 3. Dynamic Inversion Dynamic inversion is a possible alternative to gain scheduling. It operates by using control inputs to directly cancel unwanted terms in the nonlinear state equations and replaces them with desirable dynamics however, this robustness is sometimes obtained by using high loop gains, which may prove detrimental in the presence of unmodeled dynamics [4]. 3

4 4. PID Controllers with Gain Scheduling Given the difficulty of the first two options, it is appropriate to recall that similar issues with nonlinear systems are not new or even unique to aviation as, they were first encountered during the commercial construction of automated chemical plants during the 1930 s [5]. The discoveries in this industry lead to development of one of the first closed loop systems applied to aircraft, the C-1 autopilot installed on the B-17E bomber during the Second World War[6]. In both of these cases, the solution was to implement Proportional, Integral, Derivative (PID) feedback controllers which operated between different gains depending on the input conditions. Figure 3 shows typical system responses utilising a combination of the available PID controllers. The proportional controller (figure 3.a) has satisfactory settling time although suffers from steady state error (figure 3.a). Introducing an integral controller removes the steady state error however, the settling time greatly increases (figure 3.b). Implementation of a controller with full proportional, integral and derivative components achieves optimum system response in both settling time and steady state error (figure 3.c). The peak overshoot remains comparable to the previous two controllers and thus selection of a full PID controller is the most effective for system control. Figure 3: System responses to feedback controllers featuring; a) Proportional Controller, b) Proportional & Integral Controller, c) Proportional, Integral & Derivative Controller The PID controller operates as shown in figure 4, where the system error forms the data to be linearised and concurrently has the difference, integral and derivative determined. Three separate gains exist for the proportional, k p k i k d integral and derivative gains respectively,, to boost or alter the strength of the errors. These calculations are then combined and sent back to the system as an input to generate the next output response. The proportional line of the controller is responsible for the vast error correction while the integral ensures steady state error is avoided. Together, the PI response occurs at the low-frequency region and the PD response occurs at the high frequency region. PID control is appropriate for our system as it ensures that both transient and steady-state performance is accounted for [7]. With the PID controller in place, selection of the k p, k i and k d gains has traditionally been achieved by extensive, time consuming simulations [8]. Figure 4: PID Controller Interaction with the System This gain selection method essentially acknowledges that the dynamics of the system are too complicated to determine and simply treats the plant as a black box that provides the outputs for a given input. It was soon realised that by scheduling the gains to change upon the input, a control engineer could tweak the system until the desired output was achieved. This method became known as Gain Scheduling and its application has become a common engineering practice, used to control nonlinear plants such as in flight [9] and process control [10]. 4

5 Despite its popularity, gain scheduling remains an ad hoc methodology. For example, the robustness, performance, or even nominal stability properties of a global gain scheduled controller are not addressed explicitly in the design process. Rather, such properties are inferred from extensive simulations [11]. Thus despite the continued growth of gain scheduling, it remains something of an art [8]. Two prominent gain scheduling guidelines have emerged with its application in industry, the scheduling variable should capture the plant s nonlinearities and the scheduling variable should vary slowly.[11] As the plant nonlinearities are their most noticeable at the extremes of the flight envelope, such as high angle of attack or extreme speeds, these are the conditions to be weary of when checking the data. As our UAV is expected to operate at low altitude, low speed and simple flight responses than these factors result in a reduction in the extreme point(s) severity and likewise satisfies the ability to select a variable which will vary slowly. As the choice of the scheduling variable is critical, previous research has identified that the best closed-loop performance is obtained by using pitch angle as the scheduling variable, and not forward velocity [12]. For this reason, the final product will likewise aim to schedule according to pitch angle or dynamic pressure (for reasons discussed later). V. Initial Development of Gain Scheduled Technique To best learn the principles and justify the gain scheduling technique, initial development did not immediately focus on the raw data provided by the Telemaster UAV. It was decided that utilising the software Aerosim, 3 to generate simulated data and practise the processes of gain scheduling, would be an ideal first step to acquire an understanding of the gain scheduling process. The data from Aerosim was outputted to Matlab for plotting and interpretation. The aircraft feedback loop was created to track a 25m/s aircraft velocity, which remained the condition of interest for this gain scheduling investigation. The aircraft feedback loop is included in figure 10 in appendix A. Figure 5: Initial k p investigation Deflections are listed as degrees Figure 5 shows the resulting simulated motion of the aircraft after 30 seconds for five different values of k p. This initial plot is used to check that the elevator has not passed beyond saturation levels. In this case, k p = Aerosim is a computer package designed to simulate the mechanics of flight of several aircraft types including Cessna 172, Cessna 182, Cessna 310 and the U Dynamics aerosonde UAV (see figure nine in the appendices). It operates as a simulink interface (see figure ten in the appendices) based in Matlab and was developed by Unmanned Dynamics. It is available free of charge from 5

6 would clearly be discounted as the elevator deflection passes beyond the Telemaster limits of angle response for this value is likewise extreme for our model and adds reason to discount k p = The pitch Figure six reveals the forward velocity relationship to pitch angle and thus directly relates to figure five. The effectiveness of the range of k p values selected is still visible and k p =-0.5 remains invalid due to the large resulting pitch angle via its implementation. Figure 7 illustrates the intricate final response position of each of the k p values. As the system was expected to return to 25m/s then the acceptable aircraft response. k p ±40! =-0.2 gave the best broad reaction while maintaining Figure 6: Initial k Figure 7: Close Up View of Figure 6 p investigation Deflections are listed in degrees With the k p broad response determined, the simulation was run again to narrow down the responses about k p = Figure 8 shows the response of these new selections. It is interesting to note the difference in severity of response between k p =-0.5 (figure 6) and k p =-0.4. This small change creates a difference in pitch response of approximately 45 degrees, showing the importance and effect of correctly scheduling the gains. In this case, k p = -0.2 was still selected as its response combined a small aircraft response with a good settling time about 25m/s. Figure 8: Fine k p investigation This process was repeated to determine the and k d values and then repeated with all determined k values to check that the final solution remained valid. Table one summarises the results using this method. k i 6

7 Table 1: K values, generated for one Aerosim flight condition (forward velocity = 25m/s) k p = -0.2 k i = k d = -0.7 Thus the Aerosim package, coupled with the Matlab programme, enabled one k to be investigated at a time while the other two were held constant. The calculations then became an iterative process with continual simulations narrowing down the margin of error until the final k values were determined. It should be noted that this simulation was conducted for one flight condition and assuming no wind was present. VI. Future Direction While good progress has been made until this point, many aspects of the design must be addressed before the baseline controller is ready for implementation in the UAV. A proposed timeline of thesis events is included in Appendix C. 5. Full Gain Scheduling of Aerosonde model An extension of the method described in section V. (above) will be used to develop a full schedule of gains for the aerosonde model. 6. Comparison to the Modern control theory outputs By using the gains obtained by the iterative process discussed, validation of results generated from the modern control methods will be conducted. 7. Weather effects Simulation of the random effects of weather must be included in the model to verify the aircraft response during sudden weather impacts. It is anticipated that this will be possible by utilising inbuilt random generators in Matlab coupled with weather models supplied from Aerosim. 8. Separation of the raw UAV data It will be necessary to develop a computer code to sort the raw UAV input and response data into useful information, ready for analysis. 9. Dynamic Pressure Investigation of scheduling the gains according to dynamic pressure will be conducted. It is anticipated that this will be a satisfactory scheduling variable as it accounts for aircraft speed and altitude. 10. Full Gain Scheduled Baseline Controller This stage will be allocated the greatest amount of time as it forms the solution to the aim of this thesis. Far more extensive testing will be required than that demonstrated for the simulation case. Again, these results will be compared to those reactions generated by modern control theory to check for accuracy. 7

8 11. Testing with the UAV model & Development of Lateral model When the full gain scheduled baseline controller has satisfied the thorough checks, testing on the UAV model will commence. Should this stage be reached, it will require a testing schedule and risk analysis to be developed (closer to the date). Time permitting, a similar process will be used to develop appropriate gains for the UAV s lateral motion. VII. Summary This thesis seeks to develop a baseline UAV controller for use in the UNSW@ADFA Telemaster aircraft. Although a principal control system utilising modern control techniques exists, this baseline controller will enhance the UAV by; Providing redundancy should the primary control system fail, Acting as a check to verify and compare the results with the modern control theory methods, Assisting in the certification process for the final UAV package. Although various methods exist to develop the baseline controller, gain scheduling of Proportional, Integral, Derivative (PID) controllers was selected as it is simple, has proven successful industrial applications and is well suited to our low speed, low altitude UAV. Initial practice and investigation of this PID scheduled controller was conducted using simulation software Aerosim and values for the PID gains were generated. Although a simple exercise, the experience gained with this initial investigation was essential, as it will form the basis for continued development of the baseline controller. Work in the immediate future will continue to develop gain scheduling experience by developing a full series of gains, adding weather effects to the simulation and comparing the results generated with those calculated by modern control theory. Work will then focus on analysing the raw UAV data and the possibility of scheduling against dynamic pressure to develop a full range of gains for the longitudinal motion of the Telemaster aircraft. Time permitting, the gain scheduled controller will be tested in flight on the Telemaster aircraft and the lateral gains will be developed. References [1] Lt L.K.Hong, Measurement and Analysis of Angle of Attack on a Model Aircraft, UNSW, Canberra 1999 [2] CASR Part 101 UAS Operating Rules Project CS 05/01 [3] W.T.Baumann & W.J. Rugh, Feedback Control of Nonlinear Systems by Extended Linearisation, IEEE Transactions on Automatic Control, Vol. AC-31, NO. 1, January 1986 [4] S.A.Snell, P.W.Stout, Quantitative Feedback Theory with a Scheduled Gain for Full Envelope Longitudinal Control, Jounral of Guidance, Control and Dynamics, vol. 19, No. 5 September October 1996 [5] J.W. Broadhurst, T.C. Brodrick, A.W. Foster, G.E. Wheeldon, Automatic Control in the Chemical Industry, Journal of the Institution of Electrical Engineers, Vol 94, Part IIA 1947 [6] T.R.Yechout, S.L.Morris, D.E.Bossert, W.F.Hallgren, Introduction to Aircraft Flight Mechanics, AIAA Education Series, Virginia, 2003 Pg. 247 [7] K. Ogata, Modern Control Engineering, Fourth Edition, Prentice Hall, USA, 2002 pg. 725 [8] W.J. Rugh, Analytical Framework for Gain Scheduling, IEEE, January 1991 [9] G. Stein, Aadaptive flight control a pragmatic view, Applications of Adaptive Control, K.S. Narendra and R.V. Monopoli, Ids. New York: Academic 1980 [10] M.J. Whatley and D.C.Pott, Adaptive gain improves reactor control, Hydrocarbon Processing, pp , May 1984 [11] J.S. Shamma & M. Athans, Gain Scheduling: Potential Hazards and Possible Remedies IEEE, June 1992 [12] J.Wang & N. Sundararajan, A Nonlinear Flight Controller Design for Aircraft, Control Engineering Practice, Vol 3, No. 6, 1994 [13] Aerosim User s Guide Version 1.2, Unmanned Dynamics. Available online from 8

A New Perspective to Altitude Acquire-and- Hold for Fixed Wing UAVs

Student Research Paper Conference Vol-1, No-1, Aug 2014 A New Perspective to Altitude Acquire-and- Hold for Fixed Wing UAVs Mansoor Ahsan Avionics Department, CAE NUST Risalpur, Pakistan mahsan@cae.nust.edu.pk