PC-Based Human-In-the-Loop Simulation for Flight

Transcription

1 Applied Mechanics and Materials Vols. 0-2 (2008) pp Online available since 2007/Dec/06 at (2008) Trans Tech Publications, Switzerland doi:0.4028/ PC-Based Human-In-the-Loop Simulation for Flight L. Zhang, a, H.Z. Jiang, b and H.R. Li, c School of Mechatronics Engineering, Harbin Institute of Technology, Harbin, China a dr.zhanglei@26.com, b jianghz@hit.edu.cn, c lihr@hit.edu.cn Keywords: Flight simulator, COTS, Virtual prototype, Coordination. Abstract. The development of a flight simulator is a challenging work because of its complexity and tremendous cost. We implement a prototype composed of PC cluster and have proved its coordination character as a flight training device. This paper describes the architecture of the flight simulator, the software development tools and hardware platform. These software and hardware constitute a PC based simulation environment and make the expense of the simulation application affordable. We also present the simulation modeling process. For the integrity of the cueing system, we designed a virtual prototype of motion system and connected it to the flight simulator. The integrated system gave us a chance to testify the coordination of the simulator. The verification method and result are presented to show the feasibility of the design based on PC. Introduction With the rapid development of Chinese civil aviation, Chinese airline companies purchase and order a great lot of airplanes in various types to satisfy the increasing necessity of air transports. New airplanes need new pilots, and new pilots need new flight simulators for training. So requirements of plenty of flight simulators are in sight. A flight simulator is the outcome of System Engineering and is the typical human-in-the-loop simulation device. It involves many domains and integrates tremendous knowledge. These knowledge amalgamates into tens of thousands of lines of code for the simplest training simulation, even over a million lines of code for the complex, level D full flight simulator. The code runs on special and costly hardware equipped in the simulator. These all lead to a level D full flight simulator cost about twelve million US dollars. As Commercial-off-the-shelf (COTS) solutions have replaced custom development, PC systems have now supplanted proprietary designs, workstations and even mainframes for many simulations and control applications. With the advent of PC clusters, embedded PCs, high-end graphics cards and other technologies, the range of simulation applications that can benefit from the low cost of COTS hardware has grown []. Today, true COTS hardware such as PC, Ethernet and IEEE394 (Fire wire) allow large savings in cost and engineering effort. Similarly, commercial simulation software such as MATLAB/Simulink and MATRIXx/SystemBuild are now widely used in engineering simulation. They have become the modeling tools of choice because their graphic interfaces mimic the function-block methodologies taught in universities and technical schools. This reduces the learning time and increases engineering productivity. Another useful feature of these packages is automatic code generation that frees engineers from the tedious and error-prone task of writing code. They make it possible to go from concept to simulation without ever having to write code. Considering the advantage of developing simulation application based on COTS solutions, we built a prototype of Boeing flight simulator. The model and executable code in the simulator were generated by commercial software and run on a PC cluster. This system afforded a platform for key technology research of flight simulator [2]. This paper describes our work and shows the character of our flight simulator. In section two, a generic functional model of a flight simulator is presented. Section three introduces the software and hardware used for our project. In section four, we choose a main route of data flow in flight simulator and describe the methods of simulation model building along the route. This is presented as an example of rapid prototype All rights reserved. No part of contents of this paper may be reproduced or transmitted in any form or by any means without the written permission of TTP, (ID: , Pennsylvania State University, University Park, United States of America-04/06/4,:06:30)

2 Applied Mechanics and Materials Vols modeling methods. In section five, we design a method to testify the coordination of our flight simulator. Finally we conclude our work and lay a course for next study. Generic Structure A flight simulator normally contains seven sub-systems: simulation models, cockpit instruments, visual cueing system, motion cueing system, audio cueing system, instructor station and cockpit controls & force feedback system. The relationship of these components is shown in Fig.. Fig. A generic functional model of a flight simulator. Depending on these facilities, flight simulator provides pilots visual cue, audio cue, force feedback cue and motion cue. These sensory cues must all be simulated to a high degree of fidelity for training purpose and with a high degree of coordination in order to avoid simulator sickness [3]. The instructor is in charge of the pedagogical aspects of flight simulation. It is the person who decides what mission will be run, and under what conditions. Software and Hardware Based on the COTS solutions, we choose MATLAB/Simulink to build simulation models, VAPS for drawing cockpit instruments, MultiGen Creator for modeling virtual visual environment which is driven by MultiGen Vega. Not only visual, the audio files are called by Vega too. As an experimental flight simulator, we designed but didn t build a Stewart platform with six degrees of freedom which sustains the cockpit to provide pilots motion cue. However, we needed it for the integrality. So we used SimMechanics CAD Translator to transform SolidWorks geometric assembly of the designed Stewart platform into Simulink block diagram model which could perform dynamic simulation after adding controller, washout filter and actuator models. Then the dynamic model was connected with VRML virtual reality to constitute a virtual prototype of the Stewart platform which replaced the real platform to join the flight simulator system. Simulation models were transformed to C code through MATLAB/RTW and loaded by RT-Lab to QNX real-time computational nodes to be compiled and executed. Moreover, RT-Lab provided API function to IOS for model running control and monitor, such as loading model, executing model, pausing model and changing parameter values. Data acquired boards, Advantech PCL88HD and PCL733, were equipped on the QNX main computational node, A/D and DI which were in charge of receiving pilots manipulating information, D/A and DO which were in charge of sending control signals for generating feedback force cue. The whole software was running on popular PCs except the one which fixed data acquired boards. This computer was Advantech industrial control computer, whose controller was also PC based. Fig.2 shows the structure of the computers cluster. Model Construction A flight simulator contains many parts, and every part contains many sub-systems. Just aircraft model which belongs to simulation model is composed of flight control model, engine model,

3 478 e-engineering & Digital Enterprise Technology navigation model, auto flight model, fuel model, electrical power model, hydraulic power model, air conditioning model, and so on. Further more, flight model contains aerodynamic model, equation of motion, atmosphere model, crash model, ground model, height above terrain, reposition, thrust model, weight model, and so on. Model and model connect each other by data flows, which organize a complex big net. But tremendous data flows have a basic and main route, which begins from data input by pilots, and passes flight control system, flight model, finally outputs flight state to visual system, instruments system, motion system and force feedback system. For the sake of convenience to show the rapid prototype modeling methods and prepare for the next section to testify the coordination of our flight simulator, we choose some models on this typical route to describe their building process, such as Equation of Motion, Aerodynamic Table Lookup, and the virtual prototype of the motion system, which colored in Fig.3. Fig.2 Structure of the computers cluster. Fig.3 Components of the Flight Model. Fig.3 describes the components of the flight model which are enclosed in dash frame. The most important block is the Equation of Motion. It works out the flight state according to the summation of forces and moments acting on the center of gravity of the simulated aircraft (A/C). The forces and moments come from various source. Firstly, aerodynamic coefficients are defined in the Aerodynamic Table Lookup block as functions of the flight state of the A/C including such as Mach number, angle of attack, angle of sideslip, true airspeed, and geometric characteristics of the aircraft. The geometric characteristics of the aircraft include such as control surfaces deflections, high lift devices, landing gear, and center of gravity location [4]. And then, Aerodynamic block uses these coefficients, air density, true airspeed, wing platform area, wing span, and wing mean

4 Applied Mechanics and Materials Vols aerodynamic chord to work out aerodynamic forces and moments. Secondly, the Landing Gear block works out landing forces and moments according to flight state and runway conditions. Thirdly, the Thrust block works out engine thrust, engine drag and ram drag. Finally, the gravity, of course, is computed in the Weight block. The Weight block also calculates the inertia of tensor and estimates the center of gravity according to the fuel consumption, landing gear position, and deflections of the wing flaps. The equation of motion summates these forces and moments, then integrates the effect of wind and turbulence, furthermore obeys the commands sent out by IOS, and at last outputs the state of flight. The detailed descriptions of some models building methods are following. Equation of Motion. As the core of the simulation model, the Equation of Motion affects the fidelity of the flight simulation directly. So we consider many effects which influence the dynamic of the A/C, such as earth rotation, earth flattening, mass of A/C altering. And we use quaternion instead of Euler angles to describe the A/C attitude in order to avoid singularity. Meanwhile, we neglect the Earth s nutation and polar motion which affect the flight dynamic trivially. The Equation of Motion block considers the rotation of an Earth-centered Earth-fixed (ECEF) coordinate frame about an Earth-centered inertial (ECI) reference frame. The origin of the ECEF coordinate frame is the center of the Earth, additionally the body of A/C is assumed to be rigid, an assumption that eliminates the need to consider the forces acting between individual elements of mass. The representation of the rotation of ECEF frame from ECI frame is simplified to consider only the constant rotation of the ellipsoid Earth including an initial celestial longitude [5]. There are many complex equations to be transformed to Simulink model to carry out the functions. Fortunately, Aerospace Blockset [6] which is one of the Simulink components offers us a ready-made model, Custom Variable Mass 6DoF ECEF (Quaternion). So we just needed to do a little improvement on it. This is an advantage of COTS software, mature, reliable and easy to use. Fig.4 presents the Simulink diagram of the Equation of Motion. Forces 4 mass 3 m_dot 6 I 5 I_dot f orces mass mdot I I_dot f orces mass I, I_dot Out9 Determine Force, Mass & Inertia 2 Ab [XEI] [0 0 w_e] [] XEI Ab p q r we p q r rel Calculate Velocity in Body Axes [_IF] 9 p,q,r rel [] _if V_mass we XEI Calculate Position in EI [_IF] [] u T [XEI] [XECEF] 2 XECEF p q r pdot,qdot,rdot 2 Moments 7 Triger Enable [XECEF] I, I_dot Moments X f µ l h ECEF Position to LLA pdot qdot rdot Calculate omega_dot [] pm_0 3 XG s x o p,q,r [] 0 p,q,r p q r mu l _if Euler _be _ef 5 6 _be 7 _ef 4 Euler _be _ef _f b VECEF LG0 LG0 deg rad L G (0) L G Celestial Longitude of Greenwich LG _IF ECEF to Inertial [_IF] Calculate & Euler Angles [] Calculate Body to ECEF m/s m/s Velocity Conversion 8 Fig.4 Simulink diagram of the Equation of Motion. There are other functions need to be realized in the Equation of Motion block, which don t compute the state of the A/C. The functions allow IOS to freeze the A/C, or reset the state of the A/C, such as reposition. For freezing the A/C, we add an Enable block colored by orange in Fig.4 to hold the Integrators in the Equation of Motion. For reposition, we use an external signal, which

5 480 e-engineering & Digital Enterprise Technology colored by magenta in Fig.4, sent by IOS to reset the integrator in Fig.5. Once the falling edge trigger signal is coming, the integrator resets to its initial condition colored by green in Fig.5 which has been modified by IOS using the API function provided by RT-Lab. And then the position of A/C will be reset to the new value. 3 we A B Cross Product C = AxB pxwe C 4 V_mass 4 u T xg_0 Calculate Position in EI µ l h 2 _if X f LLA to ECEF Position 5 x o s p 2 XEI Fig.5 Simulink diagram of A/C s position calculation and reposition. Aerodynamic Table Lookup. The reason why an aircraft can fly is the aerodynamic force acting on it and lifting it up. The Aerodynamic Table Lookup block reflects the flight character of the A/C. To define the aerodynamics of a specific aircraft, the numerous force and moment coefficients must be known. Because we haven t achieved the complete data package of these coefficients at the prototype design stage, we use the data files in the computer game, Flight Gear, to replace it. Although the data are fallacious, we concentrate on the study to testify the methods and feasibilities of developing a flight simulator based on COTS solutions. So we add some man-made data to increase the burden of computation, yet we must ascertain these data not to influence the result of computation. For example, data which define the contribution of rudder to lift force doesn t include in the data files in Flight Gear, but we know the influence factor such as angle of sideslip, deflection of rudder, and angle of attack, as in Eq., Cl rud = f ( β δrud α) () 9 According to this, we add a table with three dimensions and set every element to 0. In this way, we try our best to make the burden of computation approximate to the real commercial flight simulator nowadays. Latter, we can evaluate the character of coordination, yet not fidelity. Lookup Tables Toolbox in Simulink is the best tool for us to build the Aerodynamic Table Lookup block. We use a pre-load function to import the data to MATLAB workspace. Every data table is expressed by a named matrix, and considered as the table data by the Lookup Tables. Virtual Prototype of the Motion System. As we know, it is difficult to analyze kinematics and dynamics of Stewart platform. If we designed and developed the virtual prototype of the platform with traditional method, it would cost a lot of time on dynamic equations [7]. Even if we worked out the dynamic equation of every part of the Stewart platform such as actuators, joints, and the platform, it was also difficult to drive these numerous parts in virtual reality by those equations. For these reason, we choose SimMechanics CAD Translator as a bridge to combine CAD assemblies of the Stewart platform s mechanical system, SimMechanics model and virtual reality. With the Translator, models of the motion system s virtual prototype are unified in the MATLAB environment and can be used to drive virtual reality. This method uses SimMechanics CAD Translator to transform SolidWorks geometric assembly into Simulink block diagram model. After adding controller s model, actuators models and washout filters, the Motion system s model can perform dynamics simulation. Then the model is combined with the flight simulation model as a whole. Finally we use RT-Lab to compile the whole model and load it to separately computational nodes. This is the source of driving signals to virtual reality. At the same time, we build a LabVIEW Virtual Instrument (VI) to contain the VRML model which is generated from SolidWorks sub geometric assemblies and added Java script node. Then UDP and TCP communication blocks are added to receive driving data and send back

6 Applied Mechanics and Materials Vols object-to-object collision information which provides by Cortona Clients (ParallelGraphics. Inc). Connecting the simulation part with the virtual reality part by Ethernet is the last step to complete the virtual prototype of the motion system. The virtual prototype has the same control interface and similar dynamic character as the real motion system. And its animation is exquisite as in Fig.2. Its major advantage is utilizing the existence resource in the process of motion platform design stage. The method decreases the difficulties of generating virtual prototype and makes the development rapid and high quality. Coordination Validation The cues which a flight simulator provides to the pilots being trained must be strictly coordinated. It wouldn t do to have the pilot execute a turn, but not to begin to see the change visually, or feel the change for even a small period of time. Even for delays which are so small that they are not consciously detectable; a lack of coordination may be a problem. Such delays may result in a phenomenon known as simulator sickness, a purely physiological reaction to imperfectly coordinated sensory inputs. Although we can t testify the fidelity of our flight simulator because of the lack of data, we must testify the coordination of our flight simulator to answer the question whether our design based on COTS solutions has its future. We solved this problem with a way as call and back. We installed a press-button on the cockpit where the pilots could touch it easily. Before we started the test, we should control the A/C to fly high above the terrain. Once the pilots pressed down the button, the message was captured by a Digital-In card and then a step signal was generated in the flight control model. The step signal had a very large final value, and replaced one of the normal control signals such as aileron, elevator, or rudder, which had defined by IOS and input by the pilots through the wheel and the pedal. Now the step signal with other control signals would enter to the Aerodynamic Table Lookup block to join the computation, and then the Equation of Motion block. Because of the large value of the step signal, there must be a large angular acceleration on the output port of the Equation of Motion block described in detail in section IV. The large and unusual angular acceleration was detected to reset four Integrators to their initial conditions 0 and to hold the state of flight which would be sent to Motion system, Visual system, cockpit instruments, and Force Feedback system. Fig.6 Diagram of the Coordination Testify. The cueing systems received the flight state at different time, and then refreshed their own output devices. To Visual system, a new frame was drawn on column screen by three projectors. To Cockpit Instruments system, new positions of pointers were updated. These two systems had a common method to deal with the process, which executed in a loop. While a loop was executed over, a refresh task was done. So we added a UDP sender at the end of the loop. Once a refresh task was done, the data which the system received from the simulation model, the state of flight, such as the position of the A/C to Virtual system, was sent back to simulation model. Currently, the state of

7 482 e-engineering & Digital Enterprise Technology flight had been held in the simulation model. Then, the feedback state was compared with the held state. Once the two states were equal, the output value of the Integrator was recorded. This value was the time consumed by updating a new cue, which we concerned. Fig.6 shows this method. To Motion system and Force Feedback system, there was a little difference to the Visual system and Cockpit Instruments system. We judged the task done by the acceleration of these systems. Because we only had a virtual prototype of Motion system, we measured the platform s acceleration by adding a body sensor block on the centre of gravity of the upper platform in the SimMechanics model. To Force Feedback system, we measured the actuators control signal. When the acceleration of the platform or the control signal of the actuators exceeded a predefined value, a flag signal was sent back to the simulation model to record the output value of the corresponding Integrators. And then, we had the consuming time too. We did the test for several times and with different control channels, roll, pitch and yaw [8]. We got the longest time which every cueing system needed to refresh a frame. The result was exciting. Every cueing system could execute to completion with in sixty milliseconds, which reached 6Hz refresh rate at the worst condition. Under this refresh rate, no one would feel uncomfortable for coordination. This result proved the design of our flight simular based on COTS solution was feasible and successful. Conclusion In this paper, the architecture of the flight simulator based on COTS solutions is presented. Some function blocks of the simulator are described. And the coordination of the simulator is testified to prove the design methods feasible. The whole design based on COTS solutions shows the advantage in the prototype development of a flight simulator. The application of mature and commercial simulation software speeds up and decreases the risk of engineering productions construction. The COTS hardware reduces the expense and expands the bound of application. The modularity of the implementation is intended to ease the modification of the simulator as better modelling becomes available or additional elements are included in the future, such as aerodynamic coefficient data replaced by real data of Boeing , and a modelling for the Flight Management System. We will pay more attention to increase the fidelity of our flight simulator in the future. References [] P. Baracos, G. Murere, C.A. Rabbath and W. Jin: Electric Machines and Drives Conference, 200. [2] F. Holzapfel, I. Sturhan and G. Sachs: Low-Cost PC Based Flight Simulator for Education and Research, AIAA, [3] R. Kazman: Distributed Flight Simulation: A Challenge for Software Architecture, in Handbook Of Parallel And Distributed Computing, A. Zomaya (ed.), McGraw-Hill, 995 [4] Christopher J. Atkinson: Development of an Aerodynamic Table Lookup System and Landing Gear Model for the Cal Poly Flight Simulator, California Polytechnic State University, San Luis Obispo, August, [5] B.L. Stevens and F.L. Lewis: John Wiley & Sons, New York, 992. [6] The Mathsworks: MATLAB USER S GUIDE,Vol.9(2005). [7] Sjirk Holger KOEKEBAKKER: Model Based Control of a Flight Simulator Motion System, Delft, Delft University of Technology, 200, pp.3~8. [8] Federal Aviation Authority: Airplane Simulator Qualification, Advisory Circular (AC) 20-40C,Vol.7 995).

8 e-engineering & Digital Enterprise Technology / PC-Based Human-In-the-Loop Simulation for Flight /