DYNAMIC MODEL DEVELOPMENT AND VALIDATION FOR AN AIRCRAFT FUEL TEST-RIG

Transcription

1 DYNAMIC MODEL DEVELOPMENT AND VALIDATION FOR AN AIRCRAFT FUEL TEST-RIG P.J.Bennett +, R.Dixon +, J.T.Pearson *, R.M.Goodall +, A.Martin *, M.C.Walsh *, and M.Khella * Systems Engineering Innovation Centre, BAE Systems, Sir Denis Rooke Building, Holywell Park, Loughborough University, Loughborough, Leicestershire, LE TU, Telephone +()9 6 + Department of Electronic and Electrical Engineering, Loughborough University, Loughborough, Leicestershire, LE TU, Telephone +()9 7 Abstract: Reliability is an important issue across industry. This is due to a number of drivers such as high safety levels within aviation, the need for mission success with military equipment, or to avoid monetary losses (due to unplanned outage) within the process industries. The research in this paper is part of the first phase of a project that will examine the application of diagnostic and prognostic techniques to a real plant. One key area of interest is model based techniques. Hence, the paper presents a laboratory scale aircraft fuel rig which has been commissioned and focuses on the development of a simulation model of that system. Test data obtained from experiments on the real system are compared with predictions from the simulation in order to validate the model. Copyright USTARTH Keywords: Fault Detection, Fault Diagnosis, Fault Isolation, Model-Based, Validation.. INTRODUCTION Aerospace and Defence Systems are becoming increasingly complex with higher component counts and ever more complicated components and subassemblies. Faults and failures are becoming harder to detect and isolate. The time that operators and maintenance technicians need to spend on faults is rising in direct relation to the complexity of the systems. With these increasing demands on reliability, maintainability and safety of systems, a wide range of fault detection and diagnostic methodologies have been proposed, and there has been considerable interest in the practical application of these fault diagnostic techniques (Leonhardt and Ayoubi, 997; Rengaswamy et al, ; Frank et al, ; Venkatsubramanium et al, ). Reliable diagnostic techniques can contribute to reduced maintenance costs and, perhaps more importantly, to increased system availability. The selection and integration of an appropriate diagnostic tool has the potential to produce a reduction in life cycle costs for both the customer and the manufacturer. For autonomous systems on board fault diagnosis is vital as the human interface is no longer available to perform the function that needs to be performed in order to ensure the safe operation of the system. The approach commonly referred to as Fault Detection and Isolation (FDI), or in a broader sense System Health Management (SHM), appears to have good potential in terms of improving human safety, monetary losses, and overall mission success capability. For example: the petrochemical industry alone incurs an estimated $ billion in losses every year due to process failure (Venkatsubramian et al. ), and the cost is much more when other industries such as pharmaceutical, speciality chemicals, power, etc., are included. In order to investigate the potential of FDI and SHM schemes, Engineers at BAE Systems and Loughborough University are working together to develop a realistic test rig on which various techniques can be implemented and evaluated. The target system is lab-scale aircraft fuel system simulator. Phase of the project involves development and commissioning of the rig and development of simulation models for the rig (which can later be used in FDI design). Figure below shows the overall validation process which is being

2 undertaken by the partners. This paper focuses on the development and validation of a physical mechanistic model of the fuel rig as illustrated by the Matlab model validation which is shown in stage. Other tasks, not discussed here, include the validation of the Matlab model against a model developed using gproms and beyond this the assessment and validation of various fault diagnostic tools. The paper is set out as follows. Section introduces the fuel rig. Section discusses the development of a set of sub-system models which are put together in Matlab/Simulink in order to model the rig. Section presents results from some open-loop tests on the fuel rig and compares these with predictions from the Simulink model. Finally, section sums up the results to date and describes the future direction of the research. Model (gproms) Fault Diagnostic Validation (gproms) Fault Diagnostic Tool Validation (gproms) System Parameters System Dynamics Model (Matlab) Fault Diagnostic Algorithm Synthesis (Matlab) Fault Diagnostic Software Fault Diagnostic Tool Validation (Matlab) Phase Model Validation Phase Fault Diagnostic Algorithm Validation Phase Practical Tool Implementation Fig. The Fuel Rig.. Fuel Rig Hydraulic System The fuel rig consists of four tanks: a Wing and a Main, each of which is connected to the Collector by two pumps (two Wing pumps and two Main pumps). The Collector supplies the engine (represented by the fourth tank) via two collector pumps. Fault Diagnostic Tool Implementation (DSP) Fig. Process of FDT Validation. FUEL RIG A photograph of the fuel rig is shown below in figure. It represents the fuel system of a modern aircraft and its associated electrical power supply and is based at the Prognostic Health Management (PHM) laboratory at the Systems Engineering Innovation Centre (SEIC) in Loughborough. The rig consists of a number of tanks, pumps, flow meters, pressure sensors, level sensors, three-phase motors, inverters, voltage sensors, current sensors and other instrumentation. This allows for the implementation and investigation of a wide range of fault diagnostic tools and techniques. A comprehensive collection of different types of faults can be injected into the testbed, which also has the capability for reconfiguring its fluid and electrical systems in the event of such faults being detected and isolated. Figure shows a the basic layout of the rig where, to aid clarity, only the four tanks, six pumps and three flow meters are shown. On top of the basic instrumentation, each tank is also equipped with a temperature sensor, a level sensor and discrete level limit switches. Each pair of pumps is connected to a flow meter. The piping arrangement between tanks and pumps is slightly more complex than shown as there are a number of dual port and triple port electrically actuated valves that provide the fuel rig with the capability to reconfigure the flow paths in the event of a fault, for example a blockage or a failure in one of the pumps. There are also a number of temperature probes, microphones and accelerometers mounted on or in close proximity to the pumps. It should be noted that this extra instrumentation, along with the reconfiguration valves, would not normally be found on a conventional aircraft fuel system, however it has been included to provide the fuel rig with capabilities that will enable a wider range of fault diagnostic tools and techniques to be implemented and investigated.

3 Wing Flow Collector Engine Flow Pumps Flow Main. Fuel Rig Fault Injection A range of faults can be injected into the fuel rig, these include faults in the fluid sub-system, faults in the power supply sub-system and faults in the sensing sub-system. These faults can be injected using the software running on the fuel rig PC, or they can be injected physically on the patch-board located on the end of the fuel rig. This patch-board enables all sensor signals to be manipulated in the following ways: i) physical disconnection, ii) set to full scale or v, and iii) summation with a secondary signal, such as additional noise, a bias, or another function.. SYSTEM MODEL The modelling will focus on the fluidic side of the rig and the approach used is mechanistic/physical modelling based-on the hardware of the fuel rig and the fluid properties. Fig. Fuel rig schematic showing the basic layout of the tanks, pumps and piping. The electrical system for the fuel rig has been designed to represent an aircrafts electrical system. For example there is a primary and a secondary busbar and the pump and sensor power supplies can be switched between either busbar using electronically controlled relays providing a reconfiguration and fault injection capability in the electrical system.. Fuel Rig Control and Logging System The fuel rig is linked to a PC that performs three primary tasks (fig. ). The first task is to log all the measurement data from the fuel rig sensors and control signals to the rig. These are stored in a database that forms part of the overall Prognostic Health Management (PHM) facility. The second task is to run the basic fuel control algorithm; this relatively simple algorithm and is not described in detail in this paper. The final task relates to the implementation of the diagnostic functions. On the rig, each of the three tanks supply fluid by means of two pumps (fig. ) with an array of valves connected to each. Each of the pumps and its associated valves will be referred to as a pump tray (hence there are six of these) and a schematic of one of these as represented for modelling purposes is shown in figure. Pump Pump Outlet Flow Pump Inlet Flow Pipe Pressure Relief Valve Fig. Fuel Rig Pump Tray. Each element in figure is modeled as a subsystem and the overall model consists of six such pump tray models to represent the system shown in Figure. It should be noted that whilst some of the pipes depicted in fig. are very short they are included in the model of the system to make the overall model equations solvable. The sub-systems models are discussed below.. Model Pipe Valve Pump Tray Outlet Flow Pump Motor F/B (7) Accelerometers (9) Pump Temp (7) Control Valves () Temp () Data Logging Pressure Sensors (6) and Measurements Level Meters () Analogue to Digital Flow Sensors () () Pressure Switches (6) Control PC Binary TTL Signals functions Processor () Limit Switches (8) Solenoid Valves (8) Digital to Analogue () Pump Control (7) Diagnostic and Control Valves (9) Prognostic Pump Control () functions Control Valves () Fig. Fuel Rig Control and Data Logging. The height and pressure of the fluid in the tank are defined by equation and equation. The equations provide a pressure output for a given height which in turn is dependant on the flow in and out of the tank. qin t ( ( ) q h( = a out( ) dt () P( = ρ g h( () Where the symbols are defined as: ρ = density (kg/m ) g = gravity (m/s ) a = tank area (m ) P = pressure (Pa) q = volumetric flow (m /s) h = height (m)

4 . Valve Model The valve equation below gives the volumetric flow rate from the valve for a given pressure differential across the valve. q(. cv yv( ρ P( v = () Where: q = volumetric flow rate (m /s) v = specific volume (m /kg) ρ = density (kg/m ) c v = valve conductance (m ) P = pressure difference (Pa) y v = proportional valve opening. Pump Model The pumps used on the fuel rig are peristaltic pumps, and these provide a flow that is directly proportional to the pump motor speed that in turn is proportional to a demand signal from the control and data logging system. The pump, motor and gearbox are represented in the model as one first order transfer function based on results found by practical experiment.. Pipe Model The defining equation for the pipe subsystem model is based on the compressibility of fluid in the system due to the pressure acting upon it, defined by the bulk modulus and is shown in equation 6. Ev P = ( qin qouδt () Vo Where: q = volumetric flow rate (m /s) E v = Bulk Modulus (Pa) P = Pressure (Pa) V o = Original Volume of pipe (m ). Overall System Model When constructing the overall model the subsystems are simply parameterized and connected together to represent the complete system. For parameterization of the component models, information was taken from measurement, data-sheets and based on experiment. For example: values for the valve conductance, c v, was not available from the manufacturer. Hence it was initially calculated assuming a. bar pressure drop across each valve. This was later refined once some flow tests had been performed on a valve.. VALIDATION RESULTS A number of experiments were carried out on the fuel rig in order to provide validation data for the models. A subset of these results is presented and discussed below. The first test involved the draining of a tank through a free flowing valve, which enabled the valve conductance to be calculated more accurately. The second and third tests are more complex and involve the running of pumps at various speeds and the associated measured flows, pressures and tank fluid heights. These experiments were designed in order to cover the full functionality of the rig. Having collected the data the model was then stepped through the same tests and the results of the model and the rig were compared.. Test Fully open valve The valve used in this test is manually actuated and is opened fully at time seconds. The tank then empties freely through the valve from a starting height of.m. Note that there are several similar automatic valves on the fuel rig, all of which are identical (in terms of aperture) but serve different purposes. The test is simulated on the model and the valve conductance calculated to be.x - m in order to obtain the same rate at which the tanks are emptied as shown in figure 6. Naturally the results match-up reasonably well between the simulation and the real response. height (m)..8 Valve opening simulation height ADT height Fig 6. Result from test, showing comparison of tank emptying rate through a fully open valve. There is one problem which is noted from these results, which is visible in the plot (fig. 6). The tank height reduction follows a straight path, whereas the model result follows a curved path. This is due to the incorrect calibration of the level sensor as fitted to the rig. A separate model developed using gproms also gave the same result as that achieved using Matlab confirming the validity of the model. The sensor has not yet been recalibrated to provide a more accurate result.. Test Variable Speed Input to two Main tank Pumps The test involves the increase and decrease in pump speed input and the resultant flows and pressures

5 seen within the model and the fuel rig. There are two pumps, PP and PP, which are run in this test at different speeds. Both pumps are supplied from the same tank and also both are supplying the same tank. The results can be seen in Figures 7 to 9. Figure 7 shows the pump speed demand voltage, as controlled by the control and data logging system, on a scale of -V, V equating to no flow and V input being full-scale output. Note this a demand voltage input and not a motor response. pump speed demand(v) pump speed demand(v) Sim PP demand ADT PP demand Sim PP demand ADT PP demand Fig 7. Result from test, showing comparison of speed demand voltage inputs. Figure 8 shows the levels in the main tank (top) and the collector tank (bottom). Clearly the main tank is emptying as the collector tank fills up. Note: The tanks have different cross sectional areas so one tank s loss in height is not the same as the other tank s gain. In both cases the simulation matches the rig response reasonably well... Flow (l/s) Pressure (Pa) Sim Flow ADT Flow Sim Pressure ADT Pressure - Fig 9. Result from test, showing results of flow and pressure.. Test Valve Opening and Closing (Pressure Tes. The test involves the increase and decrease in pump speed whilst opening and closing the isolation valve, this valve is located at the output end of the pump tray. Closing the valve causes a system overpressure situation which causes the pressure release valve to release fluid from the system and back to the main tank. The valve position command signals and pump motor speed are shown in figure. (top and bottom, respectively). Valve/ Limit Switch Position.8 Sim Limit Sim Limit ADT Limit ADT Limit Height (m).. Sim Tk Lev ADT Tk Lev Pump Demand (V) Sim Pump Demand ADT Pump Demand Height (m) Sim Tk Lev ADT Tk Lev - Fig 8. Result from test, showing changes in tank levels on the fuel rig compared with those predicted by simulation. Figure 9 (top) shows the total flow out of the system and the pressure seen within the pipes (bottom). The simulation results show a good correlation with the actual measured flows and pressures. It must be noted that the pump motor dynamics have been added to the model and can be seen in the flow and pressure waveforms seen in figure 9. All of the results produced on the test rig have been filtered using zero phase forward and reverse digital filtering to reduce the noise seen on the signals Fig. Result from test, showing comparison of valve position and pump speed demand voltage input. Flow (l/s) Pressure (Pa) 6 x Sim Flow ADT Flow Sim Pressure ADT Pressure Fig. Result from test, showing comparison of flow out of and pressure within the system.

6 The flow out of the system can be seen in figure (top) and the pressure in the pipe prior to the isolation valve is shown in the bottom plot. Both flow and pressure are affected by pump speed and valve position. The overpressure situation is reached when the pressure release valve opens at a limit of 8.6 KPa (7 PSI) and the pressure within the system saturates. Once again the overall correlation between the simulation and real results is clear, however further work is required in order to correlate with the practical results more closely. 6. CONCLUSION This paper has presented results from phase of an industrially focussed project aimed at evaluating fault diagnosis tools. The aircraft fuel system test rig has been introduced and a dynamic model for it described. The validation results presented show that the model is capturing the dominant mode dynamics of the real system. REFERENCES Frank, P.M., S.X. Ding and T. Marcu (). Modelbased Fault Diagnosis in Technical Processes, Transactions of the Institute of Measurement and Control, Vol., No., pp7-. Leonhardt, S. and M. Ayoubi (997). Methods of Fault Diagnosis, Control Engineering Practice, Vol., pp Rengaswamy, R., D. Mylaraswamy, K. Arzen and V. Venkatasubramanian (). A Comparison of Model-based and Neural Network-Based Diagnostic Methods, Engineering Applications of Artificial Intelligence, Vol., pp8-88. Venkatasubramanian, V., R. Rengaswamy, K. Yin and S. Kavuri (). A Review of Process Fault Detection and Diagnostics, Part I: Quantitative Model-based Methods, Computers and Control Engineering, Vol. 7, pp9-. Ultimately the validated model will be used for the design of fault diagnosis and prognosis tools that will be evaluated on the test-rig. This project is in phase at present and there is much more work to be done to build on and utilise the results (and model) presented here. The next stages shall be the validation of a gproms model followed by implementation of varying diagnostic and prognostic techniques.

7