Search

A model-based fault management system for a Three Mass Torsion Oscillator is described. Fault management includes fault detection, fault diagnosis and fault compensation. A process model of system dynamics is derived. Friction effects are modelled by a neuro-fuzzy Local Linear Model Tree (LOLIMOT) approach. Parameter estimation is obtained with two simple experiments. A friction estimation stage is used to support fault diagnosis and potential control applications. Parity equations are used to detect and diagnose speed sensor failures. In case of sensor failures, the fault management system initiates reconfiguration operations. The failure of one sensor is compensated by substituting the sensor signal with the output of a process model fed by other, intact sensors. The methods presented are evaluated using experimental measurements of a test rig.

Mechatronic components allow to design systems with many new functionalities and to turn away from classical design paradigms which had been required for purely mechanical designs. However, the increase in complexity and the renunciation from purely mechanical to electro-mechanical designs goes along with an increase in the possibility of unpredictable faults. Therefore, the design paradigm also has to increase from fault detection and diagnosis to fault-tolerance by redundancy at least for safety related applications. The idea for this paradigm change is that faults cannot entirely be avoided in complex systems. Rather, the occurence of faults is accepted and measures are taken as to limit the implications of the fault and maintain the operability despite these impairments caused by the fault. The paper at hand will first introduce the overall setup of a component respectively system with integrated fault management. Fault detection and diagnosis are compactly sketched before fault tolerance principles are introduced. The paper looks at both hardware tolerance and analytical tolerance. As the reaction to the fault can cause a functional degradation even in the presence of redundancy, the different degradation steps are also introduced. Then, different designs of fault tolerant electric and hydraulic actuators as well as sensors are presented. An essential requirement for the inclusion of the designs into this paper was at least an experimental realization of the concept. A total of 11 different designs is presented. Finally, the different designs are compared and future developments are shortly discussed

Driver assistance systems need more and more accurate information about the vehicle driving situation. If one of these sensors gives faulty information, driver assistance systems can not appraise the current driving situation. Therefore, a fault tolerant sensor system is introduced. Based on modelling the vehicle dynamics, a fault detection and diagnosis is developed. A Kalman filter and a two-track model is designed to create analytical signals replacing the faulty sensor. By reconfiguring the Kalman filter the yaw rate, the longitudinal acceleration or the lateral acceleration can be estimated with small discrepancies. Thus, it is possible to generate a fault tolerant sensor system of these sensors.

A new method for the estimation of process parameters based on the principal component analysis is developed. The estimator yields optimal estimation results in the case of errors in variables (EIV) problems which are characterized by corrupted measurements of input and output signals. As the residual generation in fault detection methods often feature EIV characteristics, the estimator can be used to identify linear models for residual calculation. To overcome the limitations on linear models, the developed estimator is integrated into the LOLIMOT approach which is able to identify nonlinear processes. The estimator is used as an alternative to the standard Least Squares estimator to identify the parameters of the local linear models. Comparative results show the better suitability of the developed estimator for the residual generation in EIV-setups.

A collision avoiding vehicle driver assistance system is developed. Collisions can be avoided using braking, steering or a combination of steering and braking. Within this paper only steering interventions are discussed. To guide the vehicle on a desired trajectory for emergency steering, two feedback control approaches are presented. A Kalman filter based vehicle state estimation approach to estimate the sideslip angle is shown. Simulation results with a validated nonlinear two-track model and results from test drives are shown.

After a short introduction into the topic of active vehicle suspension systems, a mathematical model of the used active vehicle suspension, which is presented in a test rig, is derived. It is shown how the unknown parameters can be obtained experimentally by parameter estimation. Using parameter estimation and LOLIMOT - a special type of neuronal networks, models of the active suspension are identified. These models are used for model based fault detection and identification, in order to obtain reliable knowledge of the system's state. All results are shown for measurements from an active suspension on a test rig.

In this paper a method for identifying and classifying hemodialysis patients based on the relative blood volume (RBV) is presented. A group of 42 hemodialysis patients were treated with a special ultrafiltration pattern. The response in RBV to this pattern was recorded and analysed. Using the steady state and time constant of a fitted and estimated function, the patients can clearly be separated into 4 different groups. The analysed parameters give information about the fluid-status, the hemodynamic stability and the physiologic and pathologic patient conditions.

This approach describes a methodology for the fault detection in machine tools by using few robust sensors and dynamic process models. Two different model based methods, parameter estimation and state estimation, are used for the detection of tool wear and tool breakage in the drilling process. >

The development of technical diagnostic systems has shown that the time has come to bridge the gap between the process-oriented engineering approach and the reasoning-oriented approach. As a contribution to this on-going task, the paper proposes a unified frame for integration of different fault detection and decision techniques via approximate reasoning on analytical and heuristic fuzzy symptoms. >

Takagi-Sugeno type fuzzy models are universal approximators for nonlinear dynamic processes. If they are trained to represent the inverse plant characteristics they can be used as feedforward controllers. The achievable control performance strongly depends on the model quality, and the simple inverse model controller does not cope with disturbances and process uncertainty. As a consequence, a hybrid control scheme is proposed which considerably improves the robustness properties. This paper reviews the identification of both forward and inverse fuzzy models. The difficulties accompanying the latter task are discussed. The control scheme based on the idea of disturbance observation is introduced and thoroughly analyzed. Finally, the controller is applied to a cooling blast with nonlinear behavior and variant dynamics.

A real time fault detection for a brushless DC motor driving a mechanical actuation system is presented. The brushless DC motor is controlled by a PWM inverter using rectangular current excitation. After an introduction into fault detection with parameter estimation, a mathematical model for the actuator with special emphasis on the motor itself is derived. The application of the estimation algorithm to detect electrical parameter changes in the motor using single phase currents is described. The fault detection scheme is implemented on a digital signal processor controlling the actuator. Finally experimental results are given.

Fault diagnosis requires a classification system that can distinguish between different faults based on observed symptoms of the process under investigation. Since the fault symptom relationships are not always known beforehand, a system is needed which can be learned from experimental or simulated data. A fuzzy logic based diagnosis is advantageous. It allows an easy incorporation of a-priori known rules and also enables the user to understand the inference of the system. In this contribution, a new diagnosis scheme is presented and applied to a DC motor. The approach is based on a combination of structural a-priori knowledge and measured data in order to create a hierarchical diagnosis system that can be adapted to different motors. Advantages of the system are its high degree of transparency and an increased robustness.

Summary form only given. The synergetic integration of mechanical processes, microelectronics and information processing opens new possibilities for process design as well as for automatic control. The solution of tasks within mechatronic systems is performed on the process side and the digital-electronic side. As the interrelations during the design play an important role, engineering must be simultaneous from the very beginning. Mechatronic systems are developed for mechanical elements, machines, vehicles and precision mechanic devices. The integration of mechatronic systems can be performed by the components (hardware-integration) and by information processing (software-integration). The information processing consists of low-level and high-level feedback control, supervision and diagnosis and general process management. Special signal processing, model based and adaptive methods are applied. With the aid of a knowledge base and inference mechanisms mechatronic systems with increasing intelligence will be developed. The main goals are to increase systems performance, reliability and economy, and production costs. Some examples for the development of mechatronic systems are shown, e.g. active damping of oscillations of mechanical drives, hydraulic pipes and vehicle suspensions, control of machine tools with force reconstruction, smart actuators and supervision and fault diagnosis for automobiles.

Providing a simple and effective way to describe the nonlinear input-output behaviour of a system, three-dimensional mappings (3-D maps) have gained a lot of importance in modern automotive technology. Applications cover a wide range from real-time control systems up to the area of vehicle simulation. Replacing the conventional look-up-tables by neural network or fuzzy logic representations offers an easy possibility to generate 3-D maps by measured data and to adapt them online using measured signals. This paper describes the modelling of engine characteristics for vehicle control and simulation purposes by multilayer perceptron and radial-basis function networks. In addition to that, a neuro-fuzzy approach is discussed as well.

In this paper a design approach for cutting force control of the milling operation is presented. The system is applicable to machine tools featuring speed controlled feed and main motors with a force dynamometer. The intention is to get a constant load of the milling tool independent of parameter changes of the cutting process. The manipulated variable is the feed speed which has the largest influence on the cutting force. An estimator determines a cutting tooth dependent low order model of the milling process for adaptation of the control system. The inverse process model is used to compute a reference feed which is adjusted by the normal cascaded feedback controller. Experimental results show the quality of the presented control system.

This paper considers the application of neural networks with distributed dynamics to the identification of nonlinear systems. The primary objective is to establish a neural model bank which generates prediction errors. These estimation residuals can be treated as analytic symptoms to supervise the plant operating state. The practical approach applicability has been illustrated using a thermal plant. Here, two sensor faults of interest are localized by means of the so-called residual pattern.

New technologies, rising customer demands and severe exhaust gas regulations led to a rapid advancement of combustion engines. Nowadays, engines are characterized by more and more complex structures, whereby maintenance and trouble shooting becomes more and more complicated. Therefore the development of suitable fault detection and diagnosis methods is necessary, whereas model based methods enhance a definite detection of faults according to type, size and location.

Investigates the possibility of computing position estimates for passenger cars based only on signals available in a production car. Two new dead reckoning approaches are compared to the well known rear axle based algorithm. One method combines the wheel revolution data calculated on the basis of raw ABS-signal measurements with the steering angle measurement. The other model based algorithm computes the position estimates using velocity and also steering angle measurement. For this approach the well known tricycle model has been extended by an object-oriented front axle model. Both newly proposed methods perform superior to the simple rear axle based algorithm and provide position data precise enough for control purposes. This was shown by experiments using a real car in connection with a real-time signal processing environment.

An accurate estimation of the vehicle velocity is one of the greatest challenges in modern vehicle systems. The velocity is needed for the Antilock Braking System (ABS) and Traction Control System (TCS) as well as for the Electronic Stability Program (ESP) and cannot be measured directly. Using at least two fist sensors, which measure the vertical frequency component of a front wheel and the related rear wheel, the phase difference between both signals gives knowledge about the velocity of the vehicle. The presented algorithm is based on cross correlation function to estimate the velocity using the sensors described above. Different methods are described which improve both the resolution and the accuracy of the output signal v ref .

Due to the increasing demands concerning reliability, safety and economy of technical processes, the on-line monitoring of induction motors is an important topic in the engineering field. In this paper, a model-based approach for fault detection and diagnosis of nonlinear processes is employed. The supervision of nonlinear systems is often very difficult in view of the lack of accurate models. However, neuro-fuzzy models may help to cope with this problem, as they can be trained from measured data. In this contribution the application of a multi-model approach for fault detection and diagnosis of induction motors is presented. For this purpose the process is decomposed in several subprocesses.

Fault detection is increasingly an essential part of vehicle development. Integrating such fault detection subsystems raises the reliability, maintainability, and safety of automobile components. Until now, the suspension system was not a safety-critical component, but with the introduction of global chassis control systems this system is getting more important. Therefore, the paper presents fault detection algorithms for the suspension system. Real measurements of a vehicle are made to compare the advantages of the different fault detection algorithms.

In 30% of all hemodialysis treatments a sudden hypotension occurs due to excessive fluid removal. Currently the rate of fluid removal (ultrafiltration rate or UFR) is set to predefined values based on expert knowledge of the doctor. This paper is a first step towards closing the open loop control by presenting a hydraulic circulation model of the patient, which can be used in an observer structure in order to keep certain state variables like venous pressure or cardiac output above critical values.

Regarding the mechanical engineering area, over the last 40 years a lot of effort has been undertaken to find very exact descriptions for the dynamic behavior of road vehicles based on mathematical models. All those models include certain parameter values which may be taken from data sheets or which have to be measured or determined by real driving tests. Using these physical models for vehicle simulation purposes, the problem arises, that some of the model parameters are time-variant. They vary over a smaller or larger time period, e.g. due to aging, different vehicle loads or changing environmental conditions like a transition from dry to wet or icy road. Parameter variations lead to systematic modeling errors which makes simulation results turn out incorrect. To overcome that problem, this paper describes the use of hybrid models to reduce modeling errors. Within hybrid models, conventional mathematical process models are combined with adaptive learning structures, e.g. neural networks. In this contribution, an extended radial basis function network called LOLIMOT (local linear model tree) is used to compensate the influences of changing road conditions affecting a vehicle dynamics simulation model.

Increasing demands of low emissions and low fuel consumption of modern spark ignition combustion engines require new ways for an optimal control of the ignition timing. Instead of classical open-loop strategies cylinder pressure sensors are used for an adaptive control of the ignition point. A linear feedback controller is designed as well as an online adaptive neural feedforward controller, the latter is trained during regular operation, i.e. no test cycles are required. The control algorithms were implemented and tested in a research automobile. Experimental results showed that the proposed neural network is very effective in learning the engine's nonlinearities and in compensating for manufacturing tolerances and aging. The designed adaptive feedforward control improves efficiency and fuel consumption.

Faced with the need to reduce development time and cost, the hardware-in-the-loop (HIL) simulation increasingly proves to be an efficient tool in the automotive industry. It offers the possibility to investigate new engine control systems with fewer expensive engine dynamometer experiments and test drives. In the scope of a research collaboration, Daimler Benz and Darmstadt University of Technology are developing a hardware-in-the loop simulator for the investigation of the electronic engine management of the new Mercedes Benz truck engine series 500 and 900. This paper first describes the necessary models for real-time simulation of the subsystems Diesel engine, turbo charger and vehicle. Then the setup of the simulator test bench is introduced and the performance of the simulator is demonstrated by several experimental results.

An approach for model-based fault detection and isolation (FDI) of sensor and process faults for nonlinear processes is presented. A fuzzy model (Takagi-Sugeno type) of the nominal process provides characteristic features like time constants and static gains in the actual region of operation. Comparing these with features derived by recursive parameter estimation leads to significant symptoms which indicate the state of the system. The practical applicability is illustrated on an industrial scale thermal plant. Here, nine different faults can be detected and isolated continuously over all ranges of operation.

Heat exchangers play an important role in chemical and process industries. In order to improve reliability and control performance, intelligent concepts for control, supervision and reconfiguration are necessary. In this paper, an approach is presented which integrates model-based adaptive control and reconfiguration based on fault detection/diagnosis applied to a heat exchanger plant. The adaptive controller and the fault detection scheme are based on a fuzzy model of the process (Takagi-Sugeno type) and the fault diagnosis is performed using a self-organizing fuzzy structure.

The synergetic integration of mechanical processes, micro-electronics and information processing opens new possibilities as well to the design of processes as for its automatic control. The solution of tasks within mechatronic systems is performed on the process side and the digital-electronic side. As the interrelations during the design play an import role simultaneous engineering from the very beginning has to take place. Mechatronic systems are developed for mechanical elements, machines, vehicles and precision mechanic devices. The integration of mechatronic systems can be performed by the components (hardware-integration) and by information processing (software-integration). The information processing consists of low-level and high-level feedback control, supervision and diagnosis and general process management. Special signal processing, model based and adaptive methods are applied. With the aid of a knowledge base and inference mechanisms, mechatronic systems with increasing intelligence will be developed. The main goals are to increase systems performance, reliability and economy, and production costs. Some examples for the development of mechatronic systems are given, such as smart actuators, adaptive suspensions, electrical brakes, adaptive cruising control of cars, and hardware-in-the-loop simulation of combustion engines. Experimental results are shown, and the improvements by the mechatronic approaches are pointed out.

The wheel speed is an important signal for modern automotive control systems. The performance of these systems is closely related to the quality of the processed wheel speed. However, due to manufacturing tolerances or corrosion respectively of the sensor gear wheel, the quality of the signal is affected negatively. In this paper a software-based method to compensate for the mechanical inaccuracy of the sensor wheel is introduced. The approach increases the signal accuracy of conventional automotive wheel-speed sensors significantly. Moreover, with the presented method, the demand for manufacturing precision of the sensor wheel is reduced, which in turn cuts down the costs for mass production considerably. The correction is obtained by online learning of a correction factor for each sensor impulse using a fuzzy model for reconstructing the actual wheel-speed and recursive parameter identification for estimating the correction factors. A method to determine the angular position of the gear wheel without a reference impulse is proposed. The introduced approach is assessed by applying it to real world data.

A method for identification of adjustable shock absorbers is presented which combines a modern QRRLS parameter estimation algorithm (DSFI) with an artificial neural network (ANN) for classification purposes. The parameter estimation algorithm is based on a discrete-time linear model. Thus, no state variable filter (SVF) as for continuous time identification problems is required. For the ANN, a multilayer feedforward perceptron trained by backpropagation is used. The method was tested by simulation and with data drawn from shock absorber test stands at UC Berkeley and TU Darmstadt.

A driver simulation used in dynamical engine test stands is discribed which may be used to control the velocity and acceleration during exhaust emission test cycles. A model of the car body including drive line, differential and wheels has been developed and implemented on the test stand to simulate the dynamics of load changes occurring in a driving vehicle. A comparison between a human driver and the control results of the driver simulation (realized as a digital controller) is presented and discussed.

In this paper, a dynamical engine test stand is presented which may be used as an advanced tool for combustion engine optimization and control. To simulate the dynamics of load changes occuring in a driving vehicle, a model of the car body including drive line, differential and wheels and a driver simulation have been developed and implemented on the test stand. Simulation results are presented and discussed.

In many cases the design of speed control has to take into account the changing behavior of the closed loop in dependence of inertia and load. For example, if the actual moment of inertia J is smaller than considered in the controller design, instability may result. This study deals with the adaptation of the controller for timevariant or operating point dependent conditions. It will be shown, that with appropriate algorithms, it is possible to fit an adaptive controller for a fast dynamic servodrive by using a standard ?-processor.

Shock absorbers with adjustable damping coefficients used in present vehicle suspension systems are only feedforward controlled The effective damping is not measured and many changes in vehicle conditions can not taken into consideration automatically. Two methods are presented which use real-time parameter estimation and feedback to build up a parameter adaptive shock absorber system. The dynamics of the vehicle's vertical motion can be modelled theoretically. Unknown parameters of the nonlinear model like damping coefficient, stiffness, load and friction are estimated on-line in real-time. Simulations and experiments on a quarter-car test stand show the results obtained until now.

Adaptive controllers have proven to work well as long as all theoretical pre-conditions are fulfilled and the free design parameters are chosen property. However in practice these pre-conditions may be violated and/or some free design parameters are not chosen properly. Both result in a lowering of the control performance. Therefore a supervision and coordination level is introduced, monitoring the adaptive controller and taking appropriate actions if malfunctions are detected. The introduced supervision and coordination level exists altogether of two parts, named start-up part and run-time part. The start-up part involves tasks for the configuration and the implementation of adaptive controllers and the run-time part involves tasks for the supervision and coordination of the adaptive controller in closed-loop operation. A sensible start-up procedure is recommended and an eigenvalue analysis is shown in order to monitor the parameter estimation. To track time-varying process parameters an eigenvalue controlled forgettig factor. is introduced. Simulation and experimental results for an a.c. motor are shown, indicating an improvement of the overall control performance.

Estimation of process coefficients of time-continuous models has turned out to be a viable method to detect failures in dynamic systems at an early stage of their development. In order to plan and configure an effective diagnostic system based on estimation methods there has to be a theory which supplies the minimum of instrumentation to estimate a given set of process coefficients. The second question which arises is the identifiability of process coefficients with given models and instrumentation. Identifiability is a concept which has been extensively studied in the past from different points of view, although the problems were mostly formulated in a state space representation. Here a different procedure was chosen: The idea was to translate the information contained in a block diagram of a linear or nonlinear process model into a system of equations in such a way, that a general method for the derivation of the estimation equations could be provided. The estimated parameters are mostly nonlinear functions of the process coefficients. On the basis of the chosen representation an identifiability theorem for the process coefficients was obtained in order to have a practical method for the design of modern diagnostic systems. Several examples are given to illustrate the procedure.