Mihiar Ayoubi

Methods of fault diagnosis

Abstract This paper gives a summary of methods that can be applied to automatic fault diagnosis. In the beginning, the focus is on classification and diagnostic reasoning using fuzzy logic. Subsequently, some of the ideas which have led to the emerging neuro-fuzzy algorithms are discussed. Finally, a new neuro-fuzzy algorithm which has recently been developed, is briefly described.

Neuro-fuzzy systems for diagnosis

Abstract Knowledge-based fault detection and diagnosis is described from the analytic and heuristic symptom generation to diagnostic reasoning. The extension of the knowledge-based approach by adaptive neural networks allows us to tune the knowledge base in order to investigate undetermined parameters just as membership functions, relevance weights of antecedents and priority factors of rules. An overview of design methodologies of neuro-fuzzy systems is provided with a special focus on a hybrid neuro-fuzzy network with a neural logical operator. Finally, an application of the neuro-fuzzy system to the on-line monitoring of air pressure in vehicle wheels is described.

Fault Diagnosis with Dynamic Neural Structure and Application to a Turbocharger

Abstract An attempt has been made to establish a nonlinear dynamic time discrete neuron model, the so called Dynamic Elementary Processor (DEP). This DEP disposes of local memory, in that it has dynamic states. Based on the DEP neuron, a dynamic Multi Layer Perceptron neural net (MLP) is proposed to identify nonparametric, multi-input single-output (MISO) models for nonlinear, dynamic systems. The identified models are used to build a bank similar to observer based schemes. The output residuals between the process and the bank models are used to detect and identify a fault in the process, if it has occurred. An empirical MISO model for the turbine of a turbosupercharger was identified to demonstrate the identification ability of the proposed DEP net with real data. The fault detection scheme was successfully applied to detect and diagnose a transient fault in the turbine waste gate.

Fuzzy systems design based on a hybrid neural structure and application to the fault diagnosis of technical processes

Abstract A novel structure which models the fuzzy inference mechanism based on neural units is proposed, to combine both the adaptive feature of neural networks and the transparency of fuzzy systems. It is shown how a perceptron with a sigmoidal activity function can perform the aggregation of premise antecedents and can thus implement conjunction or disjunction operations depending on the neurons threshold. Knowledge-base parameters such as relevance weights of antecedents and priority weights of rules are introduced and discussed. The network topology is extracted by means of a coincidence learning law, the so-called Hebbian rule, in order to limit the problem of high dimensionality known by local classifiers. Two real-world problems are reported: Monitoring of the state of a turbocharger on the basis of model-based symptoms, and the supervision of air pressure in vehicle wheels, based on physically extracted symptoms.

Dynamic Neural Units for Nonlinear Dynamic Systems Identification

An attempt has been made to establish a time-discrete neuron model which is applied to build Radial Basis Function and Multilayer Perceptron networks with distributed dynamics. The well-known delta-rule is extended to the dynamic delta-rule in order to optimize network parameters. Both network types were used to identify empirical, parametrical models of a turbocharger of a Diesel engine which comply with the demanded accuracy properties to a high degree. The performance of both network types is compared according to required number of parameters, approximation accuracy and computational effort.

Application of a hybrid neuro-fuzzy system to the fault diagnosis of an automotive electromechanical actuator

Modern automatic fault detection and diagnosis methods are based on analytic and heuristic models of the process under consideration. Usually, a lot of fault symptoms can be generated using analytic symptom generation methods like parameter estimation, state estimation and parity equations as well for the evaluation of sampled input/output signals of the process. However, some relations, especially the cause-effect relations between the underlying faults and the observable symptoms, are quite difficult to be represented by analytic models. A rule-based approach is more suitable to acquire, represent and process the diagnostic knowledge base. In order to cope with uncertainty and to allow automatic knowledge extraction from experimental data, a neuro-fuzzy-structure is applied to the classification of faults, based on symptoms generated by identifying a mathematical model. The hybrid neuro-fuzzy scheme SARAH used consists of three layers corresponding to the three fuzzy inference steps. All parameters are automatically determined based on experimental data by clustering and learning. Finally, the performance of the diagnosis scheme is illustrated on the example of an automobile actuator with several different faults.

Supervision of vehicles' tyre pressures by measurement of body accelerations

Abstract Due to the rising consciousness of safety aspects, the supervision of vehicles tyre pressures is a major aspect of improved active car safety. Therefore, in this paper a method for monitoring the tyre pressures is presented, using body acceleration signals. Analyzing the frequency spectrum of the virtual transfer function between the body acceleration at the front and rear wheel on one side of the vehicle, characteristic features are generated. Thereby, external interferences on the spectrum and their influences on the symptoms are discussed. To quantify the tyre pressure, a neuro-fuzzy classification of the characteristics is applied.

Measurement and Monitoring of Pressure Curves in Diesel Engines

A method for cylinder pressure curve analysis is presented which may be used for real time control and diagnosis of Diesel engines. By subtracting the pressure samples of a firing motor from the non firing case, a difference pressure curve can be obtained which contains hidden information on internal motor conditions. To use this information, a real time data reduction has been implemented which generates symptoms like the centre of gravity or the maximum pressure from the data samples and may be processed by a classification algorithm (i. g. an artificial neural network). With the presented method it becomes possible to separate heat transfer problems from combustion or injection failures and to adjust motor characteristics properly. The validity of our approach is supported by simulation and experimental data.

Identification of nonlinear dynamic processes based on dynamic radial basis function networks

An attempts has been made to establish a discrete-time neuron model with a radial basis function. The neuron is utilized to build RBF-networks with locally distributed dynamics to identify input/output models of dynamic nonlinear processes. The adaptation algorithm which ascertains the optimal network parameters is provided. Further, an enhanced parameter estimation algorithm is derived, the so-called compound estimation procedure, which combines elaborated least squares techniques to highly decrease the training times. The proposed neural model is applied to identify black-box models of a turbocharging process within a Diesel engine. Benefits and drawbacks of the proposed neural structure are worked out.

Real Time Supervision for Diesel Engine Injection

Abstract Some methods are presented which can be used for real time diagnosis of injection and combustion in turbocharged Diesel engines. Samples of the cylinder pressure, the speed signal and samples of temperature and pressure from the turbocharger contain hidden information on internal motor conditions. To use this information, a real time data reduction has been implemented to generate characteristic fault symptoms which may then be processed by a classification algorithm, eg. an artificial neural network. With the presented methods, it becomes possible to separate injection failures from heat transfer problems and turbocharger malfunction. The validity of the presented approach is supported by simulation and experimental data.