Derivation of Fuzzy Diagnosis Rules for Multifunctional Fuel Cell Systems


Elektrische Energieversorgung und Brennstoffzelle



Art der Publikation:



Christian Modest, F. Thielecke


PHM Society: Proceedings of the 2nd European Conference of the Prognostics and Health Management Society 2014, Juli 2014

This paper presents a model-based approach for the derivation of fuzzy diagnosis rules. These are used to classify data of faulty system behavior in order to identify root causes. The data is gained from an extended simulation model of a multi-functional fuel cell system for aircraft use. Faulty behavior is implemented into each component and a bottom up simulation is carried out. The data gained is classified according to root causes. This means that each data vector is assigned to a class representing one type of simulated fault. The classified data is then fed into an evolutionary optimization procedure. There it is weighted and separated into training and validation data. Inside the optimization procedure, the structure of the fuzzy diagnosis rule is represented by a chromosome that has a discrete and a real valued part. The discrete part describes the selection of a signal and the real valued part states parameters of the membership function for each signal. Based on training data, a genetic algorithm optimizes both parts and a set of optimal binary and real valued parameters is gained. By that, one fuzzy diagnosis rule at a time is identified that best fits a set of fitness functions. On basis of this rule, weights of the training data are updated afterwards. This is done in order to guide the genetic algorithm in the next run to data vectors that are not covered effectively yet. Each run of the algorithm gives a new fuzzy diagnosis rule. The performance of the set of all rules that are gained so far is evaluated by use of validation data. Subsequently, a new run is started. This process continues until a stop criterion is reached. A set of optimal fuzzy diagnosis rules is gained in the end.