PATTERN RECOGNITION IN MAINTAINING RELIABILITY

Abstract

In the paper a feedback control scheme for running and maintenance of industrial equipments is developed, where only input and output signals of the control loop are observable, the actual state of the system will be estimated from the output signals using pattern recognition. Technical and statistical expertise in the control loop are separated, therefore appropriate statistical tools and problems (classification, feature extraction, life statistics) can be individually accessed. The phenomenon of aging is modelled by random transitions among parameter classes with different failure rates. Optimization is reduced to a Markovian decision process. Concrete technical examples illustrate the relevance of the model.