Soft computing tools for machine diagnosing

This work is aimed at creating soft computing tools for machine diagnosing systems. There are some problems with interpretating measured data in these systems. To overcome the problems with a big number of information in a diagnosing system, a neural pre-processor was proposed. A neural network can be used for reducing the size of analysed features. The fault detection and isolation is difficult due to context and broaden relations between measured data and a machine state. Soft computing methods are helpful in solving such problems. Artificial neural networks and fuzzy logic systems were used in these studies. An approximation of the unknown diagnostic relations symptom-state was done by both created tools. The only information about these relations were hidden in measured data that illustrate an expert knowledge formulated in a natural language. Such a form of information is the basis of constructing neural networks and fuzzy systems adequatly. The case study was fault detection of a high power fan. The working correctness of soft computing tools, presented in this work, was examined in the context of results obtained by utilisation of pattern recognition methods. The comparison of their performance speed, noise robustness and early detection of failure was also made.