Generation of Fault Trees from Simulated Incipient Fault Case Data

Fault tree analysis is widely used in industry in fault diagnosis. The diagnosis of incipient or ‘soft’ faults is considerably more difficult than of ‘hard’ faults, which is the situation considered normally. A detailed fault tree model reflecting signal variations over wide range is required for diagnosing such soft faults. This paper describes the investigation of a machine learning method for the automatic generation of fault trees for incipient faults. Features based on the FFT (Fast Fourier Transform) of the time response simulations are used are used to provide a training set of examples comprising records of fault types, severity and feature list. The algorithm presented, called IFT, is derived from the ID3 algorithm for the induction of decision trees. A significant aspect of this approach is that it does not require any detailed knowledge or analysis of the application system. All that is needed is a ‘black-box’ model of the system; i.e. knowledge of what faults arise from measurable quantities taking on particular values. The proposed procedure is illustrated using detailed simulation results for a servomechanism typically found in machine tool applications and the results to date indicate the feasibility of the approach.