Diagnosis Using Fault Trees Induced from Simulated Incipient Fault Case Data

tree analysis is widely used in industry for fault diagnosis. The diagnosis of incipient or 'soft' faults is considerably more difficult than that of 'hard' faults, which is the case considered normally. A detailed fault tree model reflecting signal variations over a wide range is required in the case of soft faults. This paper presents comprehensive results describing the diagnosis of incipient faults based on fault trees derived using the IFT induction algorithm. The test system is a robot arm controlled by a pneumatic servo-mechanism. Detailed simulations using a nonlinear dynamic model were used to provide a training set of examples. The effectiveness of the diagnosis is demonstrated using comparative results based on a neural network approach.