Fault diagnosis of an automobile cylinder block with neural process of modal information

The focus of this article is fault-diagnosis of complex mechanical parts through the process of their modal information using a multi-layer perceptron (MLP), a type of artificial neural networks (ANNs). The major contribution of this work is to formulate the problem of fault diagnosis of complex mechanical parts based on their modal information so as to be solved with use of ANNs. This method consists of three major steps: (1) Extracting natural frequencies of the part with or without faults. (2) Creating the “fault signatures” by deducting the natural frequencies of some faulty specimens from the ones of the faultless part. (3) Constructing and training a mathematical model in the form of an ANN, with information obtained in previous steps, to locate (and even further characterize) the fault. The presented method was successfully adopted to estimate the location of an under-surface mechanical fault on an automobile cylinder block and is shown to have the potential to solve more sophisticated fault diagnosis problems.