Feature fusion of mechanical faults based on evolutionary computation

Automatically and accurately detecting the type and severity of faults in a machine is very important to condition monitoring as well as quality inspection. In this paper, we propose an evolutionary algorithm for mechanical fault recognition based on genetic programming (GP) and information fusion techniques. The improved approach can generate composite feature vectors with more information about machine faults, which are used in automatic recognition of different mechanical faults without prior knowledge of the machine itself. In addition, normalisation and bootstrap are used as a pre-processing approach for primitive features. We also compare the recognition performance of different composite feature vectors and show that our proposed approach can perform the task of mechanical fault recognition more effectively.