Combining bootstrap and genetic programming for feature discovery in diesel engine diagnosis
Ruishuang Sun, Fugee Tsung, LS Qu · 2004
This paper considers feature discovery in fault diagnosis using bootstrap processing and genetic programming. Bootstrap is one of the powerful computer-simulated statistical techniques. It integrates classical statistics with the numerical calculations of a computer. This paper uses bootstrap to preprocess the operational data acquired from a running machine. Then, genetic programming evolves with the preprocessed data samples. The main aim is to discover an efficient tree-like structure on the basis of a group of simple initial candidate features. The best compound feature found by genetic programming can discriminate among the four kinds of commonly operating statuses of the machine. This novel approach is demonstrated by fault diagnosis of the fuel system in a diesel engine. Significance: The proposed feature discovery and analysis approach, which has been tested in a diesel engine fuel system, will significantly enhance the classification performance of fault diagnosis by combing bootstrap processing, genetic programming, and engineering knowledge.