Gearbox pitting detection using linear discriminant analysis and distance preserving self-organizing map

Many intelligent learning methods have been successfully applied in the gearbox fault diagnosis. Self-organizing map (SOM) is one of such learning methods which have been used effectively as it preserves the topological relationships of the data. A novel distance preserving SOM is investigated in mechanical fault diagnosis, and a LDA-DPSOM (linear discrimination analysis and distance preserving SOM) based diagnosis method is presented for gear incipient fault detection. Firstly, LDA is used to realize feature selection of the data set, so the dimension of produced data is much fewer than that of original data. Then the DPSOM method is applied to classifying the selected data and visualizing the classification result. Experiment results indicate the effectiveness of LDA-DPSOM for gearbox incipient fault diagnosis.