Application of the self-organizing map to manual automotive transmission

Abstract

In recent years, research in the gearbox diagnostics has been done on the computation of threshold values for existing condition features, enabling the use of simple classification methods. This paper describes the application of advanced classification methods in gearbox diagnostics. Time domain signal evaluation is transformed to a vector classification problem. A vector composed of three amplitude features (the root mean square, skewness and kurtosis) of the synchronously averaged vibration signal, is computed for each tested gearbox. The classification is based on the self-organizing feature map algorithm (Kohonen neural network). A database containing vibration signals from four manual automotive transmissions has been used to test the performance of the proposed system. The results obtained using this approach, demonstrate the ability to discriminate among various types of fault.