Fault Identification and Classification of Spur Gearbox with Feed Forward Back Propagation Artificial Neural Network

Abstract

Summary form only given. This paper summarizes the implementation of a gear box failure classifier based on a feed forward back-propagation artificial neural network (ANN), in which four different failure conditions has been tested: gear tooth breakage, gear misalignment, pinion with face wear, and pinion piting under different load and speed conditions. A data acquisition board and an accelerometer were used to acquire the vibration signals needed to build up the database involved in training the neural network. Statistics measures like standard deviation, skew ness and kurtosis are use for time-domain analysis and preprocessing, whereas a FFT based 20 band spectrum partitions technique is used for the frequency domain, where the rms value of each band is taken in order to keep the energy shape at the spectrum peaks. Additionally, the characteristic vectors of preprocessed signals are used as the input parameters of the neural network resulting into successful failure identification and classification, which leads to a satisfactory performance of ANN in gear box failure diagnosis very suitable for this kind of tasks.