The fault diagnosis approach for gears using multidimensional features and intelligent classifier

Gear mechanisms are an important element in a variety of industrial applications and about 80% of the breakdowns of the transmission machinery are caused by the gear failure. Efficient incipient fault detection and accurate fault diagnosis are therefore critical to machinery normal operation. A new hybrid intelligent diagnosis method is proposed in this work to identify multiple categories of gear defection. In this method, wavelet packet transform (WPT), empirical mode decomposition (EMD) and Wigner-Ville distributions (WVD), combined with autoregressive (AR) model algorithm, were performed on gear vibration signals to extract useful fault characteristic information. Then, multidimensional feature sets including energy distribution, statistical features and AR parameters were obtained to represent gear operation conditions from different perspectives. The nonlinear dimensionality reduction algorithm, i.e. isometric mapping (Isomap), was employed in statistics to mine the intrinsic structure of the feature space in a low-dimensional space, and thus to speed up the training of the probabilistic neural network (PNN) classifier and enhance its diagnosis accuracy. Experiments with different gear faults were conducted, and the vibration signals were measured under different drive speeds and loads. The analysis results indicate that the proposed method is feasible and effective in the gear multi-fault diagnosis, and the isolation of different gear conditions, including normal, single crack, compound fault of wear and spalling, etc., has been accomplished. Since the recognition results are available directly from the output of PNN, the proposed diagnosis technique provides the possibility to fulfill the automatic recognition on gear multiple faults.