Fault diagnosis of gear box based on multi-weight neural network

Based on a new theory model-BPR (biomimetic pattern recognition), a multi-weight neural network model is implemented to recognized vibration signal of gear box in wind turbines. Because of the complex working condition of wind turbines the vibration signal tends to be non-stationary and complex the difference of the spectrum energy distribution under different loads is remarkable, which causes that there is no comparability between the extracted conventional characteristic parameters, and the change of vibration can not be discerned as a result of loads or faults. According to above characters, the information entropy, which outlines the overall statistical characteristic of the signal, was extracted as the characteristic parameter to judge the machinery state. The information entropy of the different state signals was applied as input vector to establish multi-weight neural network. Experiments results compared with traditional neural network demonstrates this new method is more effective.