Improved Constructive Morphological Neural Network for Fault Diagnosis of Gearbox

Abstract

Gearbox fault diagnosis is mainly based on artificial neural networks, but the accuracy is not guaranteed. Given the limitations of the constructive morphological neural network (CMNN) algorithm, we probed into the CMNN model and its deficiency, and proposed an improved algorithm for gearbox fault diagnosis. The recursive call of the function was used to avoid the local optimal solution of the network, while the inclusive measure was used to remove the redundancy of hyper-box clusters. The hyper-box clusters were obviously more streamlined with higher classification efficiency. Comparison among three algorithms showed the improved CMNN classified that was based on the maximum membership degree principle of inclusive measure. Experimental results confirm the effectiveness of the improved CMNN in gearbox fault diagnosis.