Levenberg-Marquardt neural network for gear fault diagnosis

In this study we are trying with the Levenberg-Marquardt neural network model to the problem of gear fault diagnosis. By using second derivative information, the network convergence speed is promoted and the generalization performance is enhanced. Taking a certain gearbox fault signal acquisition experimental system for instance, Matlab software and its neural network toolbox are used to model and simulate. The simulation result shows that Levenberg-Marquardt neural network has a good performance for the common gear fault diagnosis and it can identify various types of faults stably and accurately. Furthermore, compared with conventional BP neural network, the Levenberg-Marquardt neural network reduces training epochs and promotes diagnosis accuracy.