Bearing Fault Classification Using Improved Antlion Optimizer and Extreme Learning Machine

Abstract

Bearing is an important part of rotating machinery, and its early fault diagnosis and accurate classification have always been difficult in engineering application. At present, the models based on the fusion of various optimization algorithms and neural networks have become one of the emerging techniques for accurate fault identification. Firstly, an improved antlion optimizer (ALO) algorithm based on estimation of distribution algorithm (EDA) and variable-step Lévy flight strategy, abbreviated as ELALO, is proposed as a new bionic intelligence. During the initialization of population, the individuals with poor fitness are redistributed by the Gaussian probability model. In view of the stagnation of iteration, Lévy flight strategy is introduced and the adaptive change of disturbance step length is controlled. Experimental results on 4 benchmark functions show that the novel ELALO can effectively improve the solution accuracy and convergence speed, compared with the original ALO. Secondly, in order to solve the disadvantage that extreme learning machine (ELM) network is easy to fall into local optimization, this ELALO algorithm is used to initialize the weights and thresholds of its network and to form the new pattern recognition model, ELALO-ELM. Finally, the bearing data of 8 patterns from Western Reserve University are decomposed by local mean decomposition (LMD), and then the symbolic entropy (SE) of the first three product function (PF) components signals is extracted and used as the input eigenvectors. Compared with the standard ELM and ALO-ELM models, the ELALO-ELM model has better generalization and stronger robustness and it can effectively improve the efficiency of network training and the accuracy of early fault pattern classification in bearing fault diagnosis. The new ELALO-ELM model can also be used for other difficult classification problems.