Induction Motor Failure Identification Based on Multiscale Acoustic Entropy Feature Selection and Hierarchical Adaptive Neuro-Fuzzy Inference System With Localized Recurrent Input

As induction motor is one of the most preponderant driving equipment in modern industrial systems, timely detection and maintenance of motor failures can efficaciously circumvent economic losses and catastrophic accidents. Though vast majority of machinery fault diagnosis approaches are based on vibration analysis, there may be circumstances that it is impossible to install vibratory sensors in some sophisticated equipment or harsh environments. In view of this, a fault diagnosis method for induction motor based on multiscale acoustic entropy feature selection and hierarchical adaptive neuro-fuzzy inference system (HANFIS) with localized recurrent input is presented in this article. First, multiscale entropy features, which can reflect the health state of the running motor, are extracted from the collected acoustic signals. Subsequently, to avoid the dimensionality disaster problem and remove irrelevant features, the Relief-F algorithm is used for feature selection. Next, based on the fuzzy formalisms, a newly designed HANFIS is utilized for motor fault identification, where the localized recurrent input mechanism allows for injecting important features again into the model to prevent it from forgetting significant discriminative information, to further promote the overall diagnostic accuracy. Finally, experiments on two acoustic datasets of induction motor and comparisons with existing approaches sufficiently verify the effectiveness and superiority of the proposed method.