Multi-label gear composite fault diagnosis based on ML-MCNN

Aiming at practical engineering scenarios where composite faults of gearboxes interfere with each other and it is difficult to identify which single type of faults are specifically included in the composite faults, a proposed multi-label composite fault classification model based on multi-channel convolutional neural network is proposed (ML-MCNN). First of all, for the composite faults meshed by a single fault, to better obtain the composite fault characteristics, two sensors were used to collect gear composite fault acceleration data, and multiple channels of Convolutive Neural Networks (CNN) were used to extract features from the collected time domain data and the corresponding frequency domain data, and then the output features were fused. The output features are then fused; secondly, the types of composite faults are obtained by calculating the cosine similarity between the extracted features of the model and the fault attribute labels; finally, the proposed method is validated through several sets of experiments. The experimental results show that compared with other intelligent diagnostic models, the average diagnostic accuracy of ML-MCNN for multi-label gearbox composite faults reaches 99.72%, which has a higher accuracy rate and can accurately identify the types of single faults contained in the composite faults.