Multi-fault diagnosis for gearboxes based on multi-task deep learning

The field of mechanical fault diagnosis enters a big data era, and the deep learning achieves fruitful results in mechanical big data processing with its powerful adaptive feature extraction and classification capabilities. However, this method is used in a single-label system to diagnose a single target fault. Under the background of big data, the single-label system not only cuts apart connections among different target faults of mechanical equipment, but also is difficult to fully describe lots of equipment fault state information, such as, fault location, type, and degree, etc. Here, a diagnosis method based on the multi-task deep learning model was proposed to simultaneously diagnose faults of bearing and gear in gearbox. It was shown that with this method, features of different targets can adaptively be extracted from the same signal, and then these features are used to perform fault diagnosis through a separate task layer. The test results showed that the proposed method realizes simultaneous correct diagnosis of bearing and gear different faults in gearbox under multiple working conditions and a large number of samples. © 2019, Editorial Office of Journal of Vibration and Shock. All right reserved.