A zero-shot across tasks fault diagnosis method of bearings combined simulation model with experimental data

Zero-shot learning is a hot topic in fault diagnosis in recent years. In practice, it is difficult to collect a complete data set containing all fault categories under the same working condition, especially when the machine working condition changes. This study proposed a zero-shot across tasks fault diagnosis method based on simulation and experimental data. Firstly, the domain relationship model is constructed using fault category data and simulation data. Then, the task relationship model between several different working conditions is established by extracting the high order frequency contained in the simulation data. Finally, the new working condition simulation data and conversion data are used to fine-tune the domain relationship model to obtain the missing data of the new working condition, and thereby establish the final fault diagnosis model. The effectiveness of the proposed method is verified on three bearing data sets, and it has good diagnostic performance and can solve the zero-shot problem.