Fault diagnosis method and experimental research of reciprocating seal based on CFD-GAN-AE

Hydraulic impactors are crucial for oil and gas exploration, but seal failure is a common issue, having an effective technique for diagnosing sealing faults can provide dependable operational and maintenance assistance for hydraulic impactors. However, identifying wear failures is challenging and there is limited data available, there has been significant interest in intelligent defect diagnosis technology that is based on deep learning in recent years. Therefore, we propose a method to enhance the data and identify faults through deep learning. Initially, the computer fluid dynamics method was used to simulate seal leakage and determine whether factors such as pressure can indicate varying levels of leaking in the seal, this approach provides a theoretical foundation for signal gathering experiments. Next, the empirical mode decomposition approach is used to separate the non-smooth pressure signal from the seal experiment, revealing fault features that indicate the extent of leakage. Finally, the improved generative adversarial network method is suggested to balance imbalanced samples by utilizing the sample overlap rate, it is paired with the auto-encoder algorithm to categorize different levels of leakage. Furthermore, a comparative analysis is conducted between the proposed methodology and several classical fault diagnosis methods. This work investigates seal damage through the lens of computational fluid dynamics and the fault identification of uneven seal samples is accomplished.