Intelligent identification of acoustic emission internal leakage in spring full-open safety valve in complex environments using BTPAN

As a key safety pressure relief device in the pressure pipeline network, the spring full-open safety valve is widely used in the fields of petroleum, chemical industry, and energy. However, with the extension of the operating time, the continuous scouring of the medium can easily cause various forms of damage to the valve seat sealing surface, thereby leading to leakage from the inlet side to the outlet side of the valve cavity. The complex application environment causes the acoustic emission (AE) signal to often contain various background noises, making the original signal complex and messy, which in turn affects the detection accuracy of the traditional internal leakage identification model. To address this problem, this study proposes an internal leakage detection model based on Bayesian Temporal Pyramid Attention Network (BTPAN). This method realises end-to-end AE signal detection, directly inputting the one-dimensional signal as the input source into the BTPAN model, introduces the improved residual block for multi-scale feature extraction , and combines the MSCNN, TPP, BNN to not only improve the accuracy of the model in classification tasks, but also significantly enhance the adaptability to complex and dynamically changing signals. Experiments demonstrate that BTPAN achieving 98.7% accuracy in internal leakage detection.