Identification of Corrosion-Induced Optical Fiber Acoustic Emission Sensor Signals on the Bottom Plate of Oil Storage Tank Based on Multichannel Attention Mechanism (CBAM)

The long-term storage of corrosive chemical media in oil tanks makes them susceptible to corrosion. Therefore, it is crucial to conduct regular safety evaluations of the tank corrosion using non-destructive testing techniques such as acoustic emission. First, to address the limitations of traditional acoustic emission (AE) sensors, this study combines optical fiber acoustic emission systems with deep learning (DL) technology to effectively collect data on tank bottom plate corrosion and interference signals. Additionally, to address the issues of low identification accuracy and vulnerability to human interference in the recognition of AE signals resulting from corrosion on tank bottom plates, this article proposes a transfer learning (TL) strategy. It integrates the PReLU activation function and the channel attention mechanism (AM) convolutional block attention module (CBAM) to build the backbone network model, which enables intelligent identification of optical fiber AE signals. Moreover, the Grad-CAM++ interpretability algorithm is utilized to visualize the internal logical processing of the model, achieving initial localization of various AE signal characteristics on tank bottom plates and offering crucial insights for the on-site identification of diverse signal features. The experimental results demonstrate that the refined model significantly enhances recognition performance, achieving an average accuracy rate of 95.1%. Specifically, for corrosion-related signals, including metal dissolution and hydrogen evolution, the identification rates surpass 98%.