Pattern recognition of acoustic emission signals by Q235 steel corrosion in marine environment

To overcome the limitations of traditional monitoring methods, which are restricted to periodic inspections, this study proposes a real-time method for identifying metal corrosion damage patterns to monitor the condition of Q235 steel corrosion based on acoustic emission (AE). Firstly, AE technology was utilized to monitor the corrosion process of Q235 steel plates in simulated industrial marine environment in real-time. Wavelet packet energy spectrum coefficients, closely related to the damage mechanism, were extracted from the acquired signals. A feature matrix was then constructed using principal component analysis (PCA) to eliminate redundant information and enhance computational efficiency. The K-means clustering algorithm was then applied to classify the AE signals into three classifications: the signals of mode 1 correspond to bubble rupture, the signals of mode 2 to pit growth and expansion, and the signals of mode 3 to the detachment of corrosion products and oxide film rupture. A damage pattern recognition model based on a convolutional neural network (CNN) was developed, enabling the real-time recognition of other unknown AE signals generating during the corrosion process of Q235 steel, and it exhibited satisfactory performance in accurately identifying corrosion-related acoustic emission patterns.