Early warning study of acoustic emission signals from cathode carbon block based on machine learning

The electrolyte erodes the cathode carbon block, thereby impacting the electrolysis process. To identify the degree of damage to the cathode carbon block for early warning of cathode carbon block rupture, we collected acoustic emission data and stress data during the compression process of the carbon block at three different rates. We calculated various metrics from these data, including energy, counts, and rise time. The correlation coefficient as well as the pinch angle were then extracted from these metrics as characteristic parameters, discovering that specific characteristic parameters possessed clear precursory information before the carbon block’s rupture, which remained consistent at varying rates. Using these feature parameters as training data, the model was trained on acoustic emission signals at a specific loading rate using machine learning and validated on signals at other loading rates. The validation results show that the model has excellent generalisation performance and robustness, proving that the extracted parameters can characterise the damage patterns of carbon blocks at different rates. The features were also mixed to explore the effect of feature fusion on model performance. The results reveal that the model incorporating numerous characteristics has the best classification performance, with an average accuracy of 98.08%.