Machine learning method for damage classification based on acoustic emission waveform analysis in composite lap bar

Carbon fiber reinforced polymer (CFRP), known for their high strength, low density, and excellent corrosion resistance, are widely used in industries such as aerospace, automotive, and wind energy. In recent years, with the growing demand for lightweight solutions in the amusement ride industry, CFRP has gradually been used in non-primary load-bearing components. The lap bar, as a critical component used to secure passengers, has become a primary focus for lightweight design. This paper presents a preliminary study on failure analysis of a composite lap bar using acoustic emission (AE) and machine learning. The main purpose is to analyze the suitability of the prepared composite lap bar in a operational conditions using a classification model. The main challenge, however, is to be able to extract valid descriptors of the damage mechanism from the acquired AE signals. The damage modes of the basic units of the composite lap bar were first characterized individually and information of Hilbert marginal energy spectrum (HMES) about the AE signal associated with each damage mechanism was collected. These spectral features and parameters were then correlated and that is used as a dataset to train the model based on k-nearest neighbor (KNN) algorithm. The model achieved an accuracy of 92% through cross-validation. Then a destructive test was conducted on the composite lap bar, and the failure process was monitored using the AE technique. The acquired AE signals were identified by the classification model. This analysis provides information on the damage process of composite lap bar at different loading stages, with matrix cracking being the more common damage mechanisms. Additionally, the microanalysis of the fracture surface also verified the effectiveness of the classification model. Meanwhile, supervised machine learning shows its potential in handling multi-dimensional data.