Machine learning techniques for damage assessment and classification: a synergistic approach

The incorporation of machine learning techniques with structural health monitoring has demonstrated a significant performance improvement, attributed to the ability of machine learning to effectively handle large datasets. This work primarily concentrates on developing machine learning models that classify different damage types of a three-dimensional frame. Dynamic analysis and classification methods are used to explore a total of eight distinct structural states. As a part of the experimental investigation, a three-dimensional frame is excited using a 40 kg payload capacity uniaxial shake table. The responses are measured using accelerometers, a key component of vibration-based non-destructive testing (NDT) for analysis purposes. Three prominent machine learning methods: KNN, decision trees, and support vector machines are employed for classification study. Various classification metrics that include accuracy, precision, recall, F1 score, and specificity are employed to evaluate the performance of the classifiers. The KNN and decision tree models showed high accuracy levels and indicated consistency across different train-test splits. Furthermore, a notable enhancement of 15% in accuracy was observed when the number of channels was increased from one to three. These findings highlight the ability and effectiveness of using ML models in vibration-based NDT and SHM for structural damage classification.