Sandwich Debonding Detection and Size Estimation by Electro-mechanical Impedance Method and Machine Learning-based Data Evaluation

The present master thesis combines the electro-mechanical impedance method with artificial intelligence, in particular with machine learning. The electro-mechanical impedance method is used for structural health monitoring and is a promising approach, as it is cheap and rather simple to apply. As the calculation power of computers increases and new machine learning methods are discovered, the combination with structural health monitoring becomes a prosperous opportunity. Two machine learning models, in particular anomaly detection models, were used to identify if a damage in the sandwich structure is present. The two anomaly detection models were an isolation forest and a one-class support vector machine. The identification of the damage was for both anomaly detection methods successful when the training was done over a combination of the simulated and experimental data. The isolation forest could identify the damage of the experimental data by training only on simulated data. The size of the real-life damage was assessed with high accuracy by an artificial neural network, which was only trained on the simulated electro-mechanical impedance data. Furthermore, the artificial neural network achieved adequate results for the damage size assessment on the experimental data.