Multi-Modal Temperature Sensing using Electro-Mechanical Impedance (EMI) for Structural Health Monitoring

In this paper, we propose a new multimodal sensing approach in which Electromechanical impedance (EMI) signatures are used to sense temperature in addition to structural health monitoring for which the technique is commonly used. Here, we use machine learning to estimate temperature differences between signatures, enabling the EMI sensor to serve not only as a corrosion sensor but also as a temperature sensor. In this study, we collected two comprehensive datasets, each consisting of nine EMI signatures from a steel rod in a healthy and damaged state. The datasets span temperatures ranging from -10°C to +30°C with steps of 5°C. A peak finding algorithm was used to preprocess the datasets and several machine learning models were implemented via SciKit-learn, trained, and evaluated. The overall best model was the Support Vector Regressor using radial basis function kernel, which achieved a mean squared error (MSE) of 0.89 on the healthy data set and 1.24 on the damaged data set.