The detection of structural damage based on the electromechanical impedance technique is proven to have a greater sensitivity to structures based on baseline impedance signature data. These baseline signature data are also used to quantify impedance signatures statistically. However, it poses significant drawbacks when implemented on existing infrastructures and early strength gain monitoring. This study investigates and aims to utilize various machine learning techniques as effective regression approaches to predict impedance signatures of construction steel for different tensile load actions at lower ranges. In recent times, machine learning approaches have been applied to diversified domains for efficient prediction and forecasting. Various machine learning algorithms were used in this study, i.e., logistic regression, ridge regression, lasso regression, k nearest neighbors (KNN), decision tree, support vector machine (SVM), and ensemble techniques such as random forest, XGBoost, AdaBoost. The experimental admittance signatures used to build these models were collected using three different multi-piezo configurations, i.e., surface bonded, clamped, and metal wire based, to monitor the tensile pull action of construction steel rebar. The models were developed using optimized hyperparameters for better model tuning. Stratified splitting was applied to validate the model. All these models were evaluated, and their efficiencies were compared using model evaluation metrics like root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE). It was found that all the ensemble machine-learning models outperformed all other models. The MAPE value is superior for all the ML models, indicating over 90% accuracy in predicting futuristic EMI signals. This work significantly contributes to structural health monitoring (SHM) for predicting non-bonded and reusable piezo impedance data. These findings will help researchers to predict the multi-sensing impedance data, especially for complex structures, and utilize them for the futuristic health monitoring of infrastructure systems.
Read full abstract