Electrical admittance-based evaluation of piezoelectric active sensor condition using k-nearest neighbors and least-squares regression

In the electromechanical impedance-based health monitoring of structures, partial failure of piezoelectric lead zirconate titanate will result in signal changes, which may cause misjudgment of the structure state. Therefore, this paper proposes an evaluation method of the sensor condition based on k-nearest neighbors (kNNs) and least-squares regression (LSR) to make monitoring more reliable. After the analysis of the signal characteristics of three structural changes and four sensor faults, the principal components (PCs) of three indices are extracted by principal component analysis. Next, the kNN classifier is trained with the data represented by PCs and then tested by tenfold cross-validation. To determine the degree of sensor faults, LSR is used to fit the damage degree laws with multivariate nonlinear equations. The results show that the kNN model trained with three PCs has higher classification accuracy than the one trained with two PCs. The accuracy of the former reaches 100%. The R2 values of damage degree regressions of four sensor faults are all greater than 0.85, and the p-values are far less than 0.05, which denotes the effectiveness of the regression in the prediction of damage degree. This method has great application potential in evaluating sensor conditions accurately and quickly.