1. Introduction
The lightweight steel-Ultra-High-Performance Concrete (UHPC) composite bridge deck, composed of steel deck and a thin reinforced UHPC layer through stud shear connectors, is an effective way to eliminate bridge defects, such as preventing pavement cracking and reducing fatigue cracking of the orthotropic steel deck [
1,
2,
3,
4]. In practice, however, construction of this type of novel structure is very challenging due to several reasons. First, the thickness of the UHPC layer in the composite bridge is relatively thin, i.e., 35–50 mm, and the composite slab requires a high-temperature steam-curing treatment. Second, the steel-fiber volume ratio in the structure could be up to 3.5%, which may cause steel fiber clustering. Third, the high-temperature steam curing can cause debonding between steel and the UHPC layer. Finally, the performance of the steel-UHPC composite deck system may degrade due to the effects of vehicle overloading, thermal action, and fatigue load action during operation. The debonding between the steel deck and the UHPC layer may introduce risks, such as crack-induced water invasion and distinct reduction of the shear resistance of the bridge deck system. Thus, it is of vital importance to effectively detect the interface debonding of the steel-UHPC deck system.
The piezoelectric ceramic patches (PZT) sensor, made of piezoelectric ceramics with both positive and negative piezoelectric effects, can be used simultaneously as an actuator and a sensor. Due to its lightweight characteristics, the PZT sensor can be installed on the surface of existing structures or embedded into newly-built structures for damage detection. It has been demonstrated to be particularly useful within civil engineering due to its unique features, such as active sensing, high-sensitivity, low cost, quick response, among others. In recent years, the PZT-based approach has been broadly recognized as one of the most promising, non-destructive evaluation methods for local damage identification [
5,
6].
The PZT-based damage identification methods can be classified into two categories: impedance-based method and vibration-characteristic-based wave propagation (WP) method. The impedance method aims to assess the status of a structure through the mechanic-electric coupling between the piezoelectric material and the host structure. The piezoelectric coupling properties of the piezoelectric materials can combine the structural mechanical impedance with the electric impedance of the piezoelectric materials. Thus, structural damage and condition can be realized by monitoring the change of electric impedance associated with the PZT patch installed on the surface or inside a structure. Over the past decades, many studies have been conducted on the application of the PZT impedance method for structural damage detection (e.g., [
7,
8,
9,
10,
11,
12,
13,
14]). For instance, Sun et al. [
7] proposed a frequency domain impedance-signature-based technique for health monitoring of an assembled truss structure and used PZT as integrated sensor-actuators; Liang et al. [
8] developed a coupled electro-mechanical analysis of piezoelectric ceramic actuators integrated in mechanical systems to determine the power consumption and energy transfer in electro-mechanical systems; Yang et al. [
9] conducted an experimental study on local damage detection of beams and plates using PZT and demonstrated that both the location and extent of damage can be simultaneously identified; Xu et al. [
12] investigated the structural crack damage using the impedance spectra of the PZT sensor, and presented a scalar damage metric based on the impedance spectra of the PZT sensor; and Sevillano et al. [
13] proposed an innovative hierarchical clustering analysis to obtain a set of clusters based on damage patterns obtained from the PZT sensor.
With respect to the vibration -based WP method, the piezoelectric actuators generate stress wave under external excitation, which can be received by the piezoelectric sensors. Vibration features extracted from the acquired stress wave, such as changes in signal strength, arrival time, and transfer function before and after the introduction of damage, can be used for detection of structural damage. For instance, Wang et al. [
15] used an active diagnostic technique for identifying impact damage in composite plates. This technique used a built-in network of piezoelectric actuators and sensors to generate and receive propagating stress waves over a wide range of frequencies; Roh and Chang [
16] developed a diagnostic technique to detect the location and size of anomalies in isotropic plates; Wang [
17] developed an active diagnostic system to detect embedded damage in fiber-reinforced composites and steel-reinforced concrete; and Song et al. [
18] used piezoceramic transducers for damage detection of a reinforced concrete bridge bent-cap. During the experimental test, one embedded piezoceramic patch was used as an actuator to generate high frequency waves, and the other piezoceramic patch worked as a sensor to detect propagating waves; Lim et al. [
19] performed experimental studies to investigate the application of the wave propagation method for concrete curing and monitoring of strength development; Lu et al. [
20] investigated the propagation of ultrasonic waves in rebar-reinforced concrete beams for damage detection. An experimental test demonstrated that the surface-attached PZT disks were able to detect the change in material properties due to the existence of cracking; Xu et al. [
21] proposed an active interface condition monitoring approach for concrete-filled steel tube (CFST) using functional smart aggregates as an actuator and PZT patches bonded on the surface of the steel tube as sensors.
Though researchers have conducted many studies on damage detection of different types of civil structures using piezoelectric impedance technology, wave propagation method, and clustering algorithm, there are limited studies on integration of the above proposed technologies for nondestructive evaluation. The lightweight steel-UHPC composite bridge deck system, as an effective and novel structural form to prevent bridge pavement cracking and reduce fatigue cracking of the orthotropic steel deck, faces the challenge of identifying the interfacial debonding condition between the steel deck and UHPC overlay. To the best knowledge of the authors, detection of the interfacial debonding of the steel-UHPC composite slab using PZT methods has not been studied.
In this study, both impedance analysis and wave propagation method are employed to extract the debonding features of steel-UHPC composite slab with different preset debonding defects. Additionally, an improved PSO-k-means clustering algorithm is adopted to obtain the clustering centers of the feature data set, and Mahalanobis distance is then used to distinguish the debonding degree of the deck system. The proposed methodology is validated through experimental tests on two steel-UHPC composite slabs and a conventional steel-concrete composite slab with different artificial debonding defects.
5. Concluding Remarks
This paper presents a detection method integrating PZT-based impedance analysis, wave propagation technique, PSO-k-means algorithm and Mahalanobis distance to identify the debonding defects of steel-UHPC composite deck. Both impedance analysis and wave propagation method are employed to extract debonding features of the steel-UHPC composite slab with debonding defect in different sizes and thicknesses. An improved PSO-k-means clustering algorithm is then used to obtain the clustering centers of the feature data set, and the Mahalanobis distance is finally used to distinguish the debonding degree of the samples. The proposed methodology is validated through experimental tests on two steel-UHPC composite slabs and a conventional steel-concrete composite slab with different debonding defects.
The impedance test indicates that at two higher frequency ranges (i.e., 6000–8000 kHz, 10,000–12,000 kHz), the significant fluctuations and shifts of dominant peaks in impedance curves are observed. The RMSD is employed as an evaluation indicator of the debonding defect. Specially, as the debonding size of steel-UHPC interface reaches 50 mm × 50 mm, the RMSD damage index provides an effective tool to detect the debonding defect. Moreover, in such debonding size and appropriate frequency ranges, the RMSD damage index increases with the increase in debonding thickness.
The experimental tests indicate that the WP technique exhibits good performance in identifying defects of at least 50 mm × 50 mm in size and is capable of identifying early age interfacial debonding defects with minor thickness, e.g., 1.0 mm. In addition, the excitation frequency has an effect on the output voltage amplitude. The output voltage amplitudes exhibit distinct difference; however, under the excitation frequency of 10 kHz and 15 kHz, the trend of the effect associated with the interface debonding size and thickness on output voltage amplitudes is consistent.
The training and testing samples with four features extracted from impedance analysis and wave propagation method are considered for further PSO-k-means clustering analysis. The confusion matrix is employed to evaluate the overall identification accuracies of the PSO-k-means algorithm and Mahalanobis distance. It is observed that the debonding status with thickness of 0 mm and 3 mm can be effectively classified. However, the feature characteristics of the testing samples with debonding thickness of 1 mm and 2 mm are relatively close, which could result in misclassification. In addition, the wave propagation method can provide better classification results than the impedance-based PZT method. The averaged identification rates attain 86.3% when all four features are considered.