Constructional material deterioration and member damage can cause changes in the dynamic characteristics of bridge structures, and such changes can be tracked in the responses of passing vehicles via the vehicle-bridge interaction (VBI). Though data-driven methods have shown promising results in damage inspection for drive-by methods, there is still much room for improvement in their performance. Given this background, this paper proposes a novel time-domain signal processing algorithm for the raw vehicle acceleration data of data-driven drive-by inspection methods. To achieve the best data processing performance, an optimizing strategy is designed to automatically search for the optimal parameters, tuning the algorithm. The proposed method intentionally overcomes the difficulties in the application of drive-by methods, such as measurement noise, speed variance, and enormous data volumes. Meanwhile, the use of this method can greatly improve the accuracy and efficiency of Machine Learning (ML) models in vehicle-based damage detection. It consists of a filtering process to denoise the data, a pooling process to reduce data redundancy, and an optimizing procedure to maximize algorithm performance. A dataset is obtained to validate the proposed algorithm through laboratory experiments with a scale truck model and a steel beam. The results show that, compared to using raw data, the present algorithm can increase the average accuracy by 12.2–15.0%, and the average efficiency by 35.7–96.7% for different damaged cases and ML models. Additionally, the functions of filtering and pooling operations, the influence of window function parameters, as well as the performance of different sensor locations, are also investigated in the paper. The goal is to present a signal processing algorithm for data-driven drive-by inspection methods to improve their detection performance of bridge damage caused by material deterioration or structural change.

Recent decades have witnessed a rise in interest in bridge health monitoring utilizing the vibrations of passing vehicles. However, existing studies commonly rely on constant speeds or tuning vehicular parameters, making their methods challenging to be used in practical engineering applications. Additionally, recent studies on the data-driven approach usually need labeled data for damage scenarios. Still, getting these labels in engineering is difficult or even impractical because the bridge is typically in a healthy state. This paper proposes a novel, damaged-label-free, machine-learning-based, indirect bridge-health monitoring method named the assumption accuracy method (A${}^2$M). Initially, the raw frequency responses of the vehicle are employed to train a classifier, and K-folder cross-validation accuracy scores are then used to calculate a threshold to specify the bridge’s health state. Compared to merely focusing on low-band frequency responses (0–50 Hz), utilizing full-band vehicle responses can significantly improve the accuracy, meaning that the bridge’s dynamic information exists in the higher frequency ranges and can contribute to detecting bridge damage. However, raw frequency responses are generally in a high-dimensional space, and the number of features is much greater than that of samples. To represent the frequency responses via latent representations in a low-dimension space, appropriate dimension-reduction techniques are therefore, needed. It was found that principal component analysis (PCA) and Mel-frequency cepstral coefficients (MFCCs) are suitable for the aforementioned issue, and MFCCs are more damage-sensitive. When the bridge is in a healthy condition, the accuracy values obtained using MFCCs are primarily dispersed around 0.5, but following the occurrence of damage, they increased significantly to 0.89–1.0 in this study.

Currently, machine learning (ML) methods are widely adopted in structural health monitoring (SHM), yet they are still mostly black boxes. On the other hand, given the significant responsibility associated with SHM, understanding the rationale behind it is critically important. In some cases, even experienced experts have difficulties finding evidence related to structural integrity within complex structural signals. Thus, solely relying on these black-box SHM systems carries inherent risks. Trustworthiness is the key for decision-makers when planning to act on predictions or deciding whether to deploy a new model. This understanding can also offer insights about the models, transforming untrustworthy models or predictions into reliable ones. The indirect SHM method using passing vehicles, an emerging technique in the past two decades, offers a rapid and cost-effective solution for bridge monitoring. Its signal components are affected by factors such as vehicle dynamics and road roughness, making them more complex than those in the direct method. Although ML methods have shown promising results in this domain, their results require further explanation. In this work, SHAP (SHapley Additive exPlanation) tools are utilized to interpret the result prediction of ML methods in indirect SHM. The trustworthiness of models is demonstrated through simulation databases: deciding whether a prediction should be trusted, choosing between models, and determining why a classifier should not be trusted.