Performance Evaluation of Structural Health Monitoring Anomaly Data Processing Algorithms in Resource-Constrained Edge Computing

Edge computing plays an important role in structural health monitoring (SHM) for transportation infrastructure because it enables local data processing and low-latency decision support. However, SHM systems encounter challenges due to data anomalies caused by sensor faults, transmission errors, or irregular structural behavior. This study evaluates the deployment performance of a SHM anomaly data-processing workflow on resource-constrained edge devices. The workflow includes data cleaning, response separation, and anomaly detection. Missing data are processed using cubic spline interpolation. Jump points and drift are corrected using the Laida criterion, and noise is reduced using wavelet threshold denoising. Response separation is then performed using the detrending method based on time windows, the 3σ criterion, or wavelet packet decomposition. Anomaly detection is then performed using autoregressive integrated moving average with explanatory variables, support vector machines, and recurrent neural networks. Simulation data and field monitoring data from bridge displacement and highway pavement strain are used to evaluate the workflow. The evaluation focuses on runtime, memory usage, central processing unit usage, and a data-processing throughput proxy. This analysis helps to understand the performance and trade-offs of algorithms on edge devices under the resource constraints typical of SHM applications.