Hardware-software co-design for large-scale structural health monitoring

Modern Structural Health Monitoring (SHM) systems currently utilize a combination of low-cost, low-energy sensors and processing units to monitor the conditions of target facilities. However, utilizing a dense deployment of sensors generates a significant volume of data that must be transmitted to the cloud, requiring high bandwidth and consuming substantial power, particularly when using wireless protocols. The current cloud-based solutions cannot scale if the raw data has to be collected from thousands of buildings. To optimize the energy budget and generated data of the monitoring system, it is crucial to reduce the size of the raw data near the sensors at the edge. However, existing compression techniques at the edge suffer from a trade-off between compression and accuracy and long latency, resulting in high energy consumption. This work presents a full-stack deployment of efficient and scalable data reduction and anomaly detection pipelines for SHM systems, which does not require transmitting raw data to the cloud but relies on edge computation. Three lightweight algorithmic approaches of anomaly detection are benchmarked, i.e., Principal Component Analysis (PCA), Fully-Connected AutoEncoder (FC-AE), and Convolutional AutoEncoder (C-AE) implemented on the SHM node. By doing so, network traffic decreases by ≈ 8 ・ 105×, from 780KB/hour to less than 10 Bytes/hour for a single node installation, and minimizes network and cloud resource utilization, enabling the scaling of the monitoring infrastructure. Finally, we propose our last SHM application to take a step forward in Traffic Load Estimation via the SHM system. This novel signal processing and classification pipeline is able to differentiate vehicles into three categories: light, i.e., less than 10 tons; heavy, i.e., between 10 and 30 tons; and super heavy, i.e., above 30 tons, using only features extracted from vibration data with an accuracy of 96%, utilizing the mean-shift, an unsupervised clustering model.