Structural Health Monitoring of Similar Gantry Crane Based on Federated Learning Algorithm

When using Internet of Things (IoT) technology for health monitoring of similar batches of identical gantry cranes, uploading all their structural data to a cloud server would result in significant bandwidth waste and reduced real-time performance. Additionally, due to different usage scenarios and habits among similar batches of identical gantry cranes, existing algorithms can only monitor specific gantry cranes after training is completed. For the health monitoring of identical gantry cranes, this study employed the federated averaging algorithm (FedAvg) and XGBoost algorithm in an edge gateway to detect damage types in gantry crane structures. A small-scale gantry crane was used to simulate various connection damages. A IoT gateway was designed to monitor the structural condition of facilities. Upon receiving structural data, the gateway locally performed anomaly detection. When the edge gateway detected anomalies in the time series, it utilized the XGBoost algorithm for anomaly type classification. Additionally, the edge gateway employed the FedAvg to mitigate the impact of non-independent and imbalanced data caused by different environments and usage patterns. Ultimately, a unified anomaly detection model was obtained to facilitate subsequent usage of identical gantry cranes.