Page 15 - Demo

13Decentralised wireless monitoringTraditional SHM uses a centralised approach, meaning all data from sensors are sent to one central server. However, thousands of sensors collecting data every split second generate a huge volume of data traffic. The scale of a long-span bridge also means that data need to travel a long distance to reach the server. These factors can overwhelm the network and cause delays or even failures. The capacity of the central server also becomes a hurdle that limits scalability.The team%u2019s intelligent SHM system is the first of its kind to apply 5G networks and edge computing to bridge health monitoring. It adopts a decentralised approach %u2013 data collected by sensors are partly processed locally and then sent to the central server. This greatly reduces the amount of data transmitted, saving network resources and easing the burden on the server. The system uses AI and machine learning to detect anomalies on edge computing boards. This means data are screened locally and only signs of unusual activities that demand attention are sent to the server to alert users. The use of 5G networks further ensures high-speed data delivery with minimal delay or loss.Meanwhile, the machine learning and smart anomaly detection processes need to train a large amount of labelled data, which are not available for a new bridge. %u201cTo tackle this, we devised a source-free domain adaptation algorithm. This allows us to transfer models trained on other bridges by using a robust self-training mechanism and a self-knowledge distillation strategy,%u201d Prof. Xia explains. %u201cThe algorithm means we don%u2019t need to build source data from scratch, so the system can be used immediately%u201d. Shenzhen Chuangye BridgeFOREVER YOUNG