Smart sensing with edge intelligence for autonomous structural health monitoring

• Main Author: Khual, Thang Khan
• Other Authors: Fu Yuguang
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Civil engineering
• Online Access: https://hdl.handle.net/10356/172491
• Institution: Nanyang Technological University

Owing to its large contribution to the national economy, 16.8% of national GDP of Singapore[1], the construction sector plays a critical role in the transformation of Singapore into a smart nation. In fact, the construction sector has seen rapid adoption and investment in advanced technologies such as AI in the recent years. The benefits of the move have also been witnessed in many areas of the industry such as better reliability of equipment, automation of tasks and enhanced efficiency in both time and cost. However, with the use of new technologies, new challenges arise. These challenges include data privacy and limited storage capacity of edge devices, especially in the context of structural health monitoring in construction industry. In this report, a decentralized learning framework and a data classifying AI model are developed. The decentralized learning framework is based on Federated Learning (FL), a data privacy-preserving AI model training system. The AI model is based on Convolutional Neural Network concept. The framework and model are tested to understand their performance against baseline results using simulated data from a building structure. Results from testing the learning framework and model shows that the model deployed on the FL based framework has similar performance as the baseline model that uses centralized model training framework. This suggests that FL based learning frameworks and models can be the solution to overcome the data privacy concerns of AI technology for structural health monitoring application. Furthermore, the approach does not require the central server to have high data storage capacity as most activities of the AI system are carried out on edge devices. Despite its benefits of better data protection and offloading data storage to edge devices while offering comparable performance, the FL based learning framework still needs improvements. This includes flexibility for the type of AI model that can be deployed on the framework and security of the communication protocol of the framework.