Edge AI for condition monitoring

• Main Author: Hu, Zinan
• Other Authors: Hung Dinh Nguyen
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Engineering; Digital IC design; AI
• Online Access: https://hdl.handle.net/10356/175461
• Institution: Nanyang Technological University

Condition monitoring plays a critical role in safeguarding the operational integrity and efficiency of edge devices such as batteries, automobiles, and mobile phones. Presently, prevailing solutions for condition monitoring heavily rely on Artificial Intelligence (AI), with AI models trained and deployed in cloud-based or centralized computing systems. To solve the problems, this work emphasizes the Edge AI paradigm in contrast, where in AI models are executed on edge devices in close proximity to the batteries or automobiles. This approach facilitates expedited remedial actions and reduces the cost associated with data transmission. In this dissertation, I developed an accelerator for the image recognition algorithm LeNet-5 based on FPGA implementation. Subsequently, I configured a complete Edge AI system for Condition Monitoring using this accelerator. This system is capable of reading data from external Flash memory chip into internal FIFO, processing the data within the accelerator, and ultimately outputting the possibilities of the content in the image, which are the possibilities of the digits from 0 to 9. The Edge AI for Condition Monitoring System’s ability to extract information from images and accurately identify the digits within them demonstrates that embedding complex AI algorithms into hardware using ASICs is a reliable and promising direction for development.