Dissertation Title: Data-Driven Self-Adaptive Product Quality Prediction under Multi-Level Variations for Intelligent Manufacturing
Date: 2025/05/27 - 2025/05/27
Dissertation Title: Data-Driven Self-Adaptive Product Quality Prediction under Multi-Level Variations for Intelligent Manufacturing
Speaker: Tianyu Wang, Ph.D. candidate at UM-SJTU Joint Institute
Time: 4:00 PM, Tuesday, May 27, 2025 (Beijing Time)
Location: Room 503, Longbin Building
Abstract
Over the past decades, industries worldwide have embraced the opportunities of the Fourth Industrial Revolution (Industry 4.0) as well as the transformative initiatives such as 8220New Generation Artificial Intelligence8220. These frameworks have paved the way for intelligent manufacturing, which integrates big data, Artificial Intelligence (AI) and Cyber-Physical System (CPS) to bridge the physical and digital environments. A core component of CPS, Digital Twin (DT), provides real-time simulations and optimization for the modern manufacturing systems and processes, facilitating smart decision-making and high-quality production. While DT entails numerous decision tasks such as process monitoring and fault diagnosis in the practical applications, the prominent task of product quality prediction (and optimization) is considered in this dissertation since it directly affects the economic efficacy and production capacity of modern factories. Despite the significant research progress of this task in discrete manufacturing, its potential in process manufacturing, involving critical industries such as pharmaceuticals, food and renewable energy, is increasingly explored in an earlier stage. The inherent variability of process manufacturing, driven by diverse factors such as raw material fluctuations, operational changes, environmental disturbances, sensor and machine degradation, and market dynamics, poses significant challenges for maintaining stable performances on the entire factory and over a long time period. Addressing these challenges requires self-adaptive data-driven predictive models capable of automatically adapting to dynamic conditions in production. To this end, this dissertation decomposes the manufacturing variability into three logical levels (from bottom to top: product level, production line level, and plant level) and proposes systematic methods to tackle the key research issue of each level.
Biography
Tianyu Wang is a Zhiyuan Honorary PhD student at the University of Michigan-Shanghai Jiao Tong University Joint Institute. He received the Bachelor degree from the same institute in 2016. His current research focuses on self-adaptive product quality prediction and optimization towards variations in smart manufacturing systems and processes.