Research
Dr. Du's Research Group focus on an Interdisciplinary Framework:
Fusion of Domain Knowledge, Data Science, Optimization, Control, and Computational Science
1. Industrial sensing data analytics for change point detection, anomaly detection, process monitoring, prognostics, control, and fault diagnosis:
The rapid advancement in sensing and computing technologies has resulted in unprecedented data-rich environments in smart manufacturing systems, which brings great opportunities for quality, productivity, and efficiency improvements in smart manufacturing systems. However, the collected sensing data are usually high-dimensional, heterogeneous, noisy, and high volume. A significant challenge is how to develop advanced data analytics methodologies to effectively extract the important knowledge and information from those sensing data to advance the real-time monitoring of manufacturing system operations, accurate change detection and prognostics of manufacturing system status, automatic quality inspection and anomaly detection of final products, quick diagnosis of root causes, predictive control and adjustments for error compensation and variation reduction. To deal with the research challenge, we believe the interdisciplinary research that systematically incorporates domain knowledge, advanced data science models, optimization, and control is the key to success. Specifically, our research focuses on the following two areas.
(1) Profile data or functional curve analytics for change point detection, process monitoring, quality inspection, prognostics, and fault diagnosis.
Methodologies: Bayesian statistics, state-space modeling, data fusion, deep learning, reinforcement learning, sparse learning with domain knowledge
Applications: pipe tightening, solar cell manufacturing, semiconductor manufacturing, wind turbine, aircraft engines
Related publications: J1, J2, J3, J4, J5, J8, J10, J12, J13, J15, J17 in the publication list
(2) High-density 3D point cloud data modeling and analytics for anomaly detection, classification, quality inspection, and efficient sampling.
Methodologies: probabilistic graph modeling, sparse learning, tensor voting
Applications: forming process that exists in many manufacturing systems such as steel manufacturing and ceramic manufacturing.
Related publications: J11, J18 in the publication list
2. Computer experiments, surrogate modeling, process optimization, and control for small sample-size manufacturing system:
This research is motivated by the fact that many advanced manufacturing systems such as fuselage assembly and ship assembly where the data collection is limited and expensive. For such manufacturing systems, high-fidelity simulations such as Finite Element Analysis (FEA) will be first investigated to simulate/mimic the real manufacturing systems for process optimization and control. However, the high-fidelity simulation platform usually needs to be first calibrated given many setting parameters, and such simulations always need a large amount of time to simulate. Thus, methodologies need to be developed to first calibrate the simulation platform, and then establish surrogate models for further optimization and control. To deal with these challenges, our group has been focusing on novel data science methodologies guided by physics that accurately predict quality variations and optimally compensate errors based on computer experiments and surrogate modeling. In particular, our current research focuses on the development of advanced machine learning methodologies (e.g., sparse learning and reinforcement learning) to achieve optimal control strategy and process optimization. This research has a wide arrange of applications such as fuselage assembly, ship assembly, and additive manufacturing.
Methodologies: sparse learning, reinforcement learning, Gaussian process
Applications: ship assembly, fuselage assembly, automotive assembly
Related publications: J6, J7, J9, J14