The research objective is to advance predictive science, specifically prognostic modeling algorithms by integrating physical science with emerging ML-based data analytics techniques, to improve the accuracy, robustness, and reliability of probabilistic prediction of a situation that occurs in Dynamical Systems. Two of the envisioned applications are detecting machine fault occurrences and identifying fault type and severity (referred to as fault diagnosis), as well as predicting system performance degradation as well as remaining useful life (RUL) (referred to as RUL prognosis). The diagnosis and prognosis can subsequently provide the technical basis for intelligent preventive maintenance, which minimizes machine downtime, maintenance cost, and reliance on human experience for maintenance scheduling. 

While focusing on the novelty of model development (i.e., designing next-generation process sensing-ML architecture), we also heavily consider the applicability and generalizability of the developed technical solutions. By recognizing and addressing the discrepancies between model development in the lab and model deployment in plants, it is our ultimate goal to transform every model and solution from the lab to actual applications on the shop floor. Specific research tasks include:

1. Developing cost-effective edge devices that integrate sensors with microcontroller and wireless transmission modules on a board; 
2. Incorporating physical domain knowledge into the design and optimization of ML model architecture, loss, and training; 
3. Enabling ML model learning from unlabeled data and continual model updating through data streaming in shop floor applications; 
4. Improving the efficiency of data transmission, management, and processing through cloud-edge computing infrastructure;
5. Adapting and generalizing models with ease among machines with shared similarities and unique behaviors.