PI: Pingbo Tang
Co-PI(s): Yorie Nakahira
University: Carnegie Mellon University
Industry partner: Ethos Collaborative
The proposed project aims at interactive learning between humans and computers to predict inspection timings and operation strategies of the water treatment plants that can reduce maintenance time while keeping quality water production. The water industry relies on human operators to infer deterioration rates of water systems based on their experiences in operation and maintenance. However, operators could be unsure about the deterioration trends and the optimal control strategies when facing different contexts (e.g., influent water, backwash flow speeds) or plant configurations. Tracing operators’ inspection, operation, and maintenance behaviors could allow computers to discover and synthesize operators’ maintenance-aware operation strategies in different contexts. However, the lack of reliable process pattern mining methods impedes engineers from such experience synthesis. The project team and Industry partners (Ethos Collaborative and American Water) propose a maintenance-aware operation digital twin of water treatment plants for transferring balanced operation strategies between operators. The elements of this digital twin include 1) data-driven and physics-based simulations of water system deteriorations and fault scenarios; 2) Virtual Reality (VR) environments for capturing human inspection, operation, and maintenance strategies; 3) a pattern mining approach that identifies reusable inspection, maintenance, and operation strategies from human-in-the-loop simulation logs of operating aging water plants.