Adaptive sequential learning algorithms for autonomous manufacturing
Introducing adaptive digital twins powered by Bayesian optimization for efficient manufacturing processes. These twins use physics-aware surrogate models and soft sensors to optimize operations like mixing and drying, reducing waste and energy while maintaining quality.
Overview
Our technology offers an adaptive digital twin solution for autonomous manufacturing, leveraging Bayesian optimization and sequential experimental design. Unlike traditional methods, our approach continuously updates itself by integrating physics-aware models and soft sensors to emulate and optimize processes such as mixing, heating, extrusion, and drying. This results in reduced physical trials and enhanced robustness to changes in ingredients and equipment, thereby minimizing waste and energy use while ensuring high product quality.
Technical specifications
Key Features:
- Bayesian Optimization (BO): Employs a multi-objective BO controller to run fast simulations and schedule targeted trials when necessary.
- Digital Twin: Encodes ingredient/equipment/environment covariates and transfers knowledge across sites for faster adaptation.
- Soft Sensors: Utilizes real-time sensors to infer latent qualities like water activity and texture.
- Surrogate Models: Physics-aware and data-driven models simulate manufacturing processes, supported by active, sequential sampling.
- What-if Simulations: Facilitates pre-production risk assessments and recipe/set-up adjustments.
- Cross-site Priors: Accelerates adaptation to new production lines by leveraging prior knowledge.
Technology readiness level
This technology is currently at TRL 5, having been validated in a laboratory setting with plans for a pilot on an actual production line. The next steps include further validation, integration of soft sensors, and implementation of a user-friendly interface for operators.
About West Virginia University
West Virginia University is a comprehensive public land‑grant research university headquartered in Morgantown and serving the state through academics, research, and outreach. For industry, a statewide Extension network provides on‑the‑ground access to communities, facilities, and field sites, while an integrated academic medical center and health system enable clinical research and translational partnerships. The university connects corporate R&D with talent through co‑op and internship pipelines and applied engagements. Research is supported by competitive federal funding from agencies such as NIH, NSF, DOE, and USDA. WVU’s Office of Innovation and Commercialization manages IP, licensing, and startup pathways to provide a clear entry point for collaboration.
delivered to your inbox.