STTR Phase I: Industrial combustion optimization using machine learning to reduce emissions and increase fuel efficiency

The broader impact/commercial potential of this Small Business Technology Transfer (STTR) Phase I project will result from development of an artificial intelligence optimization system that can be applied to any industrial combustion system. The simulator component of the proposed system will allow operators to run fast, low-cost experiments on large facilities without disrupting operations, and the optimizer component will recommend equipment settings to reduce fuel costs and emissions while maintaining performance. As a result, the system can potentially have a major impact on US energy and manufacturing markets, making American firms more competitive worldwide while simultaneously reducing the generation of harmful pollutants. Areas of application include energy generation, cement production, steel manufacturing, and mining operations. This Small Business Technology Transfer (STTR) Phase I project promises to significantly advance the understanding of complex industrial combustion processes. Currently, control models use inaccurate assumptions that create operational inefficiencies, resulting in wasted fuel and increased emissions. This project proposes a new paradigm for combustion research that combines the theoretical rigor of chemistry and thermodynamics with the data-adaptive flexibility of machine learning. The primary contributions of the company’s novel analytics pipeline include a simulator that enables operators to run in silico experiments on their facilities and an optimizer that maximizes combustion efficiency with more accurate controls, saving operators’ time and reducing fuel costs. These models leverage causal information about the facility to train a nonparametric regression ensemble that learns the plant’s behavior over time. The proposed project will fund the development of a three pillars analytics pipeline: unsupervised data exploration, supervised plant simulation, and customized optimization. This approach combines well established scientific principles with state-of-the-art machine learning techniques to build fast, flexible, and accurate simulators of power plants and other industrial facilities. With these models, analysts can run in silico experiments to optimize thermal efficiency and train algorithms to recommend ideal plant settings in real-time. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.