Muhammad Harussani Moklis, Cries Avian, Prysathryd Sarabhorn, Phantisa Limleamthong, Chinnathan Areeprasert, Sasipa Boonyubol, Jeffrey S. Cross
The pulp and paper industry generates significant quantities of paper sludge waste, necessitating sustainable valorization strategies. This study proposes a physics-constrained machine learning optimization framework for predicting and optimizing nitrogen oxides (NOx) emissions during the combustion and co-combustion of hydrothermally treated paper sludge (HTT-PS) with coal. An experimental dataset comprising 59 data was used to develop predictive models based on fuel properties and operating parameters, including combustion temperature, flow rate, feeding rate, and excess air. Five ML models were evaluated, with XGBoost achieving the best performance (R2 = 0.997, RMSE = 0.058, MAE = 0.043). Model interpretability analysis using SHAP and LIME identified flow rate and feeding rate as the dominant factors influencing NOx formation. For inverse design, the optimized XGBoost model was integrated with particle swarm optimization (PSO), while combustion-domain constraints were incorporated through stoichiometric air–fuel relationships and bounded excess-air conditions. The proposed framework recorded approximately 22% NOx reduction relative to the optimized experimentally observed NOx emissions reported in previous studies. Optimized excess-air values exhibited strong agreement with stoichiometric calculations, with errors below 1% for HTT-PS and co-combustion systems. These results demonstrate the potential of physics-constrained ML as an efficient alternative to traditional trial-and-error optimization for thermochemical process design. © 2026 Elsevier Ltd.
Department of Transdisciplinary Science and Engineering, Institute of Science Tokyo, 2-12-1, Ookayama, Meguro-ku, Tokyo, 152-8550, Japan; Department of Chemical and Environmental Engineering, Faculty of Engineering, Universiti Putra Malaysia, Serdang, Selangor, 43400, Malaysia; Department of Electrical Engineering, Universitas Brawijaya, Jawa Timur, Malang, 65145, Indonesia; Department of Mechanical Engineering, Faculty of Engineering, Kasetsart University, 50 Ngam Wong Wan Road, Lat Yao, Chatuchak, Bangkok, 10900, Thailand; Department of Farm Mechanics, Faculty of Agriculture, Kasetsart University, 50 Ngam Wong Wan Road, Lat Yao, Chatuchak, Bangkok, 10900, Thailand; Department of Chemical Engineering, Faculty of Engineering, Kasetsart University, 50 Ngam Wong Wan Road, Lat Yao, Chatuchak, Bangkok, 10900, Thailand
