ANALYSIS AND OPTIMIZATION OF THERMAL SYSTEM EFFICIENCY IN INTERNAL COMBUSTION ENGINES USING ARTIFICIAL INTELLIGENCE TECHNIQUES

Abstract

Artificial intelligence (AI) is increasingly applied to internal-combustion engines (ICEs) to achieve higher thermal efficiency and lower emissions under real-world operating transients. This paper reviews and experimentally validates AI methods spanning supervised learning (surrogate models for brake thermal efficiency, brake-specific fuel consumption, and emissions), deep learning (temperature-field reconstruction and emission prediction), and reinforcement learning (RL) for closed-loop control. A unified optimization framework is developed that processes multi-sensor data—including air–fuel ratio, load, pressures, temperatures, and exhaust composition—to determine optimal set-points for injection and ignition timing, boost, and coolant actuation. Literature demonstrates that AI can reduce calibration effort and enhance control authority in nonlinear regimes; recent case studies report model-free RL improving idle fuel consumption and safe-RL maintaining HCCI engine stability with RMS IMEP errors near 0.14 bar. In the present case study using a spark-ignition single-cylinder platform, AI-assisted control reduced idle fuel flow and improved indicated efficiency while lowering predicted NOₓ compared with a production PID baseline. Deployment barriers—data quality, real-time computation, safety assurance, and retrofit cost—are analyzed, and practical pathways such as edge inference, model compression, and explainable policies are outlined. The findings confirm that AI is a viable co-pilot for thermal management and combustion phasing in modern ICEs and hydrogen-ready hybrids, supporting sustainable progress in transport and industrial energy systems. The paper ends by recommending how these technologies may be used to improve the future operation of engines and facilitate the sustainable development of the transport and industrial sectors.