Performance, emissions, and predictive modelling of a CI-ICE operating in LPG–diesel dual-fuel mode

This study presents an experimental assessment of a compression-ignition internal combustion engine (CI-ICE) operating in dual-fuel mode with diesel and liquefied petroleum gas (LPG), without fuel preheating or combustion enhancers. The objective is to evaluate the effect of LPG substitution on engine performance, emissions, and operational fuel costs under electric power generation conditions. Additionally, an artificial neural network (ANN) was developed using TensorFlow and Keras to capture relationships among key process variables at high altitudes; the model was trained on experimental data from the test campaign. Tests were conducted at several engine load conditions, comparing diesel operation with LPG–diesel dual-fuel blends, where diesel was partially substituted by LPG at three mass flow rates (0.18, 0.36, and 0.54 kg/h). Dual-fuel operation increased brake thermal efficiency by up to 10%, reaching 29.4% at 0.54 kg/h, while reducing brake specific fuel consumption. Specific emissions of CO and HC decreased significantly at medium and high loads, whereas NOx increased at full load. Fuel cost reductions reached 73% at full load. The ANN models achieved high predictive accuracy (R²>0.95), confirming their robustness for predicting CI-ICE performance in LPG–diesel dual-fuel mode. Therefore, ANN models provide a reliable tool for reducing experimental testing campaigns and associated operational and research costs. Finally, the findings confirm the technical feasibility of operating a CI-ICE in LPG–diesel dual-fuel mode with minimal intake manifold modifications. This approach maintains CI-ICE operational stability and power output while offering environmental and economic benefits, reinforcing its applicability for electricity generation systems.