Experimental and machine learning based investigation of performance and emission characteristics of a CI engine using fusel oil blends
Pathmanaban Pugazhendi, Gopinath Dhamodaran, Raju Munisamy, Murugu Nachippan Nachiappan, Nookaraju B.Ch, Raja Kannan
Abstract
This study explores the performance, emission characteristics, and machine learning-based predictive modeling of a compression ignition engine operating on diesel–fusel oil blends. Experiments were carried out at ten engine speeds ranging from 1000 to 3250 rpm under full-load conditions using six blends containing 0%–37% fusel oil by volume. The brake-specific fuel consumption (BSFC) rose from 296 g/kW h for neat diesel to 358 g/kW h for the 37% blend at 2250 rpm, corresponding to a 20.9% increase, while a 22.7% rise was noted at 1000 rpm. Maximum power output declined from 6.0 kW (diesel) to 5.78 kW (37% blend) at 3250 rpm, accompanied by a 12% reduction in engine torque (ET) at higher fusel proportions. Emission measurements indicated a 15%–20% decrease in nitrogen oxides (NOx) with greater fusel content, whereas carbon monoxide (CO) and unburned hydrocarbon emissions increased by 35% and 18%, respectively. Five machine learning models—extreme gradient boosting (XGBoost), random forest (RF), support vector regression, artificial neural network, and multi-layer perceptron, were employed to predict engine parameters using 58 experimental data samples. Hyperparameter optimization through a fivefold cross-validated grid search revealed that XGBoost achieved the highest predictive accuracy, with R2 values exceeding 0.90 for seven of nine target outputs, including 0.94 for BSFC, 0.99 for engine power and NOx, and 0.96 for ET and CO. The RF model exhibited similar reliability, whereas neural networks were less consistent due to data limitations. This hybrid framework effectively minimizes experimental effort by ∼88% while maintaining over 90% prediction accuracy.