Data-driven virtual engine framework for predicting performance and emissions of biomass waste-derived biofuel engine

This research study aims to develop virtual engine models with biofuel blends to forecast their characteristics in terms of combustion, performance and emission by employing multimodal data driven modelling through artificial neural networks. In addition to training the artificial neural network-based virtual engine models with experimental datasets, they were subjected to hyperparameter optimization thereby determining the correlation between input and output parameters. A combustion endoscope captured the in-cylinder combustion flames during the experiment while the digital image processing technique identified the average flame pixel intensity at maximum burning conditions and incorporated them into the machine learning models. A superior combustion efficiency, performance and exhaust emission was exhibited in a blend comprising 90% Gasoline and 10% pine oil than observed in other biofuel combinations. The artificial neural network biofuel engine models that were rigorously trained and hyperparameter optimized precisely forecasted the combustion, performance and emission characteristics of all the biofuel blends with coefficient of determination values ranging between 0.97 and 0.99, while the mean absolute percentage error values range between 0.4% and 5.2% and the root mean square error values ranging between 0.04 and 3.2. The precise performance and error metrics indicates the efficacy of the hyperparameter-optimized neural networks in predicting the characteristics of biofuel blended spark ignition engine.