Prediction of fuel properties and engine performance of biodiesel-ZnO blends using artificial neural networks

Prediction of fuel properties and engine performance of biodiesel-ZnO blends using artificial neural networks

This study considered the potential of artificial neural networks (ANNs) as analytical tools for modeling the main properties and performance of diesel engines fueled with biodiesel-zinc oxide (ZnO) nanoparticle blends. Experimental datasets were first collected from past researches to estimate kinematic viscosity, calorific value, brake thermal efficiency (BTE), and brake specific fuel consumption (BSFC) across various diesel-biodiesel ratios, ZnO dosages, engine rotational speeds, and loads. While adding biodiesel causes increasing viscosity and reducing heating value, the inclusion of ZnO nanoparticles was found to mitigate these drawbacks by stabilizing viscosity and improving energy content, which translated into improved combustion efficiency and reduced fuel consumption. To capture these complex nonlinear interactions, four independent ANN models were established: two for predicting viscosity and calorific value, and two for engine performance indicators (BTE and BSFC). The models employed feed forward backpropagation networks trained with the Levenberg–Marquardt algorithm. Statistical evaluation confirmed strong predictive capability, with the BTE and BSFC models showing the highest accuracy (R²values of 0.9 and 0.8, respectively), followed by viscosity (R²= 0.8) and calorific value (R²= 0.7). These results highlight that ANN performs best when outputs are strongly sensitive to input and operating conditions, such as in performance metrics, whereas intrinsic chemical properties remain more challenging to predict. Overall, the findings demonstrate that ZnO nanoparticles are effective additives for improving biodiesel–diesel blends and that ANN models provide reliable, computationally efficient alternatives to extensive experimental trials. This work bridges experimental evidence with machine learning, offering predictive tools and decision guidelines for optimizing biodiesel–nanoparticle formulations in diesel engines, while also identifying future research needs related to emissions modeling, dataset diversity, and real-time deployment.