Prediction of the total exhaust heat emission from motorcycles using a backpropagation neural network

Abstract

The increasing number of motorcycles in Indonesia contributes significantly to traffic congestion, noise pollution, air pollution, and thermal emissions to the surrounding environment. This study develops a rapid and accurate method to predict total exhaust heat from motorcycles in real-time using a backpropagation neural network (BPNN) optimized with a genetic algorithm. The research methodology involves measuring exhaust heat from 17 motorcycle types using thermal imaging equipment across various engine speeds (2000–5000 rpm). Input parameters include motorcycle brand, engine displacement, transmission type, manufacturing year, ambient temperature, and vehicle speed, while output parameters comprise heat from the engine surface, the exhaust surface, and exhaust gas. The BPNN model achieved a mean square error of 0.01 after training on 500 datasets (70 % training, 15 % validation, 15 % testing). Results show that engine surface heat contributes 87 % of total exhaust heat, exhaust surface contributes 12 %, and exhaust gas contributes 1 %. This BPNN module enables real-time environmental heat assessment, supporting sustainable transportation planning and vehicle design improvements.