Performance Prediction for Spark Ignition Engines Using Artificial Neural Networks: Model Design and Validation
A.A. Aderibigbe
Department of Mechanical Engineering, Ladoke Akintola University of Technology, Ogbomoso, Nigeria.
A.D. Ogunsola *
Department of Mechanical Engineering, Ladoke Akintola University of Technology, Ogbomoso, Nigeria.
E.A. Fadiji
Department of Mechanical Engineering, Bamidele Olumilua University of Education, Science and Technology, Ikere Ekiti, Nigeria.
O. Adeyi
Department of Chemical Engineering, Michael Okpara University of Agriculture, Umudike, Nigeria.
A.J. Adeyi
Department of Mechanical Engineering, Ladoke Akintola University of Technology, Ogbomoso, Nigeria and Forest Research Institute of Nigeria, Ibadan, Nigeria.
E.A. Owoo *
Department of Agricultural Engineering, Ladoke Akintola University of Technology, Ogbomoso, Nigeria.
*Author to whom correspondence should be addressed.
Abstract
The demand for efficient and environmentally friendly spark ignition (SI) engines has driven researchers to explore advanced methods for optimizing engine performance and reducing emissions. One such method is the use of Artificial Neural Networks (ANNs) to develop predictive models that can accurately estimate engine performance under various operating conditions. This study presents the design and implementation of an ANN-based performance prediction system for spark ignition engines, focusing on critical performance metrics. An ANN based model and network architecture were developed and simulated in MATLAB neural network toolbox environment. The search for efficient network architecture was performed in terms of activation function, number of hidden layers, number of neurons in the hidden layers and the type of training function using highest regression value criteria. The ANN predicted results were validated by comparing with corresponding actual values obtained from experiments using t test. The search for efficient network architectures showed that 6 – 13 – 9 – 6 – 8 network architecture gave the best predicted results for the ANN model. Logsig activation function and trainlm training function gave reliable predicted results for the model. The results of the t test and comparison of ANN predicted results with actual experimental results showed that there is no significant difference between the two sets of results at 5% level of significance. The results also showed that 28 neurons distributed into three hidden layers have capability to map and generalize the non-linear data effectively thereby predicting the results accurately. It is observed that Increasing the number of neurons in the network generally increases the ability of the network to predict accurate results but beyond a certain limit this ability decreases due to overgeneralization of the non-linear data. It is concluded that the developed ANN based prediction system for SI engines is robust and capable of giving accurate results.
Keywords: Artificial neural network, exhaust gas emissions and SI engine, activation function, regression value