Prediction of torque and specific fuel consumption of a gasoline engine by using artificial neural networks Necla Kara Togun * , Sedat Baysec Department of Mechanical Engineering, University of Gaziantep, Gaziantep, Turkey article info Article history: Received 9 April 2009 Received in revised form 22 June 2009 Accepted 11 August 2009 Available online 17 September 2009 Keywords: Gasoline engine Torque Brake specific fuel consumption Neural networks Explicit solution abstract This study presents an artificial neural network (ANN) model to predict the torque and brake specific fuel consumption of a gasoline engine. An explicit ANN based formulation is developed to predict torque and brake specific fuel consumption of a gasoline engine in terms of spark advance, throttle position and engine speed. The proposed ANN model is based on experimental results. Experimental studies were completed to obtain training and testing data. Of all 81 data sets, the training and testing sets consisted of randomly selected 63 and 18 sets, respectively. An ANN model based on a back-propagation learning algorithm for the engine was developed. The performance and an accuracy of the proposed ANN model are found satisfactory. This study demonstrates that ANN is very efficient for predicting the engine torque and brake specific fuel consumption. Moreover, the proposed ANN model is presented in explicit form as a mathematical function. Crown Copyright Ó 2009 Published by Elsevier Ltd. All rights reserved. 1. Introduction An engineering phenomenon may embed complicated physical, chemical or electrical theory and may require very complicated arithmetic to describe them, yet, arithmetic emerged may not be solvable in closed form. Artificial neural network (ANN) is an alter- native technique for providing a relationship between the variable quantities of interest. ANN requires only a set of experimental re- sults, numerical in nature and describes the relation by analyzing them. In other words, it only needs solution examples concerning the problem. ANN techniques require a lot of arithmetic basically of trial-end-error nature, involving numerical differentiation, inte- gration, noise rejection, etc., and are never feasible without fast computation facilities. Advent of digital computers providing high speed arithmetic and vast amounts of data storage has given rise to the application of ANN techniques to many engineering problems. In recent years, this method has been applied to various disciplines including automotive engineering, in forecasting of engine charac- teristics for different working conditions. The relationship between the temperature of the exhaust gases and fuel consumption of an internal consumption engine has been studied in [1]. ANN approach has been used in another study, to analyze the effect of cetane number on exhaust emissions from the engine [2], and also in [3], to model diesel particulate emission [4,5] are remarkable studies where ANN is used to forecast gaso- line consumption and the effects of intake valve timing on engine performance and fuel economy respectively. Similarly in [6] the ef- fect of throttling is studied, taking the intake manifold geometry into consideration. Numerous studies have been undertaken to predict the perfor- mance and exhaust emission characteristics of internal combus- tion engines by using ANNs [7–11]. Studies done by neural networks and genetic algorithms have been used to predict and re- duce diesel engine emissions [12]. Neural networks have been found to be the domain for numer- ous successful applications of prediction tasks, in modeling and prediction of energy-engineering systems [13], prediction of the energy consumption of passive solar buildings [14], and modeling a burner heated catalytic converter during cold start in a four stroke, spark ignition engine [15]. And also artificial neural net- work technique has been applied to control the air fuel ratio of the engine [16,17] and exhaust gas recirculation control [18]. In this study, an explicit ANN based formulation was developed to predict torque and brake specific fuel consumption of a gasoline engine in terms of spark advance, throttle position and engine speed. Experimental studies were completed to obtain training and testing data. The experimental data from totally 81 test runs was used to train and test the ANN model for predicting torque and brake specific fuel consumption. Inputs for the network were the spark advance, throttle position and engine speed, while the outputs were torque and brake specific fuel consumption. The experimental study to determine torque and fuel consumption characteristics in a gasoline engine is complex, time consuming and costly. It also requires specific instrumentation. To overcome 0306-2619/$ - see front matter Crown Copyright Ó 2009 Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.apenergy.2009.08.016 * Corresponding author. Tel.: +90 342 3172534; fax: +90 342 3601104. E-mail address: nkara@gantep.edu.tr (N. Kara Togun). Applied Energy 87 (2010) 349–355 Contents lists available at ScienceDirect Applied Energy journal homepage: www.elsevier.com/locate/apenergy