Prediction of Friction Losses in Spark Ignition Engine: An Artificial Neural Networks Approach

ABSTRACT

Multi-layered, feed forward . , back-propagation artificial neural networks (ANN) models were developed to predict friction losses in spark-ignition engines. The friction losses were modeled as friction mean effective pressure (fmep) due to: crankshaft, reciprocating parts, valve trains, auxiliary, and pumping systems. The developed models were validated in relation to existing engine friction data and empirical models of Patton et ai. (patton, K.J., Nitsche, R.G. and Heywood lB. [1989]. Development and evaluation of a friction model for spark-ignition engines, SAE paper 890836). Results have shown that, the mean absolute deviations of the ANN model predictions for crankshaft, reciprocating parts, valve trains, auxiliary, and pumping systems were, respectively, 18.78,2.10, 10.57,2.66, and 8.84% in relation to the existing engine friction data, while those of the Patton et al. model predictions were 37.50, 25.14, 60.99, 6.71, and 14.67%, respectively. The corresponding root mean square errors were for md to be 2.278, 1.157, 3.145, 0.678, and 2.118 for the ANN predictions and 7.006, 12.837, ..5.889,2.277, and 3.317, respectively, for the Patton et al. predictions. The developed ANN friction models appeared to have better end more accurate predictions, thus it could be used as tool for designing of energy-efficient spark-ignition engines