Artificial neural network–based performance modeling of a diesel engine within the whole operating region considering dynamic conditions

Abstract

Engine performance under full working conditions, especially dynamic ones, is indispensable in many vehicle-level research fields. To acquire the engine performance parameters, a novel whole-region engine model, considering both steady and dynamic conditions, was developed based on limited test data in this work. This model used throttle position, engine speed, and its acceleration as the input variables to predict torque and brake-specific fuel consumption under all practical conditions within its operating envelope. The engine bench test was first conducted under typical operating conditions to collect test data for model development and validation. Then, the backpropagation neural network with designed structure was employed to perform data fitting for test conditions. After the analysis of parameter distribution tendency, the two-step interpolation method was used to generalize performance parameters under conditions apart from those test ones. The cross-condition prediction accuracy of developed engine model was validated by test data under various operating conditions. Also, the parameter prediction error of proposed modeling method was lower compared to that of existing neural network methods, which further proved its applicability to dynamic engine modeling issues.

A Prediction Model Based on Artificial Neural Network for the Temperature Performance of a Hydrodynamic Retarder in Constant-Torque Braking Process

“…For the output layer transfer function, the Sigmoid function appears to be inappropriate as it narrows the output signal to a sensitive range [30]. And the output layer should faithfully transfer the signal from the last hidden layer to the output layer.…”

Design of a virtual test cell based on GMDH-type neural network for a heavy-duty diesel engine

“…20,21 It is widely known that engines typically have complicated, advanced and non-linear dynamic behaviors; making the investigation of their system dynamics more multifaceted. 22 In addition, ANNs are quick enough to be efficiently applied to engine production processes at a commercial level. On the other hand, ANN-based models suffer from a drawback in that any intense change in operating conditions requires large amounts of training data (compared to physical models) to make sure the model has the correct responses.…”