Development of neural network model of disc brake operation

Abstract

The quality of artificial neural network models mostly depends on a proper setting of neural network architecture i.e. learning algorithm, transfer functions, range and distribution of data used for training, validation, and testing, etc. The goal of this paper is related to the investigation on how artificial neural network architectures influence the network's generalization performance based on the same input/output parameters. The complex procedure of an artificial neural network model development has been demonstrated related to the disc brake performance. The influence of disc brake operation conditions (application pressure, initial speed, and temperature) has been modeled related to the disc brake cold, fade, and recovery performance. The artificial neural network model has been developed through investigation of how the synergy of different network's parameters, such as learning algorithm, transfer functions, the number of neurons in the hidden layers, affect the neural model abilities to predict the disc brake performance. It was shown in this paper that complex non-linear interrelations between the disc brake input/output variables can be modeled by proper analysis and setting of artificial neural network parameters. .

Keywords

• artificial neural network
• disc brake
• performance
• neural modeling