Optimisation of Clutch Disc Friction Material Using a Multi-Layer Perceptron Artificial Neural Network

This paper presents an analysis of four clutch disc friction materials (from different manufacturers) used in manual transmissions. Scanning electron microscopy and energy-dispersive X-ray spectroscopy were employed for the microstructural and chemical characterisation of the friction materials. To reveal the tribological properties of the selected clutch discs, three measurements of the friction coefficient between the material and the cast iron disc were conducted. The findings were employed to construct an artificial neural network using Easy NN software (V 14), with the objective of optimising the friction material. The chemical composition of the friction materials was employed as the input data, whereas the minimum, maximum, and average values of the friction coefficient, as well as the temperature generated during friction, were utilised as the output data. To assess the efficacy of the neural network, the correlation between the importance of input data and their sensitivity to output data was examined. It was determined that the model with three hidden layers exhibited a notable correlation between the six most influential chemical elements and their sensitivity. Based on this neural model, the chemical composition of the friction disc materials was optimised using the "Query" mode, aiming to minimise discrepancies in friction coefficients and temperature development.