APPLICATION OF A NEURAL-NETWORK TO THE ESTIMATION OF SURFACE-ROUGHNESS FROM AE SIGNALS GENERATED BY FRICTION PROCESS

The article presents a newly developed system for on-line estimation o f the roughness and hardness parameters of surfaces involved in slidin g friction process. The information about the process is extracted fro m the acoustic emission signal generated by the friction. The detected signal is first transformed into the power spectrum and then mapped i nto the process parameters by an artificial neural network simulated o n a PC. The system performance was tested in the on-line pin-in-disk e xperiment for four different classes of the surface roughness and/or p in hardness. The recognition reliability for placing given cases in co rrect classes was found to be greater than 94% for the majority of exp eriments.