Measurement of EDMed surfaces roughness using convolutional neural network

In addition to dimensions, surface roughness measurement is crucial in every manufacturing. In this study, a trustworthy method for characterising the surface roughness of electrical discharge machined surfaces was developed using a convolutional neural network. Since feature extraction is incorporated into the convolution process of the network, this technique eliminates it. Images of EDMed surfaces were taken using a mobile camera. MATLAB software was used to process a signal vector that was created from the intensity of the picture pixels. A database of specimens with recorded surface roughness values was created. When samples with known surface roughness are given, the proposed technique is a strong contender for real-time surface roughness measurement. The generated predicted values are compared with the measured values acquired from a profilometer using a stylus. The applicability and accuracy of five loss functions are considered before they are chosen and examined for the prediction models. The accuracy and performance of this digital model suggest that it has the capability to assess the surface roughness very well.