Non-contact surface roughness evaluation of milling surface using CNN-deep learning models

Machining quality control is a bottleneck operation as human inspectors and expensive equipment needed in most operations. Automated assurance the manufacturing industry has potential to replace humans lower cost of machined product. This paper presents analysis end-milled surfaces backed with experimental deep learning model investigations. The effects machining parameters like spindle speed, feed rate (table feed), depth cut, cutting duration were investigated find surface roughness using Taguchi orthogonal array. Following standard DOE, image recorded for each experiment categorized into four classes, viz. fine, smooth, rough, coarse, based on value (Ra). images used develop CNN models class prediction. Further, comparative studies among five popular optimizers performed. results showed that ‘Rectified Adam’ optimizer performed better amongst pool, training test accuracy 96.30% 92.91%, respectively. proposed features highly accurate slim structure, potentially substituting procedures employ measuring devices.