CNC Milling Tool-Wear Prediction Using Machine Learning

A worn milling insert spoils the surface and can break, while changing inserts too early wastes money. This project predicts the flank wear VB of the inserts after every face-milling pass on a CNC machining centre from the spindle- motor current, vibration and acoustic emission signals of the real NASA Ames milling data set, and turns the prediction into a tool-change decision at the usual limit VB = 0.3 mm. It compares a Taylor-style wear-versus-time law and the specification's linear time baseline with ridge regression, gradient boosting and random forest on hand- crafted signal features, Taylor + machine-learning hybrids and a small 1-D convolutional neural network on the raw signals - always with whole insert sets held out (leave-one-case-out validation). A web application monitors insert sets pass by pass, shows the remaining cutting minutes, records inspections and insert changes, raises tool-change alerts and plans cutting parameters, on five databases. A FreeCAD CAM job of the same face-milling set-up, a GNU Octave cross-check and a complete MATLAB version are included.

Category: Production Engineering

Production Engineering Project PR01 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.