Using machine learning for anomalous toolpath identification in subtractive manufacturing

Abstract

The emphasis and application of machine learning with respect to manufacturing and machining has focused primarily on tool wear or bearing health. Few studies have focused on the parts produced by these processes and how changing parameters during machining operations can affect the final outcome. Quality control is a costly but necessary step in the manufacturing process to ensure that a finished part meets specification. For a machined part, this is usually accomplished using inspection and measurement techniques. However, inspection of a machined part has typically occurred after certain predetermined milestones. This study aims to identify and classify machining phenomenon compared to a reference signal to determine if the toolpath mimics reflects the intended behavior. To accomplish this, a Computer Numerical Control (CNC) milling machine is instrumented with accelerometers to track and record vibrations. This data is collected from the spindle and processed using a machine learning algorithm that segregates signatures based on selected features and classifies them as expected behaviors or anomalous. The results of the study indicate that certain phenomena can be accurately identified and labeled as normal or abnormal with respect to feed rate or spindle speed overrides. It is a promising insight into more complex toolpath identification and integration with Computer Aided Design (CAD) and Computer Aided Manufacturing (CAM) software to anticipate and mitigate machining errors.