Diagnosis of spindle failure by unsupervised machine learning from in-process monitoring data in machining

In high-speed machining (HSM), process performance is closely linked to the optimization of cutting conditions and spindle exploitation. Keeping high levels of productivity and machine availability with limited costs is important. However, machining incidents, such as abnormal vibration or tool failure, can cause spindle failure and machine downtime. Consequently, identifying which kind and which severity of machining incident can damage an HSM spindle is critical (as well as which evolution of spindle vibration signature reveals a damage). For that purpose, in-process monitoring data and spindle condition monitoring data are analyzed by knowledge discovery in database (KDD), with a dedicated method to the machining process. Since daily spindle vibration signatures are measured, the in-process monitoring data needs to be daily aggregated. An original unsupervised co-training by genetic algorithm is then proposed for the diagnosis of an HSM spindle, in order to determine which machining events are critical for the spindle condition. Afterwards, preventive actions can be taken. The approach was applied to three spindle lifetimes, during which the monitoring data were collected for two years of machining of aeronautic structural components. Two major causes of spindle failure were then identified.