Anomaly detection in three-axis CNC machines using LSTM networks and transfer learning

There is a growing interest in developing automated manufacturing technologies to achieve a fully autonomous factory. An integral part of these smart machines is a mechanism to automatically detect operational and process anomalies before they cause serious damage. The long-short-term memory (LSTM) network has shown considerable promise in the literature, with applications in the detection of tool wear and tool breakage to name a few. However, these methods require a significant amount of machine-specific training data to be successful, which makes these networks custom to a machine, requiring new networks and new data for each machine. Transfer learning is an approach where we use a network developed with a rich data set on one machine and re-train it with a smaller data set on a target machine. We have implemented this approach for chatter detection with a LSTM network, using sensor data and a rich data set from one machine, and then use a transfer learning methodology, similar sensors, and a smaller data set for the chatter detection algorithm on another machine. This allows for the transfer of knowledge from one machine to be applied to a similar machine, with some local optimization from transfer learning.