Degradation monitoring of machine tool ballscrew using deep convolution neural network

High-value manufacturing often requires a high level of accuracy. While this may be an achievable aim, the demands of consumers and end-users are also for the often competing targets of lower cost, greater efficiency and resource-lean products. Notwithstanding the ambition for higher accuracy, increased availability of production machines is a fundamental requirement to maintain competitiveness in the manufacturing industry. Ballscrews are a fundamental part of the transmission system for most high-value machine tools. They are therefore integral to the positional accuracy and performance of the machine and also represent a weaklink in terms availability. Hence, the state of the ballscrew is essential in determining machine accuracy and availability. This work proposes a deep learning approach for ballscrew performance monitoring. The technique works such that remedial activities can be scheduled and carried out when degradation is detected before breakdown occurs. The deep learning algorithm uses convolution to distinguish between a worn and good ballscrew in a machine tool. The technique was tested on a five-axis gantrytype machine tool with two parallel axis ballscrew. The results from the test carried out indicates that an overall accuracy of 94 % can be achieved with this technique.