Development of a thermal error compensation system for a CNC machine using a radial basis function neural network

This work aimed to develop a thermal error compensation system for a three-axis computer numerical control (CNC) machine, a 20-year-old CNC milling machine. An interferometric laser was used to acquire the error measurements, and a compensation model was developed using a radial basis function neural network. The model was implemented directly via the machine programmable logic controller. The model was able to predict the observed errors under the acquired conditions (using supervised learning), replicating the trends in the errors. A further conclusion was that the model could differentiate between errors arising at room temperatures from thermal errors found during machine operation at higher temperatures. For practical validation, 360 data points were acquired and tested under conditions not used for model training. The neural network model was implemented directly in the controller of the CNC machine using ladder logic blocks as an extra routine accessed simultaneously during machine operation. The resulting calculus value was added as a correction value to the final axis position instantaneously, based on the instant axis position and instant temperatures at the bearings. The system was able to apply the corrections suggested by the model with a strong alignment between the real corrected positioning and the positioning predicted by the model. Preliminary results obtained under higher-temperature operating conditions indicated a reduction in the maximum thermal error up to 77.8%, and for room temperatures, the positioning errors were compensated at average rates of 33%.