Abstract: To improve modeling precision of CNC geometric error and error compensation, the main cutting force is simulated and measured using a dynamometer. Errors in the pitching and deflection angles are measured with a laser interferometer. Trained with practically measured error data, the force-geometric error
predicting model based on PSO-SVM is established with the key parameters optimized using the particle swarm optimization algorithm (PSO). Verification experiments show that difference between the measured error and the maximum deflection angle error using the PSO-SVM model is 0.6 μrad, and that with the maximum error  of pitching angle error is 0.21 μrad. Compared with the force-geometric error predicting model based on BP neural networks and SVM whose parameters is optimized using a conventional method, the prediction precision of the PSO-SVM error model is greatly improved. Therefore the proposed model can compensate geometric error of CNC machine tools in real-time with high-precision.

Key words: CNC machine tools, PSO-SVM, force-geometric error model, support vector machine (SVM), particle swarm optimization (PSO)