Thermal Error Modelling using Features selected from Particle Swarm Optimization

Computer Numerical Control (CNC) machines are employed in various industrial manufacturing processes to produce precise components. The CNCs are prone to thermal deviations from the heat generated during machining. Axis motors, rotary bearings, and spindle motors are the main sources of heat. In this context, temperature sensitive areas (TSAs) apply to these locations. The accuracy of the machining process is directly affected by the heat-induced deviation between the tool and the workpiece. The objective is to minimize these deviations between the tool and the workpiece. The most challenging aspect of creating thermal error compensation models is figuring out which TSAs to use. This work proposes employing particle swarm optimization (PSO), a computational intelligence (CI) technique, to choose significant features. First, information about the machine's temperature and thermal deviation is collected. Second, the outliers are eliminated by cleaning the data. Following this, PSO optimizes the error band to identify which TSAs are significant. Advanced machine learning (ML) models like polynomial regression (PR) and random forest (RF) regression are trained utilizing these TSAs as features. The proposed approach is compared with matching models developed with features selected through the brute force (BF) approach. While PSO-RF outperformed BF-RF in providing 16.60% error reduction with different features, PSO-PR and BF-PR provided 47.12% error band reduction with similar features. Based on the findings, it can be concluded that PSO-PR can be used on the machine and that PSO outperforms BF in terms of identifying the ideal subset of influential features.