An Integrated System for CNC Machine Tool Rotary-Axis Error Diagnostics with Machine Learning Inference

This paper presents the design, architecture, and Python-based implementation of a diagnostic application for monitoring rotary-axis positioning errors in five-axis CNC machine tools. The system accepts raw data files from the R-Test measurement procedure and processes them through a threemodule pipeline: (1) data ingestion with parsing and normalization; (2) machine-learning inference using one of three selectable models—Multilayer Perceptron (MLP), Kolmogorov-Arnold Network (KAN), or Multi-Output Gaussian Process (MOGP)—combined with rule-based fault classification; and (3) interactive visualization and automated PDF report generation. The graphical user interface is built with the Tkinter framework and supports bilingual (Polish/English) operation. An embedded heuristic engine distinguishes between four fault categories—mechanical backlash, thermal drift, geometric misalignment, and servo/encoder anomaliess—from displacement trajectories recorded along the X′,Y′, and Z′ axes. Experimental data collected on two five-axis milling machines (monoBLOCK 65 and Lasertec 65) confirm that MOGP achieves the highest reconstruction accuracy (average R2=0.991,MPE=2.29%), outperforming KAN (R2=0.974,MPE=4.86%) and MLP (R2=0.761,MPE=14.68%), which validates the default inference path in the application. The presented solution bridges quantitative predictive modelling with maintenance-ready reporting, filling a practical gap in condition monitoring for precision machining environments.