Identification of Rotary Axis Positioning Errors in the Machining of Mining Equipment Components Using Machine Learning Techniques
Artykuł w czasopiśmie
MNiSW
0
brak dyscyplin
| Status: | |
| Autorzy: | Józwik Jerzy, Barszcz Marcin, Tomiło Paweł, Kuric Ivan, Sałamacha Daria |
| Dyscypliny: | |
| Aby zobaczyć szczegóły należy się zalogować. | |
| Rok wydania: | 2025 |
| Wersja dokumentu: | Drukowana | Elektroniczna |
| Język: | angielski |
| Numer czasopisma: | 3 |
| Wolumen/Tom: | 30 |
| Strony: | 759 - 775 |
| Impact Factor: | 1,4 |
| Efekt badań statutowych | NIE |
| Finansowanie: | The research was funded under the project Lublin University of Technology - Excellent Science: “Investing in Potential” (grant no. 8/IP/2025/F). This work was supported by the project VEGA 1/0470/23 - “Research into methods and means of implementing artificial intelligence in automated quality control systems for products with volatile quality parameters.” |
| Materiał konferencyjny: | NIE |
| Publikacja OA: | TAK |
| Licencja: | |
| Sposób udostępnienia: | Witryna wydawcy |
| Wersja tekstu: | Ostateczna wersja opublikowana |
| Czas opublikowania: | W momencie opublikowania |
| Data opublikowania w OA: | 30 grudnia 2025 |
| Abstrakty: | angielski |
| This study addresses rotary-axis positioning errors that degrade dimensional fidelity during machining of mining-equipment components – particularly conical picks and their toolholder interfaces, where press-fit sockets and conical shanks demand tight tolerances. We propose a data-driven identification framework that learns the mapping from R-Test trajectories to the worktable’s rotation-center errors along X′/Y′/Z′. Experiments on two five-axis milling machines (monoBLOCK 65 and Lasertec 65) covered full 360° rotations (Δα = 30°), both directions of motion, radii R = 75– 300 mm, and feeds vf = 500–5000 mm/min. After statistical analysis and feature engineering, three models were benchmarked: a multilayer perceptron (MLP), a Kolmogorov-Arnold network (KAN), and a multi-output Gaussian process (MOGP). MOGP achieved the best predictive fidelity (average R2 = 0.991, MPE = 2.29%, MSE = 0.002), outperforming KAN (R2 = 0.974) and MLP (R2 = 0.761). Error distributions showed weak sensitivity to feed and motion direction (left-right correlations ≥ 0.90, lowest for Z′), indicating predominantly geometric/thermal origins. The learned model enabled a high-resolution “error map” of the worktable based on resultant displacement, supporting corrective actions that preserve press-fit tolerances and free rotation in mining-component assemblies. We further implemented an integrated diagnostic tool that ingests raw R-Test exports, validates units/ranges, performs model-aware inference, and generates bilingual (PL/EN) technical reports. Embedded rule-based logic flags likely mechanisms (backlash, thermal drift, geometric misalignment) from trajectory patterns, bridging quantitative predictions with maintenance decisions. The results demonstrate that nonparametric, uncertainty- capable multi-output modeling is a robust foundation for rotary-axis error cartography and diagnostics, with immediate applicability to quality assurance and predictive maintenance in the machining of mining-equipment components |
