Optimizing predictive maintenance with synthetic data: a case study on ABB IRB 1100 industrial robot

Résumé

Within the paradigm of industry 4.0, predictive maintenance emerges as the pivotal strategic axis foe enhancing equipment reliability and optimization operating costs. This maintenance is fundamentally predicated on the utilization of artificial intelligence tools and data analysis techniques. In this context, the present article proposes a synthetic data generation approach dedicates to detecting failures of the ABB IRB 1100 industrial robot. To address the dearth of real-world data, a sophisticated simulation model was developed, incorporating the robot’s six axes and sensors that monitor electrical, mechanical, thermal, and vibration parameters. A dataset comprising 2,000 instances and 215 explanatory characteristics is constructed from the simulated multisensory signals. The results show that ensemble tree methods outperform SVM based approaches, with LighGBM achieving the best overall performance (F1-macro = 0.7625) with a reduced subset of 100 features selected by mutual information. A detailed class-based analysis reveals strong discrimination capabilities for overheating and misalignment defects (F1 ≈ 0.97–1.00), while gear wear remains the most challenging condition to address due to partial feature overlap and imbalance effects. Furthermore, severity-level validation (three levels per defect type) demonstrates moderate intra-defect separability for misalignment (F1 ≈ 0.676) and overheating (F1 ≈ 0.613), confirming the feasibility of graded maintenance decision-making. A redundancy and dimensionality analysis, combining correlation filtering and supervised feature selection, confirms that dimensionality can be reduced by more than 50% without performance degradation. These results highlight the importance of structured feature selection to improve the generalization, computational efficiency, and interpretability of predictive maintenance systems. The proposed framework enables both fault identification and prioritization based on severity, thus contributing to optimized maintenance planning in Industry 4.0 environments, while also emphasizing the need for further validation on real-world industrial datasets.

Citer ce document

Accès au document

Texte intégral en lecture en ligne, réservé aux abonnés SPHAERO et aux membres de l'institution. Se connecter

Auteur(s)

Statistiques

Consultations : 1

Téléchargements : 0