Non-Invasive Failure Detection in Electric Parking Brake Modules
Artykuł w czasopiśmie
MNiSW
100
Lista 2024
| Status: | |
| Autorzy: | Ambrożkiewicz Bartłomiej, Syta Arkadiusz, Wójcik Łukasz, Uemura Wataru, Georgiadis Anthimos, Litak Grzegorz |
| Dyscypliny: | |
| Aby zobaczyć szczegóły należy się zalogować. | |
| Rok wydania: | 2026 |
| Wersja dokumentu: | Drukowana | Elektroniczna |
| Język: | angielski |
| Numer czasopisma: | 4 |
| Wolumen/Tom: | 148 |
| Strony: | 1 - 14 |
| Impact Factor: | 1,9 |
| Scopus® Cytowania: | 0 |
| Bazy: | Scopus |
| Efekt badań statutowych | NIE |
| Finansowanie: | This research was funded by the Lublin University of Technology program titled “Investing in Potential” (Grant No: 2/IP/2024/F). |
| Materiał konferencyjny: | NIE |
| Publikacja OA: | NIE |
| Abstrakty: | angielski |
| This article examines nondestructive diagnostics of electric parking brake (EPB) modules using piezoelectric sensors and machine learning. The proposed approach addresses a limitation of current-based regeneration testing, where defects are indicated only by the value of output current or torque, without identifying the specific damaged component. Piezoelectric sensors were mounted on the EPB housing near the electric motor, pinion gear, and planetary gearbox before disassembly. Short time-series voltage signals were recorded from eight modules with different internal damage. These eight individual faults were subsequently grouped into three component-level fault classes (motor, belt, and gears) and complemented by an additional class representing the undamaged EPB module (healthy). Linear statistical features—such as peak-to-peak and root mean square—were extracted from the sensor data. Machine learning classification models, including extra trees, multilayer perceptron, support vector machine, and XGBoost, were trained to distinguish among the four classes (motor, belt, gears, healthy), achieving up to 100% classification accuracy, particularly with data from the planetary gearbox sensor. The results confirm the effectiveness of selected statistical indicators for both damage detection and component-level fault classification. The method is designed for use in controlled factory-based laboratory environments during EPB regeneration and quality control. This noninvasive diagnostic technique enables early identification of specific component faults without disassembly, reducing downtime and associated costs. |
