Exploring predictive maintenance of bike-sharing systems using user-centric information

Bike-sharing systems are becoming increasingly important for sustainable urban mobility, but they need much maintenance, making them less efficient. Although predictive maintenance has been examined using technical and usage data, there has been insufficient focus on user-centric information as an indicator of bicycle reliability. This study examines the incorporation of user ratings into predictive maintenance models for the Lisbon GIRA bike-sharing system. We used linear mixed models and generalized additive models to examine a 1-month dataset with travel records, user feedback, and maintenance logs. Four formulations of Remaining Useful Life were developed, with riding time exhibiting the most reliable prediction capability. The findings show that user ratings are favorably related to Remaining Useful Life, and both models had an R 2 of about 0.50. Residual analysis often undervalues severe results, which successfully indicate the need for maintenance sooner. Kernel density estimations and bathtub curve analysis further demonstrated the degradation effects linked to cumulative usage. The results suggest that putting the user at the heart of information enhances predictive maintenance strategies, making operations more dependable and scheduling maintenance easier in large-scale bike-sharing system operations.