Task assignment in predictive maintenance for free-float bicycle sharing systems

Abstract

The growth of Free-Float Bike-Sharing Systems (FFBSs) is heavily impeded by faulty bike maintenance among other operational challenges. In this paper, we aim to improve the efficiency of faulty bike maintenance by predicting faulty bikes in order to make better maintenance assignment decisions. Inspired by industry practice, we identify the role of “black holes” in accurate predictions of faulty bikes: locations with morbidly high faulty rates, which can be characterized using data-driven approaches (clustering and convex hull). Based on the prediction result, we propose two maintenance policies, i.e., the pooling model and the dedicated model, for the faulty bike maintenance assignment problem with the objective of minimizing the sum of maintenance time cost and travel time cost. Finally, we provide a tractable reformulation via linear mix-integer Second-Order Conic Programming (SOCP) and conduct a case study with real data. Our analysis identifies the main trade-off between routing efficiency and maintenance efficiency in the different maintenance policies. We find that the pooling policy concentrates on routing efficiency while the dedicated policy emphasizes maintenance efficiency. Moreover, we demonstrate the importance of “black holes” in the prediction of faulty bikes. In the case study, we observe that bikes in “black holes” are about 70% more likely to be faulty than those out of “black holes.” We find that the improvement due to prediction is significant even when the prediction is imperfect. In our case study, when prediction accuracy exceeds 65%, we can observe the cost reduction by prediction in the faulty bike maintenance problem.