Low-Data Predictive Maintenance of Railway Station Doors and Elevators Using Bayesian Proxy Flow Modeling

TL;DR

This paper introduces a Bayesian-based low-data predictive maintenance framework for railway station doors and elevators, utilizing passenger flow data and expert knowledge to estimate maintenance needs and optimize scheduling.

Contribution

It presents a novel Bayesian modeling approach that infers asset loads from passenger flows without direct condition monitoring, enabling cost-effective maintenance planning.

Findings

Proxy operational data supports maintenance scheduling effectively.

The framework improves maintenance alignment over calendar-based policies.

Simulation results demonstrate practical applicability in real station scenarios.

Abstract

This paper proposes a low-data predictive maintenance framework for automatic doors and elevators in a railway station building. The method is intended for assets without direct condition monitoring, where only aggregate passenger traffic information and expert knowledge about movement patterns are available. Passenger flows are modeled on a reduced station graph using a Bayesian formulation with uncertain totals and routing shares. The inferred flows are converted into approximate operating-cycle loads for doors and elevators through simple stochastic proxy relations. These loads are combined with uncertain age- and cycle-based maintenance thresholds to estimate the probability that predefined maintenance conditions have been reached. A cost-aware scheduling model is then used to align maintenance activities while accounting for service costs, disruption, delay penalties, and grouping opportunities within each asset class. The framework is illustrated on a simulated case study reflecting a real station layout. The results show that proxy operational data can support maintenance scheduling with low incremental implementation cost and can improve alignment relative to a calendar-based policy.