Dual-Threshold Maintenance Optimisation for Hydraulic Floodgates under Runaway Stochastic Degradation

A stochastic predictive degradation modelling framework for hydraulic floodgates operating with combined environmental and operational influences is presented. A condition-based maintenance approach is optimised for a multi-gate system to identify the most cost effective negative opportunistic preventative replacement policy, in order to minimise concurrent maintenance interventions. Our research focuses on the hydropower context, where evolving operating practices and regulatory constraints have increased exposure to the highly degrading fluid-induced vibration mechanism. A coupled simulation of the system is developed integrating (i) a synthetic hydrological forcing process, (ii) a reservoir-operation control model, and (iii) a stochastic degradation model based on a gamma process with state-dependent parameters. The degradation dynamics capture the interacting feedback loops that connect seal wear, leakage, vibration and erosion, with environmental conditions and operating decisions, leading to a run-away degradation effect beyond critical thresholds. A dual-threshold preventive maintenance policy is implemented using vibration as an observable proxy for a hidden degradation state, considering for inspection intervals, degradation uncertainty, and operational constraints on gate availability. The policy performance is evaluated via long-term Monte Carlo simulation, optimising expected annual cost under variations in conditions. The expected cost is highly sensitive to the lower preventive threshold, which effectively mitigates runaway degradation, while the upper threshold shows limited influence due to increasing end-of-life uncertainty. This undermines the effectiveness of negative opportunistic maintenance strategies aimed at avoiding simultaneous interventions for systems with runaway degradation behaviour. The findings emphasise the critical role of degradation non-linearity and information limitations in maintenance decision-making, and indicate that improved monitoring may be necessary to effectively make use of condition-based maintenance policies for similar applications.