Data-driven warranty modeling for multi-state deteriorating products with stochastic repair times
针对维修时间随机的多状态退化产品，提出数据驱动方法提取保修模型，并用改进的Proxel仿真确定最优维修或更换策略，发现数据偏差会导致可靠性低估。
This paper presents a data-driven approach to extracting warranty models for multi-state deteriorating repairable products, with a focus on scenarios involving stochastic repair times. While analytical solutions exist for cases with negligible or fixed repair/replacement durations, no explicit solutions are available when these times are stochastic. To address the lack of explicit solutions for stochastic repair times, we first extract models from failure and repair data, then apply a modified Proxel-based simulation to determine optimal repair-replacement policies that minimize expected warranty servicing costs per item sold. Our results reveal that when minimal repairs are performed instantaneously, replacement is generally favored over early repairs. Conversely, when repair times are non-zero, the system tends to prefer repair over replacement. Additionally, we find that data-driven warranty models evolve with continuous data integration but often underestimate reliability due to biased failure data, highlighting the need for bias-aware modeling techniques.