[Paper Review] Inspection plan prediction for multi-repairable component systems using neural network

Nooshin Yousefi, Stamatis Tsianikas+2 more|arXiv (Cornell University)|Jan 23, 2020

Reliability and Maintenance Optimization13 references4 citations

TL;DR

This paper proposes a dynamic inspection planning method for multi-repairable component systems subject to degradation and random shock failures using a neural network to predict the next optimal inspection time. By modeling system reliability and cost rate functions, the approach dynamically accounts for varying initial ages of components after each repair, significantly reducing maintenance costs while improving system reliability through adaptive scheduling.

ABSTRACT

Implementing an appropriate maintenance policy would help us to have a more reliable system and reduce the total costs. In this paper, a dynamic maintenance plan is proposed for repairable multi-component systems, where each component is subject to two competing failure processes of degradation and random shock. For systems with individually repairable components, it is not economical to replace the whole system if it fails. At any inspection time, the failed components can be detected and replaced with a new one and the other components continue functioning; therefore, the initial age of each component at any inspection time is different from other components. Different initial ages have an effect on the optimal time that the whole system should be inspected. The optimal inspection time should be calculated dynamically considering the initial age of all the components and their configuration within the system. In this paper, a neural network method is used to predict the next optimal inspection time for systems considering the initial age of components at the beginning of the inspection. System reliability and cost rate function are formulated and used to train the prediction model. The proposed maintenance plan is demonstrated by numerical examples

Motivation & Objective

• To develop a dynamic maintenance policy that adapts to the varying initial ages of components after individual repairs in multi-component systems.
• To reduce total maintenance costs by predicting the optimal next inspection time based on current system state.
• To address the challenge of non-uniform component aging in repairable systems, where component replacement affects system-level inspection scheduling.
• To integrate system reliability and cost rate functions into a machine learning framework for real-time inspection planning.

Proposed method

• A neural network is trained using system reliability and long-term cost rate functions as input features.
• The input to the network includes the initial ages of all components at the start of each inspection cycle.
• The network predicts the optimal time for the next system inspection, minimizing expected maintenance costs.
• System configuration and failure processes (degradation and random shocks) are modeled using probabilistic reliability functions.
• The training data is generated from numerical simulations of component degradation and shock processes.
• The model dynamically updates inspection schedules based on component-specific ages post-repair, ensuring cost-effective maintenance.

Experimental results

Research questions

• RQ1 How can the optimal inspection time be dynamically determined in a multi-repairable component system with varying component ages?
• RQ2 What impact do component-specific initial ages have on the system-level inspection planning and cost efficiency?
• RQ3 Can a neural network effectively learn and predict optimal inspection intervals based on reliability and cost rate functions?
• RQ4 How does the proposed method compare to static or fixed-interval inspection policies in terms of cost and reliability?

Key findings

• The neural network model successfully predicts the next optimal inspection time with high accuracy by learning from system reliability and cost rate functions.
• The dynamic inspection policy reduces total maintenance costs compared to fixed-interval inspection strategies.
• The model accounts for heterogeneous component aging after repairs, enabling more precise and cost-effective scheduling.
• Numerical examples demonstrate that the proposed method maintains high system reliability while minimizing long-term cost rates.
• The approach is scalable to systems with multiple components and complex configurations, showing robustness under varying failure process conditions.