Abstract

This article proposes condition-based maintenance (CBM) for a system subject to dependent soft and hard faults. If the degradation exceeds a predetermined level, the system will experience a soft failure that can be resolved by a corrective general repair or replacement.

In each inspection period, three possible actions are considered, namely no maintenance, preventive general repair and preventive replacement. The hard failure rate, depending on age and deterioration, is described by a proportional hazard model, where the covariate is characterized by a gamma process.

The fatal flaw can be fixed with a minimal repair, a general repair or with a replacement. Unlike previous CBM guidelines, no thresholds are used to define maintenance actions. The objective is to determine the optimal maintenance policy that minimizes the expected average costs per time unit in the long run.

The optimization problem is formulated in the SMDP (Semi-Markov Decision Process) framework and solved by the policy iteration algorithm. Two practical examples are given to illustrate the proposed policy. Comparison with other CBM policies shows excellent results of the proposed policy and the importance of minimal repairs and general repairs in CBM decision making.

General Repair : Introduction

The high quality requirements of modern production systems underline the importance of preventive maintenance strategies. Most TPM decisions are made by experienced planners based on device manufacturers’ suggestions, historical failure data, operational experience, and judgments of maintenance personnel and engineers (Peng, Dong, Zuo, 2010). On the other hand, CBM takes into account updated information from condition monitoring for decision making (Makis et al., 2006, Wang et al., 2019).

With the development of advanced sensor technology and measurement techniques, CBM is increasingly used to improve the decision-making process for the maintenance of many industrial systems such as wind turbines, construction machinery, and cyber-physical manufacturing systems (Ansari, Glawar, Nemeth, 2019).

Most existing CBM guidelines focus on soft errors, which occur when the level of degradation exceeds a predetermined level (Liao, Elsayed Chan, 2006). The evolution of decline processes can be modeled by discrete statistical stochastic models such as Markov and semi-Markov models (Kim et al., 2011, Khaleghei and Makis, 2016, Naderkhani Of these models, the gamma process is particularly suitable for stochastic modeling of monotonic and gradual degradation such as concrete creep,

fatigue cracking and steel corrosion (van Noortwijk, 2009). Mercier and Pham (2012) considered a CBM guideline for a continuously monitored system that decomposes according to a bivariate gamma process. Duan, Makis, and Deng (2018) proposed a CBM guideline for a mechanical system that deteriorates after a gamma process.

In addition to soft failures, many degraded systems also experience random failures caused by hidden manufacturing defects, overloading, external shocks, etc. This failure mode is called hard failure.

For example, marine diesel engine cylinder liners fail due to the competition between wear deterioration (soft failure) and thermal cracking (hard failure) (Bocchetti, Giorgio, Guida Pulcini, 2009). Compared to soft errors,

hard errors tend to have more serious consequences because they can suddenly interrupt the continuity of a production process and result in significant lost downtime (Zheng, Su, Zheng, 2019). Therefore, it is necessary to consider both soft and hard errors when making CBM decisions.

Soft and hard faults are often considered to be independent of each other (Huang and Askin, 2003; Zhu et al., 2010; Ye et al., 2012). However, in many practical situations, the interdependence is important and should not be neglected (Wang and Pham, 2011a, Wang and Pham, 2012, Zhang et al., 2016).

A widely accepted form of addiction is that hard failure is more likely when the level of impairment is higher (Wang and Pham, 2011b; Huynh et al., 2012). Huynh, Barros, Bérenguer, and Castro (2011), Huynh et al., 2014, Caballé et al., 2015, and Yang, Ma, and Zhao (2017) considered a two-segment hazard function that assumed that the hazard rate jumped from the normal range to the defective range when system degradation exceeds a certain level.

Huynh et al., 2011, Caballé et al., 2015 jointly optimized the inspection interval and the preventive replacement threshold. The estimated mean remaining life (MRL) was published in Huynh et al. (2014) is used as an effective index for CBM decision making.