Preventive maintenance scheduling and replacement using parallel evolutionary algorithms

Abstract

The dependability of safety-critical systems prescribes the replace policy when the reliability of any of its components drops below a prede ned unacceptable level. Various techniques have been used in the past to draw up the perfect preventive maintenance schedule based on the meantime to failure (MTTF). This paper presents a review of parallel evolutionary algorithms to accomplish this task.

Keywords

• Dependability placement Island Model (SPEA2) Optimisation Systems

Introduction

With modern safety-critical systems displaying multiple and complex failure modes, orthodox manual analysis of systems turn out to be more and more complicated and error-prone. Even rule-based and traditional safety and reliability analysis techniques are outdated. Preventive maintenance (PM), the introduction of scheduled maintenance for all or critical components of the system, is introduced to address these limitations. Preventive maintenance is naturally expressed as a combinatorial optimisation problem [ 4, 2 ].

Typically, an appropriate preventive maintenance schedule is meant to elongate the useful life of components [ 5 ] through repairs based on the meantime between failures (MTBF). However, at a certain point, these components will become unmaintainable. In this work, we extend the preventive maintenance with a replacement policy according to the meantime to failure (MTTF). In this paper, we present a review of parallel evolutionary algorithms for perfect preventive maintenance based on a replacement policy. Evolutionary algorithms (EA) represent a family of algorithms used in arti cial intelligence (AI) where the problem space is searched by focusing on the ttest 3 3.1

Ikechukwu et al individuals over several generations. An EA starts with an initial population, which it evaluates to select the ttest. After that, the algorithm iterates through the selection of the ttest, crossover and mutation until the termination criteria (an expected tness or the maximum number of iterations has been reached) are met.

Because EAs are generally computationally costly, previous research e orts have developed parallel versions of EAs (parallel evolutionary algorithms, PEAs), using a single population or multiple sub-populations. These parallelisation efforts prove useful while executing EAs on massively parallel computers or supercomputers. Alba [ 1 ] discusses three variants of PEAs, including a master-slave PEA, a ne-grained PEA and a coarse-grained PEA. A master-slave PEA uses a single population but distributes the crossover, mutation and tness evaluation to other processors. A ne-grained PEA also uses a single population and con nes individuals within a spatial structure, where they can only interact with their neighbours. Finally, a coarse-grained PEA or island model which divides the population into several sub-populations. The latter then have an amount of migration among them.

Evolutionary Selection Algorithm

The Non-dominated Sorting Genetic Algorithm II (N SGA II) N SGA II [ 3 ] is a multi-objective, multi-directional optimisation algorithm. It is an improved version of N SGA that mitigates some its drawbacks. For example, N SGA does not have an elitist feature, which improves the performance of the search process as it retains good solution from generation to generation. This limitation is addressed in N SGA II. As well, parameter tuning is handled in In N SGA II. Finally, N SGA II is a Pareto frontier algorithm; i.e., it produces optimal solutions. 3.2

Strength Pareto Evolutionary Algorithm 2 (SP EA 2) SP EA 2 [ 6 ] is an enhanced elitist multi-objective EA which uses an improved tness allotment approach di erent from what is available in its predecessor SP EA. The tness assignment approach integrates density information and the archive. Additionally, a clustering technique, which is introduced as soon as the non-dominated front surpasses the archive border, is substituted with another truncation method with similar characteristics and nevertheless keeps the boundary details. Furthermore, in SP EA 2 the archive members are involved in the reproduction process.

References

1. Alba , E.: Parallel Metaheuristics: A New Class of Algorithms . Wiley-Interscience, New York, NY, USA ( 2005 )

2. Budai , G. , Huisman , D. , Dekker , R.: Scheduling preventive railway maintenance activities . Journal of the Operational Research Society 57 ( 9 ), 1035 { 1044 ( 2006 ). https://doi.org/10.1057/palgrave.jors. 2602085 , https://doi.org/10.1057/ palgrave.jors.2602085

3. Deb , K. , Pratap , A. , Agarwal , S. , Meyarivan , T. : A fast and elitist multiobjective genetic algorithm: Nsga-ii . Trans. Evol. Comp 6 ( 2 ), 182 {197 (Apr 2002 ). https://doi.org/10.1109/4235.996017, http://dx.doi.org/10.1109/4235.996017

4. Mahadevan , M. , Poorana , K. , Vinodh , R. , Paul, R.: Preventive maintenance optimisation of critical equipment in process plant using heuristic algorithms . In: Proceedings of the 1st International Conference on Industrial Engineering and Operations Management . pp. 647 { 653 . Dhaka, Bangladesh ( 2010 )

5. Nggada , S.H. : Reducing component time to dispose through gained life . International Journal of Advanced Science and Technology 35 ( 4 ), 103 { 118 ( 2011 )

6. Zitzler , E. , Laumanns , M. , Thiele , L. : Spea2: Improving the strength pareto evolutionary algorithm for multiobjective optimization . vol. 3242 ( 01 2001 )