Introduction

 Abstract 

In this work we are trying to reduce the computational costs of resources optimisation in Reliability, Availability & Maintainability (RAM) models of complex systems by applying Genetic Algorithms (GA) for the search process. Optimisation processes in these types of models are often prohibitively expensive since the resources⁽¹⁾ space has many degrees of freedom, consisting of a large number of solution points, where each solution requires a complete Monte Carlo Simulations (MCS), which are often slow making the entire process expensive. MCS must be applied in such problems since generally the form of the equations which govern the solution is a multidimensional integral transport equation (the System Transport Equation) or a set of simultaneous integral equations. The System Transport Equation stems from the Boltzmann Transport Equation for neutral particles. A full discussion can be found in Dubi, 2000. One way of bypassing this obstacle is to use a hybrid process which uses fewer MCS runs to provide information to an analytic algorithm using bulk parameters to search within the resources topology with reliable references. In that manner, the analytic algorithm learns from the MCS and the analytic predictions are verified and corrected.

Although the Hybrid optimisation is very effective and accurate, it requires a full understanding and study of the model to set it up. Moreover, it is limited to spare parts and repair teams as the resources and so far has been tailored only to availability as a response function. We wish to expand the optimisation to any controlled parameter of the problem's input, such as preventive maintenance (PM) plan or alternative management policies, different logistic structures etc., and to apply the search process to other functions, such as to Whole Life Cost (WLC), profit, safety related issues and so on.

We start with a definition to our optimisation problem. We then introduce the concept of Sufficiency, which will be utilised also for the GA at a later stage. After that and, before we introduce the Hybrid method we introduce and describe the studied model scenario. As a starting point for the use of the GA we take a series of random samples from the Resources space, which we follow by application of the GA, enabling a more efficient search process.

 Optimisation