Prioritizing the purchase of spare parts using an approximate reasoning model

Abstract

The complexity of a spare parts prioritization model should be consonant with the amount and quality of data available to populate it. When production processes are new and the reliability database is sparse and represents primarily expert knowledge, an approximate reasoning (AR) based model is appropriate. AR models are designed to emulate the inferential processes used by experts in making judgments. The authors have designed and tested such a model for the planned component production process for nuclear weapons at Los Alamos National Laboratory. The model successfully represents the experts' knowledge concerning the frequency and consequences of a part failure. The use of linguistic variables provides an adaptable format for eliciting this knowledge and a consistent basis for valuing the effect on production of different parts. Ranking the parts for inclusion in a spare parts inventory is a straightforward transformation of the AR output. The basis for this ranking is directly traceable to the elicitation results. AR-based models are well-suited to prioritization problems with these characteristics.