The role of prescriptive data and non-linear dimension-reduction methods in spare part classification. (January 2023)

• Record Type: Journal Article
• Authors: Sheikh-Zadeh, Alireza; Scott, Marc A.; Enayaty-Ahangar, Forough
• Is Part Of: Computers & industrial engineering. Volume 175(2023)
• Publication Date: 2023-01
• Subjects: ADU Annual dollar usage -- DC Distribution center -- DEA Data envelopment analysis -- EBO Expected number of backorders -- KPCA Kernel principal component analysis -- MIME Multi-item and multi-echelon -- OF Order frequency -- PC Principal component -- PCA Principal component analysis -- RSO Ratio of shortage and ordering cost -- SKU Stock keeping unit -- TICC Total on-hand inventory cost for classified items; Spare part replenishment -- Multi-item and multi-echelon model -- Dimension-reduction analysis -- Classification -- Service level classification; Engineering -- Data processing -- Periodicals; Industrial engineering -- Periodicals; 620.00285
• DOI: 10.1016/j.cie.2022.108912
• Languages: English
• ISSNs: 0360-8352

Abstract

Abstract: Servitization business trends have impacted spare parts management processes significantly. These trends result in the need for firms to invest in increased inventory levels to address demand driven by the growth in the long tail of spare parts assortments. This study proposes data-driven spare parts inventory ranking and classification approaches for continuous review, multi-item and multi-echelon (MIME) spare part replenishment systems that assign group-specific service levels and control measures to spare parts. We first show that any form of, even sub-optimal, prescriptive data as an input for classification significantly improves classification performance. We also propose that the stochastic nature of the MIME systems necessitates the utilization of nonlinear dimension-reduction methods for ranking items as opposed to commonly used linear methods. Further, we introduce a detailed classification performance measurement and group-specific service level assignment that enhance decision-making after classification. Finally, based on the MIME spare part management system of a large public transit agency in the United States and several carefully synthesized problem instances, our numerical study indicates that the new approach strongly outperforms the alternatives by a margin of 8.5%. Highlights: Including both prescriptive and descriptive data enhances inventory classification. Nonlinear ranking outperforms linear methods, particularly for less collinear data. The length of training does not improve the inventory classification performance. The proposed method is evaluated by real public-transit and simulated datasets.