Abordagem heurística do método bootstrapping para a previsão de demanda de itens sobressalentes : aplicação na indústria mineral

Abstract

Forecasting spare parts demand and measuring its effectiveness are major challenges in inventory management. Particularly in the mineral sector, hybrid approaches are used to incorporate specific parameters of each mine to reduce the gaps found in traditional forecasting methods when applied to intermittent data sets. Aiming at a more effective and replicable method – for different mines and sectors - this study proposes a heuristic approach to Willemain, Smart and Schwarz (2004) bootstrapping method for the forecasting of spare parts, considering 2-OPT and SWAP algorithms to model autocorrelation instead of the traditional two-state Markov process. The proposed method was compared to the original bootstrapping of Willemain, Smart and Schwarz (2004) and to the traditional methods of Exponential Smoothing, Croston, Syntetos and Boylan Approximation (SBA) and Revised Croston Method (TSB). The results obtained from a 11-year historical data set of approximately 13,000 items of the mineral industry through the MAE, MSE, breakdown cost and inventory cost metrics showed that the heuristic method results in more accurate forecasts of demand distribution over a fixed lead time and reduces stock shortages of intermittent items still in the usual cycle of use, presenting superior performance to all traditional methods without linking the forecast to stock obsolescence. Moreover, the heuristic method lead to a 35% reduction in breakage costs when compared to the classic bootstrapping method. For erratic, irregular and regular demand patterns it was possible to identify conditions in which the traditional methods outperform the bootstrapping approach. An exploratory analysis of the spare parts demand field in the mining industry as well as an analysis of the error metrics effectiveness for predictive performance measurement in extremely intermittent series are also presented in the study.