A Comprehensive Prediction Approach for Hardware Asset Management

Abstract

One of the main tasks in hardware asset management is to predict types and amounts of hardware assets needed, firstly, for component renewals in installed systems due to failures and, secondly, for new components needed for future systems. For systems with a long lifetime, like railway stations or power plants, prediction periods range up to ten years and wrong asset estimations may cause serious cost issues. In this paper, we present a prediction approach combining two complementary methods: The first method is based on learning a well-fitted statistical model from installed systems to predict assets needed for planned systems. Because the resulting regression models need to be robust w.r.t. anomalous data, we analyzed the performance of two different regression algorithms - Partial Least Square Regression and Sparse Partial Robust M-Regression - in terms of interpretability and prediction accuracy. The second method combines these regression models with a stochastic model to estimate the number of asset replacements needed for existing and planned systems in the future. Both methods were validated by experiments in the domain of rail automation.