Scholarlink

<p>Accurately estimating future demand for service parts during a product’s end-of-life (EOL) phase is essential for ensuring long-term support while avoiding costly overstock or shortages. This study proposes a decay-function-blended machine learning (DFB-ML) framework for forecasting spare-part demand across extended service horizons. The framework integrates historical consumption data, warranty failure rates, and attrition-adjusted fleet estimates to model lifecycle behaviour for 1,709 automotive part numbers. Feature engineering captures both short-term trends and long-term decay through variables such as lagged demand, replacement intensity, and vehicle dropout dynamics. Multiple regression models were evaluated, with Random Forest achieving the highest forecasting accuracy (Safe Mean Absolute Percentage Error = 4.36%). An ablation study confirmed that moderate decay blending (α = 0.2–0.4) yields stable long-horizon forecasts and sub-linear error growth over the 8-year horizon. The framework was further validated for scalability within ERP/SAP-linked distributed networks, demonstrating readiness for industrial deployment. The resulting forecasts support data-driven Last-Time-Buy (LTB) decisions through part-wise procurement recommendations and risk-adjusted buffers. This approach unifies lifecycle decay modelling with machine learning and provides a generalizable blueprint for uncertainty-aware EOL inventory forecasting in engineering and supply-chain domains.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A machine learning framework for long-term forecasting of spare part demand in end-of-life product scenarios

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月3日

查看全文
  • Sendhil Nathan B,
  • B. Veera Siva Reddy,
  • V. S. Shenbaga Sujan,
  • C. Chandrasekhara Sastry,
  • J. Krishnaiah,
  • Santi Jitpichitchai

摘要

Accurately estimating future demand for service parts during a product’s end-of-life (EOL) phase is essential for ensuring long-term support while avoiding costly overstock or shortages. This study proposes a decay-function-blended machine learning (DFB-ML) framework for forecasting spare-part demand across extended service horizons. The framework integrates historical consumption data, warranty failure rates, and attrition-adjusted fleet estimates to model lifecycle behaviour for 1,709 automotive part numbers. Feature engineering captures both short-term trends and long-term decay through variables such as lagged demand, replacement intensity, and vehicle dropout dynamics. Multiple regression models were evaluated, with Random Forest achieving the highest forecasting accuracy (Safe Mean Absolute Percentage Error = 4.36%). An ablation study confirmed that moderate decay blending (α = 0.2–0.4) yields stable long-horizon forecasts and sub-linear error growth over the 8-year horizon. The framework was further validated for scalability within ERP/SAP-linked distributed networks, demonstrating readiness for industrial deployment. The resulting forecasts support data-driven Last-Time-Buy (LTB) decisions through part-wise procurement recommendations and risk-adjusted buffers. This approach unifies lifecycle decay modelling with machine learning and provides a generalizable blueprint for uncertainty-aware EOL inventory forecasting in engineering and supply-chain domains.