Forecasting Parts Demand Using Service Data and Machine Learning Volume 1
Abstract
The Parts Forecasting Using Service Data and Artificial Intelligence (AI) short-term project (STP 9-L-04) evaluated machine learning (ML) as a potential approach for improving forecasts of part demands. Data sparsity limited the success of ML across a wide range of techniques. As a result of this research focused recommendations for follow-on R and D are identified to explore the successful elements of this investigation. Due to the sparsity of the available demand data, using Maintenance and Availability Data Warehouse (MADW) data alone is insufficient to improve Defense Logistics Agency (DLA) demand forecasts beyond those using established DLA models. The F/A-18E/F was selected as the trial platform for the project. Data sparsity was encountered across all platform component part demands. However, ML models of maintenance events, rather than part demands, delivered improved forecasting over time series modeling while addressing data sparsity concerns. Further research should be performed to determine whether this approach can enhance customer service, leading to improved readiness of the supported weapon systems. To improve further analysis, use of additional data, beyond MADW, should be explored to develop a more complete view of supply chain demands. The multi-echelon supply system should be modeled for added precision. Consumption data along with both retail and wholesale demands should be included in the follow-on models.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 2020
- Accession Number
- AD1098738
Entities
People
- Catherine M. Beazley
- Esther C. Thron
- John A. Stephenson
- Justin D. Ward
- Karl M. Kruger
- Russell S. Salley
- Sergio Posadas