Text Analytics of PDF Technical Documents
The Air Force required a logistics data crosswalk to mitigate known maintenance and supply data connection challenges limiting accurate demand planning and forecasting.
ILW data scientists used natural language processing (NLP) and unsupervised machine learning (ML) techniques to evaluate and determine an automated method to tie Work Unit Code (WUC) to related National Item Identification Numbers (NIINs). They used information extracted from Technical Orders in native PDF format as well as data captured in maintenance and supply data systems.
- Extracted master parts list (MPLs) for two Air Force weapon system programs
- Developed multiple table extraction techniques that read PDF documents and pull tabular information out with high degrees of accuracy. Techniques leverage and improve open-source libraries
- Provide enterprise search capability of Air Force technical documents
- Improves parts supportability, contract lead times, integrated repair planning
- Enables planning for predictable shifts in demands and condemnations, buying the right quantities of the right parts, avoiding overbuy on other parts
- Open-source Python solution using DoD-compatible libraries: Pandas, Tabula and Fitz, Scikit-learn, and OpenCV
- Native PDFs
- Text analytics, NLP, Machine Learning