Addressing C-130 reliability issues in a timely manner requires connecting supply and maintenance data for ordering new parts or repairing and replacing sections of the aircraft. However, no formal mapping between Work Unit Codes (WUCs) and Part Numbers (PNs) existed, which introduced inefficiencies and delays in fixing reliability problems. Using natural language processing (NLP) and machine learning. MERC developed a solution to quickly map United States Air Force supply and maintenance data for the C-130H and C-130J. Machine learning is the ideal approach for modeling complicated and time-dependent relationships between variables for large amounts of data. Since almost every WUC and PN has a description associated with them, MERC used NLP to encode these descriptions as mathematical vectors and machine learning to match WUCs and PNs with similar descriptions to bridge the gap between the supply and maintenance systems.
In developing the machine learning-based NLP model, MERC extracted textual data from over 500 aerospace documents associated with the C-130 and incorporated this data into the model. MERC used the NLP model, AirBERT, to generate WUC-to-PN and PN-to-WUC matches for the systems in both the C-130H and C-130J. The MERC team additionally used AirBERT to reduce the time and partially automate the process of correcting the WUCs assigned to C-130 maintenance records, thereby improving the speed at which these records can facilitate timely ordering of parts needed for repair or replacement.
