SIFTED: Semantic Information for Faults, Troubleshooting, and Extraction from Documents

To address the Navy’s need for maintaining naval aviation readiness by performing maintenance, repair, and overhaul (MRO) of aircraft and components, Redshred proposes to develop a new Semantic Information for Faults, Troubleshooting, and Extraction from Documents (SIFTED) system, based on advanced ML, natural language processing (NLP), and computer vision (CV) technologies. Specifically, the innovation in building maintenance archives, building semantically searchable vector databases, generating task graphs and performing quality assurance (QA) checks will provide a chatbot-style interface to generate recommended and detailed repair instructions (as both model-based task graphs and written instructions) from archives of existing repair instructions and will assemble multiple documents together from disparate data silos into a single maintenance archive and provide access via local chatbot to perform troubleshooting with the primary aim to retrieve already developed repairs that users may not know about. As a result, this technology offers to meet such specifications in areas such as providing detailed repair instructions, automating decision-making processes, interoperating with Navy maintenance databases, and performing quality assurance (QA) on maintenance actions, thus enhancing overall maintenance efficiency and effectiveness, which directly address the program requirements for helping with troubleshooting and performing autonomous decision-making for aviation maintenance. In Phase I, Redshred will fulfill all Phase I requirements from the solicitation and demonstrate the feasibility of SIFTED by assembling documents for maintenance tasks, developing AI/ML-based troubleshooting software, generating expanded instructions with AI/ML, performing QA analyses, generating new troubleshooting instructions, and beginning integration with Navy maintenance databases