140D0424C0006

Small Business Innovation Research Program (SBIR) Phase II

Deuce Coupe

The Reduce, Reuse, Recycle mantra often gets reduced to its easiest and least impactful last member --- ``just put it in the blue bin.' Deuce Coupe looks to embrace the practice of reuse to mitigate vulnerabilities in supply chains. Reusing all or some parts of discarded equipment eliminates waste and the need to produce a replacement. However, there are barriers to reusing scrapped equipment such as: keeping an inventory of scrapped parts, finding applicable replacements for a broken component amongst that inventory, and assessing the reusability of salvaged parts. Unlike recycling, which can be done in bulk by converting assemblies to raw materials, reusing is a retail business. The manpower intensive nature of such a retail business limits the reach and scale of reuse efforts. We propose to develop TreyUse, a part reuse marketplace made possible by three workflows: inventory of available components, expansion of requests, and assessment of suitability of potential matches between inventory and expanded requests Workflow 1: Inventory of reusable components. We use NLP processes to parse assembly instructions, repair manuals, and repair logs to extract components lists, part numbers, manufacturers, and generic equivalents. We learn equivalence classes at the system, sub-assemblies and component levels to facilitate reasoning about potential substitutions. We use the latest advances in computer vision to process images of parts, matching them with diagrams in documentation, automating classification and later facilitating information retrieval regarding their extraction. Workflow 2: Expansion of requests. Using the extracted ontologies and equivalence classes of assemblies from documentation we then develop a relational graph to reason part matching. We exploit commonality in adjacent parts or sub-assemblies to evaluate potential compatibility between components. We generate candidates for replacements that meet basic requirements and that can be further evaluated by the suitability model and the TreyUse operator. Workflow 3: Assess suitability for use.} A model is trained to determine part usability based on visual defects and other case by case contexts. Training data for this model is continuously generated based on TreyUse and is composed of failed parts along with acceptable, working parts. If a certain part is exceptionally rare, or has poorly cataloged images, we employ a synthetic augmentation algorithm to ensure model performance . Barnstorm Research has assembled a team of experts in the foundational technologies necessary to successful research, implement, commercialize and transition TreyUSe. Raytheon BBN, a world leader in NLP will develop the component that ingests documentation and repair logs. PARC will develop the image processing capabilities to identify and locate components, and Barnstorm Research will develop the knowledge base and reasoning models.

The goal of phase II is to continue the R&D efforts initiated in Phase I. Funding is based on the results achieved in Phase I and the scientific and technical merit and commercial potential of the project proposed in Phase II.

HR0011SB20234-07

D2D-0558

23.4

John Francis N Mergen