SBIR-STTR Award

Simple Technology Solutions (STS) proposes leveraging Google Cloud Artificial Intelligence (AI) / Machine Learning (ML) to conduct autonomous NDIs to determine or validate if pitting or uniform attack corrosion exists, and therefore if maintenance or repairs are needed. We will use Google AutoML and publicly available datasets of damaged and undamaged platforms to create and train a model that recognizes pitting and attack corrosion. We will then apply the model to high resolution RGB images gathered via American drone services, and iteratively test and re-train the model to achieve the highest possible accuracy measurement. Our hypothesis is that if you train Google AutoML to identify and flag uniform attack corrosion and pitting in aerial images (thereby prioritizing maintenance and repair needs), you can remove labor from external surface inspections in terms of both total hours and necessary on-site expertise.

Benefit:
STS's ultimate vision is removing labor to the maximum extent possible from maintenance inspections across industries. We are designing and developing the solution to be

Keywords:
aerial images, aerial images, Predictive maintenance, Artificial Intelligence / Machine Learning, Google, Modeling

For Phase I we used commercially available hardware and software to develop a computer vision AI/ML model that detects corrosion in aerial images of vessels. This required robust cloud hosting and cloud platform/applied machine learning capabilities; a fully automated data pipeline; and extensive collaboration with our drone flight services partner to obtain high quality images. To increase the business value to the Navy and commercial viability, we plan to turn the model into a platform solution in Phase II. Our goal for Phase II is to create a data fusion platform of which the existing automated data pipeline and corrosion detection model are pieces. Making this a standalone platform requires a front-end user interface (UI) and enhanced integration and modeling capabilities. The UI will turn data analyses into actionable business insights provided directly to the inspectors/business users in the field. Without a UI the model cannot serve up the results of the analyses and business users have no way to engage with the model or the findings. We will also fuse other sensor data with the RGB images to account for invisible corrosion undetectable optically. The fusion of these data sources/analyses will predict maintenance and repair needs. Based on Phase I lessons learned and MRO market research, our technical objectives for Phase II are: Enhance data capture using autonomous drones. Develop a multi-label classification algorithm. Develop end user interface and enhanced 3D modeling capabilities. Provide CONOPS for integration with existing maintenance and scheduling systems. Investigate the use of other sensors (eg. infrared, acoustic, electrochemical) and submersibles to identify

Benefit:
Maintenance inspections of Navy and Department of Defense platforms are still a highly manual process. This is also true of infrastructure maintenance inspections in adjacent industries like commercial maritime, oil and gas, transportation, constructions and facilities. Some of these industries utilize drones to capture inspection data in hard to reach places (eg. high in the air), but review of the captured images is still performed by inspectors or other subject matter experts. Our solution automates not only the data capture process but more importantly data and image analysis. The benefit is a drastic reduction in the labor hours required to perform maintenance inspections as well as the fusion of different data sources to most accurately detect and predict maintenance needs. We have validated that our solution fills a void in the aforementioned inspection markets, and there are several prospective buyer groups within each of those supply chains - the end customers that own the assets; the drone companies collecting inspection data; and the inspectors/corrosion engineers (in-house or contracted) that perform the analyses. We are partnered with representatives of several of these buyer groups (i.e. DroneUp and Elzly), who have already expressed interest in acquiring the solution themselves in addition to brokering discussions with their end customers. In 2019 the MRO market in the United States was valued at more than $115B so, as demonstrated above, theres no shortage of prospects. Regarding prospective federal clients, we are targeting DON (especially NAVSEA) and the U.S. Coast Guard (USCG). Currently the Department of Defense spends $20B annually on corrosion-related maintenance and repair, $8B of that is the Navy and the preponderance is vessels. Our solution can reduce labor costs associated with performing those maintenance inspections, and also avoid costs incurred by servicing or drydocking vessels unnecessarily. There is a similarly addressable market at USCG, which currently allocates more than $1B annually to vessel inspections, maintenance, and repairs. State and Local agencies are prospects too as they own 87% of large government infrastructure including the vast majority of bridges, transportation, highways and streets.

Keywords:
Automation, Artificial Intelligence, cloud computing, Machine Learning, Drones, data fusion, AI/ML, digitalization