Innovating Quality Control: The Journey to Creating the CV ML QC Station
In the realm of manufacturing, quality control (QC) is paramount. However, traditional QC methods often struggle to keep pace with the demands of modern production scales, particularly in situations where physical oversight is limited, such as during the COVID-19 pandemic. Faced with this challenge while scaling production for the EDN SmallGarden in Shenzhen, and hindered by travel restrictions, we noticed a concerning trend: cosmetic defects were slipping through and reaching customers. The need for a solution was urgent and clear.
The Alternatives Didn't Work
The quest for an effective and cost-efficient quality control mechanism led us down various paths, including consultations with vendors offering visual inspection services. However, the steep price tags of around $20,000 per month were far from feasible for our needs. It was within this context that the concept of the Computer Vision ML QC Station was born—a self-contained, simple yet sophisticated inspection unit designed to enhance our QC capabilities through the power of machine learning (ML) and computer vision (CV).
The Rig
The station is ingeniously designed around an enclosed box, housing five LG 4K web cameras under uniform LED lighting to capture detailed images of products from multiple angles. This setup ensures comprehensive visual coverage, crucial for identifying even the most minor cosmetic flaws. The simplicity of its operation belies the complexity of its backend—a system where images captured are analyzed using a machine learning pipeline powered by AWS technologies.
The frontend interface of the Computer Vision ML QC Station is user-friendly and intuitive. With just a click, it captures images of the product, subsequently displaying a green checkmark for a pass or a red X for a fail, based on the analysis. This immediate feedback loop is crucial for streamlining the QC process in real-time.
The Pipeline
Behind the scenes, the machine learning pipeline is robust, utilizing AWS services such as Amplify, AppSync for API routing, S3 for storage, SageMaker for running the ML models that inspect for cosmetic defects, and Amazon Lookout for Vision to enhance the detection accuracy. This setup not only automates the inspection process but does so with remarkable efficiency and precision.
The Result
The development and deployment of the Computer Vision ML QC Station has significantly improved our QC processes, allowing us to maintain high-quality standards despite the challenges posed by remote operations. By leveraging advanced ML and CV technologies, we have not only addressed a critical need within our manufacturing workflow but also set a new standard for quality control in the industry.
That, and the amount of resources needed was significantly less than available options on the market. Ultimately this solution costed <$2000 and was able to be deployed in a few weeks time!
