Azure AI Custom Vision Application

A Case Study In Continuous Improvement Using AI.

Several years ago, Credo Software collaborated with a client’s quality and logistics departments to create a digital lean solution to solve a parts kitting issue. Before the Kitting Count Tool, associates were using paper lists to collect components for each work order. These components were placed in a “kit” and transferred to the assembly area. However, assembly operators would often report incorrect components or the incorrect quantity of components. Upon further investigation it was found that on occasion, an assembly operator would accidentally drop a component, and seek to blame the “kitters”. ? This constant blame game between the two departments was slowing down production and reducing quality.?

To solve and improve the accuracy of these kits and end the “blame game”,? the Kitting Count Tool solution was developed to enable “digital counting” by kitters as components were gathered.? This was done by laying the components on a tray which was equipped with a camera and the kitter would “tap” each part on the monitor. Each tap was highlighted and quantity was then verified against the order at hand. This image was then saved showing the final result. This greatly improved the accuracy of each kit and all but eliminated finger pointing between these groups. Using this digitalization to safeguard the manual shipping processes, the Kitting Count Tool won national recognition in 2017.

Credo is excited to share the continuous improvement initiative of this Kitting Count Tool for our clients.? The second generation of this solution is utilizing and training AI to both count AND identify these components. So, no longer will the kitter need to “tap” the components on the screen. Rather, the kitter will load the components on the tray and the AI visualization will detect not only the “count”, but also verify that the part is the “correct part”. This is a game changer for the kitting personnel.?

The object detection component of Count Tool is using Microsoft Azure Custom Vision to train an AI object detection model using historical Count Tool photos. This ensures that the model is trained against “realistic” photos and not simply “staged” photos. To create the model, the solution sends Custom Vision the photos and then box and tag each instance of a part.??

Training the model is Microsoft AI magic. Consuming the model is straightforward, an image is sent to the Azure Custom Vision API in real time and results from the model are used to box and identify parts automatically and produce a “predicted accuracy” value.? If the accuracy rating is above the limits set by the Logistic engineers, the component is accepted and the kitter moves to the next part.?

This AI integration into the Kitting Count Tool has greatly elevated this work efficiency project. ? Reduced kitters fatigue and time by eliminating the need to tap the screen. Improved quality and increased production on the assembly side.??

Using this innovative AI integration, Credo Software is proud to be a Microsoft Partner.

https://azure.microsoft.com/en-us/products/ai-services/ai-custom-vision.