Introduction

The customer is a successful German business operating in the electrical engineering industry, specializing in the creation and distribution of aesthetic and functional connection points, connector strips, power distribution options, and high-performance power distribution units for various applications. The client wanted to improve the speed and quality control of their manual assembly lines, so they looked for a solution that would let them monitor the production in real time and automate data collection and reporting.

Challenges

The customer has thousands of product categories, some of which differ just by color or the brand printed on them. The production is done in small batches that are repeated at regular intervals of weeks or months.

• Data on production and quality was collected on paper, then on Excel, with reports arriving in the ERP system with delays and potential inaccuracies.
• There was no centralized and reliable data on efficiency, and using the in-hand procedure yielded no findings. Defects were often hand-noted on paper.
  
  

Overall, there was no real-time visibility into the performance and quality processes of the assembly lines, which inhibited decision-making and limited the optimization potential.

Solution

KFactory created a system that can precisely identify and track every product as it is manufactured while having no impact on actual operator work.

A cutting-edge computer vision model that employs cameras strategically positioned along the line to capture real-time images of the products was successfully developed.

The backend infrastructure is Microsoft Azure, which provides the security and power required to support a real-time manufacturing process such as this one.

It was kept in mind that, due to the large number of product categories, there were no pictures of all products; thus, the images captured in real time are compared to the known product database. If the products are not detected, the line supervisor is notified, who can identify the product code linked with the unknown product photos with the help of a local tablet application. Then, using the platform’s sophisticated infrastructure in Microsoft Azure and algorithms, we retrained the new model to recognize the newly added product categories in minutes.

The model is employed immediately in production, which means that the time between detecting an unknown product and starting to recognize it automatically is only a maximum of 30 minutes.

The local application allows operators to logon and classify actual defects, eliminating all paperwork required for quality monitoring and keeping track of actions per employee.

Results

The system has been successfully implemented inside the client organization, with all roles, from operators to supervisors and managers, seamlessly using it.

The degree of automatic product recognition is more than 99%, which is an outstanding result.

The managers have complete visibility over the process: reports are sent automatically after each shift, and the business analytics platform is fed daily with new data, updating KPIs and breaking down daily activities, improving overall efficiency and productivity.

New product categories are added daily, creating an image database that is becoming more precise every moment.

Due to automation, manual data collection is reduced to zero, and potential errors are eliminated.