Select your location
Austria

Austria

Czech Republic

Czech Republic

Germany

Germany

Italy

Italy

The Netherlands

The Netherlands

Romania

Romania

How to improve quality assurance leveraging AI

Use case for how to become a data company,

Overcome the issues of manual quality control – cost, time and effectiveness, by installing AI cameras at the conveyor belt and using continuous quality checks.


Quality assurance is all about accuracy and efficiency. It’s a sort of safeguard to evaluate processes and products. But as any manufacturing company will confirm: manual quality control is expensive and time-consuming. A waste recycling company encountered these very problems, and wanted to find a solution to improve the quality control. By using cameras and AI-driven automation, the quality assurance is now improved, faster and more complete than ever before.

Automating quality assurance

How can we improve quality control without increasing manual labor?

The price of manual labour
Quality control has long been a headache for manufacturing or processing companies. It requires an intense concentration from employees – inevitably compromising the accuracy at a certain point. Looking at a conveyor belt is not only exhausting, it’s a poor application of man hours. Moreover, this manual labor can lead to high costs.

Sample-based quality control
One common challenge with manual quality control, is that it is by definition sample-based. Since the manual quality control is already so time-consuming and expensive, it is simply impossible to check each and every product coming off the conveyor belt. This problem is usually solved with sample-based quality control. And although this will give you some idea of the general quality, these samples are not representative for the overall quality.

New types of material
And finally, when new types of products or material are introduced, this requires employees to form new habits and define new quality parameters. Again, this compromises the accuracy of the quality control and increases the workload for employees.

Although sample-based quality control will give you some idea of the general quality, these samples are not representative for the overall quality.

Automating quality control with AI-driven cameras

To maximize the accuracy of quality control, you need a continuous stream of quality checks. Instead of checking samples manually, each and every item on a conveyor belt should be checked automatically.

AI-driven cameras
Cegeka installed strategically placed cameras aimed at the conveyor belt of the waste management company. An AI algorithm allows the cameras to analyse the images for impurities. Concretely: the conveyer belt on a sorting machine divides different types of products and materials, in order to recycle them properly. For instance, plastics are recycled by colour. The AI- driven camera detects the colour of each piece of plastic on the conveyor belt, and counts how many of each colour have been sorted correctly before they are pressed to a bale.

Custom-built computer vision
The cameras are of course just a part of the equation – the hardware. To analyze the images, an AI model was needed. Since the plastic sorting process is so specific to the material, many vendor-based computer vision models were insufficient for this use case. So, Cegeka customized the model based on the available data and the requirements at hand. The biggest advantage of customizing this model? It’s easily adaptable to new materials. The hardware stays the same, but the recognition software is retrained with every new process.

A mobile app to connect the automation with human control
To easily present the results to the quality control managers, Cegeka built a mobile app that sends pictures of the sorting process in real time. The quality experts still check the bales manually and sample-based to do a detailed quality check on the end product.

The automation model can check every item, but it does so based on certain parameters. Human expertise is still invaluable to check the bales in detail. The pictures on the mobile app are a useful addition to this process, and provide a higher accuracy. As the bales have been consistently pressed thanks to the automated checks, the samples are much more representative for the sorting process.

Cegeka’s reference architecture
To build this end-to-end solution, Cegeka used its proven reference architecture in the cloud. This included the custom-built AI model, data lake and data warehouses. It also enabled the integration of the hardware, a portal where detections are presented, the development of a mobile app and multiple interfaces.

Our team has guided a number of companies on the road to becoming data-driven enterprises by applying the Cegeka reference architecture as a blueprint.

Cloud Data Platform with Cegeka Reference Architecture

A minimal cost automation solution

What does this increased effectiveness bring to our business?

Realtime analysis
Crucially, the largest portion of analysis is edge-based, since sending the video data through the cloud in real time would not perform well due to the high latency. What’s unique about this project, is the combination of edge and cloud. The AI model is trained in the cloud, based on big data. This intelligence is then pushed to the edge, close to the hardware. The low latency communication ensures a realtime analysis on the impurities. The camera then only sends quality results – the percentage of impurity over a certain period – which requires very little communication.

Automated reports
Within the same reference architecture, the waste recycling company also has Power BI and Azure set up. In line with the ‘start small, think big’ philosophy, the AI was added as an extra component to the existing architecture. The combination of the automated quality control and the BI environment have seamlessly led to automated reports of the quality control.

Increased quality in sorting process
With this solution fully operational, the accuracy and efficiency of analysing the sorting process has drastically improved. This in turn results in significantly higher quality of the sorting process itself. Not only is this beneficial to the environment (since the recycling material is more pure), the clients buying these bales to use in the next phase of the recycling process have higher quality material.

Future possibilities

With the AI model, it is easy to add new sorting processes, types of material or types of impurities. Even different machines and conveyor belts can fluently be integrated in the machine park without compromising the automated quality control. The system is thus completely futureproof and modular.

New call-to-action

Challenges

  • Increasing the accuracy and efficiency of the quality control
  • Lower cost of quality control
  • Improving the sorting process
  • Providing a solution that is adaptable to new processes and materials

Approach

  • Setup of a cloud data platform based on Cegeka reference architecture
  • Installing cameras to generate visual data
  • Customizing an AI model on the edge based on computer vision
  • Developing a mobile app to present quality results

Solutions


Results

  • Improved quality control and sorting process
  • Realtime analyses and automated reports
  • A futureproof and modular AI model
  • A cost-efficient end-to-end edge and cloud solution

Perhaps we’ve inspired you to work out your own use case?

  • Kristel Demotte

    Kristel Demotte
    Global VP Data Solutions

Then we will discover together how we can help you to become a data company.