How can you test AI for quality control without immediately investing in a new production line? In this user story, we show how Inno-group, together with Blue Engineering through AI-MATTERS, tested an AI-driven vision solution for automated final inspection. By first carrying out a feasibility study, the company was able to investigate whether AI can detect and reduce human errors in complex production processes, without making major investments straight away. The results were positive and form the basis for the next step towards implementation.
Delivering quality in an environment where people make the difference
Inno-group is a supplier specialising in high-quality sheet metal and welding work for, among others, the medical and laboratory sectors and mechanical engineering. The company has sites in Eindhoven and Zaandam and produces complex products with high accuracy requirements.
For many customers, quality requirements apply that are comparable to those in the automotive industry. At the same time, Inno-group does not operate in a fully automated factory. The production environment is a typical high-mix, low-volume setting in which many manual operations take place. For Gertjan van den Hazelkamp, Commercial Director at Inno-group, one question was therefore essential: how can we further improve quality when human errors can never be completely eliminated from a process?
The Challenge: limiting human errors in complex production processes
Gertjan: “Many of Inno-group’s products go through a large number of production steps before they are ready for delivery. Some products even pass through twenty different operations and are handled by employees twenty times. Despite continuous process improvements and training, human error remains possible. According to our own estimates, when manual operations are added up, something goes wrong in around 1% of these operations. This could be a forgotten operation, a moment of inattention or a minor deviation during assembly. In large volumes, this can have significant consequences. A customer receiving 6,000 products annually receives around 120 faulty products at an error rate of 2%. We need to reduce that number drastically; our customers are affected too much by this.”
The approach: a feasibility study through AI-MATTERS
Gertjan came into contact with Blue Engineering via LinkedIn. In addition, the company was also alerted through its own network to the opportunities offered by Blue Engineering and collaboration around vision technology. Together, Inno-group and Blue Engineering decided to carry out a feasibility study within the AI-MATTERS programme.
The aim of the study was clear:
- To investigate whether AI-driven vision technology is suitable for automated final inspection.
- To validate whether the system can recognise human errors.
- To determine whether a business case for implementation is feasible.
“AI-MATTERS made it possible to carry out this exploration in an accessible way and with limited risk,” says Gertjan. “This allows us to learn and validate first, before investments in equipment are made.” experimentation, enabling SOFIA MED to assess feasibility and real-world impact before committing to production deployment.
What has been tested?
The Technology
The pilot focused on AI-supported vision technology for automated quality inspection. The system uses cameras and artificial intelligence to analyse products and recognise known fault conditions.
The Use case
The application focused on final inspection of complex sheet metal products.
Instead of assessing whether a product is completely correct, the system learns which errors can occur. The AI then checks whether any of these known deviations are present.
Test setup
Two types of important products were selected for the pilot:
- Products with a high turnover value.
- Products where relatively many quality issues occurred.
At the end of the production process, the vision system was used to check whether all operations had been carried out correctly.
Gertjan says: “Inno-group already had an extensive library of known fault conditions. This, together with the knowledge of experienced quality inspectors, was used to teach these deviations to the AI solution. As a result, the system could learn what specifically to look out for during inspections.”
Results: AI proves promising for automated quality control
The results of the feasibility study were positive. The pilot showed that:
- AI-driven vision technology can reliably recognise known errors.
- Automated inspection is feasible for complex high-mix, low-volume production environments.
- Modern vision solutions can be trained much faster than before.
- There is sufficient potential to invest further in an operational inspection machine.
Another striking result was the impact within the organisation. An employee who had been very sceptical about the possibilities of AI beforehand became particularly enthusiastic during the project after seeing the results. This shows that a feasibility study not only provides technical insights, but also helps to create support for innovation.
The ultimate aim is to drastically reduce the number of faulty deliveries.
Where a customer may currently still receive around 120 faulty products per year, Inno-group wants to reduce this number to approximately ten units per year.
This means:
- Higher customer satisfaction.
- Lower quality costs.
- Less rework.
- Better delivery reliability.
- Greater confidence in complex production processes.
Although these results still need to be realised, the feasibility study has shown that the technology offers a realistic basis for achieving them.
Key Insights
“The speed at which modern AI systems can be trained made an impression,” says Gertjan. “This makes vision technology accessible for applications in companies like ours that would not have been commercially viable a few years ago.”
What worked well?
- Testing with real products from the production environment.
- Using the existing quality knowledge of experienced inspectors.
- The close collaboration between Inno-group and Blue Engineering.
- Carrying out a feasibility study before making investments.
The most important lesson? “Start small, validate first and invest afterwards. By testing first, you gain insight into both the technical feasibility and the potential business value.”
What are the next steps?
After the successful pilot, Inno-group wants to start the next phase:
- Develop an inspection machine based on the results.
- Put the solution into production.
- Train employees in the use of the technology.
- Further expand the number of automated inspection applications.
The feasibility study has provided sufficient confidence to take this step towards implementation.
Why AI-MATTERS?
AI-MATTERS helps manufacturing companies test AI solutions in a realistic production environment before they invest. For Inno-group, the added value lay mainly in:
- Bringing together the right partners.
- Access to expertise in AI and vision technology.
- Carrying out a feasibility study with limited risks.
- Obtaining concrete insights for future investment decisions.
By experimenting and validating together, companies can determine more quickly and with greater certainty which AI solutions genuinely add value.
FAQs
AI Feasibility Study for automated visual quality inspection. A feasibility study in which an AI-driven vision solution was tested and validated for automated final inspection of complex products.
Through AI-MATTERS, companies can carry out a proof of concept or feasibility study in a realistic production environment. This enables the technical and economic feasibility to be validated before investments are made.
For more information and a specific dive into your use case, we welcome you to contact us for a non-binding conversation.
Yes. Thanks to recent developments in AI, vision systems can be trained and adapted to different products more quickly. This makes automated quality control interesting for production environments with a high degree of variation and smaller series as well.
Would you like to know how our project managers can help your organisation?
Contact us for a non-binding conversation: https://ai-matters.eu/contact/