Damian Heimel gives an insight into the automation process and explains how system integrators help deevio to automate quality control in the factory.
In which steps are you supported by partners?
In concrete terms, this happens when the machine vision system is installed at the customer's site. Together with the system integrator, we select the right camera and the right light, but the installation of the system, especially light and camera, is carried out by the partner.
In short: we supply the intelligent machine learning software and our partner automates the respective step in the factory.
The partner installs the system above the conveyor belt and prepares it so that images of the products can be taken. With these images we then train the deep learning model and the partner in turn takes care of the integration of the model in the factory and the machine control. The partner’s responsibility is to make sure that a part that is detected as defective by our software is actually sorted out at the end.
Why are you working with external partners on this particular step?
We quickly realized that the selection of hardware and the installation of the system in the factory was very complex. And even though deep learning algorithms are somewhat more stable to external influences, image quality is still extremely important. This is a classic phenomenon from the field of artificial intelligence: if I add poor quality data to the system, the result will also be poor.
From which fields do your partners come? What expertise do they have?
The partners we work with come from mechanical engineering, image processing and factory automation. In addition, they usually have knowledge in IT, because they also have to be able to control the individual components of the system automatically. These are all classic skills of a system integrator, which our partners have been proving very successfully for many years. Not only in machine vision, but also in other production steps.
How does the cooperation between deevio and the partners look like in detail? And how do you find partners?
The cooperation is as follows: we establish contact with the end customers and together with them consider how a project can be realized, i.e. how the respective manual inspection can be automated with deep learning. Once the project is decided, we specifically select a system integrator with regard to the respective application, i.e. someone who has a strong track record in the relevant industry. There are system integrators for pharmaceuticals, automotive or food – for every industry. Then we realize the project together. The competencies are clearly defined: we do everything on the software side, the integrator takes care of the automation in the factories.
How do we find these partners? On the one hand, we ourselves have a well-developed network of system integrators through the VDMA (Verband Deutscher Maschinen- und Anlagenbauer). In addition, our supplier of machine vision parts, Stemmer Imaging, has extensive domain expertise and can recommend system integrators to us. Also, some system integrators have already approached us themselves because they have seen that there is a lot going on in the field of machine vision and deep learning at the moment and they are interested in working with us.
What is interesting for the partners about working with deevio?
On the one hand, deep learning systems are very much in demand in the machine vision industry now. Of course, we are not the only product on the market in this area, there are also a number of other deep learning providers. However, most of these companies have a different business model: they give you a tool, a so-called Software Development Kit (SDK), with which you as a system integrator have to train the deep learning models yourself. This approach works very well in some areas. But the broad mass does not have in-house AI experts in system integration. And we are convinced that experts are needed for 99.x percent accuracy. That's why our approach is to be fully responsible for training the models. The system integrators give us the images and our data scientists train a deep learning model from these pictures. As a result, neither the system integrator nor the end customer has to train a model. This is the sole responsibility of deevio. And because we are providing the deep learning models, we can work with a large number of companies.
Hence, for the system integrators, working with us is also a win-win situation because their order books are full. And one reason why they can only accept a limited number of orders is the time-consuming and costly software development. Because we take this step off their hands, we can give system integrators the chance to focus on what they are best at: factory integration and automation.
Do you benefit from working together?
Absolutely. Especially with some partners who already have expertise in deep learning and have their own AI experts, we regularly exchange information and learn from each other. This of course makes system integration even easier, because these partners already know exactly what is important when you want to bring a deep learning model into production.
In which areas can you still use partners?
We're always looking for companies that – like us – want to bring deep learning applications to the factories. We are especially looking for and talking to companies from the automotive, pharmaceutical and metal industries.