Deep Learning: with minimal effort to optimal defect detection on EL images.

MBJ Solutions optimizes image processing in production at 3S Swiss Solar Solutions in Thun with the help of neural networks.

[Translate to english:] MBJ Deep Learning

Last week, the defect detection rate in production at 3S in Thun was significantly optimized by teaching customer-specific, relevant defect images.

3S Swiss Solar Solutions has been developing and producing solar modules in Switzerland since 2001. 3S solar modules are integrated directly into the building envelope - the roof, the facade the balcony railing. Already since 2010, 3S has been using MBJ Solutions' electroluminescence inspection for quality assurance of their modules in production.

In the production process, high-resolution image data taken with an MBJ electroluminescence system are evaluated in fractions of a second by neural networks created using artificial intelligence, and the defects found in the images are marked. A incorrect module is thus quickly and reliably detected and forwarded to a repair station. There, the defect and the location of the error are displayed on a monitor and targeted repairs can be carried out. Only then is the module sent on for lamination. In this way, 3S not only saves material and time, but also ensures the high quality of its solar modules at this point. The detected defects are also used to correct and improve process steps that precede this inspection

How does it work?

MBJ Solutions provides each customer with a pre-trained neural network that reliably detects and classifies typical defects such as micro-cracks, dark areas and solder defects. This network can be further customized as needed by MBJ Solutions support with help from the customer or by the customer themselves. In this case, MBJ was on-site at 3S and the optimization was done in close cooperation with 3S' quality assurance.

How does MBJ Support work with on-site Deep Learning training?

Together with the customer, it is determined on site which defects are most critical for him. Data is then collected and analyzed directly from the production line. The networks are trained, installed and evaluated by MBJ experts to assess the defect detection rate.

Network performance was discussed with 3S and further optimized. In addition, the quality criteria defined by 3S for modules to be classified as incorrect are stored in the machine. At the end of the MBJ visit, 3S is running a highly stable defect detection system for several defect types, optimized specifically for their needs.

What advantages does MBJ offer the customer with Deep Learning?

Neural networks detect defects faster, better and always the same. At the same time, they significantly reduce the pseudo error rate.

Remarkably, artificial intelligence is very robust to variations in cellular material. The time-consuming and incorrect optimization of earlier image processing systems with long lists of parameters is completely eliminated. Only additional sample images are required for optimization. 

In contrast to classical image processing, artificial intelligence can be extended with little effort to include further defect classes according to customer requirements. Just a few sample images are enough to create a new defect class.

The customer compiles a sample catalog of images in the low to mid two-digit range according to his requirements and marks the defects in these images. The neural network is optimized on these examples and the optimized network can be used in the customer's system.

MBJ also offers in-depth hyperparameter optimization, which is especially important for the following:

  • Best detection performance
  • Minimum computational power
  • Fastest feature extraction


About MBJ Solutions GmbH

MBJ Solutions GmbH specializes in the development and sales of test and measurement systems for the photovoltaic industry. MBJ offers sun simulators, electroluminescence test systems and test systems for insulation and ground testing for the solar module production and mobile testing in the field.

The company was founded in 2009 and has since sold more than 650 test systems worldwide. The headquarters of the MBJ Group is close to Hamburg. Of course we develop and produce in Germany. Since many of our customers are based in Asia, we have a service office in Taiwan.

Innovative solutions and customer orientation are a matter of course for us. Our customers include well-known institutes and manufacturers of the PV industry. Our many years of experience and a motivated team make us the perfect partner for your PV projects.

Press contact MBJ Solutions GmbH

Annika Beck
phone +49 41 02 77 89 0 - 18