Quality Control with Deep Learning

One of the greatest challenges in the automated production of battery modules is to produce a secure and stable welded joint without damaging the battery cells. In order to meet the high quality requirements of the automotive industry in particular with regard to welding, while at the same time reducing costs for customers, Manz is working intensively on inline inspection systems that enable 100 percent control of the welding process.

Safe Welding in Battery Production

Quality inspection is not easy when it comes to batteries. Typically, welded joints were statistically inspected on individual components using destructive tests such as transverse grinds or force pull-off tests. However, this procedure is time-consuming, expensive, and does not guarantee flawless quality of every weld.

Today, Manz uses industrial image processing (machine vision) to make it possible to inspect the welding process in its modular Battery Laser System BLS 500 and to automate it end-to-end. The deep learning functions integrated in the software based on artificial intelligence (AI) offer particular added value for quality inspection.

Battery Laser System BLS 500

The Battery Laser System BLS 500 is a flexible, modular platform for different laser processes.

  • Highly improved error detection thanks to self-learning algorithms
  • Significantly reduced programming effort
  • Detection even of errors not precisely defined in advance
Cross section of welds

The top two images show cross-sections of a good weld (left) and a weld where there was hardly any mixing of material (right). Below are the corresponding inspection images, where hardly any differences can be seen. Despite the small difference and the different background, Deep Learning can be used to classify 100 percent correctly.

Success with small amounts of image data thanks to AI

In addition, the software solves another problem: Normally, AI training processes require a six-digit number of valid image data to achieve acceptable defect detection rates. Quantities that are typically not available in mechanical and plant engineering. In addition, the potential defects are often not known in all their forms. At the beginning of a new process, for example, only between 10 and 100 sample images are available.

Anomaly Detection" provides a remedy here. Not only does it get by with a very small amount of image data, but it also requires only "good images" showing defect-free objects for training.


Advantages for the customer

Manz can provide 100% quality control of the laser-based production processes in the modular laser system by using deep learning technologies. The reliable detection of errors allows the quality of battery production to be maintained at a consistently high level. In addition, the mechanical engineering company saves a lot of time and thus costs thanks to the accelerated training processes.



  • Optimization of production, faster processes
  • Less scrap and lower costs
  • Higher quality and improved safety of products
We are looking forward to your inquiry!

Contact us