Ralf Jostan and Sebastian Findeisen from the Sindelfingen plant of Mercedes-Benz AG showed how an AI-based process platform can improve collaboration between car manufacturers and suppliers in plastic painting. This is because the high demands of the car manufacturer lead to a complex process in color development until the add-on parts ultimately match the body.
"In order to leverage significant efficiency potential with digitalization and AI, we set out to bring the process world and the data world together," explained Findeisen. The result is the Coating Intelligence System (CIS): it includes a platform for collecting data across company boundaries on the one hand and AI on the other. The system links material, quality management and process data, from which the AI can develop optimized parameters. "For example, we start with a new color," said Findeisen, "the system makes an initial parameter calculation, while we paint and collect data. The CIS analyzes this data automatically. The model learns from the results, gives new parameters to the suppliers and a new iteration begins."
The current AI challenge is data exchange
Dr. Christoph Schulte from BASF Coatings in Münster spoke about how AI is influencing the future of automotive series painting. The industry is currently in a phase in which it is creating data transparency and beginning to use it for process digitalization, predictive maintenance or generative AI. To have a significant impact on quality and costs, AI would have to be developed further - towards AI-driven services or even further towards autonomous agents.
However, the current challenges are different: "We have around 200 AI projects worldwide at BASF, and they all have the same problem: how do we get data?" reported Schulte. The industry needs a way of exchanging data, and this is a task for the coming years.
Dr. Meiko Hecker, Managing Director of AOM-Systems in Heppenheim, and Dr. Oliver Tiedje from the Fraunhofer Institute for Manufacturing Engineering and Automation (IPA) in Stuttgart reported on spray monitoring, AI for error analysis and error determination in the spray.
Spray jet monitoring improves paint application
AOM-Systems offers a droplet measuring device called Sprayspy to monitor the spray jet in terms of droplet size, viscosity and speed. During application, it can record volume flow, from which the coating thickness can be calculated. In a project at Fraunhofer IPA, Tiedje and his team investigated how the device can be integrated into process monitoring and how the data can be used in other processes.
To do this, the IPA researchers deliberately introduced defects such as process deviations, worn equipment or paint material with different properties. They then evaluated the characteristic parameter values of the Sprayspy and documented the quality deviations in coating thickness, color tone or waves. "For most parameters, over 90 percent of the errors were detected. Correspondingly low. The proportion of incorrectly reported errors was less than 10 percent," explained Tiedje.
In a further step, the researchers added process data from the line and set up a behavioral model for it. The result was a model for analyzing complex time series. "It is important to not only look at the values at each point in time, but also at the history. For example, when I open the main needle of the atomizer, the amount of paint still varies for a certain amount of time," said Tiedje.
AI project for the configuration of robot cells restarted
Dr. Pavel Svejda from Dürr Systems in Bietigheim-Bissingen gave an insight into a recently launched research project on AI development for the configuration of robot cells. It is part of the joint project "Digital ecosystem for AI-based robotics" (RoX). The sub-project deals with the commissioning and optimization of path processes in static multi-robot systems. "Among other things, system integration and commissioning are to be raised to a new level through the use of AI," reported Svejda.
Launched in the third quarter of last year, the project comprises 24 project partners, eleven associated partners, five associated projects and initiatives and three European partners. The project is scheduled to run for 30 months and is being coordinated by ABB with support from Siemens, DLR and the Fraunhofer-Gesellschaft. Dürr is involved in sub-project 3, use case 4 "AI for robot operations". This involves AI support for the concept or layout, safety concept, robot path programming, parameterization of the application and PLC programming.
Manuel Haß from Dataspree in Berlin presented an AI-based system for the automatic surface inspection of high-gloss plastic parts. "Until now, this inspection has mainly been a manual process," he said. Especially for smaller parts with high cycle times, it is difficult to reliably automate this task. Existing systems have their weaknesses: although deflectometry offers high measurement accuracy, it is only of limited use for molded parts, requires a stop-and-go process with long inspection times and has its limitations with various defect types and surface properties. Shape from shading offers high measurement accuracy for flat surfaces and can also be used for matt colors. However, users must accept limitations with non-flat surfaces. In addition, like deflectometry, it requires a stop-and-go process and is also not suitable for all defect types and surface properties. Rule-based image processing can be implemented quickly with software libraries and the right imaging. However, rule-based algorithms often fail due to their complexity; this method cannot map the versatility of surface properties and defects.
AI supports surface inspection
AI-based surface analysis, on the other hand, can also be used on non-planar surfaces and maps the versatility of surface properties and defects. Other advantages include very short recording and inspection times, high detection accuracy and flexible expansion of functions using AI training. One disadvantage is the lower measurement accuracy when determining the size of defects.
You can read a more detailed report on these AI developments in issue 8/9-2025 of MO, which will be published on September 1.