Phoenix Contact is a global market leader with its corporate headquarters in Germany. The group of companies stands for advanced products and solutions for comprehensive electrification, networking, and automation of all sectors of the economy and infrastructure. A global network in more than 100 countries with 22,000 employees guarantees vital proximity to the customer. With a broad and innovative product portfolio, Phoenix Contact offers its customers sustainable solutions for different applications and industries. This is particularly true in the target markets of energy, infrastructure, industry, and mobility.
The results at a glance
- Rapid implementation: Realization of a proof of concept (PoC) for an AI assistance system within 16 days.
- Data generation: Phoenix Contact Electronics GmbH can continuously obtain detailed data about its assembly process, which can be used for monitoring and optimization.
- Enablement and knowledge transfer: Phoenix Contact Electronics GmbH's Manufacturing Data Support team is now able to independently implement computer vision-based solutions and use the developed solution in various areas.
Initial situation
Employees assemble individual components by hand in many of the production processes. Errors and irregularities occur repeatedly over time, in particular due to the repetitive nature of such activities. Such errors cause additional costs as well as quality issues, especially if they are detected late. In the worst case, entire production plans have to be revised. Phoenix Contact Electronics GmbH aims to provide its employees with the best possible support in their daily work, including making processes more efficient and ensuring high quality.
Challenge
The aim of this project was to develop a PoC for an AI assistance system that can detect errors in the assembly process in real time. This assistance system was intended to enable employees to identify errors and correct them immediately. However, employees often set up their workplaces individually, which makes standardization difficult. It is also generally not clear what an optimized procedure looks like. A secondary objective of the project was therefore to obtain data on the assembly process. It had to be possible to draw conclusions based on sound data that could then be used to make the process more efficient.
Another challenge was the feasibility of possible AI training. When developing complex AI applications, it is rarely possible (and usually very costly) to build up generic end-to-end training, as is taught in textbooks. Instead, real life calls for an intelligent combination of off-the-shelf solutions, customized algorithms, ML techniques, and pre-trained models.
Solution
Phoenix Contact and codecentric already work together in the area of intelligent applications for PLCNext technology, in particular with the ML Extension module for AI-on-the-edge applications. Since codecentric AG has already proven its computer vision expertise on several occasions, the collaboration on this project was the next logical step. In the run-up to the project, various ideas for realizing such an assistance system were developed and evaluated in initial discussions with Phoenix Contact Electronics GmbH. The favored solution, which was ultimately implemented, follows a computer vision approach that observes which components are picked in the assembly process. This allows AI-supported recognition when a component is forgotten during an assembly cycle.
A data pipeline was set up for the realization, which in the first step uses a finished computer vision solution for hand recognition in videos. The resulting time series consisting of hand positions then flows into a customized algorithm, which first performs tracking on the individual hands over several frames. A segmentation of gripping movements is then carried out. In the next step, gripping targets are extracted from the movements, which are then automatically classified by means of classical clustering algorithms. The result is a time series of target clusters which should be a periodic sequence of error-free assembly operations. A sequence can be recognized as faulty if there are deviations in the time series, and the employee can be alerted to this.
Result
The solution could be realized in a few days thanks to the intelligent use of suitable technologies and components and is currently being tested in production. Phoenix Contact Electronics GmbH was able to successfully add to its expertise and experience in the field of AI-supported camera solutions in a very short period and can also implement its own solutions in other areas with the help of the developed core solution.
Stefan Schnake
Manager Manufacturing Data Support, Phoenix Contacts Electronics GmbH
We had already gained experience with AI solutions before partnering with codecentric. In light of this, codecentric was able to integrate their solution extremely quickly and seamlessly into our existing architecture and established processes. This successful implementation enabled us to use the potential of the AI solution not only for employee support on the shop floor, but also to extend it efficiently to other use cases.
Stefan Schnake
Manager Manufacturing Data Support, Phoenix Contacts Electronics GmbH
Any questions about the project?
Are you looking for solutions to improve your production? Let's talk without obligation.
IT Consultant
Further projects of codecentric AG
Find out about other successful projects that we have completed with our customers. Perhaps you will find inspiration for a use case in your company here.
Benedikt Nordhoff
IT Consultant