to content
to footer

EU Emblem

Overall objective

The project aims to develop a multimodal, AI-supported measurement system that detects production defects at an early stage and reliably estimates the remaining service life of both new and used power modules. The focus is on developing an intelligent system that can subject test substrates to an artificial aging process. Measurement data from different sources (e.g. partial discharges, impedance spectroscopy and insulation measurements) can then be combined from these artificially aged test subjects and analyzed and evaluated with an artificial neural network. The final demonstrator of the system is intended to be able to recognize interdependent data patterns and classify power modules with regard to their probability of failure. By optimizing this AI-based solution and implementing a flexible architecture, a platform is to be created that both increases production efficiency and enables the second use of modules. The result is a powerful, adaptive and transferable solution that enables the quality assessment and lifetime prediction of power electronics and allows a potential further development of measurement technology without continuous redevelopment of the respective AI-supported modules.

Motivation and responsibilities of Develappers

Over the past few years, the engineers and software architects at Develappers GmbH have built up deep expertise in artificial intelligence, data processing and software technology and have already applied it very successfully in various industries. The aim now is to exploit the research results from the collaboration with Fraunhofer IKTS after the end of the project in order to expand our own services and products in the field of artificial intelligence in industrial environments. The joint project is intended in particular to demonstrate the technical feasibility and potential, and to explore methods for transferability to analogous problems in industry. Alongside the actual problem, methods for increasing acceptance in industrial environments are also a subject of the research. This includes, for example, methods for adapting and optimizing artificial neural networks (ANNs) so that they can be used on standard, low-cost industrial hardware and without specialist AI knowledge.