Supporting Development of AI-Based Applications in Electronic Systems Domain Using a Modular Platform Approach

This paper presents a modular platform architecture to support the development of Artificial Intelligence (AI)-based applications in the electronic systems domain. Built of freely available tools and open-source software, it tries to minimize the restrictions required while providing support to AI pipeline developers. In order to deal with unforeseen requirements, retain sovereignty and have the possibility to customize the pipeline where necessary, the presented solution comprises components and concepts such as data storage, versioning, pipeline execution, containerization or local and remote environments. Software tools implementing the aforementioned components can possibly be exchanged and adapted to the problem at hand. Furthermore, the modularized concept allows the developers to run the same pipeline seamlessly on a local machine or a remote server. Additionally, other entities can be integrated in the places where they are needed, e.g. data producers. Finally, the proposed architecture is applied to two different use cases. First, a classification model for electromagnetic compatibility is designed to identify possible root causes given measurement spectra. Second, a pipeline to derive netlists from schematics is depicted where data is represented as bitmaps only. For both use cases, excerpts of their implementation and special features using the proposed tools are shown. Although the two use cases could be implemented in the architecture, a slightly higher configuration effort is currently required.