ReBAR

The rise of machine learning offers a massive potential for applications in various fields as engineering, transport or medicine for example. By now, a wide variety of methods is available with a quite confusing number of technical implementations. The project Reducing Barriers for AI in (applied) Research – ReBAR aims for developing a modular platform to grant a uniform access to machine learning methods independent from concrete technical subtleties of individual program codes. Scientists from the DLR Institute for AI Safety and Security are collaborating with scientists from the DLR Institutes for Structures and Design, Robotics and Mechatronics, Materials Research, Software Technology, Vehicle Concepts, and Maintenance, Repair and Overhaul. In December 2023, we will not only provide a working prototype of the framework, but also create the nucleus of a user community within the DLR.

Development of an accessible machine learning platform for a broad spectrum of applications

The basic idea of ReBAR is simple: The whole workflow starting at raw data and concluding with the evaluation and verification of the results will be integrated into a single environment. Depending on the data set and the evaluations requested, suitable modules will be executed. The modules are connected by standardized interfaces to allow for a simple expansion of the methodological portfolio.

This approach has several advantages for users to enable them to concentrate on their research questions without being concerned with technicalities requiring in depth knowledge of the algorithms. ReBAR will not only reduce barriers for the development of AI and machine learning applications by simplifying the approach to it, but also provides methods to compare results and their reliability for different machine learning algorithms that can be executed and evaluated side by side due to the modularity. An extensive documentation will not only aid users in applying AI but also give developers the opportunity to dock their algorithms as additional modules into the framework. This allows for a quick spread of new technologies within the user community.

Contribution Institute for AI Safety and Security

Our institute is concerned with two main topics in this project: On the one hand, we are working on the definition of the common modular architecture where we are collaborating closely with our project partners which provide the user perspective. On the other hand, we are developing a module for verification and visualization purposes. In this topic, we are concerned with implementing a easy to use front end for interpreting models and results as well as evaluating options for further steps. Specifically, this includes metrics, the comparisons of different algorithms and graphical representations, for example in a dashboard fashion.

During the Project, five use cases provide the set of requirements for the framework’s prototype. Example one deals with the fusion of data from the operation of DLRs research planes to recognize maneuvers. In example two, our project partners identify anomalies during fully automatic additive production processes. The characterization of materials using non-invasive methods and the prediction of their properties is part of example three. The last two use cases are dealing with the simulation of jet engines and the automatic classification of configurations for novel air transport concepts.