In industrial production, the innovation driver AI has the potential to optimise manufacturing processes so that they are faster, cheaper and less prone to errors: For example, disruptions in production can be minimised through predictive machine maintenance or production can be optimised based on precise demand forecasts. The basis for this is data, as an AI system ultimately makes forecasts and decisions based on previously fed-in data sets. Artificial intelligence can thus enable the manufacturing industry to utilise its immense amounts of data profitably. Currently, however, self-learning systems are not sufficiently transparent: the path for an AI's decision-making is paved in a training process with large amounts of data - but is then hardly comprehensible to an external observer. As a result, the reasons for an AI's decision often remain hidden. The result is a lack of trust in AI systems.

Alpha version of the XMANAI AI platform aims to create transparency and build trust

Approaches and solutions for more transparency and improved accessibility of AI in industrial production are being developed in the XMANAI research project: a platform with an user-friendly interface will provide services for handling data as well as for the creation, evaluation and execution of AI models. The first test version - the so-called “alpha version” - is now available online for all project partners. Specifically, the platform enables its users to train and test AI models to solve challenges in industrial production based on corporate data. Comprehensible and interactive visualisations, e.g. in the form of knowledge graphs and data diagrams, as well as explanations in natural language show which data and algorithms influence the AI system's decisions. At the same time, the platform provides tools for data management - from importing and storing data to the (semi-)automated preparation and exploration of this data. Ultimately, the data must be accurate, up-to-date and complete throughout the entire process for AI systems to be successful and comprehensible at the same time.

Developing “Explainable AI” together with 14 project partners

In the Horizon 2020 project XMANAI, Fraunhofer FOKUS has been working with 14 other partners from all over Europe for more than two years to make artificial intelligence more accessible and comprehensible - and thus create trust in this technology. Guided by the human-centred concept of “Explainable AI”, the project aims to enable stakeholders in industrial production to understand exactly how decisions are made by an AI system. To this end, the project partners develop comprehensible algorithms for various challenges in production processes as well as the AI platform, which can now be comprehensively tested.

The Digital Public Services (DPS) business unit at Fraunhofer FOKUS is participating as a scientific partner for real-time data management. XMANAI will continue until April 2024 and will be continuously developed over the coming year.