Initial situation and problem statement

High availability and the longest possible service life of technical systems are key requirements for ensuring productivity while simultaneously reducing resource consumption in mechanical and plant engineering. To achieve this, failures must be detected early, and maintenance, spare parts procurement, and reuse must be planned proactively throughout the life cycle. Established methods such as Failure Mode and Effects Analysis (FMEA) are based primarily on empirical knowledge and static assumptions during the development phase and can only take dynamic operating conditions and aging into account to a limited extent. At the same time, companies have increasing access to operational, condition, and maintenance data. However, there is currently a lack of transferable approaches that link this data to an understanding of the system and make it usable for reliable failure and service life predictions, as well as for resource-efficient life-cycle planning.

The goal of the RELiANT collaboration project is to develop a transferable methodology for data-driven failure and service life prediction of technical systems. Real-world operational, condition, failure, and maintenance data will be used to model and refine degradation trends and failure probabilities under various usage and environmental conditions. To this end, the project will develop adaptive prediction models and link them to system models. The results are intended to be directly applicable to decisions regarding maintenance, spare parts procurement, system adaptation, and reuse. Two industrial pilot projects and an internal project demonstrator will be used to develop and validate these approaches. The insights gained will ultimately be compiled into a practice-oriented guide for industrial implementation.

The RELiANT collaboration project combines model-based system understanding with real-world lifecycle data and data-driven forecasting methods. First, the functions, structures, and causal relationships of the systems under consideration are formally described using MBSE and linked to operational, failure, and maintenance data. Building on this foundation, machine learning techniques and statistical lifetime analysis and prediction methods are employed to forecast failure probabilities and remaining lifetimes. Transfer learning supports the application of predictive knowledge to similar scenarios, while explainable AI methods ensure the transparency of the results. The models are tested using a demonstrator and two industrial pilot applications. Prediction results are then translated into concrete measures for fault prevention, maintenance, spare parts provision, and reuse. This creates a comprehensive approach ranging from system and data analysis to resource-efficient lifecycle decision-making.