Methodology for the development of a data-driven process time prediction for manual assembly systems of highly variable products

• Applicants: Professor Dr.-Ing. Bernd Kuhlenkötter; Dr.-Ing. Sascha Wischniewski
• Subject Area: Production Automation and Assembly Technology
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 570596910

Industrial assembly is of great economic importance. Sectors with a high proportion of manual work and significant process costs are no longer economically viable in Germany. The shortage of skilled workers and high market dynamics make it difficult to organize tasks and efficiently allocate personnel and materials. Thus, there are high demands on assembly planning and the corresponding accurate process time forecasting. This is fundamental for efficient assembly work, as it is the only way to avoid waiting times and maintain delivery dates. Simultaneously, process time forecasting is particularly challenging in this volatile environment. Industrially established procedures are reaching their limits due to the significant adaptation effort required due to customer-specific product configurations and variant order volumes. In (partially) automated series production, ML methods are already used today for production planning, so that a transfer to manual assembly is promising. However, there is still a lack of data acquisition and evaluation methods, while the data acquisition technologies have already been developed. The reasons for this are the often low level of standardization, the human influences and a frequently low level of acceptance among employees. Preliminary work has shown that the evaluation of only production-related data is not adequate to derive a sufficiently accurate measure of production times. Thus, human-related aspects have a significant influence on process times, so that their consideration is necessary. This results in a system in which the interdependence of factors of influence can hardly be mapped in an analytical model. Machine learning methods offer increasingly better forecasting possibilities here. Therefore, the project continuously investigates the use of ML methods during data acquisition. Thus, already in the laboratory phase and later in the operating phase for data acquisition, the data status will be evaluated regarding the suitability of ML methods. Based on this, a model for the prediction of process times will be developed. This model considers the correlations between the human factor and the productivity indicators of the assembly system and is iteratively improved with data from real assembly operations. The knowledge of the interrelationships in work systems thus not only provides the possibility of flattening the personnel workload based on a more accurate forecast of process times, but also of making assembly work in general more human-centered. The project thus pursues 3 development goals: 1) A methodology for regular subjective stress measurement; 2) A model for forecasting process times; 3) A method for introducing human-centered work planning.

• DFG Programme: Research Grants