Data-supported determination and prediction of the effective surface condition and intervention in conveyor belt flow processes

Tool wear occurs particularly in forming processes with high effective surface loads such as sheet metal forming (BMF). However, the production of faulty components due to worn tools leads to rising production costs and environmental pollution. Early detection of signs of wear offers the potential to improve product quality during the production of a batch and positively influence the environmental balance of the overall process. The combination of the data-driven analysis of product, process and machine resource (PPR) with the causal expert knowledge formally described as ontology enables the early detection of the onset and causes of signs of wear based on machine and process data. This leads to a reduction in scrap quantities and thus to an improvement in the material and energy efficiency of BMU processes.

The aim of the project is to research an early warning system for tool wear based on a cause-effect graph with the greatest possible integration into the machine automation using a full reverse extrusion process from the belt. By adapting the existing tool system at the LFT and implementing process-specific active elements for use on the high-speed press, an extrusion process from the belt is being set up. The correlation strengths between target variables and process variables are quantitatively determined by process monitoring using sensor technology, optical analysis and microscopic examination of tools and component samples. In combination with a causal model developed by experts, these allow the creation of a quantified, uncertain cause-effect graph. In addition, experts can use strong correlations to uncover previously unknown causal relationships, which are validated experimentally and simulatively.

The expected outcome of the research project is initially the creation of a database by combining experimental and numerical methods and the development of an information model formatted by ontology, which contains both a PPR-compliant description of the BMU and a cause-effect graph for tool wear (see Fig. 1). The cause-effect graph contains qualitative correlations that are learned from the developed prediction model for tool wear, as well as causal relationships between process parameters that are formalised from expert knowledge. Based on this, an assistance system is developed that supports ‘backward’ queries of influencing factors on observed quality deviations on the one hand and enables ‘forward’ predictions of deviating component properties or machine and process data on the other.