maQinto

Machine-trained quality sensor, intelligent process control, and an ML framework for resource-efficient, customized carbon fiber production

Research into new carbon fiber precursors and novel fiber property profiles is primarily conducted through an iterative process, with the process parameters of the thermal conversion stages, namely stabilization and carbonization, being adjusted step by step. Achieving satisfactory fiber properties therefore requires numerous iterations, resulting in considerable time and resource expenditure. In the “maQinto” project, an automated property analysis for the carbon fiber (CF) manufacturing process was therefore developed by combining high-frequency inline measurement technology with machine learning (ML) models. To this end, eddy current sensors were designed for the continuous monitoring of the CF manufacturing process, as existing sensors were only available for intermittent, non-contact testing of CF components. Using ML approaches based on sensor data, plant parameters, and experimentally identified fiber properties, two models were designed: one maps the eddy current signals to the material properties. The other model analyzes correlations between plant parameters and material properties to predict the resulting fiber properties. These outputs can then be used to generate recommendations for the plant operator that ensure the specified fiber properties are achieved.