VECTOR - Virtual Engine for Composite Thermo-mechanical Online Response

Kurzfassung

PredictECO is a three-year DLR research project (May 2023 to April 2026) that develops a platform for assessing and optimising the sustainability of resource-intensive manufacturing processes already at the design stage. A central element is the coupling of process simulation with real production data. For composite components cured in an autoclave, this concerns quantities such as degree of cure, glass-transition temperature, residual stress and process-induced distortion. These determine part quality and feed directly into the eco-efficiency assessment, but they are not covered by the few thermocouples available on a real part.

Detailed Abaqus cure simulation predicts these fields with good accuracy, but it runs far longer than the manufacturing cycle itself and uses planned rather than actually realised process parameters. It is therefore unsuited to support decisions during an ongoing cure or to compare many production runs. What is needed is a fast, field-resolved substitute that can be evaluated in near-real time and corrected against measured temperatures.

A comparison of direct surrogate regression, proper orthogonal decomposition (POD) and neural networks showed the highest prediction accuracy for neural networks. The best-performing solution, however, combines POD and neural networks: POD reduces each field to a small set of spatial coefficients, which the network then predicts. This combination forms the basis of the digital twin VECTOR (Virtual Engine for Composite Thermo-mechanical Online Response). The networks are trained offline on an Abaqus cure-simulation database and, during production, are combined with a sequential correction step. The underlying process inputs are adjusted so that predicted and measured thermocouple temperatures agree, and the corrected inputs update the fields across the entire part, including those not measured directly. This requires neither re-running nor retraining the simulation model.

Within PredictECO, VECTOR serves two purposes: a live dashboard that makes quality-relevant fields visible which would otherwise only be known after time-consuming or destructive inspection, and a data basis for coupling manufacturing outcomes to the eco-efficiency assessment, since it reflects the process as it actually ran rather than as planned.