A neural-network-assisted method for flow-front estimation in resin transfer molding using pressure sensors

Abstract

The resin transfer molding (RTM) process shows considerable advantages in composite manufacturing. Nevertheless, the part quality manufactured by RTM is sensitive to material and process variations during the preform impregnation. To improve the process robustness and achieve better process control, a methodology for resin flow monitoring based on a combination of a sensing system and a neural network model is proposed, which can be easily implemented into a generic RTM process. Using pressure data provided by a limited number of sensors distributed over the mold surface, the proposed method allows the prediction of flow-front patterns at any impregnation time. The dataset for training is generated by physical-based simulations. Considering the permeability changes caused by uncertainty conditions, the permeability tensor is modeled with random variations. The network parameters are obtained by trial-and-error. Furthermore, the sensor distribution scheme and the dataset size are identified as the sensitive factors of the model. Finally, the predicted results are verified by numerical solutions. This method can be used to avoid the formation of voids and improve the final part quality.

Graphene nanoparticles as data generating digital materials in industry 4.0

“…Finally, the machine/deep learning-based surrogate/predictive models can be used for process simulations [155][156][157] as well as for failure predictions in diagnostic and prognostic maintenance [158][159][160] . Using the data provided by a set of pressure sensors, Zhu et al 161 implemented a neural network model for the prediction of flow-front patterns at any impregnation time. Similar predictive models were also presented for forecasting resin cure 162 and flow front progression 163 .…”