Scholarlink

<p>This study developed seven comparative machine learning (ML) frameworks to predict optimal processing parameters and mechanical properties for automotive-grade carbon fibre-reinforced plastics (CFRPs) produced via Liquid Composite Molding (LCM). Multi-linear- (MLR), support vector- (SVR), random forest- (RFR), and gradient boosting regression (GBR) algorithms were coupled with artificial neural network (ANNs) using multi-layer perceptron (MLP), single and dual-path functional application programming interface (FAPIs) models to increase the prediction outcomes. The models developed herein can be tailored to the case-specific needs of the application. The results demonstrated that the RFR model, enhanced with hyperparameter tuning, achieved the highest predictive accuracy, explaining 45% of the variability in the data for the linear regression case. The FAPI model using Keras® and Tensorflow® exhibited superior performance for the non-linear case, with test predictions of 78.20% for flexural strength, 73.8% for flexural modulus and 99.97% for binder use. Additionally, a thermal and demolding analysis expanded the complexity of the predictions by enhancing the correlation to part quality, ultimately accelerating product development and providing industry with accurate predictive modelling tools.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An Artificial Intelligence (AI) approach to predicting processing parameters for Liquid Composite Molded (LCM) Carbon Fibre-Reinforced Plastics (CFRPs)

  • 期刊论文

  • 发表日期：

    2025年8月16日

查看全文
  • Jennifer Sears,
  • Jennifer Johrendt

摘要

This study developed seven comparative machine learning (ML) frameworks to predict optimal processing parameters and mechanical properties for automotive-grade carbon fibre-reinforced plastics (CFRPs) produced via Liquid Composite Molding (LCM). Multi-linear- (MLR), support vector- (SVR), random forest- (RFR), and gradient boosting regression (GBR) algorithms were coupled with artificial neural network (ANNs) using multi-layer perceptron (MLP), single and dual-path functional application programming interface (FAPIs) models to increase the prediction outcomes. The models developed herein can be tailored to the case-specific needs of the application. The results demonstrated that the RFR model, enhanced with hyperparameter tuning, achieved the highest predictive accuracy, explaining 45% of the variability in the data for the linear regression case. The FAPI model using Keras® and Tensorflow® exhibited superior performance for the non-linear case, with test predictions of 78.20% for flexural strength, 73.8% for flexural modulus and 99.97% for binder use. Additionally, a thermal and demolding analysis expanded the complexity of the predictions by enhancing the correlation to part quality, ultimately accelerating product development and providing industry with accurate predictive modelling tools.