Void content classification for acceleration sensor embedded glass-fiber reinforced components made by resin transfer molding

Resin transfer molding (RTM) is a production technique for fiber reinforced polymer components (FRPCs). It allows the embedding of sensors into components that enable production and structural health monitoring. Since void content is one of the most critical quality criteria, this study investigates the capability of embedded acceleration sensors to capture relevant information during RTM monitoring. For this purpose, a local void content classification based on sensor signals was considered, employing Machine learning algorithms for data labeling and classification. The final classifier’s performance was estimated using nested Cross-Validation (nested CV) and a hold-out test set. While the mean CV-estimates exceeded values over 89% for the main measures F 1.14 , F 1 , Recall and Precision, the overall performance on the hold-out set decreased to F 1.14 = 79%, F 1 = 78%, Recall = 88% and Precision = 70%. Model analysis revealed that the small, imbalanced data set led to instability, and that generalization errors from labeling affected the classifier. Nonetheless, class structure was visible in the frequency domain, confirming that the sensors recorded void content information. Furthermore, the main limitations can be addressed by transitioning from laboratory-scale to series production. This study hence provides a foundation for quality control of acceleration sensor embedded FRPCs produced via RTM.