Early detection of crack vulnerability in foldable displays through critical angle curvature analysis

Abstract

Foldable displays have become integral to consumer electronics, yet they remain susceptible to mechanical failures, particularly cracks in the hinge region caused by repeated mechanical stress. This study introduces an artificial intelligence (AI)-based method to predict and prevent crack formation by analyzing hinge surface curvature measurements captured within 0.2 seconds of unfolding to 160°, at critical vulnerable angles, using Principal Component Analysis (PCA) and uncertainty quantification with k-Nearest Neighbors (k-NN). A mathematical model incorporating exponential distance-based weighting quantified classification uncertainties, distinguishing confidently classified samples from uncertain cases. Furthermore, an AI-driven scoring model was validated through leave-one-out cross-validation (LOOCV) on an expanded dataset of 200 samples. This model successfully translates complex curvature data into numerical scores, achieving an F1 score of 0.9692 under conditions ensuring zero false positives, thus preventing defective products from reaching customers. This approach significantly enhances quality control in foldable display manufacturing.