Nondestructive quantitative model for predicting the residual tensile strength of silk textile cultural heritage based on machine learning

Abstract

Silk is susceptible to deterioration induced by light, heat, and moisture, which leads to reduced structural stability and progressive strength loss. However, because destructive analysis is fundamentally unsuitable for silk textile cultural heritage, residual tensile strength is typically inferred through subjective evaluation. This study aims to quantitatively analyze deterioration behavior and examine the feasibility of nondestructive prediction of the residual tensile strength of silk textile cultural heritage by employing indicators that exhibit significant correlations with tensile strength reduction. As indicators of tensile strength, protein secondary structure, pH, moisture regain, yellow index, and K/S value were measured in artificially aged silk textiles. Results showed that tensile strength was not significantly influenced by differences in deterioration environments but correlated with crystallinity and protein secondary structure composition. Although linear regression demonstrated limited explanatory power (R² = 0.576), XGBoost demonstrated improved predictive performance (R² = 0.654–0.751) by capturing complex multivariate interactions, suggesting the potential for nondestructive estimation of residual tensile strength in textile cultural heritage. The prediction model using crystallinity, protein secondary structure contents, pH, yellow index, and K/S value showed the highest predictive accuracy. Further, with the accumulation of data encompassing textiles with diverse properties and the refinement of relevant indicators, the development of a machine learning–based nondestructive precision prediction model is anticipated. Such an approach can overcome the limitations of subjective assessment and contribute to systematic conservation management through the quantitative prediction of residual tensile strength in textile cultural heritage.