Machine learning for rapid quantitative stucco phase analysis in plasterboard

Abstract

Stucco phase composition critically influences the mechanical properties of plasterboard, a cornerstone in modern construction. Traditional complete phase analysis (CPA) methods, while accurate, are hindered by prolonged processing times exceeding 12 h, impeding real-time quality control. This study introduces a machine learning-assisted CPA (ML-CPA) method, utilizing artificial neural networks (ANNs) to enable rapid, quantitative analysis of industrial stucco compositions. By integrating calcination weight loss profiles and hydration temperature curves—collected within 40 min—the method circumvents the need for extended conditioning periods. A dataset of 490 synthetic stucco samples, covering typical industrial phase ranges, was used to train the ANN model. The model achieved a root-mean-square error (RMSE) of 2.2 % in phase prediction and 87.7 % accuracy in free moisture detection. In particular, this approach reduces analysis time by 96 %, offering a scalable solution for online industrial quality control. By bridging the gap between laboratory accuracy and production-line efficiency, ML-CPA represents a transformative advancement in gypsum product manufacturing, with potential annual cost reduction and rapid quality control capability.