A new path to intelligent quantitative prediction of ceramic strength: Machine vision combined with machine learning

Abstract Complex and disordered pores are one of the fundamental physical causes underlying the random distribution of fracture strength in structural ceramics, presenting a dual challenge for quantitative strength prediction: accurate pore detection/extraction, and the identification of scientific descriptors. Using Si 3 N 4 ceramics as a case study, this work proposed a straightforward, cost‐effective, and precise methodology for quantitative strength prediction based on machine vision and machine learning techniques. First, pore features were extracted through microscopic observation and image processing‐based machine vision techniques. Feature space was then reduced to five key features using SRCC and PCC. After comparing seven classical algorithms, an AdaBoost‐based model was established, achieving an R 2 > 0.96 with approximately 3% error in the validation set. Through interpretable machine learning, the pore convexity ratio (with a critical threshold of 1.18) was identified as a decisive descriptor. This study provides theoretical support and methodological guidance for the reliable application of structural ceramics, with a focus on applied research.