Estimation of Water-cement Ratio of Hardened Cement Paste Based on Microstructure Image and Convolutional Neural Network

As a determining factor for cement-based materials, water cement ratio (W/C) plays a crucial part in controlling their mechanical strength as well as durability. Whereas once solidified, the original water cement ratio of the cement paste cannot be determined effortlessly, which will lead to inadequate accuracy of the W/C estimation based on physicochemical method. In comparison, although the accuracy of W/C estimation based on microscope is improved, it requires processing of samples and calculating numerical data of image. In this paper, a novel approach for estimating W/C of hardened cement paste based on microstructure image and convolutional neural network (CNN) is proposed. Furthermore, the microstructure image of cement paste is obtained by micro computed tomography (micro-CT), and the W/C is estimated by CNN model subsequently. When adopting the proposed method to estimate cement W/C, no additional treatment is required for the sample. In addition, the proposed model directly uses the image of cement microstructure to predict W/C, avoiding the analysis and calculation of image numerical data. The results of the experiment show that the proposed method has excellent accuracy.