The prediction of cement compressive strength based on gray level images and neural network

Abstract

The prediction of cement compressive strength is a multi-variable and non-linear problem. In order to solve this problem preferably, a large amount of data should be collected. The testing of the cement strength takes a long time during hydration process. However, it is timesaving that using the gray level images of cement hydration process of different periods predicts cement strength. The images of the cement reflect the micro-structure of cement. Gray level histogram of an image shows the quantity of different phase, and gray level co-occurrence matrix shows the texture structure. Using the micro-structure of cement to describe the macroscopic properties is feasible. Therefore, a novel method using features of cement gray level images and neural network to predict the cement compressive strength is proposed. The cement images are gotten by microtomography. The eigenvalues of gray level histogram and gray level co-occurrence matrix are seen as the input to train the neural network. The value of cement compressive strength is seen as the output. Comparing with the multiple linear regression method and gaussian process to predict the cement compressive strength, the neural network model shows a lower error through the experiment.