Estimating Of Compressive Strength Of Cement-based Material Using Microstructure Images Features In GPU Environment

Cement,concrete and other cement-based materials are widely used in industry,transportation,urban construction and other fields.The quality of cement-based materials is an important factor affecting the safety of building.Strength is an important indicator of cementbased materials quality and the compressive strength is crucial.Therefore,the strength estimation of cement-based materials plays a positive role in quality detection,design and engineering application.Generally,physical experiment is a method measuring the strength of cement-based materials,which consumes a lot of manpower and material.Moreover,this method is destructive.Recently,with the continuous development of computational intelligence,scholars have broken the traditional physical experiment method,and simulated the cement hydration process with computer.Furthermore,they have studied the strength of cement using the method of computational intelligence.Because the studying of cement strength is multivariable and nonlinear,using the traditional linear regression and clustering analysis method to estimate the strength have a low accuracy.However,the artificial neural network model has achieved good results.There are still some defects.Thus,improving the model structure and the estimation accuracy are the focus of the study.Improving the previous research methods,using the images of the cement hydration achieves nondestructively estimation.Extracting features from images can save a certain measurement time.Furthermore,this method can get real-time cement strength.In addition,because the training of numerous samples is time-consuming,GPU acceleration the training process is necessary.Therefore,the cement compressive strength can be estimated quickly and accurately.In addition,it is a very time-consuming and materialconsuming work to estimate the 28 day compressive strength of concrete with different material ratio.Therefore,it is of great significance to use the method of computational intelligence to guide the production of high performance concrete by changing the ratio of different materials.This paper mainly studies the compressive strength of cement-based materials from the following aspects.(1)Cement compressive strength estimation based on the features of two-dimensional Micro-CT imagesIn this paper,an estimation algorithm for compressive strength based on the characteristics of two-dimensional Micro-CT images of cement is proposed.The Micro-CT images of cement are obtained from the linear attenuation coefficient of different substances based on micro tomographic scanner.Therefore,different gray values represent different phases.Furthermore,it can describe the spatial structure information of phases.On the one hand,the image features which are used to describe the phase features during hydration are extracted from the gray level histogram and gray level co-occurrence matrix,and then the neural network is used to estimate the compressive strength of cement.On the other hand,the convolutional neural network is adopted to estimate the compressive strength of cement through images directly without the extraction of image features by manpower.Through the experiments,it is proved that the proposed method has a good estimation results.(2)Cement compressive strength estimation based on three-dimensional Micro-CT images of cement in GPU environmentIn this paper,a compressive strength estimation algorithm based on the three-dimensional Microstructure images of cement is proposed.At first,the phase features during hydration are extracted from the three-dimensional Microstructure images by gray level histogram and gray level co-occurrence matrix.Furthermore,the deep belief network is used to estimate the cement strength.In addition,the parallelization of the algorithm is realized by GPU.This method shows a lower estimation error,and the GPU speeds up the training process about 10 times through the experiments.(3)Estimation of 28 day compressive strength of concreteThe experimental data used in this study is the parameters of concrete material ratio,and is closely related to the concrete strength.The 28 day strength of concrete is estimated by stacked auto-encoders.Compared with other methods,it is proved that the method achieves a low estimation error and saves a lot of experiment time.