Research On Rapid Prediction Of Mechanical Parameters And Quality Level Of Granite After High Temperature Based On Deep Learning

With the implementation of China’s "strong transportation" strategy,the scale and number of tunnel construction in China is growing rapidly,however,there are many adverse factors inside the tunnel that can easily lead to tunnel fire accidents.The study of physical and mechanical properties of rocks subjected to high temperatures has important engineering applications for post-fire assessment and repair of tunnels.However,in the existing research,scholars mainly obtain rock physical and mechanical parameters by means of indoor experiments,which requires high precision of experimental equipment,long experimental period and large cost,and is difficult to provide reasonable suggestions for site construction in time.In order to achieve rapid prediction of rock mechanical parameters and quality levels,this paper uses deep learning methods and takes granite after high temperature as the research object.Firstly,according to the importance of rock physical and mechanical parameters for the evaluation of the stability of the surrounding rock,the rock quality evaluation system with physical and mechanical parameters as evaluation index is proposed,and its change law and internal mechanism are analyzed and discussed.Secondly,the correlation between the change of physical surface characteristics and the change of mechanical properties was explored from three aspects: experimental phenomena,internal mechanism and statistical methods of analysis and calculation.Finally,based on deep learning image feature extraction technology,the image features of granite after high temperature are automatically extracted and analyzed,and then the images are retrieved and classified,and the rapid prediction of mechanical parameters of granite after high temperature and its quality level is achieved by the above method.The main research results achieved are as follows:(1)Combined with the practical application of engineering,the granite quality evaluation system was constructed,and an evaluation method combining the improved combination of assignment and ideal point method was proposed.The obtained physical and mechanical parameters and quality evaluation grades of granite were imported into My SQL database to establish a database of granite after high temperature.(2)By analyzing the macroscopic characteristics and physical and mechanical parameters of granite at different temperatures,and combined with the Pearson correlation coefficient method in statistics for quantitative calculations.It is concluded that the quartz content,surface cracks,rock integrity and overall color variation degree in granite have strong correlation with compressive strength.(3)Three convolutional neural network models are built for feature extraction and classification of granite images after high temperature.Select one of the feature extraction models that best meets the actual needs,and introduce the attention module into it,which can greatly improve its accuracy rate and other model evaluation metrics.And the new model is more suitable for the network configuration of field portable devices.(4)On the basis of granite feature extraction and classification after high temperature,we developed a function for granite identification and localization in the field.And the granite image retrieval function is implemented by the cosine similarity algorithm.The mechanical parameters and quality level of granite after high temperature are quickly predicted by the most similar image.