Research On Automatic Recognition Technology In TBM Construction

Geological advance warning is a very important link in tunnel construction.Timely understanding of geological information will be of great help to the construction.If early know a geological risk ahead,can be prepared for,take refuge in casualties and equipment damage,so as to avoid delay the progress.This paper takes the EH superlong diversion tunnel constructed by TBM as the background.The tunnel extends for hundreds of kilometers,and 18 sets of TBM are used for interzone construction.Up to now,the number of TBM is the world`s first,and the tunneling mileage of TBM single hold is the world`s first.The tunnel span is large,and the geological conditions are changeable,and that traditional advance geological warning is of great time and effort,which cannot meet demand in super long tunnel construction.Therefore,this paper focuses on the study of the automatic recognition technology of rock slag as the object,and establishes the automatic recognition model of rock slag in the way of deep learning,and reflects the geological information through different rock slag patterns,so as to judge whether there are risks such as landslides and faults in the geological.Rock slag contains a lot of geological information,which can be extracted to predict geological changes.This paper takes the rock slag of SS section in EH water diversion project as the research object,combined with the design parameters and geological data of TBM,summarizes the statistical theory of rock slag.The theoretical statistics of rock slag are verified with the specific conditions of TBM slag discharge in a fault fracture zone in the standard section.The results show that the content of rock mass in this section is about 90%,and the content of rock slice is about 5%,which conforms to the statistical law of rock slag corresponding to the surrounding rock of class V.In addition,the rock slag results of a certain section of class III surrounding rock are about 20% rock mass and 80% rock slice,which conforms to the statistical law of slagging.On the premise of statistical law of rock slag,the automatic recognition technology of rock slag is studied.The deep learning model of image recognition is designed by convolutional neural network,and the sample set is processed by image processing technology.For the classification of block and flake rock slag,the convolution neural network model with 8 layers of pre-training is used.And the model is optimized by gradient descent algorithm,sparse representation,early stopping and dropout.After the training,the accuracy rate on the test set reached 92.6%,and it could successfully identify rock blocks and pieces,but the effect on risk warning was not obvious.In order to realize the geological information warning more quickly and accurately,the model is applied to the new classification problem by using the transfer learning method.The experiment proves that the transfer learning can make the network converge faster,and the prediction accuracy on the test set is up to 92.7%.Through the practical engineering image verification,the rock slag can be successfully divided into two categories of risk and risk-free.