In the process of road excavation,the cutter head extrudes and cuts the rock mass,which is easy to cause cutter head wear and damage,resulting in economic losses.Therefore,the theory and technology of cutter head wear detection is needed to guide the con-struction.Rock slag is a direct product of the excavation process,and carries rich information that can reflect the current construction status.Therefore,the cutter head can be monitored indirectly by using this information through rock slag identification.A slag identifica-tion algorithm based on convolutional neural network is proposed,which achieves a classification accuracy of 96.5%on the slag dataset.Then,in order to facilitate the deployment of FPGA hardware,a network compression method is proposed,which compresses the network size to 2.28%of the original network,while the classification accuracy is only 0.9%lower than the original network.Finally,the algo-rithm is deployed on the Intel Arria 10 GX1150 platform using OpenCL,achieving a throughput rate of 224.54 GOP/s and an energy ef-ficiency ratio of 11.23 GOP/s/W.
关键词
岩渣分类/FPGA/卷积神经网络/OpenCL/硬件加速
Key words
rock slag identification/FPGA/convolutional neural networks/OpenCL/hardware acceleration
维普
万方数据