Research On Acceleration Of Optical Fiber Sensing Data Based On Edge Heterogeneous Computing

In recent years,in order to more comprehensively monitor the health and operation of infrastructure such as subway,airport and bridge,large-scale fiber Bragg grating array sensors with long service life,electrical insulation,corrosion resistance,large capacity,high precision and long monitoring distance have been started in engineering buildings.Due to the characteristics of high sampling rate,many monitoring points and wide monitoring range,such sensors generate environmental data very quickly,These data often need to be analyzed and processed in real time.If the data processing speed can not keep up with the data generation speed,it will lead to a large backlog of data and lose timeliness.Therefore,accelerating the calculation of optical fiber sensing big data to achieve the goal of real-time processing is of great research significance for the application of optical fiber sensing big data technology in practical engineering.In order to achieve the above objectives,this paper studies from three aspects: intelligent algorithm research,heterogeneous computing acceleration and edge computing platform,so as to realize the real-time analysis of optical fiber sensing big data.The main contents are as follows:1.In the aspect of algorithm research,this paper studies the corresponding data analysis algorithm for the two application scenarios of subway tunnel drilling rig intrusion detection and airport intelligent runway vehicle trajectory tracking.In the scene of tunnel drilling rig intrusion detection,according to the range characteristics of drilling rig vibration frequency,the frequency domain analysis method is used,and the effectiveness of the detection algorithm is verified by MATLAB script simulation.In the track tracking scene of airport intelligent runway vehicles,this paper uses the neural network method,first constructs the data set,and then uses Python to build and train the network,which has achieved good results.After completing the simulation verification,this paper further uses C + + language to implement the above two algorithms on the CPU.2.Although the big data analysis algorithms for the above two application scenarios have been successfully deployed on the CPU,the CPU has a slow computing speed and the data processing speed is far lower than the data generation speed.To solve this problem,this paper adopts CPU / FPGA heterogeneous computing architecture and designs a hardware accelerator dedicated to the above algorithm to improve the data processing speed of the algorithm.Firstly,the parallel characteristic of the algorithm is used to calculate multiple data at the same time,which significantly speeds up the speed of data operation.For example,using the parallelism of accumulation calculation to design the addition tree,using the parallelism of convolution calculation to design the multiplication and addition tree of high-order wide data,etc.Secondly,pipeline technology is adopted to further realize the parallel processing of multiple groups of data and speed up the data processing speed.Finally,the floating-point number in the general-purpose processor is transformed into a fixed-point number with low bit width,so as to effectively reduce the occupation of hardware resources and speed up the calculation speed.3.Due to the large amount of data generated by optical fiber sensors,the communication cost of cloud processing is high,and congestion delay and other problems may be caused,edge computing technology is used.Multiple nodes are used on the edge side to process the data of multiple optical fiber sensors at the same time,and the data processing results are transmitted to the cloud for summary.The edge computing platform in this paper is based on Kube Edge.Because Kube Edge does not support optical fiber sensing data transmission protocol,it is difficult for optical fiber sensors to access to Kube Edge edge computing platform.To solve this problem,this paper optimizes the design of Kube Edge’s southbound device access layer,and develops a southbound access SDK that can quickly and conveniently access devices with different protocols.The SDK extracts the same part of the protocol access layer and constructs it into a shared library to improve the reusability of the code,and abstracts the driving protocols of different devices into a set of unified interfaces,so as to reduce the coupling between Kube Edge South access layer and various devices.Based on the SDK,this paper successfully accesses the optical fiber sensing data to Kube Edge,and completes the edge side processing and cloud aggregation of optical fiber sensing big data.In the experimental part,this paper uses Xilinx’s zynq 7020 chip as the experimental hardware platform,which includes CPU and FPGA.In the FPGA part,this paper uses Verilog hardware description language to realize the special hardware accelerator of big data processing algorithm in two application scenarios.The results show that the hardware accelerator can accelerate the calculation of data on the premise of ensuring the correct function.At the clock frequency of 100 MHz,the acceleration effect of the two algorithm hardware accelerators is obvious,which is about 321 times and 107 times faster than zynq’s arm cortex A9 CPU respectively.The power consumption of the two FPGA accelerators is low,only 0.681 w and0.452 w respectively.In terms of edge computing,the SDK developed in this paper realizes the access of optical fiber sensing data to Kube Edge,completes the efficient processing of optical fiber sensing data on the edge side,and reduces the data transmitted to the cloud in the two application scenarios by 5000 times and 20 times respectively.