Research On Data Anomaly Detection Method Based On Heterogeneous System

With the development of science and technology,the product structure produced and used by people is becoming more and more complex,the stability of equipment operation is also facing more complicated factors,and the economic cost of product testing and repair and maintenance has also increased.The emergence and development of fault prediction and health management technology has become an important method for people to maintain product reliability and reduce economic losses caused by faults.Under the wave of the era of artificial intelligence and big data,research and progress in data processing and feature extraction have made the data-based fault monitoring and health management technology have a good algorithm knowledge base.Thanks to the development of the Internet of Things technology,the acquisition and recording of device status data has become faster,and at the same time,the fault monitoring abnormality identification system can be deployed at a low cost and easily.Under this background,this subject designed and completed an abnormal data recognition system based on deep learning algorithms on a PYNQ development board carrying CPU + FPGA heterogeneous chips.This paper first introduces the research of fault monitoring and health management technology and the important step of fault abnormal data identification.The identification of abnormal data in real problems can often be attributed to the problem of data classification,and then combined with deep learning algorithms in machine learning for abnormal data Classification identification.Based on the Tensorflow open source algorithm library,the deep learning algorithm is used to build and train a model suitable for data abnormal recognition classification.Then use the Vivado HLS software tool to complete the design and export of the IP core of the abnormal data recognition model based on the deep neural network algorithm.Then use the Vivado synthesis tool to interconnect the designed IP core with the CPU through the bus to complete a system that can be loaded with data for abnormal data identification.The system can be run remotely by connecting a network cable,and through the Juppter Notebook software service interface to complete the data transmission nuclear operation,using the high-level language Python language to complete the call to the data anomaly recognition module.The data anomaly recognition system finally completed in the experiment,the data is imported for testing and verification,the experimental results are consistent with the model based on software simulation,that is,the hardware implementation of the data anomaly recognition algorithm model is completed,and compared with the traditional machine learning algorithm in the recognition of certain features There is a relatively high increase in the rate,and the estimated power consumption of the system is 1.437 W,which has a very low additional economic cost,and has good flexibility and room for expansion.