Research On Pipeline Defect Detection Based On Multi-modal Sensor Information Fusion

As an important part of the national infrastructure network construction,pipelines provide a lot of convenience for people’s lives,but accidents caused by pipeline leakage due to poor pipeline maintenance often occur.Therefore,it is necessary to inspect the pipeline.In-pipe detection refers to a detection method in which the detector is equipped with a storage device,moves forward and detects inside the pipeline,and uniformly processes and analyzes the stored data after the detection is completed.The internal structure of the pipeline is changeable,and problems such as the jitter caused by the operation of the detector lead to a large amount of noise and interference signals in the collected signal.Therefore,it is difficult to identify the defect signal by traditional methods.On the other hand,after the defects are detected,traditional positioning methods such as mileage wheels are not accurate enough,which makes it difficult for the detected defects to correspond in real pipelines.Thesis takes the detection and location of defects in the pipeline as the research goal,uses the internal detector to collect data,fuses the multimodal sensing signals,and focuses on how to identify the defect signal and realize the defect location from the collected complex signals.The main research contents are as follows:1.In view of the problem of complex signal composition and interference defect identification in the acquisition signals,a defect detection framework named Feature Boosting as the main idea is designed.This framework can realize the hierarchical identification and extraction of different signals in the collected signal,so as to avoid the interference between different signals.At the same time,the framework can flexibly change the structure and output results according to different detection requirements,which improves the flexibility and scalability of the algorithm model.2.Aiming at the problem that the noise signal in the collected signal is large and affects the accuracy of defect identification,a time series feature mapping module is designed.This module comprehensively considers the global and local features of the time series,and maps them to the polar coordinate system.The abnormal points in the time series can be distinguished by the polar diameter and polar angle.After the feature is added,the result of anomaly detection algorithm is improved obviously and the false detection phenomenon has been significantly suppressed.3.Aiming at the problem of poor localization effect of traditional defect localization methods in pipelines,a two-step localization method is designed by fusing multimodal sensor signals.Considering that the pipeline is a man-made system,the information provided by the design drawing and the positioning points are used to achieve the preliminary positioning of the defect position,and then the precise restoration of the specific position between the positioning points is achieved through other sensor information.This method improves the accuracy of defect location.Thesis conducts extensive experiments and verifications in three different environments.The defect detection and localization effects in the three environments are better than the current typical algorithms.