Research On Tracking And Positioning System Of Pipeline Inner Detector Based On CNN-SVM

Long-distance pipeline transportation is a very important mode of transportation.Compared with other modes of transportation,pipeline transportation has obvious advantages.However,due to various reasons,wear and tear may occur inside the pipeline.However,pipeline transportation is closed,and it is difficult to observe these conditions from outside,so it is necessary to detect the wall of the pipeline by the detector inside the pipeline.In actual operation,the internal detector sometimes gets stuck in the pipeline,which has a great impact on the normal transportation of the pipeline.In order to restore the pipeline operation status as soon as possible,it is necessary to locate the internal detector quickly and accurately,so it is particularly important to study the tracking and positioning of the internal detector.On the basis of studying the relevant literature,the following work has been done in this thesis:Firstly,in view of the operation of the internal detector in the field,this thesis designs a tracking and positioning model of the internal detector which combines flow rate and negative pressure wave,and proposes an improved negative pressure wave calculation method.Firstly,the internal detector is roughly positioned based on flow balance,and then the internal detector is corrected and positioned according to weld seam recognition,so as to realize the accurate positioning of the internal detector.Secondly,aiming at the problem of how to identify weld seam,a weld seam recognition model based on convolution neural network and support vector machine is proposed in this thesis.Firstly,the pressure data is preprocessed,including the selection of time window,data normalization and data unification at both ends of pipeline.Then,the features of the preprocessed data are extracted by convolutional neural network,and the extracted features are classified by support vector machine to identify the samples containing weld information.At the same time,in order to improve the recognition accuracy and update rate of the model,an adaptive learning rate update method and weight initialization method suitable for the network are proposed.Thirdly,aiming at the problem of how to locate the welding seam according to the pressure curve with seam information,this thesis presents a localization method based on empirical mode decomposition and split wavelet transform.Firstly,the pressure curve is denoised by the filtering method based on empirical mode decomposition.Then,the abrupt points of the pressure curve after filtering are extracted accurately by the split wavelet transform,and the precise location of the weld is realized.Finally,the effectiveness of the algorithm is verified by the simulation of a number of groups of first and last station data containing weld information.Fourthly,according to the above positioning method,this thesis designs an internal detection,tracking and positioning system,including hardware design and software implementation.The hardware part designs the data acquisition module;the software part designs and implements the whole software system,and then introduces the workflow of each functional module,and shows the design of several modules.