Abnormal Signal Detection Method For Slurry Pipeline Based On Deep Learning

Due to the characteristics of large ups and downs,great changes in elevation difference,and the working characteristics of plunger pump and the frequent occurrence of regulating pump operation,abnormal signals frequently occur in the pipeline.The existence of these abnormal signals seriously interferes with the accurate identification of pipeline leakage signals,resulting in many false positives and missed positives.It can provide effective technical support for reducing false positives and missing signals to accurately identify the types of abnormal signals and their time interval.However,there is no effective method to accurately identify and locate frequent abnormal signals.In this context,this paper studies the classification and location method of abnormal signals in pulp conveying pipeline based on deep learning.For abnormal signal classification and positioning problem,this paper analyzes the one-dimensional time domain signal,gray image,one dimensional time domain signal,time and frequency domain image and one dimensional time domain signals,binary image characteristics of three kinds of transformation methods,puts forward a three-dimensional time-domain signal-2 d image conversion method and target detection algorithm with the combination of abnormal signal detection method.By converting the one-dimensional signal time-domain waveform into binary image,the abnormal signal detection problem is transformed into an image processing problem.On this basis,the abnormal signal in the image is classified and located by using the target detection algorithm.In order to select a target detection algorithm that is most suitable for the application background of the subject,Label Img software was used to manually mark images to build a data set,and then an experimental platform was built.Many classical algorithms in the field of target detection were tested and compared.The performance of Faster-RCNN algorithm,SSD algorithm,Centernet algorithm,YOLOv3 algorithm and YOLOv4 algorithm in the current application scenario,the best YOLOv4 algorithm is selected as the basic algorithm of this topic.Aiming at the problems of low accuracy and many repeated alarms in YOLOv4 algorithm,the prior frame information clustering algorithm in YOLOv4 algorithm was optimized,a new prior frame information acquisition method was proposed,and the information post-processing algorithm was improved.The test results show that the improved measures proposed in this paper achieve better results,and effectively reduce repeated alarms and false positives.In order to realize the efficient call of the abnormal signal detection program based on Python language by the leak monitoring program based on Qt,the paper designs the realization method of information exchange between the monitoring program and abnormal signal detection program,and develops the testing platform of abnormal signal detection algorithm based on Qt framework.The real-time performance of information exchange method is tested.The test results show that the algorithm invocation method designed in this paper has high real-time performance and can meet the requirements of practical applications.The abnormal signal detection method of pulp conveying pipeline studied in this paper provides a new solution to the complex engineering problem of classification and location of abnormal signals frequently occurring in pulp conveying pipeline,and helps to reduce the occurrence of false alarms and missing alarms.