(3) (PDF) CLUSTERING OF LOCALIZED ACOUSTIC EMISSION SOURCES BY THE DBSCAN ALGORITHM IN SEPARATORS

Page 1

1 Corresponding author: Pavel MakarikhinEmail: St054466@student.spbu.ru 495Vol. 05, No. 3 (2023) 495-502, doi: 10.24874/PES05.03.013Proceedings on EngineeringScienceswww.pesjournal.netCLUSTERING OF LOCALIZED ACOUSTICEMISSION SOURCES BY THE DBSCANALGORITHM IN SEPARATORSPavel Makarikhin1Anastasiya GrigorievaMargarita MaksimenkoReceived 21.01.2023.Accepted 11.06.2023.UDC – 620.179.17Keywords:Acoustic Emission, Location, Defects,Approximate Calculation,Clusterization, Pressure VesselA B S T R A C TIn this paper, the method for locating sources of acoustic emission byapproximate calculation of potential coordinates using a grid superimposedon the area around the sensors that detected the wave was proposed. Variousnoises such as electromagnetic pulses and other external noises were removedfrom input data. The DBSCAN clustering algorithm was applied. The resultswere compared with the real state of the object under study after additionalcontrol. Analytical and practical research points to the possibility of using thepresented method as a tool for determining the coordinates of defectdevelopment points and wave velocity.© 2023 Published by Faculty of Engineering1. INTRODUCTIONWhen operating objects, it is very important to knowtheir condition in order to avoid emergencies. Objectssuch as bridge columns, tanks for the transport ofliquids, various structures in industry and many othersare under constant loads during their use (Balagurin etal., 2020; Rastegaev et al., 2018; Nosov et al., 2017).There is a wide range of non-destructive testingmethods that allow monitoring objects withoutdestroying or dismantling it, such as xray inspection (Liet al., 2022), ultrasonic method (Smoqi et al., 2023) Oneof the most popular approaches is the acoustic emission(AE) method due to its accuracy and environmentalfriendliness. The AE method finds its application invarious areas of technological process monitoring. Forexample, AE method is allowing to predict the structureof the metal during casting (Yaroslavkina et al., 2020),in (Ser’eznov et al., 2020) AE method was applied inaircraft construction, (Makhnutov et al., 2020) appliedAE methods to monitor composite fiber failures. Theessence of the AE method is to detect the acousticwaves that arise in the object when defects occur duringthe operation of the object. To perform this method, AEsensors are installed on the investigated area of theobject, then during the continuous loading of the object,various defects begin to appear in it: cracks, splits,material crumbling, and others. When they occur andduring their growth, defects emit acoustic wavespropagating through the object, which are recorded bysensors. Various parameters of the acoustic wave arerecorded: amplitude, energy, time of arrival at thesensor, and many others. This is similar to complexsystems with non-periodical dynamics (Pyko et al.,2018). Assessing the condition of an object can be verytime consuming due to the size of the object underexamination, so it is extremely important to reduce theareas that need to be checked by a specialist. Acoustic

Page 2

Makarikhin et al., Clustering of localized acoustic emission sources by the DBSCAN algorithm in separators496emission data sets are usually quite large, but thecorrelation of some of the parameters allows them to bereduced for more convenient and faster processing.Analysis of the data allows obtaining information aboutthe defects in the object, their size, developmentprogress and location. At the moment, many methodshave been developed for analyzing acoustic emissiondata for various materials. However, there are still nouniversal automatic methods of analysis that give anaccurate result without the participation of a specialist,therefore, this area is relevant to this day.At the moment, there are quite a lot of differentsoftware for analyzing acoustic emission data, however,the operation of their methods is often hidden fromusers by trade secrets, and the algorithms used are oftennot updated along with the development oftechnologies. (Vallen Systeme, Interunis, Diaton, etc.)Analysis of large areas of the object under study takes alot of time for specialists, so location and clusteringmethods are needed to indicate the areas in whichdefects are most likely to be located.There are various approaches to source location:frequency and wave mode analysis (Jiao et al., 2008),laser-based reversal time concept (Park et al., 2012),analytical approach (Grigorieva et al., 2022). The resultof the work of location methods are areas in which therewill be a defect with the highest degree of probability.The location accuracy is influenced by many factors: thestructure and shape of the object under study, itscondition, the presence of cavities in it, design features,and much more. Different location methods take thesefactors into account to varying degrees.After using location methods, various machine learningmethods are often applied to the results obtained todetermine the type of signal source and predict thefurther appearance of defects. In recent years, the fieldof machine learning has been developing rapidly,various approaches, such as neural networks, clusteringand classification algorithms, and others, allow you toget more and more accurate results. Clustering andclassification algorithms are used to group data fromlocated sources and form many types of defects. Todate, there are many approaches to data clustering thatdiffer in data type and clustering algorithm. Differentalgorithms have different advantages and disadvantages,as well as different requirements for input parameters.In (Feifei et al., 2011) K-means algorithm was usedwith cluster centers initializing with random uniformdistribution to separate different types of AE signalssources. In (Calabrese et al., 2010) two differentunsupervised clustering approaches were used to reduceset of data parameters and identify clusters of differentAE signals: Kohonen map and principal componentanalysis. In (Feifei at al, 2012) DBSCAN (Density-based spatial clustering of applications with noise)(Ester et al., 1996) and K-means were applied tospecific material 2.25Cr-1Mo experiment data and wereused to separate different types of signals and to extractburst cracking signals.Solving the problem of localization and clustering ofAE signals when examining thin-walled cylindricalvessels used for transporting and storing explosivecombustible substances is relevant, as it helps to preventtheir early failure, downtime, and even a technogenic orenvironmental disaster.2. PRIMARY DATA PROCESSINGWhen fixing the signals of acoustic emission to sensorson thin-walled vessels, the task goes from spatial to flat- to development of a cylindrical object on a plane. Foranalysis, the size of the object, the coordinates of thesensors placed on it, various signal parameters, thesignal arrived on the sensor and others detected bysensors are used. Various design features of object suchas hatches, manholes, supports, unions and others arealso taken into account. In Vallen, a single propagationvelocity is set for all signals of the experiment, whichdoes not take into account the design and materialfeatures, therefore, it is not used in the location andclustering algorithms of this work.Among the unfavorable factors that have the mostnegative impact on the result of applying the method arenoise-like signals that accompany all modes ofoperation of most industrial equipment, especiallydynamically loaded equipment, and therefore noise is anintegral part of any AE diagnostic signals. A high levelof noise can lead to a failure in the correct operation ofAE signal detectors, which is accompanied by: skips insignal registration; errors in calculating the time of theirarrival; the appearance of false or displacement of reallocation events; incorrect assessment of the hazard classof acoustic sources and, in general, an incorrectassessment of the technical condition of dangerousproduction facilities. (автореферат , цитированиеИгоря Анатольевича)To solve the problem of detecting signals at the noiselevel and the possibility of recognizing from several,simultaneously acting acoustic sources, a method waschosen that is used by almost all major scientific groupsinvolved in AE methods, namely: filtering (noisesuppression) of the recorded signals in order to bringthem to an impulse form for estimation by the amplitudethreshold method.In this work, filtering is carried out: by the amplitudethreshold, by the number of sensors that recorded thesignal, by the electromagnetic attribute. There are alsotechnical limitations of AE sensors that do not capturethe low frequency range below 40 kHz. These technicalcharacteristics lead to forced filtering on a frequencybasis. This work uses the VS150-RSC. The VS150-RSCis a piezoelectric AE sensor with integrated Vallen