KR102138279B1 - Apparatus and method for monitoring vibration of rotating equipment based on deep learning time series analysis - Google Patents

Abstract

Disclosed are an apparatus and a method for monitoring a rotating equipment vibration state using deep learning-based time series analysis. According to one embodiment of the present application, the apparatus for monitoring a rotating equipment vibration state using deep learning-based time series analysis comprises: a data receiving unit receiving vibration data of rotating equipment from a vibration sensor installed on the rotating equipment; a data preprocessing unit removing noise from the vibration data and extracting a feature variable from the vibration data from which noise has been removed; a learning unit generating a predicted value associated with the vibration data through a deep learning-based time series analysis on the extracted feature variable, and generating error information based on the predicted value and the actual value of the vibration data; a margin setting unit setting a margin for determining the presence or absence of a defect associated with the vibration data based on the error information; and a defect detection unit receiving new vibration data for the rotating equipment and determining whether the new vibration data is defective based on the margin.

Description

Device and method for monitoring vibration condition of rotating equipment using deep learning based time series analysis{APPARATUS AND METHOD FOR MONITORING VIBRATION OF ROTATING EQUIPMENT BASED ON DEEP LEARNING TIME SERIES ANALYSIS}

The present invention relates to an apparatus and method for monitoring the vibration condition of a rotating facility using time-series analysis based on deep learning. More specifically, a vibration state monitoring device and method for monitoring the state of a rotating facility in real time and setting an optimal alarm margin to analyze the vibration sensor based on a deep learning time series analysis model to grasp the current situation of the rotating facility in real time It is about.

Conventional vibration monitoring system and alarm system for rotating equipment has been established through a method of processing a vibration signal and setting a specific threshold value based on expert experience and engineering knowledge on the vibration signal collected based on business rules. However, the factors that indicate the condition of the equipment are very diverse and the number of equipment managed by the equipment manager is large, making it difficult to set a suitable threshold value considering the characteristics of the equipment. Signal processing and engineering for the interpretation of vast data Due to the essential background knowledge, it is difficult for non-experts to know what defects have occurred.

In addition, the alarm method using the conventional threshold value is difficult to distinguish between the signal temporarily exceeding the threshold value and the signal due to a defect due to an external shock, but the warning is triggered by an external shock, etc., even though a defect has not actually occurred in the equipment. This could lead to economic losses, such as downtime and maintenance.

Therefore, it is difficult to provide an accurate monitoring performance only by determining whether the alarm associated with the vibration signal exceeds a certain threshold, and it is important to consider the time series characteristics of the vibration signal together to overcome this. Currently, in Korea, the vibration monitoring system has been partially implemented, but the development of the monitoring system that reflects the time-series characteristics of the vibration signal is insufficient. Accordingly, there is an increasing need for a vibration condition monitoring technology for a rotating facility in consideration of time-series characteristics of a vibration signal based on a deep learning time series analysis technology.

The background technology of the present application is disclosed in Korean Patent Registration No. 10-2027389.

The present application is intended to solve the problems of the prior art described above, and monitors and analyzes a vibration signal representing the state of a rotating facility in real time to detect and signal a defect signal. It is an object to provide a monitoring device and method.

The present application is intended to solve the problems of the prior art described above, and is based on deep learning that enables a more rigorous diagnosis of defects by applying a sequence algorithm that performs classification for a time zone rather than a general AI algorithm that performs classification for a specific time. An object of the present invention is to provide an apparatus and method for monitoring the vibration condition of a rotating facility using time series analysis.

However, the technical problems to be achieved by the embodiments of the present application are not limited to the technical problems as described above, and other technical problems may exist.

As a technical means for achieving the above technical problem, a vibration monitoring system for a rotating facility using a deep learning-based time series analysis according to an embodiment of the present application receives vibration data of the rotating facility from a vibration sensor installed with respect to the rotating facility. The receiving data receiving unit, a data pre-processing unit for removing noise of the vibration data and extracting characteristic variables from the vibration data from which the noise has been removed, and the vibration data and the vibration data through deep learning based time series analysis on the extracted characteristic variables A learning unit generating an associated predicted value and generating error information based on the predicted value and the actual value of the vibration data, and a margin setting for setting a margin for determining whether there is a defect associated with the vibration data based on the error information It may include a defect detection unit for receiving the new vibration data for the unit and the rotating equipment, and determines the presence or absence of defects in the new vibration data based on the margin.

In addition, the learning unit generates a prediction sequence for estimating a change in the vibration data during a predetermined unit time as the prediction value, and an error corresponding to a difference between the actual sequence and the prediction sequence of the vibration data during the unit time. Vectors can be generated from the error information.

In addition, the defect detection unit may determine the presence or absence of the defect through an algorithm based on supervised learning using the error vector and the margin as inputs.

In addition, the defect detection unit may determine whether a defect occurs by analyzing the new vibration data in units of the unit time through the supervised learning-based algorithm.

In addition, the learning unit may perform the deep learning-based time series analysis based on at least one of an LSTM algorithm, an Attention algorithm, a Transformer algorithm, and a BERT algorithm.

In addition, the apparatus for monitoring vibration state of a rotating facility using time-series analysis based on deep learning according to an embodiment of the present application may include an alarm unit that outputs a preset alarm when it is determined that the new vibration data is defective.

In addition, the margin setting unit may set the margin based on at least one of a maximum likelihood estimation algorithm and a multivariate Gaussian distribution analysis.

In addition, the data pre-processor may remove the noise by applying a low pass filter or a band pass filter to the vibration data.

In addition, the data pre-processing unit applies the FFT (Fast Fourier Transform) or STFT (STFTL: Short Time Fourier Transform) to the vibration data from which the noise has been removed, and the noise-removed vibration data is frequency domain data. Can be converted to

On the other hand, a method for monitoring a vibration condition of a rotating facility using deep learning-based time series analysis according to an embodiment of the present application includes receiving vibration data of the rotating equipment from a vibration sensor installed with respect to the rotating equipment, and reducing noise of the vibration data. Removing, extracting a characteristic variable from the vibration data from which noise has been removed, and generating a prediction value associated with the vibration data through deep learning based time series analysis on the extracted characteristic variable, and generating the prediction value and the vibration data Generating error information based on an actual value of, setting a margin for determining the presence or absence of a defect associated with the vibration data based on the error information, and receiving new vibration data for the rotating equipment, and generating the margin And determining whether or not the new vibration data is defective.

In addition, the generating of the error information may include generating a prediction sequence for estimating a change in the vibration data during a predetermined unit time as the prediction value, and an actual sequence and the prediction sequence of the vibration data during the unit time. It may include generating an error vector corresponding to the difference of the as the error information.

In addition, in the determining of the presence or absence of the defect, the presence or absence of a defect may be determined by analyzing the new vibration data in units of the unit time through an algorithm based on supervised learning using the error vector and the margin as inputs. .

In addition, the method for monitoring a vibration condition of a rotating facility using deep learning-based time series analysis according to an embodiment of the present application may include outputting a preset alarm when it is determined that the new vibration data is defective.

The above-described problem solving means are merely exemplary and should not be construed as limiting the present application. In addition to the exemplary embodiments described above, additional embodiments may exist in the drawings and detailed description of the invention.

According to the above-described problem solving means of the present application, a vibration monitoring device and method using a deep learning-based time series analysis capable of detecting and notifying a signal of a defect by monitoring and analyzing a vibration signal representing a state of a rotating facility in real time Can provide

According to the above-described problem solving means of the present application, since the presence or absence of a defect in the rotating equipment is classified for a unit time rather than a specific time point, it is possible to grasp a precursor phenomenon that does not exceed a threshold value, and accordingly, a strict classification of the defect may be possible. .

According to the above-described problem solving means of the present application, since the collected data is automatically processed and learned to classify the defects, it can be applied even when the characteristics of the collected data change at a specific point in time due to changes in the environment around the facility, Even if you do not rely on experts, you can easily maintain, repair and manage rotating equipment.

However, the effects obtainable herein are not limited to the effects described above, and other effects may exist.

1 is a schematic configuration diagram of a vibration condition monitoring device for a rotating facility using time-series analysis based on deep learning according to an embodiment of the present application.
2 is a diagram visualizing a result of feature extraction in a time domain of vibration data according to an embodiment of the present application.
FIG. 3 is a diagram visualizing a result of converting vibration data into a frequency domain according to an embodiment of the present application.
4 is a diagram showing visualization of a result of time-frequency domain conversion of vibration data according to an embodiment of the present application.
5 is a diagram exemplarily showing prediction values generated through deep learning-based time series analysis according to an embodiment of the present application.
6 is a graph showing error information of predicted values and actual values generated through deep learning-based time series analysis according to an embodiment of the present application.
7 is an operation flowchart of a method for monitoring a vibration condition of a rotating facility using time-series analysis based on deep learning according to an embodiment of the present application.

Hereinafter, embodiments of the present application will be described in detail with reference to the accompanying drawings so that those skilled in the art to which the present application pertains may easily practice. However, the present application may be implemented in various different forms and is not limited to the embodiments described herein. In addition, in order to clearly describe the present application in the drawings, parts irrelevant to the description are omitted, and like reference numerals are assigned to similar parts throughout the specification.

Throughout this specification, when a part is "connected" to another part, it is not only "directly connected", but also "electrically connected" or "indirectly connected" with another element in between. "It includes the case where it is.

Throughout this specification, when one member is positioned on another member “on”, “on top”, “top”, “bottom”, “bottom”, “bottom”, it means that one member is on another member This includes cases where there is another member between the two members as well as when in contact.

Throughout the present specification, when a part “includes” a certain component, it means that the component may further include other components, not to exclude other components, unless otherwise stated.

The present invention relates to an apparatus and method for monitoring the vibration condition of a rotating facility using time-series analysis based on deep learning. More specifically, a vibration state monitoring device and method for monitoring the state of a rotating facility in real time and setting an optimal alarm margin to analyze the vibration sensor based on a deep learning time series analysis model to grasp the current situation of the rotating facility in real time It is about.

1 is a schematic configuration diagram of a vibration condition monitoring device for a rotating facility using time-series analysis based on deep learning according to an embodiment of the present application.

Referring to FIG. 1, a rotation facility vibration state monitoring device 100 (hereinafter referred to as'vibration state monitoring device 100') using a deep learning-based time series analysis according to an embodiment of the present application includes a data receiving unit. It may include a 110, a data pre-processing unit 120, a learning unit 130, a margin setting unit 140, a defect detection unit 150 and an alarm unit 160.

The vibration state monitoring device 100 and the vibration sensor (not shown) installed with respect to the rotating equipment can communicate with each other through a network (not shown). The network (not shown) refers to a connection structure capable of exchanging information between each node, such as terminals and servers, in an example of such a network (not shown), a 3GPP (3rd Generation Partnership Project) network, LTE (Long Term Evolution) network, 5G network, World Interoperability for Microwave Access (WIMAX) network, Internet, Local Area Network (LAN), Wireless Local Area Network (LAN), Wide Area Network (WAN), PAN ( Personal Area Network), wifi network, Bluetooth network, satellite broadcasting network, analog broadcasting network, DMB (Digital Multimedia Broadcasting) network, and the like.

The data receiving unit 110 may receive vibration data of the rotating equipment from a vibration sensor installed with respect to the rotating equipment. According to one embodiment of the present application, the vibration sensor installed with respect to the rotating equipment may be a sensor that measures vibration data including vibration displacement information, acceleration information, and the like of the rotating equipment. In addition, according to the embodiment of the present application, the vibration sensor may be provided with various types of sensors, such as a piezoelectric acceleration method, a cantilever vibration method, and an optical fiber method.

The data pre-processing unit 120 may remove noise of the vibration data and extract characteristic variables from the vibration data from which the noise has been removed. According to one embodiment of the present application, the characteristic variables include a shape, shape (amplitude), phase, etc. of a waveform extracted from a waveform representing a change in vibration or acceleration over time for a rotating facility. can do. In other words, the data preprocessing unit 120 may extract characteristic variables for at least one of shape, size, and phase from the vibration data.

Specifically, according to an embodiment of the present application, the data pre-processing unit 120 applies noise to the received vibration data by applying a preset low-pass filter or a band-pass filter. Can be removed.

2 is a diagram visualizing a result of feature extraction in a time domain of vibration data according to an embodiment of the present application.

Referring to FIG. 2, FIG. 2(a) is a graph exemplarily showing a time-series change of RMS (Root Mean Squre) values of vibration data, and FIG. 2(b) is a center frequency (CF) of vibration data. 2(c) is a graph exemplarily showing a time-series change in the value of Var (Variance, variance) of vibration data, and FIG. 2(d) is a vibration 2(e) is a graph exemplarily showing a time-series change of skewness value of vibration data, and FIG. 2 (F) is a graph exemplarily showing a time-series change of MF (Medium Frequency) values of vibration data, and FIG. 2(g) exemplarily shows a time-series change of Kurtosis values of vibration data. FIG. 2(h) is a graph exemplarily showing a time-series change of a peak to peak (PTP) value of vibration data.

Referring to FIG. 2, it can be seen that in general, characteristics in the time domain of vibration data indicate a trend that gradually increases or decreases over time. As described above, in the case of vibration data (vibration signal), it is necessary to monitor the presence or absence of a defect or the vibration state by comprehensively considering the time-series characteristics of the vibration data (vibration signal) as well as whether the threshold value applied only at a specific time point is exceeded . In this regard, the present application enables the determination of defects reflecting the time-series characteristics of the vibration data of the rotating equipment that is continuously monitored by the vibration sensor through deep learning-based time series analysis, as described later, so that it is in a normal state and abnormal state. Strict classification of the state (defect generation state) is possible, and it is possible to grasp the precursor phenomenon preceding the occurrence of the defect in a certain period of time.

In addition, referring to Figure 2, the data pre-processing unit 120 according to an embodiment of the present application, RMS analysis for vibration data, center frequency (CF) analysis, variance (Var) analysis, intermediate frequency (IF) analysis, One or more characteristic variables may be extracted from vibration data based on at least one of a skewness analysis, a medium frequency (MF) analysis, a kurtosis analysis, and a PTP analysis.

3 is a diagram visualizing a result of conversion of vibration data to a frequency domain according to an embodiment of the present application, and FIG. 4 is a diagram showing a result of time-frequency domain conversion of vibration data according to an embodiment of the present application. .

Referring to FIGS. 3 and 4, the data pre-processing unit 120 according to an embodiment of the present application may perform FFT (Fast Fourier Transform) or STFT (Short Time Fourier Transform) with respect to vibration data from which noise has been removed. By applying, the vibration data from which noise has been removed can be converted into data in the frequency domain or data in the time-frequency domain. In other words, the data pre-processing unit 120 performs pre-processing on the collected vibration data, but converts the collected time waveform data (vibration data in the time domain) into a frequency domain through fast Fourier transform (FFT) or Conversion to a time-frequency domain through a spectrogram is performed, and it is possible to operate to extract features (trait variables) in each transformed domain. Further, according to an embodiment of the present application, vibration data converted into a frequency domain or a time-frequency domain by the data pre-processing unit 120 may be transmitted to the learning unit 130 described later. In other words, the learning unit 130 may perform deep learning-based time series analysis on vibration data converted into a frequency domain or a time-frequency domain by the data preprocessing unit 120.

FIG. 4 can be understood as a graph representing the acquired vibration data in the form of a spectrogram, and the spectrogram is a visualization tool for visualizing and grasping waves, and characteristics of a waveform and a spectrum. It is characterized by being able to confirm together. In these spectrograms, the amplitude change according to the change of the time axis (horizontal axis) and the frequency axis (vertical axis) can be confirmed as a difference in density or display color. In particular, the upper graph illustrated in FIG. 4 indicates a normal state (in other words, when there is no defect in the rotating equipment) and the lower graph represents an abnormal state (in other words, when a specific defect occurs in the rotating equipment), referring to FIG. 4. , Significant differences can be confirmed in the spectrum in the normal state and the spectrum in the abnormal state (for example, a state in which degradation has progressed).

The learning unit 130 may generate predicted values associated with vibration data through deep learning-based time series analysis on the extracted characteristic variables.

In addition, the learning unit 130 may generate error information based on the generated prediction value and the actual value of the vibration data. According to an embodiment of the present application, the learning unit 130 may generate error information for each extracted characteristic variable (eg, shape, amplitude, phase, etc. of vibration data).

According to one embodiment of the present application, the learning unit 130 may be to perform deep learning-based time series analysis based on at least one of an LSTM algorithm, an Attention algorithm, a Transformer algorithm, and a BERT algorithm.

5 is a diagram exemplarily showing prediction values generated through deep learning-based time series analysis according to an embodiment of the present application. Particularly, FIG. 5 shows prediction values generated as a result of time-series analysis based on deep learning performed based on the LSTM algorithm.

The orange graph of FIG. 5 shows the predicted value generated as a result of deep learning based time series analysis for a predetermined unit time, and the blue graph of FIG. 5 shows the actual value of vibration data during the unit time.

6 is a graph showing error information of predicted values and actual values generated through deep learning-based time series analysis according to an embodiment of the present application.

Referring to FIG. 6, the learning unit 130 may generate a prediction sequence that estimates a change in the vibration data during a predetermined unit time as a prediction value. Also, the learning unit 130 may generate an error vector corresponding to the difference between the actual sequence of the vibration data during the unit time and the generated prediction sequence as error information.

According to an embodiment of the present application, the preset unit time, which is a time range in which the learning unit 130 generates a predicted value and generates error information through comparison of the generated predicted value and the actual value, is a preset unit time of the received vibration data. It may be determined by properties. According to one embodiment of the present application, the learning unit 130, RMS analysis of the vibration data of the data pre-processing unit 120, center frequency (CF) analysis, variance (Var) analysis, intermediate frequency (IF) analysis, edition Unit time may be determined based on the results of at least one of the grape analysis, midwave (MF) analysis, kurtosis analysis, and PTP analysis.

In addition, according to an embodiment of the present application, the learning unit 130 predicts (prediction sequence) generated for a predetermined unit time through deep learning-based time series analysis and an actual value (actual sequence) acquired for the unit time ) Can be repeated in the direction of reducing the error. In other words, as the learning unit 130 repeatedly performs deep learning-based time series analysis, the size of the error vector that is the output of the learning unit 130 may be reduced.

Hereinafter, each type of the deep learning based time series analysis algorithm applied to the present application will be briefly described.

The Long-Short Term Memory (LSTM) algorithm is one of the artificial recursive neural network (RNN) architectures used in deep learning. Unlike the feed-forward neural network, there is a feedback connection. Therefore, according to the LSTM algorithm, learning and processing of not only a single data point but also an entire data sequence can be performed.

These LSTM algorithms are suitable for classifying, processing, and predicting predictions based on time series data, and LSTMs have the advantage of solving the vanishing gradient problems that can occur in training through traditional RNNs.

Attention algorithm is a technique that increases the performance of a model by making a deep learning model pay attention to a specific vector, and can be understood as an algorithm implemented to concentrate only the important part of the model in the learning process. In other words, the Attention algorithm allows the decoder to focus the different parts of the input sequence at each step when generating the output, and instead of encoding it as one fixed context vector, it creates a context vector at each step of the output. How to learn. This can be understood as determining what part of the model to focus on in the subsequent learning through the input sequence and the result generated so far, and determining the part to focus on can be performed through Softmax.

The Transformer algorithm may include an encoder-decoder structure that receives an input sequence from an encoder and outputs an output sequence from a decoder, like a conventional seq2seq architecture. However, unlike the conventional seq2seq architecture, a plurality of encoders and decoders may be provided. In addition, while the above-mentioned Attention algorithm determines which of the encoder's hidden states to focus on and utilizes in decoder calculation, the Transformer algorithm uses the Attention algorithm only internally without connecting the encoder and decoder, and is not an RNN such as LSTM or GRU. It has the characteristic that it consists of Attention Neural Network.

The BERT (Bidirectional Encoder Representations from Transformers) algorithm is an algorithm that trains the model in advance with large amounts of unlabeled data and then performs Transfer learning with labeled data that has a specific task. BERT's architecture uses some of the above-mentioned Transformer algorithms, but the pre-training and fine-tuning steps have different architectures to make transfer learning easier, and using only the Encoder part of the above-described Transformer algorithm structure. It is characterized by.

The margin setting unit 140 may set a margin for determining the presence or absence of a defect associated with vibration data based on the error information obtained by the learning unit 130. Here, the margin may function as a criterion for outputting an alarm, which will be described later, and may be differently referred to as an alarm margin, a defect classification margin, etc. in consideration of this. In addition, according to an embodiment of the present application, the margin setting unit 140, the error information for each of the characteristic variables (eg, shape, size, phase, etc. of vibration data) acquired by the learning unit 130 Based on this, the margin for each characteristic variable can be set individually. In addition, according to an embodiment of the present application, the margin setting unit 140 may determine a margin set to output a stepwise alarm according to the severity of a defect to a plurality of levels.

According to an embodiment of the present application, the margin set by the margin setting unit 140 may be determined in a sequence form for a unit time in consideration of time-series characteristics, but is not limited thereto, and predetermined characteristic variables For can be determined by a predetermined numerical value or numerical range.

In addition, according to one embodiment of the present application, defects in the rotating equipment may include a plurality of types such as misalignment, unbalance, crack, and abrasion, but are limited to this. It does not work. According to one embodiment of the present application, the margin setting unit 140 may be implemented to individually set a margin for each type of defects that may occur in a rotating facility. In addition, according to an embodiment, the defect detection unit 150 determines the type of the defect according to the characteristics of the new vibration data, based on the margin set for each defect type, and determines the type of the defect according to the characteristics of the new vibration data, 160), an alarm set individually for each defect type may be output. In other words, the alarm output from the alarm unit 160 may be determined differently according to the detected defect type of the rotating equipment.

For reference, misalignment among defect types may mean a defect that occurs when the center of rotation of the rotating equipment is inconsistent with the center of gravity of the rotating equipment, and Unbalance includes a plurality of rotating shafts of the rotating equipment. , When driven with respect to a plurality of rotating shafts, it may mean a defect that occurs when the axis centerline between the plurality of rotating shafts is not located on the same line, and cracks may be due to the long-term repetitive operation of the rotating equipment. It may mean a defect that occurs in the crack of the region. According to an embodiment of the present application, since characteristics for each characteristic variable of vibration data in a case where a defect occurs may be different for each type of defect described above, the learning unit 130 considers vibration data for each defect type To generate error information (for example, receiving vibration data when a predetermined type of defect occurs, generating error information using the received vibration data as an actual sequence, and optimizing the predicted sequence). , The margin setting unit 140 sets the margin for each defect type differently based on the error information generated for each defect type, so that the defect detection unit 150 not only determines the presence or absence of a defect from the new vibration data, but also the generated defect It can be implemented to understand the type of.

According to one embodiment of the present application, the margin setting unit 140 may set a margin based on at least one of a maximum likelihood estimation algorithm and a multivariate Gaussian distribution analysis. As another example, the margin setting unit 140 may be to estimate an optimal margin from an error vector that is a learning result of a time series model obtained by the learning unit 130 based on an arbitrary statistical technique.

According to one embodiment of the present application, the maximum likelihood estimation is an algorithm suitable for selecting an optimal margin based on data obtained by a method of finding a parameter having the greatest likelihood for a given data.

The defect detection unit 150 may receive new vibration data for the rotating equipment, and determine whether there is a defect in the new vibration data based on the set margin. It can be understood that the new vibration data received by the defect detection unit 150 is transmitted from the vibration sensor to the data reception unit 110 and then transmitted to the defect detection unit 150.

In other words, the defect detection unit 150 may determine the presence or absence of defects in the new vibration data based on the defect classification model generated to classify the presence or absence of defects in the new vibration data input in real time from the data receiving unit 110.

According to one embodiment of the present application, the defect detection unit 150 may determine the presence or absence of a defect through an algorithm based on supervised learning using the above-described error vector and a set margin as input. In addition, the defect detection unit 150 may determine whether or not a defect occurs by analyzing new vibration data in units of time through a supervised learning-based algorithm.

For reference, supervised learning may mean a method of learning in a state in which a label, which is an explicit correct answer information for data, is given. In the present application, various supervised learning algorithm models that have been known or developed in the future may be applied. According to an embodiment of the present application, the defect detection unit 150 receives, as input, an error vector and new vibration data generated through vibration data at a time before the new vibration data is received by the learning unit 130, and receives Based on the error vector and the new vibration data, it is possible to operate to predict a possibility of a defect occurring in the new vibration data or a possibility of a defect occurring after a new vibration data reception time point in consideration of a margin set by the margin setting unit 140.

The alarm unit 160 may output a preset alarm when it is determined that the new vibration data is defective as a result of the determination by the defect detection unit 150. That is, the vibration state monitoring apparatus 100 of the present application generates a alarm through the alarm unit 160 when vibration data of a specific time period is classified as a defect in the defect detection unit 150, so that a person in charge, such as a facility manager, can quickly It can act to be aware of the situation and take necessary actions.

Hereinafter, based on the details described above, the operation flow of the present application will be briefly described.

7 is an operation flowchart of a method for monitoring a vibration condition of a rotating facility using time-series analysis based on deep learning according to an embodiment of the present application.

The method for monitoring the vibration state of a rotating facility using time-series analysis based on deep learning shown in FIG. 7 may be performed by the vibration state monitoring device 100 described above. Therefore, even if omitted, the description of the vibration state monitoring apparatus 100 may be equally applied to a description of a vibration monitoring method for a rotating facility using deep learning based time series analysis.

Referring to FIG. 7, in step S710, the data receiving unit 110 may receive vibration data of the rotating equipment from the vibration sensor installed with respect to the rotating equipment.

Next, in step S730, the data pre-processing unit 120 may remove noise of the vibration data.

In addition, in step S720, the data pre-processing unit 120 may extract a characteristic variable from the vibration data from which noise has been removed.

Next, in step S730, the learning unit 130 may generate a predicted value associated with vibration data through deep learning based time series analysis on the extracted feature variables. In addition, in step S730, the learning unit 130 may generate error information based on the generated prediction value and the actual value of the vibration data.

Specifically, according to an embodiment of the present application, in step S730, the learning unit 130 may generate a prediction sequence for estimating a change in the vibration data during a predetermined unit time as a prediction value. In addition, in step S730, the learning unit 130 may generate an error vector corresponding to the difference between the actual sequence of the vibration data for a unit time and the generated prediction sequence as error information.

Next, in step S740, the margin setting unit 140 may set a margin for determining the presence or absence of a defect associated with vibration data based on the error information.

Next, in step S750, the defect detection unit 150 may receive new vibration data for the rotating equipment.

Specifically, according to an embodiment of the present application, in step S750, the defect detection unit 150 analyzes new vibration data in units of unit time through a supervised learning-based algorithm using an error vector and a set margin as inputs. It is possible to determine whether a defect has occurred.

Next, in step S760, the defect detection unit 150 may determine whether there is a defect in the new vibration data based on the set margin.

Next, in step S770, if the alarm unit 160 determines that the new vibration data is defective, it may output a preset alarm.

In the above description, steps S710 to S770 may be further divided into additional steps or combined into fewer steps, according to an embodiment of the present application. In addition, some steps may be omitted if necessary, and the order between the steps may be changed.

The method for monitoring the vibration state of a rotating facility using time-series analysis based on deep learning according to an embodiment of the present application may be implemented in a program command form that can be executed through various computer means and may be recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, or the like alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the present invention or may be known and usable by those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs, DVDs, and magnetic media such as floptical disks. -Hardware devices specially configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include high-level language code that can be executed by a computer using an interpreter, etc., as well as machine language codes produced by a compiler. The hardware device described above may be configured to operate as one or more software modules to perform the operation of the present invention, and vice versa.

In addition, the above-described method for monitoring the vibration state of a rotating facility using time-series analysis based on deep learning may be implemented in the form of a computer program or application executed by a computer stored in a recording medium.

The above description of the present application is for illustrative purposes, and those skilled in the art to which the present application pertains will understand that it is possible to easily modify to other specific forms without changing the technical spirit or essential features of the present application. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive. For example, each component described as a single type may be implemented in a distributed manner, and similarly, components described as distributed may be implemented in a combined form.

The scope of the present application is indicated by the claims, which will be described later, rather than the detailed description, and all modifications or variations derived from the meaning and scope of the claims and equivalent concepts should be interpreted to be included in the scope of the present application.

100: vibration monitoring device for rotating equipment using time-series analysis based on deep learning
110: data receiving unit
120: data pre-processing unit
130: learning department
140: margin setting unit
150: defect detection unit
160: alarm unit

Claims (12)

In a vibration monitoring system for rotating equipment using deep learning based time series analysis,
A data receiving unit that receives vibration data of the rotating equipment from a vibration sensor installed with respect to the rotating equipment;
Remove the noise of the vibration data, RMS analysis, center frequency (CF) analysis, variance (Var) analysis, intermediate frequency (IF) analysis, skewness analysis, medium wave (MF) from the vibration data from which the noise is removed ) Analysis, kurtosis analysis and data pre-processing unit for extracting one or more characteristic variables based on at least one of PTP analysis;
Based on the received vibration data through deep learning based time series analysis on the extracted characteristic variables, a predicted value, which is a prediction sequence that estimates a change in vibration data for a predetermined unit time after the vibration data is received, is obtained. A learning unit generating and generating error information based on the predicted value and the actual value of the vibration data;
A margin setting unit configured to set a margin for determining the presence or absence of a defect associated with the vibration data based on the error information;
A defect detection unit that receives new vibration data for the rotating equipment and determines whether the new vibration data is defective based on the margin; And
An alarm unit that outputs a preset alarm based on the determination result of the defect detection unit,
Including,
The learning unit,
An error vector represented by a sequence corresponding to the difference between the actual sequence of the vibration data and the prediction sequence during the unit time is generated as the error information,
The time series analysis is repeatedly performed so that the size of the error vector becomes small,
The unit time corresponds to the properties of the received vibration data, the RMS analysis, the center frequency (CF) analysis, the variance (Var) analysis, the intermediate frequency (IF) analysis, the skewness analysis, the It is determined based on an analysis result value for at least one of a medium wave (MF) analysis, the kurtosis analysis, and the PTP analysis,
The margin setting unit,
The margin is determined in the form of a sequence for the unit time to reflect the time series characteristics of the vibration data until the time point when the new vibration data is received, but considering the type of the characteristic variable, the margin for each of the characteristic variables is determined. Set individually and set the above margins separately for each type of defect that can occur in rotating equipment, including misalignment, unbalance, crack and abrasion. Accordingly, the margin is determined to a plurality of levels,
The alarm unit,
A vibration state monitoring device that outputs a different alarm according to the defect type of the defect identified as a result of the determination, and outputs a stepwise alarm based on the plurality of levels. delete According to claim 1,
The defect detection unit,
Vibration state monitoring device for determining the presence or absence of the defect through an algorithm based on supervised learning using the error vector and the margin as input. According to claim 3,
The defect detection unit,
Vibration state monitoring device for determining whether or not a defect occurs by analyzing the new vibration data in units of the unit time through the supervised learning-based algorithm. According to claim 1,
The learning unit,
Vibration state monitoring device for performing the deep learning based time series analysis based on at least one of LSTM algorithm, Attention algorithm, Transformer algorithm and BERT algorithm. delete According to claim 1,
The margin setting unit,
Vibration state monitoring device for setting the margin based on at least one of a maximum likelihood estimation algorithm and a multivariate Gaussian distribution analysis. According to claim 1,
The data pre-processing unit,
A low-pass filter or a band-pass filter is applied to the vibration data to remove the noise,
Vibration of the vibration data from which the noise is removed is transformed into data in the frequency domain by applying Fast Fourier Transform (FFT) or Short Time Fourier Transform (STFTTL) to the vibration data from which the noise has been removed. Status monitoring device. In the method of monitoring vibration condition of rotating equipment using deep learning based time series analysis,
Receiving vibration data of the rotating equipment from a vibration sensor installed with respect to the rotating equipment;
Remove the noise of the vibration data, RMS analysis, center frequency (CF) analysis, variance (Var) analysis, intermediate frequency (IF) analysis, skewness analysis, medium wave (MF) from the vibration data from which the noise is removed A) extracting one or more characteristic variables based on at least one of analysis, kurtosis analysis, and PTP analysis;
Based on the received vibration data through deep learning based time series analysis on the extracted characteristic variables, a predicted value, which is a prediction sequence that estimates a change in vibration data for a predetermined unit time after the vibration data is received, is obtained. Generating and generating error information based on the predicted value and the actual value of the vibration data;
Setting a margin for determining the presence or absence of a defect associated with the vibration data based on the error information;
Receiving new vibration data for the rotating equipment, and determining whether the new vibration data is defective based on the margin; And
Outputting a preset alarm based on the determination result,
And includes
Generating the error information,
Generating an error vector represented by a sequence corresponding to the difference between the actual sequence of the vibration data and the predicted sequence during the unit time as the error information,
Including,
Generating the error information,
The time series analysis is repeatedly performed so that the size of the error vector becomes small,
The unit time corresponds to the properties of the received vibration data, the RMS analysis, the center frequency (CF) analysis, the variance (Var) analysis, the intermediate frequency (IF) analysis, the skewness analysis, the It is determined based on an analysis result value for at least one of a medium wave (MF) analysis, the kurtosis analysis, and the PTP analysis,
The step of setting the margin,
The margin is determined in the form of a sequence for the unit time to reflect the time-series characteristic of the vibration data up to the point in time at which the new vibration data is received, but considering the type of the characteristic variable, the margin for each of the characteristic variables is determined. Set individually and set the margin separately for each type of defect that can occur in rotating equipment, including misalignment, unbalance, crack, and abrasion. Accordingly, the margin is determined to a plurality of levels,
The step of outputting the alarm,
According to the defect type of the detected defect as a result of the determination, a different alarm is output, and a stepwise alarm is output based on the plurality of levels. delete The method of claim 9,
Determining the presence or absence of the defect,
The vibration vector monitoring method is to determine whether or not a defect occurs by analyzing the new vibration data in units of the unit time through an algorithm based on supervised learning using the error vector and the margin as inputs. delete

Images