US-20260126130-A1 - VALVE FAULT DETECTION METHOD AND APPARATUS

Abstract

A valve fault detection method includes: constructing a valve fault indicator system reflecting characteristics of vibration signals of valve states; establishing an LSTM-AE model by using an LSTM deep model and according to an autoencoder principle, and setting a fault discrimination threshold; inputting training data into the LSTM-AE model based on the valve fault indicator system, to train the LSTM-AE model; and inputting detection data of a to-be-detected valve into the trained LSTM-AE model based on the valve fault indicator system, and comparing an output value obtained by the LSTM-AE model with the set fault discrimination threshold, to judge whether the detected valve is faulty and outputting a first judgment result.

Inventors

• Xinmeng Wang
• Zongwen Wang
• Tao Li
• Minghua Sun

Assignees

• YANTAI JEREH OILFIELD SERVICES GROUP CO., LTD.

Dates

Publication Date

20260507

Application Date

20251229

Priority Date

20211210

Claims (20)

1. 1 . A method of detecting faults of valves in a pump, the method comprising: detecting vibration signals with one or more vibration sensors mounted on a first valve; calculating time-domain statistical indicators of time-domain characteristics of the vibration signals; calculating frequency-domain indicators of frequency spectrum difference characteristics of the vibration signals; storing the time-domain statistical indicators and the frequency-domain indicators in a database of valve fault indicators; and inputting the database of valve fault indicators of the first valve into a pre-trained long short-term memory network-aggregate expenditure (LSTM-AE) model, and comparing an output value obtained by the pre-trained LSTM-AE model with a fault discrimination threshold, to judge whether the first valve is faulty, and outputting a first judgment result indicating whether the first valve is faulty.
2. 2 . The method according to claim 1 , further comprising: performing sensitivity comparison on the time-domain statistical indicators to obtain valve fault sensitive statistical indicators; screening the frequency-domain indicators to select a frequency-domain characteristic indicator by calculating a frequency-domain energy sum and a frequency-domain energy sum ratio of the frequency-domain indicators of the vibration signals; and constructing the database of valve fault indicators based on the valve fault sensitive statistical indicators and the frequency-domain characteristic indicator.
3. 3 . The method according to claim 1 , further comprising: storing text data in an expert experience library, wherein the text data comprises a valve structure routine maintenance log, a device maintenance specification, an after-sales maintenance record, and/or a technician experience summary for the first valve; processing the text data in the expert experience library into a normalized text; judging whether maintenance for the first valve is due according to the expert experience library, and outputting a second judgment result; and obtaining, based on the first judgment result and the second judgment result, a final judgment result for judging whether the first valve is faulty.
4. 4 . The method according to claim 3 , further comprising: outputting the first judgment result or the second judgment result as the final judgment result if the first judgment result is the same as the second judgment result.
5. 5 . The method according to claim 3 , further comprising: outputting the first judgment result as the final judgment result if the first judgment result indicates that the first valve is normal and the second judgment result indicates that the first valve is faulty.
6. 6 . The method according to claim 3 , further comprising: in response to that the first judgment result indicates that the first valve is faulty and the second judgment result indicates that the first valve is normal, determining whether a fault of the first valve is detected continuously for a period of time.
7. 7 . The method according to claim 6 , further comprising: in response to determining that the fault is detected continuously for the period of time, judging that the first valve is faulty as the final judgment result.
8. 8 . The method according to claim 2 , wherein the valve fault sensitive statistical indicators comprise a kurtosis indicator, a root mean square value, a peak indicator, and a pulse indicator.
9. 9 . The method according to claim 2 , wherein the valve fault sensitive statistical indicators comprise a root mean square value, a peak indicator, a pulse indicator, and a skewness indicator.
10. 10 . The method according to claim 1 , wherein the calculating time-domain statistical indicators of time-domain characteristics of the vibration signals comprises: constructing the time-domain statistical indicators by calculating indicators comprising a mean value, an absolute mean value, a variance, a standard deviation, a square root amplitude, a root mean square value, a peak, a maximum value, a minimum value, a waveform indicator, a peak indicator, a pulse indicator, a margin indicator, a skewness indicator, and a kurtosis indicator for the time-domain characteristics of the vibration signals.
11. 11 . The method according to claim 1 , wherein the calculating frequency-domain indicators of frequency spectrum difference characteristics of the vibration signals comprises: performing Ensemble Empirical Mode Decomposition (EEMD) on the vibration signals, to obtain a plurality of intrinsic mode function (IMF) components; and performing fast Fourier transform (FFT) on the IMF components, to obtain component frequency-domain signals.
12. 12 . The method according to claim 11 , further comprising: calculating a frequency-domain energy sum and for each IMF component, a frequency-domain energy sum ratio of an energy sum of the IMF component to the frequency-domain energy sum; and ranking the component frequency-domain signals according to the frequency-domain energy sum ratio.
13. 13 . A computer device, comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, and the processor, when executing the computer program, performs operations comprising: detecting vibration signals with one or more vibration sensors mounted on a first valve; calculating time-domain statistical indicators of time-domain characteristics of the vibration signals; calculating frequency-domain indicators of frequency spectrum difference characteristics of the vibration signals; storing the time-domain statistical indicators and the frequency-domain indicators in a database of valve fault indicators; and inputting the database of valve fault indicators of the first valve into a pre-trained long short-term memory network-aggregate expenditure (LSTM-AE) model, and comparing an output value obtained by the pre-trained LSTM-AE model with a fault discrimination threshold, to judge whether the first valve is faulty, and outputting a first judgment result indicating whether the first valve is faulty.
14. 14 . The computer device according to claim 13 , wherein the operations further comprise: performing sensitivity comparison on the time-domain statistical indicators to obtain valve fault sensitive statistical indicators; screening the frequency-domain indicators to select a frequency-domain characteristic indicator by calculating a frequency-domain energy sum and a frequency-domain energy sum ratio of the frequency-domain indicators of the vibration signals; and constructing the database of valve fault indicators based on the valve fault sensitive statistical indicators and the frequency-domain characteristic indicator.
15. 15 . The computer device according to claim 13 , wherein the operations further comprise: storing text data in an expert experience library, wherein the text data comprises a valve structure routine maintenance log, a device maintenance specification, an after-sales maintenance record, and/or a technician experience summary for the first valve; processing the text data in the expert experience library into a normalized text; judging whether maintenance for the first valve is due according to the expert experience library, and outputting a second judgment result; and obtaining, based on the first judgment result and the second judgment result, a final judgment result for judging whether the first valve is faulty.
16. 16 . The computer device according to claim 15 , wherein the operations further comprising: outputting the first judgment result or the second judgment result as the final judgment result if the first judgment result is the same as the second judgment result.
17. 17 . The computer device according to claim 15 , wherein the operations further comprising: outputting the first judgment result as the final judgment result if the first judgment result indicates that the first valve is normal and the second judgment result indicates that the first valve is faulty.
18. 18 . The computer device according to claim 15 , wherein the operations further comprising: in response to that the first judgment result indicates that the first valve is faulty and the second judgment result indicates that the first valve is normal, determining whether a fault of the first valve is detected continuously for a period of time.
19. 19 . The computer device according to claim 18 , wherein the operations further comprising: in response to determining that the fault is detected continuously for the period of time, judging that the first valve is faulty as the final judgment result.
20. 20 . The computer device according to claim 14 , wherein the valve fault sensitive statistical indicators comprise a kurtosis indicator, a root mean square value, a peak indicator, and a pulse indicator.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS This application is a continuation application of U.S. patent application Ser. No. 18/179,928, filed Mar. 7, 2023, which is a continuation application of International Patent Application No. PCT/CN2022/090145 filed Apr. 29, 2022, which claims priority to Chinese Patent Application No. 202111507167.3, filed with China National Intellectual Property Administration on Dec. 10, 2021. The content of all of the above-referenced applications is incorporated herein by reference in their entirety. TECHNICAL FIELD This application relates to the field of fault detection, and more particularly, to a valve fault detection method. BACKGROUND Reciprocating pumps such as plunger pumps are usually complex in structure, have relatively large sources that generate vibration signals, and are relatively difficult to detect faults and to maintain. A suction valve and a discharge valve in a pump device are usually components that need to be maintained most. A valve fault is considered as a main reason for causing unexpected shutdown of the pump device. Currently, some existing technical researches on valve fault detection relate to construction of an algorithm model through a pressure signal, a temperature signal, a vibration signal, and the like to perform fault detection. Because of factors such as a high pressure environment in a pump body, it is relatively difficult to mount a sensor inside a cylinder. A sensor outside the cylinder usually captures a vibration signal, and an algorithm model is created to perform fault detection. Therefore, accuracy could be higher for such models than those using data captured by a sensor inside a cylinder. SUMMARY Technical Problems to be Resolved In the existing technical researches, a model for performing fault detection through a vibration signal mainly includes: setting a threshold based on a signal acquired by a vibration sensor to perform fault detection, performing fault detection with the aid of neural network structures such as naive Bayes classification, BP, a convolutional neural network (CNN), an autoencoder, and an Long Short-Term Memory (LSTM), and the like. The related researches are mostly based on laboratory environments, but are not applied to actual production operations. For construction of an algorithm model in the related researches, indicator systems are not considered comprehensively, and prediction methods are not diversified. As a result, problems such as low prediction precision exist. In the foregoing existing technology for detecting a fault of a valve such as a plunger pump, the following problems exist: 1. In the detection method constructed based on a neural network structure, when a model is inputted and an indicator system is constructed, consideration is mostly not given simultaneously from two aspects including time-domain characteristics and frequency-domain characteristics, and the indicator system is not constructed comprehensively enough.2. In most detections, time sequence fluctuation changes of characteristic indicators in a signal fault occurrence are not considered, and are not added to the signal characteristic indicator system.3. The detection is generally based on a single neural network model, and consideration is not given to a combination of a plurality of network structures to improve prediction precision.4. Many researches are based on laboratory environments, and there is a lack of device operation data of an on-site real environment. As a result, contribution factors are not considered comprehensively, and a model prediction result cannot be applied to an actual industrial production operation.5. There is a lack of a valve replacement log related to an on-site operation of a real device and practical expert experience such as an experience summary of technical backbones and experts to assist in judgment. To resolve the foregoing problems, this application provides a valve fault detection method. The method can establish a more complete and accurate signal characteristic indicator system, to improve prediction precision of a model, and the detection method can improve universality and accuracy of actual application of the model in consideration of signal characteristic time sequence fluctuation trend factors and in combination with two network structures such as an LSTM and an autoencoder. Technical Solutions To achieve the foregoing objectives, a valve fault detection method is provided. The method includes: constructing a valve fault indicator system reflecting characteristics of vibration signals of valve states; establishing an LSTM-AE model by using an LSTM deep model and according to an autoencoder principle, and setting a fault discrimination threshold; inputting training data into the LSTM-AE model based on the valve fault indicator system, to train the LSTM-AE model; and inputting detection data of a to-be-detected valve into the trained LSTM-AE model based on the valve fault indicator syst