Pulse valve fault online diagnosis method and system based on machine learning

Abstract

The present application relates to the field of data processing technology, and specifically to a pulse valve fault online diagnosis method and system based on machine learning, the method comprising: collecting pulse valve vibration signals of each section under various working conditions, and each section of pulse valve vibration signals to be diagnosed; for each section of pulse valve vibration signals under various working conditions, decomposing the pulse valve vibration signals into each sub-signal, dividing each sub-signal into windows of each preset initial size, determining the adaptive size of each window based on the distribution characteristics of the amplitude in each window of each sub-signal in the time domain and the distribution characteristics of the amplitude of each sub-signal; determining each noise confidence interval and selection confidence of each sub-signal; determining the characteristic vector under any working condition; and performing fault diagnosis on the pulse valve vibration signal to be diagnosed based on the characteristic vectors under all working conditions. The present application aims to improve the accuracy of the fault diagnosis classification results of the pulse valve.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a pulse valve fault online diagnosis method and system based on machine learning. Background Art

[0002] Pulse valves play a vital role in industrial automation systems. Their working principle is based on electromagnetic control. The airflow through the throttle hole enters the rear air chamber, which in turn pushes the diaphragm to achieve rapid opening and closing of the valve. However, due to factors such as long-term operation or improper operation, pulse valves may have a variety of faults, including valve core obstruction, throttle hole obstruction or diaphragm damage. These faults not only affect system performance, but may also pose a threat to the stability of the entire production line.

[0003] When a pulse valve fails, abnormal vibration may occur. These abnormal vibration signals are key clues for fault diagnosis. By analyzing the vibration signal, the characteristics and type of the fault can be identified, so that corresponding maintenance or repair measures can be taken. The analysis of vibration signals is an effective means to realize pulse valve fault diagnosis, which is of great significance for ensuring the normal operation of industrial automation systems. However, traditional fault diagnosis methods usually only perform simple feature extraction on vibration signals, and the diagnostic results of different pulse valve faults are of low accuracy and cannot meet the needs of modern industry. Summary of the invention

[0004] In view of the above, it is necessary to provide a pulse valve fault online diagnosis method and system based on machine learning, which improves the accuracy of the pulse valve fault diagnosis classification results compared with the traditional pulse valve fault online diagnosis method and system:

[0005] In a first aspect, an embodiment of the present application provides a pulse valve fault online diagnosis method based on machine learning, the method comprising the following steps:

[0006] Collect vibration signals of each section of pulse valve to be diagnosed and vibration signals of each section of pulse valve under various working conditions;

[0007] For each pulse valve vibration signal under various working conditions, a signal decomposition algorithm is used to decompose the pulse valve vibration signal into sub-signals, and each sub-signal is divided into windows of preset initial sizes. Based on the distribution characteristics of the amplitude within each window of each sub-signal in the time domain and the distribution characteristics of the amplitude of each sub-signal, the adaptive size of each window is determined;

[0008] Determine each noise confidence interval of each sub-signal based on the change of the amplitude in all windows of each sub-signal;

[0009] Determine the selection confidence of each sub-signal by combining the amplitude fluctuation degree and energy distribution within all noise confidence intervals of each sub-signal in the frequency domain;

[0010] Determine a characteristic vector under any working condition based on the selection confidence of all pulse valve vibration signals and all sub-signals of all pulse valve vibration signals under any working condition;

[0011] Fault diagnosis is performed on the vibration signal of the pulse valve to be diagnosed based on the characteristic vectors under all working conditions.

[0012] In one embodiment, the process of determining the adaptive size is as follows:

[0013] Determining a window adjustment factor for each window based on fluctuations in the amplitude within each window and fluctuations in the amplitude of the sub-signal in which each window is located;

[0014] For each window of each sub-signal, when the mean value of the amplitude in the window is greater than the mean value of the amplitude of the sub-signal where the window is located, the size of the window is increased; when the mean value of the amplitude in the window is equal to the mean value of the amplitude of the sub-signal where the window is located, the size of the window remains unchanged; otherwise, the size of the window is reduced;

[0015] When the size of the window is increased, the adaptive size of the window is positively correlated with the preset initial size of the window and the window adjustment factor respectively;

[0016] When the size of the window is reduced, the adaptive size of the window is positively correlated with the preset initial size of the window and negatively correlated with the window adjustment factor.

[0017] In one embodiment, the window adjustment factor of each window is: a calculation result of dividing the standard deviation of the amplitude in each window by the standard deviation of the amplitude of the sub-signal where each window is located.

[0018] In one embodiment, the noise confidence interval is determined as follows:

[0019] Determine the noise selectivity index of each window of each sub-signal based on the change of the amplitude in each window of each sub-signal;

[0020] For each sub-signal, the average value of the noise selectivity indexes of all windows of the sub-signal is calculated, and the windows whose noise selectivity indexes are greater than the average value are taken as the noise confidence interval of the sub-signal.

[0021] In one embodiment, the noise selectivity index is determined as follows:

[0022] For each window of each sub-signal, the median of all amplitudes in the window is counted, the absolute value of the difference between each amplitude in the window and the median is calculated, and the sum of the absolute values ​​of the differences of all amplitudes in the window is calculated. The noise selection index of the window is positively correlated with the sum of the window and the adaptive size of the window, respectively.

[0023] In one embodiment, the process of determining the selection confidence is:

[0024] Determine the selection factor of each noise confidence interval by combining the amplitude fluctuation degree and energy distribution of each sub-signal in each noise confidence interval in the frequency domain;

[0025] The average value of the selection factors of all noise confidence intervals of each sub-signal is taken as the selection confidence of each sub-signal.

[0026] In one embodiment, the selection factor is determined by:

[0027] For each noise confidence interval of each sub-signal, the difference between the maximum amplitude and the minimum amplitude in the noise confidence interval in the frequency domain is calculated and recorded as the fault identification factor;

[0028] Calculate the mean of the energy at all frequencies in the noise confidence interval, record it as the energy mean; record the frequency in the noise confidence interval whose energy is less than the energy mean as the low energy frequency; record the mean of the energy at all the low energy frequencies in the noise confidence interval as the noise identification factor;

[0029] The selection factor of the noise confidence interval is positively correlated with the fault identification factor of the noise confidence interval, and negatively correlated with the noise identification factor of the noise confidence interval.

[0030] In one embodiment, the process of determining the feature vector is as follows:

[0031] For each pulse valve vibration signal under any working condition, the mean of the selected confidence of all sub-signals of the pulse valve vibration signal is calculated, recorded as the confidence mean, and all sub-signals with a selected confidence greater than the confidence mean are combined into a confidence set of the pulse valve vibration signal;

[0032] Perform independent component analysis on each sub-signal in the confidence set to extract a preset number of independent components;

[0033] All independent components of all sub-signals in the confidence set are combined into a sample component set of the pulse valve vibration signal;

[0034] Calculate the absolute value of the correlation coefficient between the pulse valve vibration signal and each independent component in its sample component set, and calculate the cumulative sum of all the absolute values ​​of the pulse valve vibration signal;

[0035] The cumulative sum of all pulse valve vibration signals under any working condition constitutes a characteristic vector under any working condition.

[0036] In one embodiment, the process of performing fault diagnosis on the pulse valve vibration signal to be diagnosed is:

[0037] The feature vectors under all working conditions are used as the input of the classification model to train the classification model;

[0038] For each pulse valve vibration signal to be diagnosed, a characteristic vector calculation method under any working condition is used to calculate the characteristic vector of the pulse valve vibration signal to be diagnosed;

[0039] The characteristic vector of the pulse valve vibration signal to be diagnosed is input into the trained classification model, and the classification diagnosis result of the pulse valve fault is output.

[0040] In the second aspect, an embodiment of the present application also provides an online diagnosis system for pulse valve faults based on machine learning, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, the steps of any one of the above-mentioned online diagnosis methods for pulse valve faults based on machine learning are implemented.

[0041] This application has at least the following beneficial effects:

[0042] The present application decomposes the pulse valve vibration signal under any working condition to obtain different sub-signals, calculates the selection confidence of the sub-signal according to the time domain and frequency domain characteristics of each sub-signal, and characterizes the prominence of the influence of the fault on the pulse valve vibration signal relative to the influence of noise on the pulse valve vibration signal;

[0043] Selecting sub-signals with smaller noise levels for subsequent analysis can reduce the impact of modal aliasing caused by noise, which is conducive to obtaining more accurate fault diagnosis results; performing independent component analysis on the selected sub-signals, extracting independent components in the sub-signals, and calculating the characteristic vector under any working condition based on the correlation coefficient between the pulse valve vibration signal and the independent components under any working condition, taking the correlation between the vibration signals of the same type of faults as the characteristics of the fault, and performing fault diagnosis based on the characteristics of the fault, can improve the accuracy of the fault diagnosis classification results of the pulse valve. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0045] Figure 1 A flowchart of the steps of a pulse valve fault online diagnosis method based on machine learning provided in one embodiment of the present application;

[0046] Figure 2Schematic diagram of empirical mode decomposition of original vibration signal;

[0047] Figure 3 Schematic diagram of independent component analysis for sub-signals;

[0048] Figure 4 Get a flow chart for the feature vector. DETAILED DESCRIPTION

[0049] In the description of the embodiments of the present application, words such as "exemplary", "or", "for example" and the like are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary", "or", "for example" and the like is intended to present related concepts in a concrete manner.

[0050] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art in the present application. The terms used in the specification of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. It should be understood that, unless otherwise specified, " / " means or.

[0051] It should also be noted that the terms "first" and "second" in the specification, claims and drawings of this application are used to distinguish similar objects, rather than to describe a specific order or sequence. The method disclosed in the embodiments of this application or the method shown in the flowchart includes one or more steps for implementing the method. Without departing from the scope of the claims, the execution order of multiple steps can be interchanged with each other, and some steps can also be deleted.

[0052] The specific scheme of the pulse valve fault online diagnosis method and system based on machine learning provided by the present application is described in detail below with reference to the accompanying drawings.

[0053] See also Figure 1 , which shows a flowchart of a pulse valve fault online diagnosis method based on machine learning provided by an embodiment of the present application, the method comprising the following steps:

[0054] Step 1: collecting vibration signals of each section of the pulse valve to be diagnosed and vibration signals of each section of the pulse valve under various working conditions.

[0055] A vibration sensor is used to collect the vibration signal of the pulse valve to be diagnosed, and the pulse valve vibration signals under different working conditions are collected respectively, including the pulse valve vibration signals during normal operation, valve core blocking, throttle hole blockage and diaphragm damage. n segments of pulse valve vibration signals to be diagnosed are collected, and n segments of pulse valve vibration signals are collected under each working condition.

[0056] In one embodiment of the present application, when collecting the pulse valve vibration signal, the sampling frequency is 4000 Hz, the sampling time is 1 min, and the value of n is 50. The implementer can set the sampling frequency, sampling time and the value of n according to the actual situation, and this application does not impose any special restrictions.

[0057] The collected pulse valve vibration signal is baseline corrected to eliminate the baseline offset caused by sensor drift or environmental changes.

[0058] In one embodiment of the present application, a wavelet transform method is used for baseline correction. As another implementation method, on the basis of being able to perform baseline correction on the pulse valve vibration signal, the implementer may use other existing technologies to perform baseline correction on the pulse valve vibration signal, and the present application does not impose any special restrictions.

[0059] Step 2, for each pulse valve vibration signal under various working conditions, a signal decomposition algorithm is used to decompose the pulse valve vibration signal into sub-signals, and each sub-signal is divided into windows of preset initial sizes. The adaptive size of each window is determined based on the distribution characteristics of the amplitude of each sub-signal in each window in the time domain and the distribution characteristics of the amplitude of each sub-signal.

[0060] As the service time increases, various components of the pulse valve may fail, such as valve core obstruction, throttle hole blockage and diaphragm damage. The vibration signal of the pulse valve contains rich information about its operation. By analyzing the vibration signal, the health of the pulse valve can be evaluated and fault diagnosis can be performed. Usually, when a pulse valve fails, the amplitude of the vibration signal changes significantly in the time domain, and the changes in amplitude caused by different faults are also different; in the frequency domain, the amplitude and energy contained in different faults are also different. However, the traditional method only extracts features from the time domain and frequency domain to perform fault diagnosis, and the accuracy of the diagnostic results is low.

[0061] Therefore, this application proposes an online diagnosis method for pulse valve faults based on machine learning. First, a signal decomposition algorithm is used to decompose the original vibration signal to obtain sub-signals, and then the selection confidence of each sub-signal is calculated. According to the selection confidence, the sub-signal that is more relevant to the vibration caused by the fault is selected, and then independent component analysis is performed on these sub-signals. Finally, the characteristic vector of the vibration signal is determined based on the independent components of the sub-signals, and the classification model is trained.

[0062] In one embodiment of the present application, the original vibration signal is decomposed into IMFs and a trend term by using the empirical mode decomposition method. As other implementations, the implementer may use other existing technologies to decompose the original vibration signal on the basis of being able to decompose the original vibration signal, and the present application does not impose any special restrictions. Figure 2 shown.

[0063] For each pulse valve vibration signal under each working condition, a signal decomposition algorithm is used to decompose the pulse valve vibration signal into various sub-signals.

[0064] The signal decomposition algorithm is a time series decomposition method that can decompose the complex pulse valve vibration signal into a series of sub-signals. Each sub-signal represents the vibration characteristics of the pulse valve vibration signal within a specific frequency band, which helps to identify and extract different vibration modes in the signal. However, the impact caused by the pulse valve failure will cause instantaneous signals to appear in the pulse valve vibration signal, resulting in modal aliasing problems, making adjacent sub-signals difficult to identify due to the influence of each other. This application calculates the selection confidence and selects the sub-signals that are more relevant to the vibration caused by the fault for independent component analysis (ICA), thereby eliminating modal aliasing.

[0065] Since there is some noise in the pulse valve vibration signal, and the noise will interfere with the analysis of the faulty pulse valve vibration signal, the noise selection index is calculated to characterize the size of the noise level in the sub-signal, and the sub-signal with a small noise level is selected for ICA, thereby reducing the impact of noise on the result. From the time domain, the noise is not periodic, so the overall analysis of the noise level of the sub-signal is not effective. The sub-signal can be divided into multiple windows, and the noise level of the sub-signal can be measured by the noise level in multiple windows. Each sub-signal is divided into windows of preset initial sizes, and the process of determining the adaptive size of the window is:

[0066] The result of dividing the standard deviation of the amplitude in each window by the standard deviation of the amplitude of the sub-signal where each window is located is recorded as the window adjustment factor of each window;

[0067] For each window of each sub-signal, when the mean value of the amplitude in the window is greater than the mean value of the amplitude of the sub-signal where the window is located, the size of the window is increased; when the mean value of the amplitude in the window is equal to the mean value of the amplitude of the sub-signal where the window is located, the size of the window remains unchanged; otherwise, the size of the window is reduced;

[0068] When the size of the window is increased, the adaptive size of the window is positively correlated with the preset initial size of the window and the window adjustment factor respectively;

[0069] When the size of the window is reduced, the adaptive size of the window is positively correlated with the preset initial size of the window and negatively correlated with the window adjustment factor.

[0070] It should be understood that positive correlation means that the changing trends of the dependent variable and the independent variable are in the same direction, such as a multiplication relationship, an addition relationship, etc., which can be set by the implementer according to the actual situation; negative correlation means that the changing trends of the dependent variable and the independent variable are in opposite directions, such as a subtraction relationship, a division relationship, etc., which can be set by the implementer according to the actual situation.

[0071] In one embodiment of the present application, when the size of the window is increased, the sum of 1 and the normalized value of the window adjustment factor of the window is calculated, and the product of the sum of the window and the preset initial size is rounded off and the calculation result is used as the adaptive size of the window; when the size of the window is reduced, the difference between 1 and the normalized value of the window adjustment factor of the window is calculated, and the product of the difference of the window and the preset initial size is rounded off and the calculation result is used as the adaptive size of the window.

[0072] In one embodiment of the present application, the preset initial size of the window is 10, and the value of the initial size is artificially preset and can be set by the implementer. This application does not impose any special restrictions.

[0073] It should be noted that the purpose of setting an adaptive window size is to reduce the impact of the noise-free segment on the overall noise level. When the amplitude mean in the window is greater than the amplitude mean of the entire sub-signal, it means that the amplitude change in the window is relatively large, which may be caused by a large noise impact or a fault. At this time, the window size should be increased to capture a wider range of signal features; on the contrary, the window size needs to be reduced to avoid the window from having too much impact on the overall situation. The degree of window change is determined by the ratio of the standard deviation of the amplitude in the window to the standard deviation of the amplitude of the entire sub-signal. The standard deviation reflects the degree of discreteness of the data. The larger the value of the window adjustment factor, the greater the difference between the window and the overall situation, so the greater the degree of adjustment.

[0074] Step 3: determine each noise confidence interval of each sub-signal based on the change of the amplitude in all windows of each sub-signal.

[0075] The noise selectivity index of each window of each sub-signal is determined based on the change of the amplitude in each window of each sub-signal. The specific process is:

[0076] For each window of each sub-signal, the median of all amplitudes in the window is counted, the absolute value of the difference between each amplitude in the window and the median is calculated, and the sum of the absolute values ​​of the differences of all amplitudes in the window is calculated. The noise selection index of the window is positively correlated with the sum of the window and the adaptive size of the window, respectively.

[0077] In one embodiment of the present application, a normalized value of the product of the adaptive size of the window and the sum value is used as the noise selection index of the window.

[0078] It should be noted that the noise selectivity index of a window characterizes the stability of the amplitude in the window. The larger the value of the noise selectivity index, the higher the noise level in the window, and the greater the impact on separating the fault component in the subsequent independent component analysis. Therefore, the possibility of selecting the window for independent component analysis is smaller.

[0079] Some noise signals are not obvious in the time domain, and it is impossible to eliminate the interference of noise by the stability of the pulse valve vibration signal in the time domain window. In addition, the low stability of the pulse valve vibration signal in the window may also be caused by the impact caused by the fault, and cannot be simply considered to be caused by noise. Therefore, in order to improve the accuracy of the noise level analysis in the window, it is also necessary to combine the frequency domain signal of the sub-signal where the window is located for analysis.

[0080] For each sub-signal, the average value of the noise selectivity indexes of all windows of the sub-signal is calculated, and the windows whose noise selectivity indexes are greater than the average value are taken as the noise confidence interval of the sub-signal.

[0081] Step 4: Determine the selection confidence of each sub-signal by combining the amplitude fluctuation degree and energy distribution within all noise confidence intervals of each sub-signal in the frequency domain.

[0082] Perform a fast Fourier transform on a sub-signal corresponding to each noise confidence interval to obtain the frequency domain signal of each noise confidence interval. The noise generated by the pulse valve when it is working is generally broadband noise, which is not concentrated on any specific frequency, but is distributed in the entire frequency band, and the amplitude has no obvious peak value; while the abnormal vibration caused by the fault often appears as an impact, which is manifested in the spectrum as some amplitudes with obvious peak values.

[0083] By analyzing the relationship between the impact of fault shock on the frequency domain of the pulse valve vibration signal and the impact of noise on the frequency domain of the pulse valve vibration signal, a sub-signal with less noise impact is selected.

[0084] Combined with the amplitude fluctuation degree and energy distribution of each noise confidence interval of each sub-signal in the frequency domain, the selection factor of each noise confidence interval is determined. The specific process is:

[0085] For each noise confidence interval of each sub-signal, the difference between the maximum amplitude and the minimum amplitude in the noise confidence interval in the frequency domain is calculated and recorded as the fault identification factor;

[0086] Calculate the mean of the energy at all frequencies in the noise confidence interval, record it as the energy mean; record the frequency in the noise confidence interval whose energy is less than the energy mean as the low energy frequency; record the mean of the energy at all the low energy frequencies in the noise confidence interval as the noise identification factor;

[0087] The selection factor of the noise confidence interval is positively correlated with the fault identification factor of the noise confidence interval, and negatively correlated with the noise identification factor of the noise confidence interval.

[0088] In one embodiment of the present application, the sum of the noise identification factor of the noise confidence interval and a tuning parameter constant preset to be greater than 0 is calculated and recorded as a first tuning parameter sum value, and the ratio of the fault identification factor of the noise confidence interval to the first tuning parameter sum value is used as a selection factor for the noise confidence interval; wherein, the purpose of the tuning parameter constant is to avoid the denominator being 0, and the value of the tuning parameter constant is preset artificially and can be set by the implementer, and this application does not impose any special restrictions.

[0089] It should be noted that: when the pulse valve vibration signal in the noise confidence interval has an impact caused by a fault, peak values ​​will appear in some amplitudes in the frequency domain of the pulse valve vibration signal. The fault identification factor characterizes the size of the peak value that appears. The larger the value of the fault identification factor, the greater the amplitude difference in the frequency domain, and the greater the possibility of a fault in the noise confidence interval. The noise identification factor measures the energy distribution of the window, selects frequencies with lower energy in the energy spectrum, which may mainly contain noise, and calculates the average energy of these frequencies as an estimate of the noise level. The larger the value, the greater the noise level. The larger the value of the selection factor, the greater the impact of the fault on the pulse valve vibration signal in this window than the impact of noise on the pulse valve vibration signal.

[0090] The average value of the selection factors of all noise confidence intervals of each sub-signal is taken as the selection confidence of each sub-signal. The selection confidence represents the prominence of the influence of the fault on the pulse valve vibration signal relative to the influence of the noise on the pulse valve vibration signal. The larger the value of the selection confidence, that is, the smaller the noise level, the greater the possibility of the sub-signal for subsequent independent component analysis.

[0091] Step 5: Determine a characteristic vector under any working condition based on the selection confidence of all pulse valve vibration signals and all sub-signals of all pulse valve vibration signals under any working condition.

[0092] For each pulse valve vibration signal under any working condition, the mean of the selected confidence of all sub-signals of the pulse valve vibration signal is calculated, recorded as the confidence mean, and all sub-signals with a selected confidence greater than the confidence mean are combined into a confidence set of the pulse valve vibration signal;

[0093] Perform independent component analysis on each sub-signal in the confidence set to extract a preset number of independent components;

[0094] All independent components of all sub-signals in the confidence set are combined into a sample component set of the pulse valve vibration signal;

[0095] Calculate the absolute value of the correlation coefficient between the pulse valve vibration signal and each independent component in its sample component set, and calculate the cumulative sum of all the absolute values ​​of the pulse valve vibration signal;

[0096] The cumulative sum of all pulse valve vibration signals under any working condition constitutes a characteristic vector under any working condition.

[0097] Among them, independent component analysis is a well-known technology, and the specific process is not repeated here. Figure 3 shown.

[0098] In an embodiment of the present application, the value of the preset number is 5. The value of the preset number is preset manually and the implementer can set it according to the actual situation. This application does not impose any special restrictions.

[0099] In one embodiment of the present application, the correlation coefficient may be the Pearson correlation coefficient. As other implementations, on the basis of the measurable correlation coefficient between the pulse valve vibration signal and the independent component, the implementer may adopt other existing technologies to measure the correlation coefficient between the pulse valve vibration signal and the independent component. This application does not impose any special restrictions.

[0100] It should be noted that the correlation coefficient can reflect the degree of correlation between the two. The independent component extracted based on the fault vibration signal has a higher correlation with the same type of fault vibration signal, but a lower correlation with other faults or normal vibration signals. Therefore, the correlation between the same type of fault vibration signals is used as a feature. The schematic diagram of the feature vector acquisition process is shown in Figure 4 shown.

[0101] Step 6: Perform fault diagnosis on the vibration signal of the pulse valve to be diagnosed based on the characteristic vectors under all working conditions.

[0102] The feature vectors under all working conditions are used as the input of the classification model to train the classification model;

[0103] For each pulse valve vibration signal to be diagnosed, a characteristic vector calculation method under any working condition is used to calculate the characteristic vector of the pulse valve vibration signal to be diagnosed;

[0104] The characteristic vector of the pulse valve vibration signal to be diagnosed is input into the trained classification model, and the classification diagnosis result of the pulse valve fault is output.

[0105] In one embodiment of the present application, the classification model is an SVM model, and the SVM model is trained using a one-to-one SVM multi-classification method. The input of the SVM model is a feature vector under all working conditions, the label is the fault category corresponding to the working condition, the loss function is a mean square error loss function, and the optimizer is an Adam optimizer. The training process of the SVM model is a well-known technology and the specific process will not be repeated here. As other implementation methods, on the basis of being able to classify pulse valve faults based on feature vectors under all working conditions and feature vectors of pulse valve vibration signals to be diagnosed, the implementer may choose other existing classification models, and the present application does not impose any special restrictions on this.

[0106] Based on the same inventive concept as the above method, an embodiment of the present application also provides a pulse valve fault online diagnosis system based on machine learning, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the steps of any one of the above-mentioned pulse valve fault online diagnosis methods based on machine learning are implemented.

[0107] In summary, the present application decomposes the pulse valve vibration signal under any working condition to obtain different sub-signals, calculates the selection confidence of the sub-signal according to the time domain and frequency domain characteristics of each sub-signal, and characterizes the prominence of the influence of the fault on the pulse valve vibration signal relative to the influence of noise on the pulse valve vibration signal;

[0108] Furthermore, selecting a sub-signal with a smaller noise level for subsequent analysis can reduce the impact of modal aliasing caused by noise, which is conducive to obtaining more accurate fault diagnosis results; performing independent component analysis on the selected sub-signal, extracting independent components in the sub-signal, and calculating the characteristic vector under any working condition based on the correlation coefficient between the pulse valve vibration signal and the independent component under any working condition, taking the correlation between the vibration signals of the same type of fault as the feature of the fault, and performing fault diagnosis based on the fault features, can improve the accuracy of the fault diagnosis classification results of the pulse valve.

[0109] The flowchart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to the embodiment of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. In some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, which can depend on the functions involved. In the description corresponding to the flowchart and the block diagram in the accompanying drawings, the operations or steps corresponding to different boxes can also occur in a different order from the order disclosed in the description, and sometimes there is no specific order between different operations or steps. For example, two consecutive operations or steps can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, which can depend on the functions involved. Each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented by a dedicated hardware-based system that performs the specified functions or actions, or may be implemented by a combination of dedicated hardware and computer instructions.

[0110] It is obvious to those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and that the present application can be implemented in other specific forms without departing from the basic features of the present application. Therefore, no matter from which point of view, the above embodiments of the present application should be regarded as exemplary and non-restrictive.

Claims

1. A pulse valve fault online diagnosis method based on machine learning, characterized in that: The method comprises the following steps: Collect vibration signals of each section of pulse valve to be diagnosed and vibration signals of each section of pulse valve under various working conditions; For each pulse valve vibration signal under various working conditions, a signal decomposition algorithm is used to decompose the pulse valve vibration signal into sub-signals, and each sub-signal is divided into windows of preset initial sizes. Based on the distribution characteristics of the amplitude within each window of each sub-signal in the time domain and the distribution characteristics of the amplitude of each sub-signal, the adaptive size of each window is determined; Determine each noise confidence interval of each sub-signal based on the change of the amplitude in all windows of each sub-signal; Determine the selection confidence of each sub-signal by combining the amplitude fluctuation degree and energy distribution within all noise confidence intervals of each sub-signal in the frequency domain; Determine a characteristic vector under any working condition based on the selection confidence of all pulse valve vibration signals and all sub-signals of all pulse valve vibration signals under any working condition; Fault diagnosis is performed on the vibration signal of the pulse valve to be diagnosed based on the characteristic vectors under all working conditions; The process of determining the selection confidence is as follows: Determine the selection factor of each noise confidence interval by combining the amplitude fluctuation degree and energy distribution of each sub-signal in each noise confidence interval in the frequency domain; The average value of the selection factors of all noise confidence intervals of each sub-signal is taken as the selection confidence of each sub-signal; The determination process of the selection factor is: For each noise confidence interval of each sub-signal, the difference between the maximum amplitude and the minimum amplitude in the noise confidence interval in the frequency domain is calculated and recorded as the fault identification factor; Calculate the mean of the energy at all frequencies in the noise confidence interval, record it as the energy mean; record the frequency in the noise confidence interval whose energy is less than the energy mean as the low energy frequency; record the mean of the energy at all the low energy frequencies in the noise confidence interval as the noise identification factor; The selection factor of the noise confidence interval is positively correlated with the fault identification factor of the noise confidence interval, and negatively correlated with the noise identification factor of the noise confidence interval; The process of performing fault diagnosis on the pulse valve vibration signal to be diagnosed is as follows: The feature vectors under all working conditions are used as the input of the classification model to train the classification model; For each pulse valve vibration signal to be diagnosed, a characteristic vector calculation method under any working condition is used to calculate the characteristic vector of the pulse valve vibration signal to be diagnosed; The characteristic vector of the pulse valve vibration signal to be diagnosed is input into the trained classification model, and the classification diagnosis result of the pulse valve fault is output.

2. The pulse valve fault online diagnosis method based on machine learning as claimed in claim 1 is characterized in that: The process of determining the adaptive size is as follows: Determine the window adjustment factor of each window based on the fluctuation of the amplitude in each window and the fluctuation of the amplitude of the sub-signal in which each window is located; For each window of each sub-signal, when the mean value of the amplitude in the window is greater than the mean value of the amplitude of the sub-signal where the window is located, the size of the window is increased; when the mean value of the amplitude in the window is equal to the mean value of the amplitude of the sub-signal where the window is located, the size of the window remains unchanged; otherwise, the size of the window is reduced; When the size of the window is increased, the adaptive size of the window is positively correlated with the preset initial size of the window and the window adjustment factor respectively; When the size of the window is reduced, the adaptive size of the window is positively correlated with the preset initial size of the window and negatively correlated with the window adjustment factor.

3. The pulse valve fault online diagnosis method based on machine learning as claimed in claim 2 is characterized in that: The window adjustment factor of each window is: the calculation result of dividing the standard deviation of the amplitude in each window by the standard deviation of the amplitude of the sub-signal where each window is located.

4. The pulse valve fault online diagnosis method based on machine learning as claimed in claim 1, characterized in that: The process of determining the noise confidence interval is: Determine the noise selectivity index of each window of each sub-signal based on the change of the amplitude in each window of each sub-signal; For each sub-signal, the average value of the noise selectivity indexes of all windows of the sub-signal is calculated, and the windows whose noise selectivity indexes are greater than the average value are taken as the noise confidence interval of the sub-signal.

5. The pulse valve fault online diagnosis method based on machine learning as claimed in claim 4 is characterized in that: The process of determining the noise selectivity index is as follows: For each window of each sub-signal, the median of all amplitudes in the window is counted, the absolute value of the difference between each amplitude in the window and the median is calculated, and the sum of the absolute values ​​of the differences of all amplitudes in the window is calculated. The noise selection index of the window is positively correlated with the sum of the window and the adaptive size of the window, respectively.

6. The pulse valve fault online diagnosis method based on machine learning as claimed in claim 1, characterized in that: The process of determining the feature vector is as follows: For each pulse valve vibration signal under any working condition, the mean of the selected confidence of all sub-signals of the pulse valve vibration signal is calculated, recorded as the confidence mean, and all sub-signals with a selected confidence greater than the confidence mean are combined into a confidence set of the pulse valve vibration signal; Perform independent component analysis on each sub-signal in the confidence set to extract a preset number of independent components; All independent components of all sub-signals in the confidence set are combined into a sample component set of the pulse valve vibration signal; Calculate the absolute value of the correlation coefficient between the pulse valve vibration signal and each independent component in its sample component set, and calculate the cumulative sum of all the absolute values ​​of the pulse valve vibration signal; The cumulative sum of all pulse valve vibration signals under any working condition constitutes a characteristic vector under any working condition.

7. A pulse valve fault online diagnosis system based on machine learning, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the pulse valve fault online diagnosis method based on machine learning as described in any one of claims 1-6 are implemented.

Images