Multi-way valve element jamming fault diagnosis method and system based on acceleration sensor signals

Abstract

The invention provides a multi-way valve element jamming fault diagnosis method and system based on acceleration sensor signals, and the method comprises the following steps: S1, converting the valve element state of a multi-way valve element into a numerical value, and obtaining a valve element state coded value; s2, models are trained through acceleration sensor signals, the accuracy of valve element state coded values is diagnosed through the models, diagnosis results are obtained, and the models comprise a DT model, a PCA + SVM model and a CNN model; and S3, fusing diagnosis results of the DT model, the PCA + SVM model and the CNN model to obtain a diagnosis result of the comprehensive algorithm. By utilizing the characteristics of convenient installation and higher sampling rate of the acceleration sensor, the problem that a traditional sensor is inconvenient to install is solved, and the data analysis precision is improved; meanwhile, data diagnosis is carried out by fusing multiple algorithms, errors caused by a single algorithm are avoided, and therefore the accuracy and reliability of diagnosis are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-way valve fault diagnosis, and in particular to a multi-way valve spool stuck fault diagnosis method and system based on acceleration sensor signals. Background Art

[0002] Multi-way valves are crucial control components in hydraulic systems, widely used in engineering machinery, automated production lines, aerospace, metallurgy, mining, and other fields. Their primary function is to achieve various hydraulic system functions by regulating the direction, flow rate, and pressure of fluids. However, over time, multi-way valves can be affected by factors such as oil contamination, mechanical wear, and seal failure, leading to performance degradation or failure. Common failures include valve core sticking, valve core wear, and hydraulic oil contamination. These failures not only affect the normal operation of the equipment but can also cause serious consequences such as equipment damage and production halts.

[0003] Traditional multi-way valve fault diagnosis methods mainly rely on manual inspection, regular maintenance, oil analysis and pressure testing. Although they can detect faults to a certain extent, they have the following shortcomings:

[0004] (1) Fault detection has poor accuracy and timeliness. Diagnosis can often only be performed after a fault occurs, making it difficult to achieve real-time monitoring and predictive maintenance.

[0005] (2) Manual inspection relies on the experience of technicians and is easily affected by subjective factors, which may lead to missed or misdiagnosis and cannot effectively prevent the occurrence of potential faults;

[0006] With the advancement of sensor technology and data analysis, modern fault diagnosis methods are gradually moving towards intelligent and automated approaches. By installing pressure sensors, temperature sensors, flow sensors, and other devices on multi-way valves, the operating status of the multi-way valves can be monitored in real time. Signal processing and analysis of the collected data can help detect early signs of faults. Further incorporating machine learning algorithms, pattern recognition based on historical data can be performed, enabling early diagnosis and prediction of faults, thereby reducing the risk of failure, extending the life of the equipment, and reducing maintenance costs.

[0007] Although existing technologies have improved fault diagnosis capabilities to a certain extent, the following challenges still exist:

[0008] (1) Traditional sensor data is limited by the sampling rate, resulting in insufficient accuracy of data analysis and inconvenient installation;

[0009] (2) The modes of multi-way valve failure are diverse and complex, and the diagnostic accuracy and reliability need to be improved.

[0010] In the prior art, the Chinese invention patent application number CN202411074413.4 discloses an online diagnosis method for solenoid valve faults based on time-frequency domain characteristic analysis, which uses a machine learning (isolation forest) algorithm for fault diagnosis and evaluation. This method is simple and efficient, but has certain randomness problems and is not as robust as a deep learning algorithm.

[0011] Therefore, there is an urgent need for a more efficient, accurate and intelligent multi-way valve fault diagnosis method to address the shortcomings of existing technologies and improve the reliability and operating efficiency of hydraulic systems. Summary of the Invention

[0012] The present invention is made to solve the above-mentioned problem, and aims to provide a method and system for diagnosing a multi-way valve spool stuck fault based on acceleration sensor signals.

[0013] The present invention provides a multi-way valve spool stuck fault diagnosis method based on acceleration sensor signals, which has the following characteristics and includes the following steps: S1, converting the spool state of the multi-way valve spool into a numerical value, thereby obtaining a spool state coding value; S2, training a predetermined model through the acceleration sensor signal, diagnosing the accuracy of the spool state coding value through the predetermined model, and obtaining a diagnosis result, wherein the predetermined model includes a DT model, a PCA+SVM model and a CNN model; S3, fusing the diagnosis results of the DT model, the PCA+SVM model and the CNN model to obtain a diagnosis result of a comprehensive algorithm, thereby avoiding the diagnosis error caused by a single algorithm and improving the accuracy and reliability of the diagnosis.

[0014] The multi-way valve spool stuck fault diagnosis method based on acceleration sensor signals provided by the present invention may also have the following features: wherein, step S1 specifically includes the following sub-steps: encoding the valve spool stuck state according to the position of the multi-way valve spool relative to the valve cavity; when the valve spool is stuck, encoding the state as the first position value of the valve spool relative to the valve cavity at this time; when the valve spool is normal, programming the state as the second position value, which is much larger than the first position value.

[0015] The multi-way valve spool stuck fault diagnosis method based on acceleration sensor signals provided by the present invention may also have the following features: wherein, when the predetermined model is a DT model, step S2 specifically includes the following sub-steps: S2-1, dividing the acceleration signal used to train the DT model into a training set and a test set, then performing time domain and frequency domain feature extraction on the training set to obtain a training set feature vector group; S2-2, inputting the extracted training set feature vector group and the corresponding spool state code value of the training set into the C45 algorithm, and generating the corresponding DT model by the C45 algorithm;

[0016] S2-3, extract time domain and frequency domain features from the test set to obtain a test set feature vector group, diagnose the test set feature vector group through the DT model, compare the diagnosis results with the valve core state code value of the test set, and judge the diagnostic accuracy of the DT model; S2-4, if the accuracy of step S2-3 is not greater than 95%, repeat steps S2-1, S2-2 and S2-3; S2-5, if the accuracy of step S2-3 is greater than 95%, retain the DT model and its diagnostic accuracy.

[0017] The multi-way valve core stuck fault diagnosis method based on acceleration sensor signals provided by the present invention may also have the following features: wherein, when the predetermined model is a PCA+SVM model, step S2 specifically includes the following sub-steps: (1) dividing the acceleration signal used to train the PCA+SVM model into a training set and a test set, and then performing time domain and frequency domain feature extraction on the training set to obtain a training set feature vector group; (2) selecting appropriate parameters and performing dimensionality reduction processing on the training set feature vector group using the PCA algorithm to obtain reduced dimensionality data; (3) performing dimensionality reduction processing on the valve core state of the corresponding training set according to the reduced dimensionality data and the test set. The coding value is grid searched to find the optimal parameters of the SVM diagnostic model and generate the SVM model; (4) the time domain and frequency domain features of the test set are extracted to obtain the test set feature vector group, and then the PCA algorithm is used for dimensionality reduction. The diagnosis is performed through the SVM model, and the diagnosis results are compared with the valve core state coding value of the test set to determine the diagnostic accuracy of the SVM model; (5) if the accuracy of step (4) is not greater than 95%, the training steps (1)-(4) are repeated; (6) if the accuracy of step (4) is greater than 95%, the SVM model, PCA algorithm parameters and its diagnostic accuracy are retained.

[0018] The multi-way valve spool stuck fault diagnosis method based on acceleration sensor signals provided by the present invention may also have the following features: wherein, time domain and frequency domain feature extraction refers to extracting time domain features and frequency domain features from the acceleration signal and forming them into feature vectors,

[0019] Where Y(t) represents the collected acceleration signal, F(f) is the signal spectrum of Y(t), and the power spectrum P(f) of the signal frequency is:

[0020] P(f)=F 2 (f)

[0021] The calculation formulas for each time domain feature and frequency domain feature are as follows:

[0022] Time domain mean T mean for:

[0023]

[0024] Time domain peak T peakfor:

[0025] T peak =max|Y(t)|

[0026] Time domain median T med for:

[0027]

[0028] Time domain standard deviation T std for:

[0029]

[0030] Time domain RMS amplitude T sra for:

[0031]

[0032] The time domain root mean square is T rms :

[0033]

[0034] The time domain skewness is T skew :

[0035]

[0036] The time domain kurtosis is T kurt :

[0037]

[0038] Time domain crest factor T peakFac for:

[0039]

[0040] Time domain margin factor T marFac for:

[0041]

[0042] Time domain waveform factor T waveFac for:

[0043]

[0044] Frequency domain mean P mean for:

[0045]

[0046] Frequency domain center frequency P center for:

[0047]

[0048] Frequency domain root mean square P rms for:

[0049]

[0050] Frequency domain variance P var for:

[0051]

[0052] Frequency domain skewness P skew for:

[0053]

[0054] Frequency domain kurtosis P kurt for:

[0055]

[0056] The multi-way valve spool stuck fault diagnosis method based on acceleration sensor signals provided by the present invention may also have the following features: wherein, when the predetermined model is a CNN model, step S2 specifically includes the following sub-steps: I. dividing the acceleration signal used to train the CNN model into a training set and a test set, performing Gram angle field conversion on the training set to obtain Gram angle field data; II. inputting the Gram angle field data and the spool state code value of the corresponding training set into the CNN algorithm, and generating the corresponding CNN model by iterative algorithm; III. performing Gram angle field conversion on the test set to obtain feature data, performing diagnosis by the CNN model, comparing the diagnosis result with the spool state code value of the test set, and judging the diagnostic accuracy of the CNN model; IV. if the accuracy of step III is not greater than 95%, repeating steps I, II and III; V. if the accuracy of step III is greater than 95%, retaining the CNN model and its diagnostic accuracy.

[0057] The multi-way valve core stuck fault diagnosis method based on acceleration sensor signals provided by the present invention may also have the following features: wherein, step S3 specifically includes the following sub-steps: S3-1, using three predetermined models of DT model, PCA+SVM model and CNN model to diagnose the actual data and obtain diagnosis results respectively; S3-2, performing weighted voting on the diagnosis results to determine the final comprehensive diagnosis result.

[0058] The multi-way valve core stuck fault diagnosis method based on acceleration sensor signals provided by the present invention may also have the following features: wherein, step S3-1 specifically includes the following sub-steps: S3-1-1 inputs the original acceleration signal into the comprehensive algorithm, performs time domain and frequency domain feature extraction and Gram angle field conversion on the original acceleration signal, and obtains feature vectors and Gram angle field data respectively; S3-1-2, inputs the feature vectors into the DT model and PCA+SVM model respectively for diagnosis, and inputs the Gram angle field data into the CNN model for diagnosis.

[0059] The multi-way valve core stuck fault diagnosis method based on acceleration sensor signals provided by the present invention may also have the following features: wherein step S3-2 specifically includes the following sub-steps: S3-2-1, after the three algorithms output various diagnostic results, weighted voting is performed on each diagnostic result, and the weighted voting result V out for:

[0060]

[0061] Among them, V DT 、V PCA+SVM and V CNN They are the diagnosis results of DT model, PCA+SVM model and CNN model, Acc DT 、Acc PCA+SBM and Acc CNN They are the accuracy rates of DT model, PCA+SVM model and CNN model respectively; S3-2-2, the result of weighted voting V out The absolute difference is compared with each valve core state code value, and then the code value with the smallest difference is selected as the diagnosis result of the comprehensive algorithm.

[0062] The present invention also provides a multi-way valve spool stuck fault diagnosis system based on acceleration sensor signals, which has the following characteristics: a spool state coding module, which converts the spool state into a numerical value, thereby obtaining a spool state coding value; a model training module, which trains a predetermined model through the acceleration sensor signal, diagnoses the accuracy of the spool state coding value through the predetermined model, and obtains a diagnosis result, wherein the predetermined model includes a DT model, a PCA+SVM model, and a CNN model; and a comprehensive algorithm diagnosis module, which integrates the diagnosis results of the DT model, the PCA+SVM model, and the CNN model to obtain the diagnosis result of the comprehensive algorithm, thereby avoiding the diagnosis error caused by a single algorithm and improving the accuracy and reliability of the diagnosis.

[0063] Functions and effects of the invention

[0064] According to the present invention's method and system for diagnosing a stuck multi-valve spool based on acceleration sensor signals, the acceleration sensor can be mounted directly on the housing, eliminating the need for connection to the hydraulic system, as is required for flow sensors and pressure sensors. This overcomes the inconvenience of traditional sensor installation. Furthermore, the acceleration sensor's higher sampling rate improves the accuracy of data analysis. By integrating multiple algorithms for data diagnosis, errors introduced by a single algorithm can be avoided, thereby improving diagnostic accuracy and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 This is the time domain and frequency domain feature extraction process in an embodiment of the present invention;

[0066] Figure 2 This is the decision tree (DT) model training process in an embodiment of the present invention;

[0067] Figure 3 It is the support vector machine (PCA+SVM) model training process;

[0068] Figure 4 is the Convolutional Neural Network (CNN) model training process; and

[0069] Figure 5 This is a comprehensive algorithm diagnostic process of a multi-way valve core stuck fault diagnosis method based on acceleration sensor signals in an embodiment of the present invention. DETAILED DESCRIPTION

[0070] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the following embodiments, in conjunction with the accompanying drawings, specifically illustrate the multi-way valve spool stuck fault diagnosis method and system based on acceleration sensor signals of the present invention.

[0071] The method for diagnosing a stuck multi-way valve core fault based on acceleration sensor signals in this embodiment includes the following steps:

[0072] S1, converting the valve core state of the multi-way valve core into a numerical value, thereby obtaining the valve core state coding value.

[0073] The essence of a multi-way valve is a combination of multiple reversing valves. Each reversing valve itself contains a valve core and a valve cavity, and the valve core can move relative to the valve cavity. Therefore, the state of the valve core being stuck can be encoded according to the position of the multi-way valve valve core relative to the valve cavity. When the valve core is stuck, the state is encoded as the first position value of the valve core relative to the valve cavity at this time; when the valve core is normal, the state is encoded as the second position value, which is much larger than the first position value.

[0074] S2, trains a predetermined model through the acceleration sensor signal, diagnoses the accuracy of the valve core state coding value through the predetermined model, and obtains the diagnosis result. The predetermined model includes a decision tree model (DT), a support vector machine model (PCA+SVM) and a convolutional neural network model (CNN).

[0075] Figure 1 This is the time domain and frequency domain feature extraction process in an embodiment of the present invention.

[0076] like Figure 1 As shown in , time domain and frequency domain feature extraction refers to extracting time domain features and frequency domain features from the acceleration signal and forming them into feature vectors. The extracted time domain features and frequency domain features are shown in Table 1 below.

[0077] Table 1

[0078]

[0079] Where Y(t) represents the collected acceleration signal, F(f) is the signal spectrum of Y(t), and the power spectrum P(f) of the signal frequency is:

[0080] P(f)=F 2 (f)

[0081] The calculation formulas for each time domain feature and frequency domain feature are as follows:

[0082] Time domain mean T mean for:

[0083]

[0084] Time domain peak T peak for:

[0085] T peak =max|Y(t)|

[0086] Time domain median T med for:

[0087]

[0088] Time domain standard deviation T std for:

[0089]

[0090] Time domain RMS amplitude T sra for:

[0091]

[0092] The time domain root mean square is T rms :

[0093]

[0094] The time domain skewness is T skew :

[0095]

[0096] The time domain kurtosis is T kurt :

[0097]

[0098] Time domain crest factor T peakFac for:

[0099]

[0100] Time domain margin factor T marFac for:

[0101]

[0102] Time domain waveform factor T waveFac for:

[0103]

[0104] Frequency domain mean P mean for:

[0105]

[0106] Frequency domain center frequency P center for:

[0107]

[0108] Frequency domain root mean square P rms for:

[0109]

[0110] Frequency domain variance P var for:

[0111]

[0112] Frequency domain skewness P skew for:

[0113]

[0114] Frequency domain kurtosis P kurt for:

[0115]

[0116] Figure 2This is the decision tree (DT) model training process in an embodiment of the present invention.

[0117] like Figure 2 As shown in the figure, the process of training the DT model includes the following steps:

[0118] S2-1, the acceleration signal used to train the DT model is divided into a training set and a test set, and then the time domain and frequency domain features of the training set are extracted to obtain a training set feature vector group.

[0119] S2-2, the extracted training set feature vector group and the corresponding training set valve core state code value are input into the C45 algorithm, and the C45 algorithm generates the corresponding DT model.

[0120] S2-3, extract the time domain and frequency domain features of the test set to obtain the test set feature vector group, diagnose the test set feature vector group through the DT model, compare the diagnosis results with the valve core state code value of the test set, and judge the diagnosis accuracy of the DT model.

[0121] S2-4: If the accuracy of step S2-3 is not greater than 95%, repeat steps S2-1, S2-2 and S2-3.

[0122] S2-5, if the accuracy of step S2-3 is greater than 95%, the DT model and its diagnostic accuracy are retained.

[0123] Figure 3 It is the support vector machine (PCA+SVM) model training process.

[0124] like Figure 3 As shown in the figure, the process of training the PCA+SVM model includes the following steps:

[0125] (1) The acceleration signal used to train the PCA+SVM model is divided into a training set and a test set. The time domain and frequency domain features of the training set are then extracted to obtain a training set feature vector group.

[0126] (2) Select appropriate parameters and use the PCA algorithm to reduce the dimensionality of the training set feature vector group to obtain reduced dimensionality data.

[0127] (3) A grid search is performed based on the dimensionality reduction data and the valve core state code values of the corresponding training set to find the optimal parameters of the SVM diagnostic model and generate an SVM model.

[0128] (4) The test set is subjected to time domain and frequency domain feature extraction to obtain the test set feature vector group, which is then reduced in dimension by the PCA algorithm. The SVM model is used for diagnosis, and the diagnosis results are compared with the valve core state coding value of the test set to determine the diagnostic accuracy of the SVM model.

[0129] (5) If the accuracy of step (4) is not greater than 95%, repeat the training steps (1)-(4).

[0130] (6) If the accuracy of step (4) is greater than 95%, the SVM model, PCA algorithm parameters and their diagnostic accuracy are retained.

[0131] Figure 4 It is the convolutional neural network (CNN) model training process.

[0132] like Figure 4 As shown in the figure, the process of training the CNN model includes the following steps.

[0133] I. Divide the acceleration signal used to train the CNN model into a training set and a test set, and perform Gram angular field conversion on the training set to obtain Gram angular field data.

[0134] II. Input the Gram angle field data and the corresponding valve core state encoding value of the training set into the CNN algorithm, and the algorithm iteratively generates the corresponding CNN model.

[0135] III. The test set is converted into feature data through Gram angle field, and the CNN model is used for diagnosis. The diagnosis results are compared with the valve core state code values of the test set to determine the diagnosis accuracy of the CNN model.

[0136] IV. If the accuracy of step III is not greater than 95%, repeat steps I, II and III.

[0137] V. If the accuracy of step III is greater than 95%, retain the CNN model and its diagnostic accuracy.

[0138] S3, integrates the diagnostic results of the DT model, PCA+SVM model and CNN model to obtain the diagnostic results of the comprehensive algorithm, avoids the diagnostic errors caused by a single algorithm, and improves the accuracy and reliability of diagnosis.

[0139] Figure 5 This is a comprehensive algorithm diagnostic process of a multi-way valve core stuck fault diagnosis method based on acceleration sensor signals in an embodiment of the present invention.

[0140] like Figure 5 As shown, step S3 specifically includes the following sub-steps:

[0141] S3-1, use three predetermined models, DT model, PCA+SVM model and CNN model, to diagnose the actual data and obtain the diagnosis results respectively.

[0142] Step S3-1 specifically includes the following sub-steps:

[0143] S3-1-1 inputs the original acceleration signal into the integrated algorithm, performs time domain and frequency domain feature extraction and Gram angular field conversion on the original acceleration signal, and obtains the feature vector and Gram angular field data respectively.

[0144] S3-1-2, the feature vectors are respectively input into the DT model and PCA+SVM model for diagnosis, and the Gram angle field data is input into the CNN model for diagnosis.

[0145] S3-2, perform weighted voting on the diagnosis results to determine the final comprehensive diagnosis result.

[0146] Step S3-2 specifically includes the following sub-steps:

[0147] S3-2-1, after the three algorithms output various diagnostic results, each diagnostic result will be weighted voted, and the weighted voting result V out for:

[0148]

[0149] Among them, V DT 、V PCA+SVM and V CNN They are the diagnosis results of DT model, PCA+SVM model and CNN model, Acc DT 、Acc PCA+SBM and Acc CNN They are the accuracy rates of the DT model, PCA+SVM model and CNN model respectively.

[0150] S3-2-2, the weighted voting result V out The absolute difference is compared with each valve core state code value, and then the code value with the smallest difference is selected as the diagnosis result of the comprehensive algorithm.

[0151] The present invention also discloses a multi-way valve core stuck fault diagnosis system based on acceleration sensor signals, comprising a valve core state coding module, a model training module and a comprehensive algorithm diagnosis module.

[0152] The valve core state encoding module converts the valve core state into a numerical value, thereby obtaining a valve core state encoding value.

[0153] The model training module trains a predetermined model through the acceleration sensor signal, diagnoses the accuracy of the valve core state coding value through the predetermined model, and obtains the diagnosis result. The predetermined model includes DT model, PCA+SVM model and CNN model.

[0154] The comprehensive algorithm diagnosis module integrates the diagnostic results of the DT model, PCA+SVM model and CNN model to obtain the diagnostic results of the comprehensive algorithm, avoiding the diagnostic errors caused by a single algorithm and improving the accuracy and reliability of diagnosis.

[0155] Those skilled in the art will appreciate that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for diagnosing a stuck multi-way valve core fault based on acceleration sensor signals, characterized in that: The steps include: S1, converting the valve core state of the multi-way valve core into a numerical value, thereby obtaining the valve core state code value; S2, training a predetermined model using the acceleration sensor signal, and diagnosing the accuracy of the valve core state code value using the predetermined model to obtain a diagnosis result, wherein the predetermined model includes a DT model, a PCA+SVM model, and a CNN model; S3, fusing the diagnostic results of the DT model, the PCA+SVM model and the CNN model to obtain the diagnostic result of the comprehensive algorithm, thereby avoiding the diagnostic error caused by a single algorithm and improving the accuracy and reliability of the diagnosis.

2. The method for diagnosing a stuck multi-way valve core fault based on acceleration sensor signals according to claim 1, characterized in that: in, The step S1 specifically includes the following sub-steps: encoding the state of the valve core being stuck according to the position of the valve core of the multi-way valve relative to the valve cavity; when the valve core is stuck, encoding the state as the first position value of the valve core relative to the valve cavity at this time; when the valve core is normal, encoding the state as the second position value, which is much larger than the first position value.

3. The method for diagnosing a stuck multi-way valve core fault based on acceleration sensor signals according to claim 1, characterized in that: in, When the predetermined model is a DT model, step S2 specifically includes the following sub-steps: S2-1, dividing the acceleration signal used to train the DT model into a training set and a test set, and then performing time-domain and frequency-domain feature extraction on the training set to obtain a training set feature vector group; S2-2, inputting the extracted training set feature vector group and the corresponding training set valve core state code value into the C45 algorithm, and generating a corresponding DT model by the C45 algorithm; S2-3, performing time-domain and frequency-domain feature extraction on the test set to obtain a test set feature vector group, diagnosing the test set feature vector group using the DT model, comparing the diagnosis result with the valve core state code value of the test set, and determining the diagnosis accuracy of the DT model; S2-4, if the accuracy of step S2-3 is not greater than 95%, repeat steps S2-1, S2-2 and S2-3; S2-5: If the accuracy of step S2-3 is greater than 95%, retain the DT model and its diagnostic accuracy.

4. The method for diagnosing a stuck multi-way valve core fault based on acceleration sensor signals according to claim 1, characterized in that: in, When the predetermined model is a PCA+SVM model, step S2 specifically includes the following sub-steps: (1) dividing the acceleration signal used to train the PCA+SVM model into a training set and a test set, and then performing time-domain and frequency-domain feature extraction on the training set to obtain a training set feature vector group; (2) selecting appropriate parameters and performing dimensionality reduction processing on the training set feature vector group using the PCA algorithm to obtain dimensionality-reduced data; (3) performing a grid search based on the dimension-reduced data and the valve core state code values of the corresponding training set to find the optimal parameters of the SVM diagnostic model and generate an SVM model; (4) extracting time-domain and frequency-domain features from the test set to obtain a test set feature vector group, then performing dimensionality reduction using a PCA algorithm, performing diagnosis using the SVM model, and comparing the diagnosis results with the valve core state code value of the test set to determine the diagnostic accuracy of the SVM model; (5) If the accuracy of step (4) is not greater than 95%, repeat the training steps (1)-(4); (6) If the accuracy of step (4) is greater than 95%, the SVM model, PCA algorithm parameters and their diagnostic accuracy are retained.

5. The method for diagnosing a stuck multi-way valve core fault based on acceleration sensor signals according to claims 3-4, characterized in that: in, The time domain and frequency domain feature extraction refers to extracting time domain features and frequency domain features from the acceleration signal and forming them into feature vectors. Where Y(t) represents the collected acceleration signal, F(f) is the signal spectrum of Y(t), and the power spectrum P(f) of the signal frequency is: P(f)=F 2 (f) The calculation formulas for each time domain feature and frequency domain feature are as follows: Time domain mean T mean for: Time domain peak T peak for: T peak =max|Y(t)|time domain median T med for: Time domain standard deviation T std for: Time domain RMS amplitude T sra for: The time domain root mean square is T rms : The time domain skewness is T skew : The time domain kurtosis is T kurt : Time domain crest factor T peakFac for: Time domain margin factor T marFac for: Time domain waveform factor T waveFac for: Frequency domain mean P mean for: Frequency domain center frequency P center for: Frequency domain root mean square P rms for: Frequency domain variance P var for: Frequency domain skewness P skew for: Frequency domain kurtosis P kurt for:

6. The method for diagnosing a stuck multi-way valve core fault based on acceleration sensor signals according to claim 1, Its characteristics are: in, When the predetermined model is a CNN model, step S2 specifically includes the following sub-steps: I. Dividing the acceleration signal used to train the CNN model into a training set and a test set, performing Gram angle field conversion on the training set to obtain Gram angle field data; II. The Gram angle field data and the corresponding training set spool state code value are input into the CNN algorithm, and the algorithm iteratively generates the corresponding CNN model; III. The test set is converted into feature data by Gram angle field, and the CNN model is used for diagnosis. The diagnosis result is compared with the valve core state code value of the test set to determine the diagnostic accuracy of the CNN model; IV. If the accuracy of step III is not greater than 95%, repeat steps I, II, and III; V. If the accuracy of step III is greater than 95%, the CNN model and its diagnostic accuracy are retained.

7. The method for diagnosing a stuck multi-way valve core fault based on acceleration sensor signals according to claim 1, characterized in that: in, The step S3 specifically includes the following sub-steps: S3-1, use the DT model, PCA+SVM model and CNN model to diagnose the actual data and obtain the diagnosis results respectively; S3-2, performing weighted voting on the diagnosis results to determine the final comprehensive diagnosis result.

8. The method for diagnosing a stuck multi-way valve core fault based on acceleration sensor signals according to claim 7, characterized in that: in, The step S3-1 specifically includes the following sub-steps: S3-1-1 inputs the original acceleration signal into the integrated algorithm, performs time-domain and frequency-domain feature extraction and Gram angle field conversion on the original acceleration signal, and obtains feature vectors and Gram angle field data respectively; S3-1-2, input the feature vectors into the DT model and PCA+SVM model respectively for diagnosis, and input the Gram angle field data into the CNN model for diagnosis.

9. The method for diagnosing a stuck multi-way valve core fault based on acceleration sensor signals according to claim 7, characterized in that: in, The step S3-2 specifically includes the following sub-steps: S3-2-1, after the three algorithms output various diagnostic results, each of the diagnostic results will be weighted voted, and the weighted voting result V out for: Among them, V DT 、V PCA+SVM and V CNN They are the diagnosis results of DT model, PCA+SVM model and CNN model, Acc DT 、Acc PCA+SBM and Acc CNN They are the accuracy rates of the DT model, PCA+SVM model, and CNN model respectively; S3-2-2, the weighted voting result V out The absolute difference is compared with each valve core state code value, and then the code value with the smallest difference is selected as the diagnosis result of the comprehensive algorithm.

10. A multi-way valve core stuck fault diagnosis system based on acceleration sensor signals, characterized in that: include: The valve core state encoding module converts the valve core state into a numerical value, thereby obtaining the valve core state encoding value; A model training module, which trains a predetermined model through the acceleration sensor signal, diagnoses the accuracy of the valve core state code value through the predetermined model, and obtains a diagnosis result. The predetermined model includes a DT model, a PCA+SVM model, and a CNN model; as well as The comprehensive algorithm diagnosis module integrates the diagnosis results of the DT model, the PCA+SVM model and the CNN model to obtain the diagnosis result of the comprehensive algorithm, thereby avoiding the diagnostic errors caused by a single algorithm and improving the accuracy and reliability of the diagnosis.

Images