A hydraulic pump fault prediction method and system

Abstract

The present invention discloses a hydraulic pump fault prediction method and system, which relates to the technical field of hydraulic system fault diagnosis and prediction, including collecting vibration sensor signals of hydraulic pumps, converting the vibration sensor signals into time domain vibration data according to a preset sampling frequency; performing wavelet packet decomposition on the time domain vibration data to obtain energy coefficients of different frequency bands, and selecting the frequency bands corresponding to the first three maximum energy coefficients as characteristic frequency bands; extracting characteristic indicators within the characteristic frequency bands, constructing fault feature vectors, training the fault feature vectors, and establishing a hydraulic pump fault prediction model; inputting the fault feature vectors of the hydraulic pump to be tested into the hydraulic pump fault prediction model to predict the fault type of the hydraulic pump. The present invention accurately predicts the fault type and degree of the hydraulic pump, effectively solving the challenges of data imbalance and nonlinear feature extraction faced in hydraulic pump fault diagnosis, thereby realizing early prediction and early warning of hydraulic pump faults.

Description

Technical Field

[0001] The invention relates to the technical field of hydraulic system fault diagnosis and prediction, and in particular to a hydraulic pump fault prediction method and system. Background Art

[0002] Hydraulic pumps are widely used in the industrial field, especially in the engineering machinery, aerospace, petrochemical and other industries, and they play an important role in power transmission and control. As the operation time of the equipment increases, hydraulic pumps often face failure problems caused by the use environment, workload, wear and other factors. Traditional hydraulic pump fault detection methods mainly rely on manual inspection, regular maintenance and experience-based fault diagnosis. However, these methods often have defects such as untimely diagnosis, low efficiency and difficulty in real-time monitoring, and cannot meet the high requirements for the safety and stability of hydraulic pump operation in modern industry. Therefore, the development of efficient and accurate hydraulic pump fault prediction methods has become a key direction of current technological development.

[0003] In recent years, with the rapid development of sensing technology, data acquisition technology and machine learning technology, fault diagnosis methods based on vibration signals have gradually become a hot topic in the research of hydraulic pump fault prediction. As a reflection of the operating status of the hydraulic pump, the vibration signal can effectively capture the abnormal conditions of the components in the pump. Although traditional fault diagnosis methods, such as time domain analysis and frequency domain analysis, can provide certain fault information in some occasions, their adaptability and accuracy are limited. Especially in complex working environments, the accuracy of traditional methods often cannot meet the requirements of real-time fault prediction. In recent years, advanced signal processing methods such as wavelet transform have been gradually applied to vibration signal analysis. By extracting multi-scale features, it can better reveal the subtle changes of hydraulic pump faults, thereby improving the accuracy and reliability of fault prediction.

[0004] However, although the wavelet analysis method based on vibration signals in the prior art has improved the accuracy of fault diagnosis to a certain extent, it still faces problems such as difficulty in feature extraction and selection, complex training models, and poor real-time performance. In addition, traditional methods often lack in-depth analysis of energy distribution in different frequency bands after feature extraction, resulting in the inability to effectively distinguish multiple types of faults, reducing the accuracy and generalization ability of fault prediction. Therefore, the prior art still has certain shortcomings in the practical application of hydraulic pump fault prediction, and new methods and systems are urgently needed to improve the accuracy and real-time performance of prediction. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention proposes a hydraulic pump fault prediction method and system.

[0006] Therefore, the present invention provides a hydraulic pump fault prediction method, which can solve the problems mentioned in the background technology.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, an embodiment of the present invention provides a hydraulic pump fault prediction method, which comprises:

[0009] Collecting vibration sensor signals of the hydraulic pump, and converting the vibration sensor signals into time domain vibration data according to a preset sampling frequency;

[0010] Performing wavelet packet decomposition on the time domain vibration data to obtain energy coefficients of different frequency bands, and selecting frequency bands corresponding to the first three maximum energy coefficients as characteristic frequency bands;

[0011] Extracting characteristic indicators within the characteristic frequency band, constructing a fault characteristic vector, training the fault characteristic vector, and establishing a hydraulic pump fault prediction model;

[0012] The fault feature vector of the hydraulic pump to be tested is input into the hydraulic pump fault prediction model to predict the fault type of the hydraulic pump.

[0013] As a preferred solution of the hydraulic pump fault prediction method of the present invention, the characteristic frequency band is obtained including:

[0014] Decomposing the time-domain vibration data into low-frequency and high-frequency parts, and obtaining sub-signals of multiple frequency bands by recursive decomposition;

[0015] Calculating the sub-signal energy coefficients of the multiple frequency bands respectively;

[0016] Arrange the sub-signal energy coefficients in descending order according to their numerical values, and select the frequency bands corresponding to the first three largest energy coefficients after sorting as preliminary characteristic frequency bands;

[0017] Using frequency band correlation evaluation indicators to further verify and optimize the effectiveness of the preliminary characteristic frequency bands;

[0018] The verified and optimized preliminary characteristic frequency bands are reconstructed to obtain the characteristic frequency bands.

[0019] As a preferred solution of the hydraulic pump fault prediction method of the present invention, the calculation of the sub-signal energy coefficient is shown in the following formula:

[0020] ,

[0021] in, is the frequency band weighting coefficient, which is determined by the following formula:

[0022] ,

[0023] in, is the center frequency of the ith frequency band, is the characteristic frequency of the hydraulic pump, is the bandwidth adjustment parameter, is the sub-signal energy coefficient, is the wavelet packet decomposition coefficient, To decompose the scale, is the translation parameter, is the frequency band number;

[0024] The verified and optimized preliminary characteristic frequency bands are reconstructed and expressed by the following formula:

[0025] ,

[0026] in, is the characteristic frequency band number, is the adaptive reconstruction coefficient, which is calculated by the following formula:

[0027] ,

[0028] in, To reconstruct the optimization parameters, is the energy coefficient of the first characteristic frequency band Power, is the energy coefficient of the second characteristic frequency band Power, is the energy coefficient of the third characteristic frequency band Power, For the The energy coefficient of each frequency band Power, is the wavelet packet basis function, To reconstruct the signal.

[0029] As a preferred solution of the hydraulic pump fault prediction method of the present invention, wherein: the characteristic index includes the vibration root mean square value, the peak factor and the waveform index;

[0030] The vibration root mean square value is calculated as follows:

[0031] ,

[0032] in, is the number of sampling points, Weighting coefficient for the time window:

[0033] ,

[0034] in, is the signal mean, is the standard deviation, For the The first characteristic frequency band reconstructs the signal The amplitude of the sampling points, For the The root mean square value of the vibration of the reconstructed signal of the characteristic frequency band.

[0035] As a preferred solution of the hydraulic pump fault prediction method of the present invention, the peak factor is calculated as shown in the following formula:

[0036] ,

[0037] in, is the correction factor:

[0038] ,

[0039] in, is the signal skewness, is an adjustment parameter with a value range of [0.5,1.5], For the The peak factor of the reconstructed signal in each characteristic frequency band;

[0040] The calculation of the waveform indicator is shown below:

[0041] ,

[0042] in, is the entropy correction factor:

[0043] ,

[0044] in, is the Shannon entropy of the signal:

[0045] ,

[0046] in, is the signal amplitude probability distribution, For the The reconstructed signal of the characteristic frequency bands, is the total number of sampling points.

[0047] As a preferred solution of the hydraulic pump fault prediction method of the present invention, the steps of constructing the hydraulic pump fault prediction model include:

[0048] Input the constructed fault feature vector into the pre-trained support vector machine (SVM) fault prediction model;

[0049] The classifier calculates the similarity between the fault feature vector and the support vector using a preset kernel function;

[0050] According to the kernel function result of the input feature, the decision function value is calculated by combining the weight, label and bias of the support vector;

[0051] Output The sign of determines the fault type of the hydraulic pump, and completes the construction of the hydraulic pump fault prediction model.

[0052] As a preferred solution of the hydraulic pump fault prediction method of the present invention, the kernel function adopts an adaptive weight combination, as shown in the following formula:

[0053] ,

[0054] in, is a Gaussian kernel, used to capture nonlinear features. is a polynomial kernel, used to capture multidimensional linear combination features, is the order of the polynomial, is the hyperbolic tangent kernel, which is used to simulate the activation characteristics of neural networks. is the kernel function, , , is the kernel function weight coefficient, satisfying:

[0055] .

[0056] As a preferred solution of the hydraulic pump fault prediction method of the present invention, the determination of the fault type includes:

[0057] According to the output The symbol of If the value of is positive, it means that the hydraulic pump corresponding to the fault feature vector is in normal working condition and the fault type is "normal"; if If the value of is negative, it means that the hydraulic pump has a certain fault, and the fault type is the corresponding fault mode;

[0058] The failure mode is determined as follows:

[0059] For each fault category , calculate its decision function value ,if A negative value indicates that the input feature vector is related to this failure mode, and the one with the smallest negative value is selected. The corresponding fault category is used as the final output fault type.

[0060] In a second aspect, an embodiment of the present invention provides a hydraulic pump fault prediction system, which includes:

[0061] A data acquisition module, used to collect vibration sensor signals of the hydraulic pump, and convert the vibration sensor signals into time domain vibration data according to a preset sampling frequency;

[0062] A feature construction module is used to perform wavelet packet decomposition on the time domain vibration data to obtain energy coefficients of different frequency bands, select frequency bands corresponding to the first three maximum energy coefficients as feature frequency bands, extract feature indicators within the feature frequency bands, and construct a fault feature vector;

[0063] A model building module, used for training the fault feature vector and establishing a hydraulic pump fault prediction model;

[0064] The fault prediction module is used to input the fault feature vector of the hydraulic pump to be tested into the hydraulic pump fault prediction model to predict the fault type of the hydraulic pump.

[0065] The beneficial effect of the present invention is that the accuracy and real-time performance of hydraulic pump fault diagnosis are significantly improved by comprehensively applying vibration signal acquisition, wavelet packet decomposition, feature extraction and machine learning algorithms. First, the high-precision vibration sensor and high-sampling rate data acquisition card are used to ensure the high-quality acquisition of vibration data, providing accurate time-domain vibration data for subsequent analysis. Secondly, the characteristic frequency band is selected by wavelet packet decomposition and energy coefficient, and the key information in the hydraulic pump vibration signal is effectively extracted. In addition, the frequency band correlation evaluation index R is introduced to optimize the selection of characteristic frequency bands, improve the complementarity of frequency band information, and avoid the interference of redundant features. Furthermore, by introducing adaptive reconstruction coefficients and improved time-domain feature extraction methods, the various features of the vibration signal are optimized and calculated to ensure the accuracy of the fault feature vector. Finally, the support vector machine (SVM) is combined with the adaptive kernel function and sample equalization technology to accurately predict the fault type and degree of the hydraulic pump, effectively solving the challenges of data imbalance and nonlinear feature extraction faced in the fault diagnosis of the hydraulic pump, thereby realizing the early prediction and early warning of the hydraulic pump fault. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0067] Figure 1 The figure is a flow chart of a hydraulic pump fault prediction method. DETAILED DESCRIPTION

[0068] Example 1, reference Figure 1, which is the first embodiment of the present invention, and provides a hydraulic pump fault prediction method, comprising:

[0069] S1: collecting vibration sensor signals of the hydraulic pump, and converting the vibration sensor signals into time domain vibration data according to a preset sampling frequency;

[0070] High-precision piezoelectric vibration sensors are installed at the drive shaft end, pump body and outlet end of the hydraulic pump respectively. The sensitivity of the vibration sensor is 100mV / g and the measurement range is 0.5Hz-10kHz. The vibration sensor converts the mechanical vibration generated by the hydraulic pump during operation into an electrical signal, and the analog electrical signal output by the vibration sensor is collected by a data acquisition card. The sampling frequency of the data acquisition card is set to 20kHz, the sampling accuracy is 16 bits, and the sampling time is 60s. The collected analog electrical signal is converted into a digital signal through A / D to obtain the time domain vibration data of the hydraulic pump; wherein the time domain vibration data includes vibration displacement, velocity and acceleration information.

[0071] S2: performing wavelet packet decomposition on the time domain vibration data to obtain energy coefficients of different frequency bands, and selecting frequency bands corresponding to the first three maximum energy coefficients as characteristic frequency bands;

[0072] The db4 wavelet basis is used to perform a three-layer wavelet packet decomposition on the time domain vibration data. The time domain vibration data is recorded as , the wavelet packet decomposition coefficient calculation formula is:

[0073] ,

[0074] in, To decompose the scale, is the translation parameter, is the frequency band number, is the wavelet packet basis function.

[0075] The decomposition process first decomposes the original signal into low-frequency and high-frequency parts, and finally obtains 8 frequency bands of sub-signals through recursive decomposition. The 8 frequency bands of sub-signals are obtained through wavelet packet decomposition, and the energy coefficient of the sub-signal of the i-th frequency band is The calculation of adopts the improved normalization method:

[0076] ,

[0077] in, is the frequency band weighting coefficient, which is determined by the following formula:

[0078] ,

[0079] in, is the center frequency of the ith frequency band, is the characteristic frequency of the hydraulic pump, The purpose of introducing the frequency band weighting coefficient is to highlight the frequency band close to the characteristic frequency of the hydraulic pump and weaken the influence of the frequency band far away from the characteristic frequency.

[0080] The energy coefficients of the eight frequency bands { } Sort by value in descending order, and select the top three largest energy coefficients after sorting , , The corresponding frequency bands are used as characteristic frequency bands, and the frequency ranges of the characteristic frequency bands are 0-2.5kHz, 2.5-5kHz and 5-7.5kHz respectively. The screening criteria include arranging the energy coefficients in descending order, selecting the frequency bands corresponding to the first three largest energy coefficients, and determining their frequency ranges. In this step, the frequency bands that are most representative of the vibration characteristics of the hydraulic pump are preliminarily screened out through the size and distribution of the energy coefficients.

[0081] Furthermore, the frequency band correlation evaluation index is introduced Further verify and optimize the effectiveness of the selected feature frequency bands:

[0082] ,

[0083] when When it is less than the preset threshold of 0.85, it is considered that the selected characteristic frequency bands have good complementarity. Specifically, after selecting the three frequency bands of 0-2.5kHz, 2.5-5kHz and 5-7.5kHz by the energy coefficient, the relationship between these frequency bands needs to be evaluated. The index reflects the degree of complementarity of information between different frequency bands by calculating the product and normalization of the energy coefficients of adjacent characteristic frequency bands. The value means that the vibration characteristic information contained in these frequency bands is relatively independent, which can reflect the operating status of the hydraulic pump from different angles and avoid the redundancy of frequency band information.

[0084] Finally, the sub-signal of the characteristic frequency band is reconstructed to reconstruct the signal The calculation formula is:

[0085] ,

[0086] in, is the characteristic frequency band number, is the adaptive reconstruction coefficient, which is calculated by the following formula:

[0087] ,

[0088] in, is the reconstruction optimization parameter, the value range is [1,2], is the energy coefficient of the first characteristic frequency band Power, is the energy coefficient of the second characteristic frequency band Power, is the energy coefficient of the third characteristic frequency band Power, is the energy coefficient of the mth frequency band Power.

[0089] The reconstruction method selects the sub-signal of the characteristic frequency band and introduces the adaptive reconstruction coefficient , the reconstruction coefficient is weighted exponentially according to the energy of each frequency band. The calculation of the reconstruction coefficient takes into account The parameter (range [1,2]) can adjust the weight ratio of different frequency bands. This reconstruction process not only integrates the characteristic frequency bands previously screened by the energy coefficient and R index, but also gives each frequency band a differentiated weight, so that the reconstructed signal can more accurately reflect the vibration characteristics of the hydraulic pump.

[0090] S3: extracting the vibration root mean square value, peak factor and waveform index within the characteristic frequency band to construct a fault feature vector;

[0091] The time domain statistical features are extracted from the reconstructed signal of the characteristic frequency band. For each reconstructed signal of the characteristic frequency band m, , calculate the improved time domain characteristic index. Including vibration root mean square value, peak factor and waveform index, among which the root mean square value Adopt adaptive segment weighted calculation method:

[0092] ,

[0093] in, is the number of sampling points, Weighting coefficient for the time window:

[0094] ,

[0095] in, is the signal mean, is the standard deviation, For the The first characteristic frequency band reconstructs the signal The amplitude of the sampling points.

[0096] The introduction of the adaptive segmented weighting method is to calculate the RMS value more accurately. The traditional RMS calculation treats all sampling points equally, while this method introduces an exponentially decaying weighting coefficient. , so that the points that deviate more from the signal mean get higher weights. This means that for data points far away from the average level of the signal, their impact on The contribution of is more significant, so it can more sensitively reflect the fluctuation characteristics of the signal.

[0097] The crest factor Optimize calculation by peak-to-valley ratio:

[0098] ,

[0099] in, is the correction factor:

[0100] ,

[0101] in, is the signal skewness, The peak factor is no longer a simple ratio of the peak value to the RMS value, but a correction factor based on the signal skewness is introduced. .pass In the form of, when the signal skewness is larger, the correction coefficient is larger, which means that for signals with asymmetrical distribution, the calculation of the peak factor will give more weight.

[0102] The Waveform Indicators Introducing entropy correction:

[0103] ,

[0104] in, is the entropy correction factor:

[0105] ,

[0106] in, is the Shannon entropy of the signal:

[0107] ,

[0108] in, is the probability distribution of signal amplitude. Shannon entropy It reflects the randomness and complexity of the signal amplitude distribution. The larger the entropy value, the more irregular the signal. The entropy value is normalized and converted into a correction factor in the form of The algorithm will adjust accordingly to better capture the complexity and abnormal characteristics of the signal.

[0109] The nine time domain characteristic indicators of the three characteristic frequency bands are combined into a fault feature vector :

[0110] ,

[0111] in, , , is the RMS value of the reconstructed signal for the 1st, 2nd and 3rd characteristic frequency bands, , , The peak factors of the reconstructed signals for the 1st, 2nd and 3rd characteristic frequency bands, , , Waveform indicators of the reconstructed signals for the 1st, 2nd and 3rd characteristic frequency bands.

[0112] S4: using a support vector machine to train the fault feature vector to establish a hydraulic pump fault prediction model, wherein the training samples include feature vectors under normal working conditions and typical fault working conditions;

[0113] The training sample set contains feature vectors collected under five working conditions of the hydraulic pump: normal working condition, bearing failure, blade erosion, cavitation and seal leakage, and each working condition contains 200 sets of sample data. The selection of samples fully considers the main types of failures that may occur in the hydraulic pump to ensure that the model has wide applicability.

[0114] The decision function of the multi-core fusion support vector machine is:

[0115] ,

[0116] in, is the Lagrange multiplier, is the sample label, is the kernel function, is the bias term.

[0117] The hybrid kernel function Adopt adaptive weight combination:

[0118] ,

[0119] in, is a Gaussian kernel, used to capture nonlinear features. is a polynomial kernel, used to capture multidimensional linear combination features, is the order of the polynomial, is the hyperbolic tangent kernel, which is used to simulate the activation characteristics of neural networks. , , is the kernel function weight coefficient, satisfying:

[0120] ,

[0121] The kernel function weight coefficient is solved by the particle swarm optimization algorithm. The particle swarm optimization algorithm simulates the evolution of swarm intelligence and searches for the optimal kernel function combination by continuously updating the weights of each kernel function. The iterative process of the algorithm takes into account the current optimal solution and the global optimal solution, so that the kernel function weight can be adjusted adaptively. The optimization process can be expressed as follows:

[0122] ,

[0123] ,

[0124] in, For the The weight of the kernel function is The value after iterations, is the weight update speed, is the inertia weight, , is the learning factor, , is a random number, The particle itself is The optimal position in the iteration, For all particles in The global optimal position in iterations.

[0125] Among them, the fitness function J is defined as:

[0126] ,

[0127] in, is the classification accuracy, is the area under the ROC curve, is the mean square error, , , is the weight coefficient.

[0128] Since hydraulic pump fault data usually have the problem of class imbalance, a sample balancing processing mechanism is introduced to perform SMOTE oversampling on minority class samples. The SMOTE method generates new minority class samples through interpolation, which effectively alleviates the deviation caused by imbalanced training data.

[0129] S5: Inputting the fault feature vector of the hydraulic pump to be tested into the hydraulic pump fault prediction model, and outputting the fault type and fault degree of the hydraulic pump.

[0130] The constructed fault feature vector F is input into the pre-trained support vector machine (SVM) fault prediction model. The classifier uses the preset kernel function to calculate the similarity between the input feature vector and the support vector. SVM calculates the decision function value based on the kernel function result of the input feature, combined with the weight, label and bias of the support vector, and outputs The sign of the pump determines the type of fault (such as normal or a certain failure mode) of the hydraulic pump.

[0131] According to the output The symbol of If the value of is positive, it means that the hydraulic pump corresponding to the input feature vector is in normal working condition and the fault type is "normal"; if If the value of is negative, it means that some kind of fault has occurred in the hydraulic pump, and the fault type is the corresponding specific fault mode.

[0132] For each fault category , calculate its decision function value ,if A negative value indicates that the input feature vector is related to this failure mode, and the one with the smallest negative value is selected. The corresponding fault category is used as the final output fault type.

[0133] Furthermore, this embodiment also provides a hydraulic pump fault prediction system, comprising:

[0134] A data acquisition module, used to collect vibration sensor signals of the hydraulic pump, and convert the vibration sensor signals into time domain vibration data according to a preset sampling frequency;

[0135] A feature construction module is used to perform wavelet packet decomposition on the time domain vibration data to obtain energy coefficients of different frequency bands, select frequency bands corresponding to the first three maximum energy coefficients as feature frequency bands, extract feature indicators within the feature frequency bands, and construct a fault feature vector;

[0136] A model building module, used for training the fault feature vector and establishing a hydraulic pump fault prediction model;

[0137] The fault prediction module is used to input the fault feature vector of the hydraulic pump to be tested into the hydraulic pump fault prediction model to predict the fault type of the hydraulic pump.

[0138] This embodiment also provides a computer device suitable for the case of a hydraulic pump fault prediction method, comprising a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the hydraulic pump fault prediction method proposed in the above embodiment.

[0139] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0140] This embodiment further provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for predicting hydraulic pump faults as proposed in the above embodiment is implemented.

[0141] The storage medium proposed in this embodiment and the data storage method proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be found in the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0142] Embodiment 2 is the second embodiment of the present invention. This embodiment provides a hydraulic pump fault prediction method. In order to verify the beneficial effects of the present invention, a scientific demonstration is carried out through simulation experiments.

[0143] The vibration data of hydraulic pumps under four working conditions, including normal working condition, wear failure, cavitation failure and leakage failure, were selected from a construction machinery construction site. 200 groups of samples were collected for each working condition, with a total of 800 groups of data. The sampling frequency was set to 10 kHz, and the sampling time for each group was 1 second. The training set and the test set were randomly divided into training set and test set with a ratio of 7:3.

[0144] By comparing the method of the present invention, traditional wavelet transform, conventional feature extraction, single kernel SVM and empirical mode decomposition (EMD), conventional feature extraction and BP neural network methods, the present invention is referred to as method A, traditional wavelet transform, conventional feature extraction and single kernel SVM are referred to as method B, and empirical mode decomposition (EMD), conventional feature extraction and BP neural network are referred to as method C.

[0145] Comparing the fault recognition accuracy, calculation time, noise resistance (performance under different signal-to-noise ratios) and model generalization ability of different methods, the experimental results are shown in the following table:

[0146] Table 1 Fault identification accuracy of different methods on the test set (%)

[0147]

[0148] Table 2 Computing performance comparison

[0149]

[0150] Table 3 Fault recognition accuracy under different signal-to-noise ratios (%)

[0151]

[0152] Table 4 Comparison of model generalization performance (% accuracy under different working conditions)

[0153]

[0154] The results of comparative tests show that the hydraulic pump fault prediction method proposed in this invention is superior to the traditional method in all performance indicators. In terms of fault identification accuracy, the identification accuracy of the method of this invention for four working conditions (normal, wear, cavitation, leakage) is above 96.9%, and the average accuracy reaches 97.6%, which is more than 7.5% higher than the traditional method.

[0155] In terms of computing performance, the feature extraction time, model training time and single prediction time of the present invention are reduced by 42%, 29% and 52% respectively, which significantly improves the practicality of the algorithm.

[0156] In the anti-noise performance test, when the signal-to-noise ratio is reduced to 5dB, the method of the present invention still maintains a recognition accuracy of 90.1%, while the traditional method drops to about 70-73%, reflecting excellent anti-interference ability;

[0157] In the verification of the model generalization performance, in the face of fluctuations in load, speed and temperature, the accuracy of the method of the present invention still remained above 93.5%, while the accuracy of the traditional method dropped significantly under the same conditions, indicating that the present invention has stronger environmental adaptability and generalization ability, and can meet the needs of actual engineering applications.

[0158] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A hydraulic pump fault prediction method, characterized in that: include: Collecting vibration sensor signals of the hydraulic pump, and converting the vibration sensor signals into time domain vibration data according to a preset sampling frequency; Performing wavelet packet decomposition on the time domain vibration data to obtain energy coefficients of different frequency bands, and selecting frequency bands corresponding to the first three maximum energy coefficients as characteristic frequency bands; Extracting characteristic indicators within the characteristic frequency band, constructing a fault characteristic vector, training the fault characteristic vector, and establishing a hydraulic pump fault prediction model; Inputting the fault feature vector of the hydraulic pump to be tested into the hydraulic pump fault prediction model to predict the fault type of the hydraulic pump; Decomposing the time-domain vibration data into low-frequency and high-frequency parts, and obtaining sub-signals of multiple frequency bands by recursive decomposition; Calculating sub-signal energy coefficients of the multiple frequency bands respectively; Arrange the sub-signal energy coefficients in descending order according to their numerical values, and select the frequency bands corresponding to the first three largest energy coefficients after sorting as preliminary characteristic frequency bands; Using frequency band correlation evaluation indicators to further verify and optimize the effectiveness of the preliminary characteristic frequency bands; Reconstructing the verified and optimized preliminary characteristic frequency band to obtain the characteristic frequency band; The calculation of the sub-signal energy coefficient is shown in the following formula: ， in, is the frequency band weighting coefficient, which is determined by the following formula: ， in, is the center frequency of the ith frequency band, is the characteristic frequency of the hydraulic pump, is the bandwidth adjustment parameter, is the sub-signal energy coefficient, is the wavelet packet decomposition coefficient, To decompose the scale, is the translation parameter, is the frequency band number; The verified and optimized preliminary characteristic frequency bands are reconstructed and expressed by the following formula: ， in, is the characteristic frequency band number, is the adaptive reconstruction coefficient, which is calculated by the following formula: ， in, To reconstruct the optimization parameters, is the energy coefficient of the first characteristic frequency band Power, is the energy coefficient of the second characteristic frequency band Power, is the energy coefficient of the third characteristic frequency band Power, For the The energy coefficient of each frequency band Power, is the wavelet packet basis function, To reconstruct the signal.

2. The hydraulic pump fault prediction method according to claim 1, characterized in that: The characteristic indicators include vibration root mean square value, peak factor and waveform index; The vibration root mean square value is calculated as follows: ， in, is the number of sampling points, Weighting coefficient for the time window: ， in, is the signal mean, is the standard deviation, For the The first characteristic frequency band reconstructs the signal The amplitude of the sampling points, For the The root mean square value of the vibration of the reconstructed signal of the characteristic frequency band.

3. The hydraulic pump fault prediction method according to claim 2, characterized in that: The calculation of the peak factor is shown below: ， in, is the correction factor: ， in, is the signal skewness, is an adjustment parameter with a value range of [0.5,1.5], For the The peak factor of the reconstructed signal in each characteristic frequency band; The calculation of the waveform indicator is shown below: ， in, is the entropy correction factor: ， in, is the Shannon entropy of the signal: ， in, is the signal amplitude probability distribution, For the The reconstructed signal of the characteristic frequency bands, is the total number of sampling points.

4. The hydraulic pump fault prediction method according to claim 3, characterized in that: The steps of constructing the hydraulic pump fault prediction model include: Input the constructed fault feature vector into the pre-trained support vector machine (SVM) fault prediction model; The classifier calculates the similarity between the fault feature vector and the support vector using a preset kernel function; According to the kernel function result of the input feature, the decision function value is calculated by combining the weight, label and bias of the support vector; Output The sign of determines the fault type of the hydraulic pump, and completes the construction of the hydraulic pump fault prediction model.

5. The hydraulic pump fault prediction method according to claim 4, characterized in that: The kernel function adopts an adaptive weight combination, as shown in the following formula: ， in, is a Gaussian kernel, used to capture nonlinear features. is a polynomial kernel, used to capture multidimensional linear combination features, is the order of the polynomial, is the hyperbolic tangent kernel, which is used to simulate the activation characteristics of neural networks. is the kernel function, , , is the kernel function weight coefficient, satisfying: 。 6. The hydraulic pump fault prediction method according to claim 5, characterized in that: The determination of the fault type includes: According to the output The symbol of If the value of is positive, it means that the hydraulic pump corresponding to the fault feature vector is in normal working condition and the fault type is "normal"; if If the value of is negative, it means that the hydraulic pump has a certain fault, and the fault type is the corresponding fault mode; The failure mode is determined as follows: For each fault category , calculate its decision function value ,if A negative value indicates that the input feature vector is related to this failure mode, and the one with the smallest negative value is selected. The corresponding fault category is used as the final output fault type.

7. A hydraulic pump fault prediction system, based on the hydraulic pump fault prediction method according to any one of claims 1 to 6, characterized in that: include: A data acquisition module, used to collect vibration sensor signals of the hydraulic pump, and convert the vibration sensor signals into time domain vibration data according to a preset sampling frequency; A feature construction module is used to perform wavelet packet decomposition on the time domain vibration data to obtain energy coefficients of different frequency bands, select frequency bands corresponding to the first three maximum energy coefficients as feature frequency bands, extract feature indicators within the feature frequency bands, and construct a fault feature vector; A model building module, used for training the fault feature vector and establishing a hydraulic pump fault prediction model; The fault prediction module is used to input the fault feature vector of the hydraulic pump to be tested into the hydraulic pump fault prediction model to predict the fault type of the hydraulic pump.

Images