Engine vibration data acquisition and storage method, medium and system

Abstract

The invention provides an engine vibration data collection and storage method, medium and system, and belongs to the technical field of engine vibration collection, and the method comprises the steps: collecting vibration data to obtain a vibration original data matrix, decomposing the vibration original data matrix to obtain a vibration preprocessing data matrix and vibration unconventional data, respectively carrying out time domain decomposition and wavelet transformation to obtain a time domain feature data matrix and a frequency domain feature data matrix, carrying out feature extraction and dimension reduction to obtain a dimension reduction conventional feature matrix, and carrying out calculation and analysis to obtain a vibration unconventional feature matrix; calculating the feature correlation, feature contribution degree and feature weight of the dimensionality reduction conventional feature matrix and the vibration unconventional feature matrix, performing feature fusion to obtain a vibration comprehensive feature matrix, outputting the vibration comprehensive feature matrix, and performing parameter adaptive adjustment of the sampling rate and the filtering parameter according to the vibration comprehensive feature matrix to obtain the vibration characteristic matrix. The technical problem that an existing method lacks effective acquisition of some instant irregular and sudden abnormal vibration signals and is easily neglected in the acquisition process is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of engine vibration collection, and in particular, relates to a method, medium and system for collecting and storing engine vibration data. Background Art

[0002] The engine will generate various vibration signals during long-term operation. These vibration signals contain rich information about the engine status. Using advanced signal processing technology to conduct in-depth analysis can achieve effective evaluation of the engine's operating status and fault diagnosis.

[0003] Existing engine vibration monitoring technologies mainly include the following methods: 1. Time domain analysis method. By analyzing the statistical characteristics of the vibration waveform, such as peak value, root mean square value, skewness, kurtosis, etc., the operating status of the engine can be reflected. This method is simple in basis, but its ability to diagnose complex faults is limited.

[0004] 2. Frequency domain analysis method. By using fast Fourier transform (FFT) or other spectrum analysis techniques to convert the vibration signal to the frequency domain for analysis, characteristic frequency components can be found and fault characteristics can be identified. This method can obtain richer fault diagnosis information, but its analysis ability for non-stationary signals is weak.

[0005] 3. Time-frequency analysis method. Using time-frequency analysis methods such as wavelet transform and Hilbert-Huang transform, the non-stationary characteristics of vibration signals can be characterized from both time domain and frequency domain perspectives to improve the accuracy of fault diagnosis. This method is computationally complex and has poor real-time performance.

[0006] 4. Feature extraction and pattern recognition method. Use feature extraction techniques such as principal component analysis and independent component analysis to extract the most critical features for fault diagnosis from the original vibration data, and then combine pattern recognition algorithms such as neural networks and support vector machines to determine the fault type. This method can effectively improve the diagnostic accuracy, but requires a large amount of historical data for model training.

[0007] Although the above methods have improved the effect of engine vibration monitoring to varying degrees, the existing methods lack effective collection of some instantaneous irregular and sudden abnormal vibration signals, which are easily overlooked during the collection process. Summary of the invention

[0008] In view of this, the present invention provides a method, medium and system for collecting and storing vibration data of an engine, which can solve the technical problem that the existing methods lack effective collection of some instantaneous irregular and sudden abnormal vibration signals and are easily ignored during the collection process.

[0009] The present invention is achieved in that: A first aspect of the present invention provides a method for collecting and storing vibration data of an engine, comprising the following steps: S01, collecting vibration sensor data, temperature sensor data, and speed sensor data during engine operation at a preset sampling rate to obtain a vibration raw data matrix; S02, digitally filter the vibration raw data matrix, use the preset high-frequency threshold and the preset low-frequency threshold as initial filtering parameters, remove the high-frequency noise signal and the low-frequency interference signal, obtain the vibration preprocessing data matrix, and the abnormal mutation data in the vibration raw data matrix discharged during the filtering process, which are recorded as vibration unconventional data; S03, performing time domain decomposition on the vibration preprocessing data matrix, extracting vibration amplitude data, vibration frequency data, and vibration phase data, and forming a time domain feature data matrix; S04, performing wavelet transform on the vibration preprocessing data matrix, extracting vibration energy distribution data, and forming a frequency domain feature data matrix; S05, extracting features from the time domain feature data matrix and the frequency domain feature data matrix to obtain a vibration conventional feature matrix, and inputting the vibration conventional feature matrix into a feature dimension reduction equation to obtain a dimension-reduced conventional feature matrix; S06. Performing time domain decomposition and wavelet transform on the vibration unconventional data to obtain a vibration unconventional feature matrix, and aligning the dimension-reduced conventional feature matrix with the vibration unconventional feature matrix based on the dimension-reduced conventional feature matrix; S07, inputting corresponding characteristic components of the dimension-reduced conventional characteristic matrix and the vibration unconventional characteristic matrix into the characteristic correlation equation to obtain a characteristic correlation coefficient matrix; S08, inputting the characteristic correlation coefficient matrix and the vibration energy distribution data into the characteristic contribution equation to obtain the characteristic contribution matrix; S09, inputting the feature contribution matrix, the preset feature threshold, and the feature correlation coefficient matrix into the feature weight equation to obtain the feature weight coefficient; S10, inputting the dimension-reduced conventional feature matrix, the vibration unconventional feature matrix, the feature weight coefficient, and the preset feature combination parameter into the feature fusion equation to obtain the vibration comprehensive feature matrix; S11, performing standardization processing on the vibration comprehensive feature matrix, and writing the standardized vibration comprehensive feature matrix into a database according to a preset storage format; S12. Input the vibration comprehensive feature matrix, preset evaluation threshold, historical iteration number, and preset sampling rate into the feature evaluation equation to obtain the parameter adjustment indication mark; when the parameter adjustment indication mark is 1, input the feature correlation coefficient matrix, preset high-frequency threshold, preset low-frequency threshold, and preset update step coefficient into the filter parameter update equation to obtain the updated high-frequency threshold and the updated low-frequency threshold, and return to step S02 to re-execute the data processing flow; when the parameter adjustment indication mark is 2, multiply the preset sampling rate by the preset adjustment coefficient as the updated sampling rate, and return to step S01 to restart data collection; when the parameter adjustment indication mark is 0, this round of data processing is completed.

[0010] On the basis of the above technical solution, the vibration data collection and storage method of an engine of the present invention can also be improved as follows: Among them, the characteristic dimension reduction equation is used to perform principal component analysis on the vibration conventional feature matrix, the input is the vibration conventional feature matrix, the preset sampling rate, and the preset dimension reduction ratio, and the output is the reduced dimension conventional feature matrix.

[0011] Furthermore, the feature correlation equation is used to calculate the degree of correlation between the reduced-dimensional conventional feature matrix and the vibration unconventional feature matrix, and the input is the reduced-dimensional conventional feature matrix, the corresponding feature components of the vibration unconventional feature matrix, the preset high-frequency threshold, and the preset low-frequency threshold, and the output is the feature correlation coefficient matrix.

[0012] Furthermore, the characteristic contribution equation is used to evaluate the influence of characteristic components on the engine vibration state, and the input is the characteristic correlation coefficient matrix, the vibration energy distribution data, the preset importance threshold, and the preset attenuation coefficient, and the output is the characteristic contribution matrix.

[0013] Furthermore, the feature weight equation is used to determine the weight of the feature component in the fusion process, the input is the feature contribution matrix, the preset feature threshold, the feature correlation coefficient matrix, the preset weight coefficient, and the output is the feature weight coefficient.

[0014] Furthermore, the feature fusion equation is used to realize weighted fusion of feature components, and the input is the dimension reduction conventional feature matrix, the vibration unconventional feature matrix, the feature weight coefficient, and the preset feature combination parameter, and the output is the vibration comprehensive feature matrix.

[0015] Furthermore, the feature evaluation equation is used to evaluate the vibration feature processing effect and determine the parameter adjustment strategy, the input is the vibration comprehensive feature matrix, the preset evaluation threshold, the historical number of iterations, the preset sampling rate, and the output is the parameter adjustment indication flag.

[0016] Furthermore, the filtering parameter update equation is used to dynamically adjust the filtering parameters according to the feature processing effect, and the input is the feature correlation coefficient matrix, the preset high-frequency threshold, the preset low-frequency threshold, and the preset update step coefficient, and the output is the updated high-frequency threshold and the updated low-frequency threshold.

[0017] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the above-mentioned engine vibration data collection and storage method.

[0018] A third aspect of the present invention provides an engine vibration data acquisition and storage system, which includes the above-mentioned computer-readable storage medium.

[0019] Compared with the prior art, the engine vibration data acquisition and storage method, medium and system provided by the present invention make full use of the dual characteristics of time domain and frequency domain, comprehensively consider conventional vibration signals and unconventional vibration signals, and adopt advanced signal processing and feature extraction technologies, such as digital filtering, Hilbert transform, wavelet transform, principal component analysis, etc., which can more comprehensively characterize the complex characteristics of the engine vibration state. At the same time, the method adopts innovative algorithm design in key links such as feature fusion and parameter adaptive adjustment, which greatly improves the accuracy and robustness of vibration data processing, and provides strong support for engine state monitoring and fault diagnosis.

[0020] Specifically, the technical solution of the present invention includes the following innovative features: 1. Identification and utilization of unconventional vibration signals. In the digital filtering preprocessing stage, not only high-frequency noise and low-frequency interference are removed, but also the mutation data detected during the filtering process are recorded and classified, laying the foundation for subsequent feature extraction and fusion. This unified consideration of conventional and unconventional vibration signals is conducive to better characterizing the overall vibration characteristics of the engine.

[0021] 2. Deep fusion of time domain and frequency domain features. This paper proposes a feature dimensionality reduction method based on principal component analysis, which can extract the key features that have the greatest impact on the engine state from the original high-dimensional features; at the same time, it adopts the idea of ​​weighted fusion, comprehensively considers the correlation between conventional features and unconventional features, energy distribution characteristics and importance weights, realizes the deep fusion of time domain and frequency domain features, and further improves the accuracy of feature expression.

[0022] 3. Adaptive parameter adjustment mechanism. The present invention has designed corresponding adaptive adjustment algorithms in key links of data processing, such as digital filtering and sampling rate setting, which can dynamically optimize relevant parameters according to historical processing effects, ensure stable and efficient data processing performance under different working conditions, and avoid some momentary irregular and sudden abnormal vibration signals being ignored by filtering. This ability of autonomous learning and optimization further enhances the versatility and robustness of the method.

[0023] Therefore, the present invention solves the technical problem that the existing methods lack effective collection of some momentary irregular and sudden abnormal vibration signals and are easily overlooked during the collection process. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 A flow chart of the method provided by the present invention; Figure 2 The time domain waveform diagram of the vibration signal collected by the three-axis acceleration sensor in Example 2; Figure 3 This is a comparison diagram of the original vibration signal and the signal after being processed by the filter in Example 2; Figure 4 Graph showing the frequency domain characteristic analysis result of the vibration signal in Example 2; Figure 5 This is a graph showing the changing trend of the evaluation index during the parameter adaptive adjustment process in Example 2. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0026] like Figure 1 FIG. 1 is a flow chart of a vibration data collection and storage method for an engine provided by the present invention. The specific implementation method of each step is described in detail below: The specific implementation of step S01 is: first, a preset sampling rate is used to collect data from the vibration sensor, temperature sensor and speed sensor during the operation of the engine. The vibration sensor collects engine vibration waveform data, the temperature sensor collects engine temperature data, and the speed sensor collects engine speed data. The collected raw data are combined into a three-dimensional data matrix, namely the vibration raw data matrix. The purpose of this step is to obtain basic data on the engine's operating status.

[0027] The specific implementation method of step S02 is: digital filtering is performed on the vibration raw data matrix obtained in step S01. First, a preset high-frequency threshold and a low-frequency threshold are set as initial filtering parameters. Then a bandpass filter is used to filter the raw data matrix in order to remove high-frequency noise signals and low-frequency interference signals. The data matrix obtained after filtering is called a vibration preprocessing data matrix. During the filtering process, if abnormal mutation data points are found in the raw data matrix, they are recorded as vibration unconventional data. This step aims to eliminate the impact of external interference on the data and improve the quality and reliability of the data.

[0028] The specific implementation method of step S03 is: perform time domain analysis on the vibration preprocessing data matrix obtained in step S02. First, use Hilbert transform to calculate the vibration amplitude, vibration frequency and vibration phase of each data point. The vibration amplitude reflects the intensity of the vibration, the vibration frequency reflects the periodicity of the vibration, and the vibration phase reflects the initial phase of the vibration. These time domain feature parameters are combined into a feature data matrix. The purpose of this step is to extract vibration feature information from the time domain perspective.

[0029] The specific implementation of step S04 is: perform frequency domain analysis on the vibration preprocessing data matrix obtained in step S02. Use continuous wavelet transform to analyze the data matrix and extract the vibration energy distribution characteristics of each data point. Wavelet transform can reflect the energy distribution of the signal at different scales and capture the time-frequency characteristics of non-stationary signals. These frequency domain feature parameters are combined into another feature data matrix. The purpose of this step is to extract vibration feature information from the frequency domain perspective.

[0030] The specific implementation method of step S05 is: first, the time domain feature data matrix and the frequency domain feature data matrix obtained in step S03 and step S04 are subjected to feature extraction and dimensionality reduction processing. The feature extraction adopts the principal component analysis method, and the purpose is to extract the main features that have the greatest impact on the engine state from the original features. Feature dimensionality reduction is to map the high-dimensional feature space to the low-dimensional space to reduce the complexity of subsequent processing. The dimensionality reduction conventional feature matrix obtained through this step is the input of steps S06, S07, S08, and S09.

[0031] The specific implementation of step S06 is: perform time domain analysis and frequency domain analysis on the vibration unconventional data recorded in step S02, extract its time domain features and frequency domain features, and form a vibration unconventional feature matrix. Then align the dimension-reduced conventional feature matrix and the vibration unconventional feature matrix to prepare for subsequent feature fusion. The purpose of this step is to make full use of the information contained in the unconventional data.

[0032] The specific implementation method of step S07 is: calculate the correlation between the reduced-dimensional conventional feature matrix and the vibration unconventional feature matrix. Specifically, the Pearson correlation coefficient is calculated for the corresponding feature components in the two feature matrices to obtain a feature correlation coefficient matrix. The larger the correlation coefficient, the higher the degree of correlation between the two features. The purpose of this step is to quantify the correlation between conventional features and unconventional features.

[0033] The specific implementation method of step S08 is: comprehensively consider the feature correlation and feature energy distribution characteristics, and evaluate the influence of each feature component on the engine vibration state. Specifically, the feature correlation coefficient matrix and the vibration energy distribution data obtained by wavelet transform are input into the feature contribution equation to calculate the contribution of each feature component. The higher the contribution, the greater the influence of the feature on the engine vibration state. The purpose of this step is to determine the importance weight of each feature component.

[0034] The specific implementation method of step S09 is: according to the feature contribution, correlation and preset feature importance threshold, the weight coefficient of each feature component in the subsequent fusion process is calculated. Specifically, the feature contribution matrix, the correlation coefficient matrix and the preset feature importance threshold are input into the feature weight equation to obtain the final feature weight coefficient. The larger the weight coefficient, the more important the feature is in the vibration state assessment. The purpose of this step is to determine the fusion weight of each feature component during feature fusion.

[0035] The specific implementation method of step S10 is: using the weighted fusion method, the dimension-reduced conventional feature matrix, the vibration unconventional feature matrix and the feature weight coefficients calculated previously are fused to obtain the final vibration comprehensive feature matrix. The fusion equation not only considers the weight of each feature component, but also considers the cross terms between features, and can capture the nonlinear relationship between features. The purpose of this step is to comprehensively utilize conventional features and unconventional features to extract the most representative vibration features.

[0036] The specific implementation of step S11 is: standardize the vibration comprehensive feature matrix obtained in step S10 to meet the storage format requirements of the database, and then write the standardized feature matrix into the database. Standardization can eliminate the dimension effect and make the eigenvalue fall within the appropriate numerical range. The purpose of this step is to facilitate the subsequent storage and call of data.

[0037] The specific implementation method of step S12 is: evaluate the effect of current feature extraction and fusion, and determine whether the parameters of data processing need to be adjusted according to the evaluation results. Specifically, the vibration comprehensive feature matrix, the preset feature evaluation threshold, the number of historical iterations and the sampling rate are input into the feature evaluation equation to obtain the parameter adjustment indication flag. If the flag is 1, it means that the filter parameters need to be adjusted, and the correlation coefficient matrix, the high and low frequency thresholds and the update step coefficient are input into the filter parameter update equation to obtain a new filter threshold, and then return to step S02 to reprocess the data; if the flag is 2, it means that the sampling rate needs to be adjusted, and the current sampling rate is multiplied by the adjustment coefficient as the new sampling rate, and then return to step S01 to re-collect data; if the flag is 0, it means that the current parameter settings are reasonable and no adjustment is required, and the current round of data processing is completed. The purpose of this step is to ensure the continuous optimization of the effect of data processing.

[0038] The equations or calculation processes involved in the present invention are described in detail below: First, the expression of the vibration raw data matrix is: ; In the formula, is the number of sampling points, is the number of sensors, For the Sampling point The value of a sensor.

[0039] The digital filtering process uses a bandpass filter, whose transfer function is: ; In the formula, is the signal frequency, is the low frequency cutoff frequency, is the high frequency cut-off frequency, is the filter order.

[0040] Time domain feature extraction uses Hilbert transform: ; In the formula, is the original signal, is the transformed signal.

[0041] Vibration amplitude calculation formula: ; In the formula, is the instantaneous amplitude.

[0042] Phase calculation formula: ; In the formula, is the instantaneous phase.

[0043] Frequency calculation formula: ; In the formula, is the instantaneous frequency.

[0044] Wavelet transform uses continuous wavelet transform: ; In the formula, is the scale parameter, is the translation parameter, is the wavelet basis function.

[0045] The feature dimension reduction equation uses principal component analysis: ; In the formula, is the eigenvector matrix, is the original data matrix, is the matrix after dimension reduction.

[0046] Characteristic correlation equation: ; In the formula, For the Features and The correlation coefficient of the features.

[0047] Feature contribution equation: ; In the formula, For the The contribution of a feature, is the correlation coefficient, is the energy ratio, is the importance weight, is the adjustment coefficient.

[0048] Feature weight equation: ; In the formula, For the The weight of the feature, is the characteristic value, is the threshold value, is the attenuation coefficient.

[0049] The feature fusion equation uses weighted fusion: ; In the formula, is the fused feature, For the Features, is the corresponding weight, is the correlation coefficient between features, is the cross-term coefficient.

[0050] Characteristic evaluation equation: ; In the formula, As the evaluation index, The first Quantity, is the autocorrelation coefficient, is the mutual correlation coefficient, is the weight coefficient.

[0051] Filter parameter update equation: ; ; In the formula, are the updated high and low frequency thresholds, is the target evaluation value, To adjust the parameters.

[0052] Sampling rate update equation: ; In the formula, is the updated sampling rate, is the learning rate.

[0053] in: 1) Initial sampling rate : Determined according to Nyquist sampling theorem; 2) Filter threshold : Determined by spectrum analysis; 3) Wavelet parameters : Determined based on signal feature scale; 4) Evaluation threshold : Obtained through historical data statistics; 5) Update parameters : Determined by convergence analysis.

[0054] The range and acquisition method of each parameter are described as follows: 1) Filter parameters: The range is 100Hz1000Hz, The range is 1Hz10Hz, determined by spectrum analysis; 2) Wavelet transform parameters: scale The value range is 1 to 64, with intervals of 2; 3) Feature contribution coefficient: The sum is 1, optimized by cross-validation; 4) Weight decay coefficient The value range is 0.1~1.0; 5) Evaluation equation weights: The sum is 1, and it is adjusted according to actual needs; 6) Update equation parameters: The value is 0.10.5. The value is 15. The value range is 0.01~0.1.

[0055] Explanation of the equation construction principle: 1) Hilbert transform is used to obtain analytical signals and can accurately extract instantaneous features; 2) Wavelet transform has multi-resolution characteristics and is suitable for non-stationary signal analysis; 3) The feature correlation uses the Pearson coefficient to measure the degree of linear correlation; 4) The feature contribution equation comprehensively considers three aspects: relevance, energy and importance; 5) An exponential decay term is introduced into the weight equation to make the weight change smoothly with the degree of deviation of the eigenvalue; 6) The fusion equation takes into account the characteristic cross terms and can capture nonlinear relationships; 7) The evaluation equation evaluates the feature quality from three dimensions: variance, autocorrelation and cross-correlation; 8) The parameter update equation adopts exponential form to ensure convergence and smoothness.

[0056] The following is a detailed explanation of the derivation process and principle of each equation: 1. Derivation of characteristic correlation equation: Starting from the covariance matrix: ; After standardization, we get: ; The discretization result is: .

[0057] 2. Feature contribution equation construction process: First consider the impact of correlation: ; Add energy contribution: ; Consider feature importance: ; Through weighted combination, we get: ; The weight coefficient is determined by the following steps: 1) Initialization ; 2) Using grid search optimization; 3) Use cross validation to determine the optimal value.

[0058] 3. Derivation of feature weight equation: The basic weight is obtained by normalizing the contribution: ; Consider the influence of eigenvalue deviation: ; in Determination process: 1) Set the initial value ; 2) Optimize by gradient descent; 3) Adjust based on convergence.

[0059] 4. Derivation of feature fusion equation: Linear combination terms: ; Consider feature interactions: ; Optimization steps: 1) Initialization ; 2) Minimize the reconstruction error; 3) Iterative optimization parameters.

[0060] 5. Parameter update equation construction: Use exponential form to ensure smoothness: ; Introduce a symbolic function to control the direction: ; The final result is: .

[0061] In the field of traditional engine vibration data collection and storage, engineers usually adopt relatively simple and fixed solutions. Such solutions mainly rely on basic data acquisition equipment and signal processing processes with fixed parameters. Specifically, traditional methods usually use accelerometers with fixed sampling rates for data collection, and the sampling frequency is generally set between 2 kHz and 20 kHz. In the signal preprocessing stage, engineers often only use the most basic low-pass filtering or band-pass filtering for noise elimination. Although this processing method is simple to operate, it is difficult to cope with complex and changing working conditions. In the feature extraction link, traditional methods are mainly limited to calculating some basic time domain statistical features, such as mean, standard deviation, peak value, etc., and are supplemented by frequency domain features obtained by fast Fourier transform. Although this feature extraction method is intuitive, it often cannot fully reflect the deep characteristics of engine vibration signals. In terms of data storage, traditional methods often use the method of directly storing raw data with basic features. Although this method retains complete information, it has low storage efficiency and is difficult to conduct subsequent in-depth analysis.

[0062] In contrast, the innovative method proposed in the present invention has achieved significant technological breakthroughs in the data acquisition and preprocessing links. First, the present invention introduces an adaptive sampling rate update mechanism, and the system can dynamically adjust the sampling frequency according to the signal evaluation index calculated in real time. This adaptive mechanism can optimize the sampling efficiency while ensuring signal quality. In terms of signal filtering, the present invention designs a bandpass filter with an adjustable order and is equipped with an automatic update algorithm for filter parameters based on evaluation indicators. This design significantly improves the system's anti-noise ability. It is particularly worth mentioning that the adaptive update mechanism of the filter parameters can adjust the high and low frequency cutoff frequencies in real time according to the signal quality evaluation results. This dynamic adjustment capability enables the system to better adapt to the vibration characteristics under different working conditions.

[0063] In terms of feature extraction and processing, the present invention adopts a variety of advanced signal processing technologies to form a comprehensive feature analysis framework. By introducing the Hilbert transform, the system can accurately extract the instantaneous characteristics of the vibration signal, including the instantaneous amplitude, phase and frequency information, which is of great significance for analyzing the vibration characteristics of the engine under non-stationary conditions. At the same time, the present invention also uses continuous wavelet transform for time-frequency analysis. This multi-resolution analysis method can capture signal characteristics at different time scales and is particularly suitable for processing the vibration signal of the engine under variable speed conditions. In addition, by performing dimensionality reduction processing through principal component analysis, the system can effectively reduce data redundancy and improve the efficiency of feature expression.

[0064] In the feature selection and fusion links, the present invention has established a complete feature evaluation and optimization system. First, through feature correlation analysis, the system can accurately evaluate the degree of correlation between different features and avoid information redundancy. The feature contribution calculation mechanism comprehensively considers the three dimensions of correlation, energy distribution and importance to ensure that the selected features are well representative. The feature weight allocation adopts a dynamic calculation method with an exponential decay term. This design makes the weight allocation more reasonable and can better reflect the actual importance of the feature. In the feature fusion stage, the present invention not only considers the linear combination of features, but also introduces feature cross terms. This design can effectively capture the nonlinear relationship between features.

[0065] The innovation of the present invention is also reflected in its complete quality evaluation and parameter optimization mechanism. The design of the evaluation index comprehensively examines the feature quality from three dimensions: variance, autocorrelation and cross-correlation. This multi-dimensional evaluation method can fully reflect the effectiveness of the features. The parameter update adopts an exponential smoothing adjustment mechanism, which not only ensures the convergence of the algorithm, but also maintains the stability of the system response. This complete set of evaluation and optimization mechanisms enables the system to continuously adjust and improve its performance.

[0066] In general, the advantages of the present invention over traditional methods are mainly reflected in the following aspects: first, the innovative adaptive sampling and filtering mechanism significantly improves the quality and efficiency of data acquisition; second, the advanced feature extraction method enhances the system's ability to analyze non-stationary signals; third, the perfect feature selection and fusion mechanism improves the expression ability and discriminability of features; finally, the scientific evaluation and optimization mechanism ensures the continuous improvement of system performance. These advantages make the present invention have broad application prospects in the field of engine vibration monitoring, and can provide more reliable technical support for engine status monitoring and fault diagnosis.

[0067] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the above-mentioned engine vibration data collection and storage method.

[0068] A third aspect of the present invention provides an engine vibration data acquisition and storage system, which includes the above-mentioned computer-readable storage medium.

[0069] Specifically, the principle of the present invention is: 1. More comprehensive characterization of the engine vibration state. By considering both conventional and unconventional vibration signals and combining the deep fusion of time domain and frequency domain features, this method can more comprehensively reflect the vibration characteristics of the engine and provide richer and more accurate information support for subsequent state monitoring and fault diagnosis.

[0070] 2. Greatly improved the accuracy and robustness of data processing. The innovative feature extraction and fusion algorithm, as well as the introduction of adaptive parameter adjustment mechanism, make this method significantly superior to existing technologies in key indicators such as anti-interference ability and adaptability, greatly improving the overall performance of vibration data processing.

[0071] 3. Lay the foundation for engine status monitoring and fault diagnosis applications. The vibration data acquisition and storage method proposed in the present invention can effectively extract the most representative features from massive vibration signals, which not only provides reliable data support for subsequent status monitoring, but also creates good conditions for the training and application of fault diagnosis algorithms, which is of great significance in improving the level of engine operation and maintenance management.

[0072] A specific embodiment 1 of the present invention is provided below. The specific implementation of each step in this embodiment 1 is described in detail as follows: The specific implementation of step S01 is as follows: First, a preset sampling rate is used. Collect data from vibration sensors, temperature sensors and speed sensors during engine operation. Initial sampling rate The determination of can refer to the Nyquist sampling theorem, that is, ,in The highest frequency of the vibration signal. The collected raw data are combined into a three-dimensional data matrix ,in is the number of sampling points, is the number of sensors, Indicates Sampling point The purpose of this step is to obtain the basic data of the engine operating status.

[0073] The specific implementation of step S02 is as follows: First, set a preset high-frequency threshold and low frequency threshold As the initial filtering parameters. Through spectrum analysis, the main frequency range of the vibration signal can be determined, generally taking exist between, exist Then, a bandpass filter is used to filter the vibration raw data matrix obtained in step S01. For digital filtering, the filter transfer function is ,in is the order of the filter. The purpose of filtering is to remove high-frequency noise signals and low-frequency interference signals and obtain the vibration preprocessing data matrix During the filtering process, if the original data matrix There are abnormal mutation data points in the data, which are recorded as vibration abnormal data. This step aims to improve the quality and reliability of the data and lay the foundation for subsequent feature extraction.

[0074] The specific implementation of step S03 is as follows: the vibration preprocessing data matrix obtained in step S02 Perform time domain analysis. First, use the Hilbert transform to calculate the vibration amplitude of each data point , vibration frequency and vibration phase ,in is the signal after Hilbert transformation. These time domain feature parameters are combined into a time domain feature data matrix The purpose of this step is to extract vibration feature information from the time domain perspective and prepare for subsequent feature fusion.

[0075] The specific implementation of step S04 is as follows: the vibration preprocessing data matrix obtained in step S02 Perform frequency domain analysis. Use continuous wavelet transform Analyze the data matrix and extract the vibration energy distribution characteristics of each data point In wavelet transform, the scale parameter The value range is ,by The exponential form of the translation parameter Determined according to actual needs; is the wavelet basis function, and Gaussian wavelet, Mexican hat wavelet, etc. can be selected. Wavelet transform can reflect the energy distribution of the signal at different time scales, which is conducive to capturing the time-frequency characteristics of non-stationary signals. The purpose of this step is to extract vibration feature information from the frequency domain perspective to prepare for subsequent feature fusion.

[0076] The specific implementation of step S05 is as follows: First, the time domain feature data matrix obtained in step S03 and step S04 is And frequency domain feature data matrix Perform feature extraction and dimensionality reduction. Feature extraction uses principal component analysis, that is, ,in is the eigenvector matrix, is the original data matrix, is the feature matrix after dimensionality reduction. The dimensionality reduction ratio can be set according to actual needs, usually taking the first The principal component analysis can extract the main features that have the greatest impact on the engine state from the original high-dimensional features, reducing the complexity of subsequent processing. The reduced-dimensional conventional feature matrix obtained through this step That is the input of steps S06, S07, S08, and S09.

[0077] The specific implementation of step S06 is as follows: the vibration irregular data recorded in step S02 Perform time domain analysis and frequency domain analysis to extract its time domain features and frequency domain characteristics , forming a vibration unconventional characteristic matrix . Then reduce the dimension of the regular feature matrix and vibration unconventional characteristic matrix Alignment is performed to prepare for subsequent feature fusion. The purpose of this step is to make full use of the information contained in unconventional data and improve the accuracy of feature fusion.

[0078] The specific implementation of step S07 is as follows: Calculate the dimension reduction conventional feature matrix Non-conventional characteristic matrix with vibration The specific method is to calculate the Pearson correlation coefficient for the corresponding feature components in the two feature matrices. , get the feature correlation coefficient matrix . Correlation coefficient The larger the Features and The purpose of this step is to quantify the correlation between conventional features and unconventional features, and provide a basis for subsequent feature fusion.

[0079] The specific implementation of step S08 is as follows: Comprehensively consider the feature correlation , characteristic energy distribution , to evaluate the influence of each characteristic component on the engine vibration state. The specific method is to transform the characteristic correlation coefficient matrix Vibration energy distribution data obtained by wavelet transform Enter the feature contribution equation , calculate the contribution of each feature component .in, is the adjustment coefficient, satisfying Contribution The higher the The greater the influence of each feature on the engine vibration state, the greater the influence of each feature on the engine vibration state. The purpose of this step is to determine the importance weight of each feature component in the subsequent fusion process.

[0080] The specific implementation of step S09 is as follows: , Correlation And the preset feature importance threshold , calculate the weight coefficient of each feature component in the fusion process The specific approach is to transform the feature contribution matrix , correlation coefficient matrix And the preset feature importance threshold Enter the feature weight equation , get the final feature weight coefficient .in, For the The characteristic value of a feature, is the preset feature threshold, is the attenuation coefficient, and its value range is . Weight coefficient The larger the The more important a feature is in vibration state assessment, the more important it is. The purpose of this step is to determine the fusion weights of each feature component when fusion is performed.

[0081] The specific implementation of step S10 is as follows: using the weighted fusion method, the reduced-dimensional conventional feature matrix , vibration unconventional characteristic matrix And the feature weight coefficient calculated previously Fusion is performed to obtain the final vibration comprehensive feature matrix The fusion equation is ,in For the Features, is the cross-term coefficient. This equation not only considers the weight of each feature component, but also the cross-correlation between features, which can better capture the nonlinear relationship between features. The purpose of this step is to comprehensively utilize conventional features and unconventional features to extract the most representative vibration features.

[0082] The specific implementation of step S11 is as follows: the vibration comprehensive feature matrix obtained in step S10 Standardization is performed to meet the storage format requirements of the database, and then the standardized feature matrix is ​​written into the database. Standardization can be done by ,in is the mean of the feature matrix, is the standard deviation of the feature matrix. Standardization can eliminate the dimension effect and make the eigenvalue fall within the appropriate numerical range. The purpose of this step is to facilitate the subsequent storage and call of data.

[0083] The specific implementation of step S12 is as follows: evaluate the effect of current feature extraction and fusion, and determine whether the parameters of data processing need to be adjusted according to the evaluation results. , preset feature evaluation threshold , historical iteration times And the sampling rate Enter the feature evaluation equation , get the parameter adjustment indicator .in, is the weight coefficient, satisfying .like , it means that the filtering parameters need to be adjusted and the correlation coefficient matrix , high and low frequency thresholds And update the step size coefficient Input filter parameter update equation and , get the new filtering threshold , then return to step S02 to reprocess the data; if , it means that the sampling rate needs to be adjusted. Multiply by the adjustment factor As the new sampling rate , then return to step S01 to re-collect data; if , it means that the current parameter settings are reasonable and no adjustment is needed, completing this round of data processing. The purpose of this step is to ensure the continuous optimization of the data processing effect.

[0084] In order to better understand and implement the present invention, the following provides Example 2 of a specific application scenario of the present invention: A generator set is a key power equipment, and real-time monitoring and accurate diagnosis of its vibration state are crucial to ensure the safe operation of the equipment. The engine vibration data acquisition and storage method proposed in the present invention has been successfully applied to the vibration monitoring system of the generator set, achieving obvious technical advantages.

[0085] Data acquisition: A three-axis acceleration sensor is used to collect the vibration signal of the generator set. The initial sampling rate is set to , which meets the requirements of Nyquist sampling theorem. After spectrum analysis, it is found that the main frequency components of the vibration signal of the generator set are concentrated within 100Hz, so the high-frequency cutoff frequency of the digital filter is set to Set to 1kHz, low frequency cutoff frequency Set to 5Hz, filter order Pick .like Figure 2 As shown in the figure, the time domain waveform of the vibration signal collected by the triaxial acceleration sensor is displayed. The changes of the vibration signals in the X, Y, and Z directions over time are shown respectively.

[0086] Data preprocessing: collect the original vibration data matrix Perform digital filtering to remove high-frequency noise and low-frequency interference to obtain the preprocessed data matrix During the filtering process, it is found that Row, No. Row and The data has a large mutation, which is recorded as vibration irregular data. . Figure 3 The blue curve shows the original signal containing noise, and the red curve shows the smoothed signal after filtering.

[0087] Feature extraction: First, preprocess the data matrix Perform time domain feature extraction and calculate the vibration amplitude of each data point , vibration frequency and vibration phase , forming a time domain feature data matrix Then, continuous wavelet transform is used to transform Perform frequency domain analysis to extract vibration energy distribution characteristics , scale parameter Pick , translation parameters Pick .

[0088] Next, the time domain feature matrix and frequency domain feature matrix Perform principal component analysis to extract the components that have the greatest impact on the generator vibration state. Features, forming a conventional feature matrix after dimensionality reduction . For the non-routine data recorded above , and also extracted the time domain and frequency domain features to form an unconventional feature matrix . Figure 4 The frequency domain characteristic analysis results of the vibration signal are displayed, clearly showing the distribution of the main frequency components.

[0089] Feature fusion: First calculate the dimensionality reduction conventional features With unconventional features The correlation coefficient matrix between , calculated using the Pearson correlation coefficient. Then, based on the feature correlation, energy distribution characteristics, and importance weights, the influence of each feature component on the generator vibration state is evaluated to obtain the feature contribution matrix .

[0090] Next, the feature contribution , correlation coefficient And the preset feature importance threshold Substitute the feature weight equation to calculate the weight coefficient of each feature component in the fusion process Finally, the weighted fusion method is used to , and The final vibration comprehensive feature matrix is ​​obtained by fusion .

[0091] Parameter adaptive adjustment: In order to ensure the continuous optimization of data processing effect, the present invention introduces an adaptive adjustment mechanism in the key parameter setting link. , preset evaluation threshold , historical iteration times and the current sampling rate Substitute into the characteristic evaluation equation to get the evaluation index and parameter adjustment indicator .

[0092] according to As a result, the filtering parameters need to be adjusted dynamically. , the current high and low frequency thresholds , And adjustment parameters , , , Input the filter parameter update equation and calculate the new high and low frequency thresholds: , Then return to step (2) and re-execute the data preprocessing process.

[0093] After adjusting the above parameters, we evaluate again. , , indicating that the current parameter settings are reasonable and no further adjustment is needed. Finally, the final vibration comprehensive feature matrix Standardize the process to form , and write it into the database according to the preset data storage format. Figure 5 The changing trend of the evaluation index E during the parameter adaptive adjustment process is shown, as well as the comparison with the target threshold E0.

[0094] Compared with the traditional engine vibration monitoring method, the solution proposed by the present invention has the following advantages: 1. More comprehensive vibration feature extraction. Traditional methods mainly rely on basic time domain statistical features and frequency domain FFT features, which are difficult to capture the deep characteristics of vibration signals. The present invention combines time domain analysis and frequency domain analysis to extract not only time domain features such as vibration amplitude, frequency, and phase, but also frequency domain features such as vibration energy distribution, which can more comprehensively reflect the vibration state of the engine. At the same time, special attention is also paid to unconventional vibration signals, further enriching the content of feature extraction.

[0095] 2. More accurate feature fusion. Traditional methods usually use a simple weighted average or cascade method to fuse features, which makes it difficult to fully explore the intrinsic relationship between features. The present invention proposes a fusion method based on feature correlation, energy distribution and importance weight, which not only considers the independent contribution of each feature component, but also introduces feature cross terms, which can more accurately characterize the complex relationship between vibration features, thereby improving the accuracy of feature fusion.

[0096] 3. More robust adaptive parameter adjustment. Traditional methods usually use fixed parameter settings in key links such as data acquisition and signal preprocessing, which are difficult to adapt to complex and changing working conditions. The present invention introduces an adaptive adjustment mechanism in the key parameter setting link, which can dynamically optimize parameters such as sampling rate and filter threshold according to historical processing effects, ensuring stable and efficient data processing performance under various working conditions.

[0097] The following table compares the performance of the solution of the present invention and traditional engine vibration monitoring technology in terms of key indicators: Table 1 Comparison table of key indicators of traditional method and the present invention index Solution of the present invention Traditional methods Feature extraction dimension Time domain + frequency domain Main time domain Unconventional signal processing Focus neglect Feature fusion accuracy high generally Parameter Adaptability powerful weak Data processing efficiency high generally Monitoring diagnostic performance Excellent generally As can be seen from the table above, the solution of the present invention is significantly superior to the traditional method in terms of feature extraction dimension, unconventional signal processing, feature fusion accuracy, parameter adaptive capability, etc., and the overall data processing efficiency and monitoring and diagnostic performance are also greatly improved. These advantages reflect the innovation and practicality of the present invention compared with the existing technology.

[0098] The following takes the test run of a generator set as an example to specifically illustrate the application effect of the solution of the present invention: During the trial operation, the acceleration sensor collected the vibration data of the generator set at a sampling rate of 5kHz. After digital filtering, it was found that Second, Seconds and At 15.3 seconds, the vibration signal showed an obvious abnormal mutation (as shown in Table 2).

[0099] Table 2 Unconventional data records of generator set vibration

[0100] Next, the preprocessed vibration data is subjected to time domain feature extraction and frequency domain feature extraction to obtain the conventional feature matrix and unconventional feature matrices By calculating the feature correlation coefficient matrix ,It is found that the unconventional features have a strong correlation with the ,conventional features.

[0101] In the feature fusion stage, according to the feature contribution , Correlation and the preset importance threshold , calculate the fusion weight of each feature component .Will , and Substitute into the weighted fusion equation to obtain the final vibration comprehensive characteristic matrix .

[0102] right Evaluate and calculate the indicators , below the preset target , parameter adjustment indicator If is 1, the filter parameters need to be adjusted. , Current filter threshold , and adjustment parameters , , , Substitute into the filter parameter update equation and calculate the new high and low frequency thresholds: , .

[0103] After adjusting the parameters, we evaluate again. , , indicating that the data processing effect has reached the standard and no further adjustment is needed. Standardize the process to form , and stored in the database.

[0104] Compared with the traditional engine vibration monitoring method, the scheme of the present invention performs as follows in this trial operation scenario: 1. The comprehensive extraction of time domain and frequency domain features fully reflects the complex characteristics of the vibration state of the generator set, providing richer information support for subsequent state assessment and fault diagnosis.

[0105] 2. The identification and utilization of unconventional vibration signals not only enhances the robustness of data processing, but also creates conditions for discovering potential fault hazards. Traditional methods usually ignore these abnormal data, resulting in deviations in diagnostic results.

[0106] 3. The fusion method based on feature correlation and energy distribution effectively captures the intrinsic connection between features, improves the accuracy of feature fusion, and thus enhances the reliability of overall data processing.

[0107] 4. By adaptively adjusting the filtering parameters, the data preprocessing process can be dynamically optimized to ensure stable and efficient processing performance under complex and changing working conditions. Traditional methods are difficult to cope with such changes due to fixed parameters.

[0108] 5. The final comprehensive vibration characteristics Not only can the vibration state of the generator set be described more comprehensively and accurately, but the data format also meets the storage requirements, laying a good foundation for subsequent state monitoring and fault diagnosis and analysis.

[0109] It should be noted that the variables involved in the present invention are explained in detail as shown in Table 3 below.

[0110] Table 3 Variable explanation table

[0111] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for collecting and storing vibration data of an engine, characterized in that: The following steps are involved: S01, collecting vibration sensor data, temperature sensor data, and speed sensor data during engine operation at a preset sampling rate to obtain a vibration raw data matrix; S02, digitally filter the vibration raw data matrix, use the preset high-frequency threshold and the preset low-frequency threshold as initial filtering parameters, remove the high-frequency noise signal and the low-frequency interference signal, obtain the vibration preprocessing data matrix, and the abnormal mutation data in the vibration raw data matrix discharged during the filtering process, which are recorded as vibration unconventional data; S03, performing time domain decomposition on the vibration preprocessing data matrix, extracting vibration amplitude data, vibration frequency data, and vibration phase data, and forming a time domain feature data matrix; S04, performing wavelet transform on the vibration preprocessing data matrix, extracting vibration energy distribution data, and forming a frequency domain feature data matrix; S05, extracting features from the time domain feature data matrix and the frequency domain feature data matrix to obtain a vibration conventional feature matrix, and inputting the vibration conventional feature matrix into a feature dimension reduction equation to obtain a dimension-reduced conventional feature matrix; S06. Performing time domain decomposition and wavelet transform on the vibration unconventional data to obtain a vibration unconventional feature matrix, and aligning the dimension-reduced conventional feature matrix with the vibration unconventional feature matrix based on the dimension-reduced conventional feature matrix; S07, inputting corresponding characteristic components of the dimension-reduced conventional characteristic matrix and the vibration unconventional characteristic matrix into the characteristic correlation equation to obtain a characteristic correlation coefficient matrix; S08, inputting the characteristic correlation coefficient matrix and the vibration energy distribution data into the characteristic contribution equation to obtain the characteristic contribution matrix; S09, inputting the feature contribution matrix, the preset feature threshold, and the feature correlation coefficient matrix into the feature weight equation to obtain the feature weight coefficient; S10, inputting the dimension-reduced conventional feature matrix, the vibration unconventional feature matrix, the feature weight coefficient, and the preset feature combination parameter into the feature fusion equation to obtain the vibration comprehensive feature matrix; S11, performing standardization processing on the vibration comprehensive feature matrix, and writing the standardized vibration comprehensive feature matrix into a database according to a preset storage format; S12, input the vibration comprehensive feature matrix, the preset evaluation threshold, the number of historical iterations, and the preset sampling rate into the feature evaluation equation to obtain a parameter adjustment indication flag; when the parameter adjustment indication flag is 1, input the feature correlation coefficient matrix, the preset high-frequency threshold, the preset low-frequency threshold, and the preset update step coefficient into the filter parameter update equation to obtain an updated high-frequency threshold and an updated low-frequency threshold, and return to step S02 to re-execute the data processing flow; When the parameter adjustment indication flag is 2, the preset sampling rate is multiplied by the preset adjustment coefficient as the updated sampling rate, and the process returns to step S01 to start collecting data again; When the parameter adjustment indication flag is 0, this round of data processing is completed.

2. The method for collecting and storing vibration data of an engine according to claim 1, characterized in that: The characteristic dimension reduction equation is used to perform principal component analysis on the vibration conventional feature matrix, the input is the vibration conventional feature matrix, the preset sampling rate, and the preset dimension reduction ratio, and the output is the dimension reduction conventional feature matrix.

3. The method for collecting and storing vibration data of an engine according to claim 2, characterized in that: The feature correlation equation is used to calculate the degree of correlation between the reduced-dimensional conventional feature matrix and the vibration unconventional feature matrix. The input is the reduced-dimensional conventional feature matrix, the corresponding feature components of the vibration unconventional feature matrix, the preset high-frequency threshold, and the preset low-frequency threshold. The output is the feature correlation coefficient matrix.

4. The method for collecting and storing vibration data of an engine according to claim 3, characterized in that: The characteristic contribution equation is used to evaluate the influence of characteristic components on the engine vibration state. The input is the characteristic correlation coefficient matrix, the vibration energy distribution data, a preset importance threshold, and a preset attenuation coefficient. The output is the characteristic contribution matrix.

5. The method for collecting and storing vibration data of an engine according to claim 4, characterized in that: The feature weight equation is used to determine the weight of the feature component in the fusion process, and the input is the feature contribution matrix, the preset feature threshold, the feature correlation coefficient matrix, and the preset weight coefficient, and the output is the feature weight coefficient.

6. The method for collecting and storing vibration data of an engine according to claim 5, characterized in that: The feature fusion equation is used to realize weighted fusion of feature components. The input is the dimension reduction conventional feature matrix, the vibration unconventional feature matrix, the feature weight coefficient, and the preset feature combination parameter. The output is the vibration comprehensive feature matrix.

7. The method for collecting and storing vibration data of an engine according to claim 6, characterized in that: The characteristic evaluation equation is used to evaluate the vibration characteristic processing effect and determine the parameter adjustment strategy. The input is the vibration comprehensive characteristic matrix, the preset evaluation threshold, the historical iteration number, and the preset sampling rate. The output is the parameter adjustment indication flag.

8. The method for collecting and storing vibration data of an engine according to claim 7, characterized in that: The filtering parameter update equation is used to dynamically adjust the filtering parameters according to the feature processing effect. The input is the feature correlation coefficient matrix, the preset high-frequency threshold, the preset low-frequency threshold, and the preset update step coefficient. The output is the updated high-frequency threshold and the updated low-frequency threshold.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the engine vibration data acquisition and storage method according to any one of claims 1 to 8.

10. A vibration data collection and storage system for an engine, characterized in that: A computer-readable storage medium comprising the computer-readable storage medium of claim 9.

Images