High-rotating-speed bearing lubrication temperature rise data processing method and system

Abstract

The invention belongs to the field of data processing, and particularly relates to a high-rotating-speed bearing lubrication temperature rise data processing method and system. The method comprises the steps of obtaining original lubrication temperature rise time sequence data of the high-rotation-speed bearing, and performing N-layer decomposition through fractional order B-spline wavelets to obtain corresponding approximation coefficients and detail coefficients; estimating the noise standard deviation of each layer according to the detail coefficient of the first layer, and adaptively matching the double thresholds of the detail coefficient of each layer in combination with the energy concentration ratio and the interlayer correlation coefficient; processing the detail coefficient of each layer by adopting a nonlinear threshold function, constructing a self-adaptive Savitzky-Golay filter, and carrying out smoothing processing on an approximation coefficient; and obtaining initial temperature rise data from the reconstructed coefficient, constructing an extended Kalman filter based on a dynamic heat production physical model, and outputting final temperature rise data through state estimation and updating. According to the method, the problems that real temperature rise data of the high-rotation-speed bearing is difficult to accurately obtain in a traditional data processing mode, and the analysis of the real operation state of the high-rotation-speed bearing is inaccurate are solved.

Description

Technical Field

[0001] This invention belongs to the field of data processing, and in particular relates to a method and system for processing data on the temperature rise of lubrication in high-speed bearings. Background Technology

[0002] High-speed bearings, as supporting components of high-end equipment, significantly impact the performance and reliability of the entire system. Bearing lubrication and temperature rise are key physical quantities characterizing bearing operating conditions. Accurate monitoring and analysis of temperature rise data are crucial for predicting early bearing failures, optimizing lubrication strategies, and ensuring equipment safety.

[0003] In real industrial environments, the temperature rise signal of high-speed bearings is typically weak and susceptible to severe interference from various complex noise sources such as mechanical vibration, electromagnetic environment, and transmission system impact. This noise couples with the actual temperature fluctuation signal, forming a typical non-stationary low signal-to-noise ratio signal. Due to noise interference with the raw data, direct analysis using the raw data makes it difficult to accurately extract effective information reflecting the true temperature change patterns of high-speed bearings. While filtering methods such as moving average filtering or low-pass filtering can smooth the data to some extent, their processing lacks flexibility and adaptability, easily obscuring the detailed characteristics of rapid temperature changes, or introducing signal delay and distortion during noise removal, failing to meet the requirements of high-precision analysis.

[0004] Wavelet transform technology possesses excellent time-frequency localization characteristics and multi-resolution analysis capabilities, effectively handling non-stationary signals. Denoising methods based on wavelet transform, after wavelet decomposition of the signal, process wavelet coefficients at various scales by setting appropriate thresholds to distinguish between signal and noise. However, wavelet transform-based denoising methods still have several limitations: In the selection of threshold functions, hard threshold functions produce discontinuous results and are prone to oscillations; soft threshold functions, while continuous, produce constant deviations for larger signal coefficients, causing signal amplitude attenuation and loss of detail. Furthermore, in determining the threshold, general or fixed threshold rules ignore the inherent characteristics of the signal and the differences in noise distribution at different scales, resulting in poor adaptability and difficulty in achieving an ideal balance between denoising and fidelity. Most existing methods remain at the pure signal processing level, failing to incorporate prior knowledge of the physical processes of bearing heat generation and dissipation into the data processing flow. This leads to potential deviations between the denoising results and the actual physical laws of the system, limiting the physical interpretability and accuracy of the analysis results. Summary of the Invention

[0005] Therefore, the purpose of this invention is to provide a method and system for processing high-speed bearing lubrication temperature rise data, so as to solve the technical problem that traditional data processing methods are difficult to accurately obtain the true temperature rise data of high-speed bearings, which easily leads to inaccurate analysis results of the true operating state of high-speed bearings.

[0006] To solve the above problems, the technical solution of the present invention, a method for processing high-speed bearing lubrication temperature rise data, is as follows: A method for processing lubrication temperature rise data of high-speed bearings includes the following steps: Obtain the original lubrication temperature rise time series data of high-speed bearings under operating conditions; use a fractional B-spline wavelet basis of a preset order to perform N-level wavelet decomposition on the data to obtain N-level approximation coefficients and N-level detail coefficients. Based on the detail coefficients of the first layer, the noise standard deviation of each layer is estimated using the median absolute deviation method; for each detail coefficient, its energy concentration is calculated; for detail coefficients with more than 1 layer, the interlayer correlation coefficient between the detail coefficient and the detail coefficients of adjacent high-frequency layers is calculated; combining the energy concentration, interlayer correlation coefficient, and noise standard deviation, two different threshold values, the first threshold and the second threshold, are determined for each detail coefficient. A nonlinear threshold function is used to process the detail coefficients of each layer. Then, the global signal-to-noise ratio is estimated based on the noise standard deviation between the original data energy and the first layer detail coefficients. The parameters of the Savitzky-Golay filter are adaptively determined based on the global signal-to-noise ratio, and the Nth layer approximation coefficients are smoothed. N-level wavelet inverse reconstruction is performed using the processed detail coefficients and smoothed approximation coefficients to obtain preliminary temperature rise data. A nonlinear lumped parameter thermal network model of the high-speed bearing is established, and parameters are identified using the preliminary temperature rise data to obtain a dynamic heat generation physical model that matches the current operating conditions. An extended Kalman filter is constructed with the dynamic heat generation physical model as the state prediction model and the preliminary temperature rise data as the measurement value. The final temperature rise data is obtained through the state estimation and update process of the extended Kalman filter.

[0007] Furthermore, define For the first Layer detail factor, It is a positive integer. For the first Standard deviation of noise per layer; Standard deviation of noise per layer The methods for obtaining it include: Calculate the median absolute deviation of the first level detail coefficients. The calculation formula is: ; Based on the median absolute deviation of the obtained first-level detail coefficients Using the formula Calculate the standard deviation of the first layer of noise ; Based on the statistical law of the approximate attenuation of Gaussian white noise energy between layers in wavelet decomposition, the standard deviation of noise in subsequent layers is estimated using the following formula: ,in The preset attenuation constant is greater than 1. The number of layers, and ; This indicates the median operation.

[0008] Furthermore, define For the processed first Layer detail factor, For the first The first threshold of the layer detail factor, For the first The second threshold for the layer detail coefficients; the method of processing the layer detail coefficients using a nonlinear threshold function is as follows: when hour, ;when hour, ;otherwise, Where k, α, and β are preset parameters.

[0009] Furthermore, the first Energy concentration of layer detail factor The calculation formula is: ,in, Let q be the detail coefficients of the qth level, where q = 1, 2, ..., N; Represents the L2 norm; No. Layer detail coefficient and the first Interlayer correlation coefficient of layer detail coefficient The calculation formula is: ,in For covariance, Let the standard deviation be set. ; The first threshold corresponding to the detail level of each layer Second threshold The calculation method is as follows: First combine the first Energy concentration of layer detail factor and the Layer detail coefficient and the first Interlayer correlation coefficient of layer detail coefficient Calculate the adaptive factor The calculation formula is: ; Then use the formula Calculate the first The first threshold corresponding to the layer detail factor ,in, For the first The length of the layer detail factor, It is the square root; Using formula Calculate the first The second threshold corresponding to the layer detail factor ,in, The preset multiplier is greater than 1.

[0010] Furthermore, the preset parameters k, α, and β in the nonlinear threshold function satisfy the following condition: 0.5 <k<1;1<α≤3;1≤β≤2。

[0011] Furthermore, the construction process of the Savitzky-Golay filter includes: Based on the noise standard deviation of the original data energy and the first layer detail coefficient. Estimate the global signal-to-noise ratio ; Preset first signal-to-noise ratio threshold Second signal-to-noise ratio threshold ,and ; Based on global signal-to-noise ratio Compared with the first signal-to-noise ratio threshold Second signal-to-noise ratio threshold The comparison results determine the window length M and polynomial order P of the Savitzky-Golay filter piecewise: when When, it is set as the first parameter group (M1, P1); when At that time, it is set as the second parameter group (M2, P2); when At that time, it is set as the third parameter group (M3, P3); M1 <M2<M3,P1> P2>P3, and the window length M is an odd number greater than the polynomial order P.

[0012] Furthermore, the process of establishing a nonlinear lumped-parameter thermal network model for the high-speed bearing and identifying parameters using the preliminary temperature rise data includes: Establish a lumped parameter thermal network model that includes at least three thermal nodes from the inner ring, outer ring, rolling elements, cage, and lubricant; Based on the actual heat generation mechanism of high-speed bearings, heat source terms are set in the lumped parameter heat network model. The heat generation power of high-speed bearings is related to the speed and / or load of high-speed bearings. Using the least squares method, with the goal of minimizing the sum of squared errors between the temperature rise predicted by the lumped parameter thermal network model and the preliminary temperature rise data, the equivalent thermal resistance and equivalent heat capacity parameters in the lumped parameter thermal network model are identified.

[0013] Furthermore, the construction process of the extended Kalman filter includes: Construct a state vector and use its components as the temperature of each thermal node in the lumped parameter thermal network model; The discretized set of nonlinear differential equations representing the temperature change of each thermal node over time is used as the state transition equation of the extended Kalman filter. The temperature sequence corresponding to the outer ring or other measurable location nodes in the preliminary temperature rise data is used as the measurement vector of the extended Kalman filter, and the measurement matrix is ​​set according to the measurement vector.

[0014] Furthermore, the number of wavelet decomposition layers N is determined based on the signal length of the original lubrication temperature rise time series data.

[0015] Furthermore, the technical solution of the high-speed bearing lubrication temperature rise data processing system provided by this invention is as follows: A high-speed bearing lubrication temperature rise data processing system includes a processor and a memory. The memory stores a computer program, and the processor executes the computer program to implement the above-mentioned high-speed bearing lubrication temperature rise data processing method.

[0016] The beneficial effects of this invention are: The high-speed bearing lubrication temperature rise data processing method of this invention effectively filters out noise such as mechanical vibration and electromagnetic interference by fractional-order B-spline wavelet decomposition, setting adaptive dual thresholds for each decomposition layer, and combining them with a nonlinear threshold function, while retaining the detailed features in the temperature rise signal. In addition, this invention adapts to high / low signal-to-noise ratio data under different operating conditions of high-speed bearings by adjusting the global signal-to-noise ratio adaptively, improving the versatility and accuracy of the preliminary temperature rise data. Furthermore, this invention introduces a dynamic heat generation physical model of the bearing and extended Kalman filtering, combining pure signal processing with the thermodynamic principles of the bearing, so that the final output temperature rise data conforms to the actual temperature change law of the high-speed bearing, improving the accuracy of the analysis results of the actual operating state of the high-speed bearing, and thus helping to improve the reliability of equipment operation. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the steps of a high-speed bearing lubrication temperature rise data processing method according to the present invention. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0019] A specific embodiment of the high-speed bearing lubrication temperature rise data processing method of the present invention: like Figure 1 As shown, a method for processing high-speed bearing lubrication temperature rise data includes the following steps: S1. Obtain the original lubrication temperature rise time series data of the high-speed bearing under operating conditions; use a fractional B-spline wavelet basis of preset order to perform N-level wavelet decomposition on the data to obtain N-level approximation coefficients and N-level detail coefficients.

[0020] In this step, a K-type thermocouple temperature sensor is placed on the outer ring surface of the high-speed bearing. The sampling frequency of the data acquisition card is set to 100 Hz, and the temperature signal of the high-speed bearing is continuously collected from startup, stable operation to shutdown, forming a time series containing tens of thousands of data points, which serves as the raw lubrication temperature rise time series data. This raw lubrication temperature rise time series data accurately reflects the temperature changes of the high-speed bearing, but it also contains a large amount of noise interference from sources such as mechanical vibration and electromagnetic environment. Since the raw lubrication temperature rise time series data is a mixture of noise and real temperature signals, the excellent time-frequency analysis capability of wavelet transform is used to separate the signal at different frequency scales. Among them, the approximation coefficient represents the low-frequency part of the signal, that is, the overall trend and outline of the temperature rise; the detail coefficient represents the high-frequency part of the signal, which contains the details of rapid temperature fluctuations and most of the noise.

[0021] In an optional embodiment, a fractional-order B-spline function is constructed, where the order v is a preset non-integer, such as v=0.5. The fractional-order B-spline function is defined by the Fourier transform. Based on this fractional-order B-spline function, corresponding wavelet functions and scaling functions are generated. The number of decomposition levels N is determined according to the signal length L. The Mallit fast wavelet transform algorithm is applied, and the generated filter bank is used to decompose the original data obtained in S1 layer by layer until the Nth level, obtaining the approximation coefficients of the Nth level and the detail coefficients from the 1st to the Nth level.

[0022] S2, based on the detail coefficients of the first layer, estimate the noise standard deviation of each layer using the median absolute deviation method; calculate the energy concentration of each layer detail coefficient; for detail coefficients with more than 1 layer, calculate the interlayer correlation coefficient between the detail coefficient and the detail coefficients of adjacent high-frequency layers; combining the energy concentration, interlayer correlation coefficient and noise standard deviation, determine a first threshold and a second threshold for each layer detail coefficient, with the second threshold being greater than the first threshold.

[0023] In this step, since noise is mainly concentrated in the first-level detail coefficients at high frequencies, the noise level can be estimated based on these coefficients. The median absolute deviation method is used to avoid interference from accidental outliers in the data, thus obtaining a more accurate noise standard deviation. This step overcomes the shortcomings of traditional fixed threshold methods, allowing for a more accurate distinction between noise and valid signals.

[0024] In an optional embodiment, define For the first Layer detail factor, It is a positive integer. For the first The first threshold of the layer detail factor, For the first The second threshold of the layer detail factor, For the first The noise standard deviation of the layer.

[0025] Based on the first level of detail coefficient The standard deviation of noise for each layer was estimated using the median absolute deviation method. The methods include: Calculate the median absolute deviation of the first level detail coefficients. The calculation formula is: ; Based on the median absolute deviation of the obtained first-level detail coefficients Using the formula Calculate the standard deviation of the first layer of noise ; Based on the statistical law of the approximate attenuation of Gaussian white noise energy between layers in wavelet decomposition, the standard deviation of noise in subsequent layers is estimated using the following formula: ,in, The preset attenuation constant is greater than 1. The number of layers, and , This indicates the median operation.

[0026] Using the median absolute deviation method to assess noise levels in a signal is a more robust statistical method than the standard deviation method because it is less susceptible to the influence of individual extreme outliers in the data. For example, for a set of first-level detail coefficients containing ten data points... Suppose the values ​​of the ten data points are 0.1, 0.2, -0.1, 5.8, 0.3, -0.2, 0.1, 0.2, -0.3, and 0.1. The value of 5.8 is a clear outlier caused by a random shock. The median of this data set is calculated to be 0.15. The absolute deviation of the median is obtained by calculating the absolute value of the difference between each data point and the median. The value is 0.15. Based on the formula for the noise standard deviation, the noise standard deviation of the first layer can be calculated. It is approximately equal to 0.222. The constant 0.6745 in the formula for the noise standard deviation is a correction factor that makes the result of the median absolute deviation statistically equivalent to the standard deviation for Gaussian distributed data.

[0027] Obtaining the noise standard deviation of the first layer Then, an important characteristic of wavelet decomposition can be used to estimate the noise standard deviation of other layers. This important characteristic is that for Gaussian white noise signals, the signal energy decays systematically with increasing decomposition layer number. This decay is described by an attenuation constant C. Only one accurate estimate is needed in the first layer where the noise is most significant, and the noise levels of all subsequent decomposition layers can be calculated. This improves computational efficiency and ensures consistency in noise estimation across layers.

[0028] In an optional embodiment, the first Energy concentration of layer detail factor The calculation formula is: ,in, Let q be the detail coefficients of the qth level, where q = 1, 2, ..., N; Represents the L2 norm; No. Layer detail coefficient and the first Interlayer correlation coefficient of layer detail coefficient The calculation formula is: ,in For covariance, Let the standard deviation be set. ; The first threshold corresponding to the detail level of each layer Second threshold The calculation method is as follows: First combine the first Energy concentration of layer detail factor and the Layer detail coefficient and the first Interlayer correlation coefficient of layer detail coefficient Calculate the adaptive factor The calculation formula is: ; Then use the formula Calculate the first The first threshold corresponding to the layer detail factor ,in, For the first The length of the layer detail factor, It is the square root; Using formula Calculate the first The second threshold corresponding to the layer detail coefficient, where, The preset multiplier is greater than 1.

[0029] An adaptive factor is dynamically generated for each decomposition layer. Adjust the values ​​of the first and second thresholds. Energy concentration. It reflects the proportion of the signal energy of this layer in the total signal energy. If the energy norm value of the second-layer detail coefficients is 50, and the sum of the energy norm values of all-layer detail coefficients is 100, then the energy concentration of the second-layer detail coefficients is 0.5, indicating that the possibility of this layer containing important signal information is very high. The inter-layer correlation coefficient measures the similarity of the signal characteristics between adjacent layers. A correlation coefficient close to 1, for example is 0.8, indicating that the signal characteristics are continuous and stable at different scales rather than random noise. The adaptive factor combines these two. For the above example is approximately 2.97. This relatively large adaptive factor means that the signal-to-noise ratio of the second-layer signal is relatively high.

[0030] After obtaining the adaptive factor, it is used to adjust a basic threshold, which is a general threshold calculated based on the noise standard deviation and data length. The adjusted threshold is higher than that without using the adaptive factor, and it can more effectively retain strong signals and remove noise. It is used in the subsequent non-linear threshold function for fine denoising of the signal.

[0031] S3, Process the detail coefficients of each layer using a non-linear threshold function, and then estimate the global signal-to-noise ratio based on the original data energy and the noise standard deviation of the first-layer detail coefficients; adaptively determine the parameters of the Savitzky-Golay filter according to the global signal-to-noise ratio, and smooth the approximation coefficients of the Nth layer.

[0032] In an optional embodiment, the method for processing the detail coefficients of each layer using a non-linear threshold function is: Define as the processed detail coefficients of the layer; When , ; when , ; otherwise, ; where k, α, and β are preset parameters.

[0033] In an alternative embodiment, when , , to prevent the signs from being opposite before and after processing.

[0034] In an optional embodiment, the preset parameters k, α, and β in the non-linear threshold function satisfy the following conditions: 0.5 < k < 1; 1 < α ≤ 3; 1 ≤ β ≤ 2. The three parameters k, α, and β jointly determine the specific form of the non-linear threshold function and control the processing method for wavelet coefficients of different magnitudes. The parameter k mainly acts on the numerical values between the first threshold and the second threshold The coefficient between these values ​​controls the degree of compression for these smaller, but potentially not noise, coefficients. For example, choosing k as 0.6, which is just greater than the first threshold... The coefficients are not directly set to zero after processing, but are reduced to a smaller value. Compared to traditional hard or soft thresholding functions, this is smoother and can effectively preserve weak feature details of the signal, avoiding the introduction of distortions such as ringing in the reconstructed signal. Parameters α and β mainly affect those values ​​greater than the second threshold. The coefficients are typically those of strong signals. The parameter α controls how quickly the processed coefficients approximate their original values. A larger α value, such as 2.8, makes the function curve steeper, allowing for smaller adjustments to these strong signal coefficients and better preserving the original signal amplitude. The parameter β fine-tunes the overall function, affecting its nonlinearity. For example, choosing β as 1.2 allows for further optimization of the function's compression characteristics based on α. ​​By limiting these three parameters within a given experimental optimal range, a good balance can be achieved between maintaining signal continuity and preserving important features, resulting in a higher quality signal after denoising.

[0035] In an optional embodiment, the construction process of the Savitzky-Golay filter includes: Based on the noise standard deviation of the original data energy and the first layer detail coefficient. Estimate the global signal-to-noise ratio ; Preset first signal-to-noise ratio threshold Second signal-to-noise ratio threshold ,and ; Based on global signal-to-noise ratio Compared with the first signal-to-noise ratio threshold Second signal-to-noise ratio threshold The comparison results determine the window length M and polynomial order P of the Savitzky-Golay filter piecewise: when When, it is set as the first parameter group (M1, P1); when At that time, it is set as the second parameter group (M2, P2); when At that time, it is set as the third parameter group (M3, P3); M1 <M2<M3,P1> P2>P3, and the window length M is an odd number greater than the polynomial order P.

[0036] In this step, based on the overall signal quality, the most suitable filter parameters are intelligently selected to smooth the low-frequency trend components of the signal, which are the Nth-level approximation coefficients. This is achieved by calculating the total energy of the original signal and the standard deviation of the first-level noise. The ratio of these two values ​​can be used to obtain a global signal-to-noise ratio. The estimated value. For example, if the global signal-to-noise ratio... Approximately 38 dB, with two preset signal-to-noise ratio thresholds, such as a high threshold. 30 decibels, low threshold It is 15 dB, used to classify signal quality into three levels: high, medium, and low.

[0037] Based on the calculated global signal-to-noise ratio (SNR) range, a set of preset Savitzky-Golay filter parameters is selected. In the example above, the SNR is higher than the high threshold, indicating a high-quality signal. In this case, the first parameter group is chosen, for example, a window length M1 of 5 and a polynomial order P1 of 3. This short window and high-order fitting can accurately follow subtle changes in the signal, preserving the details of the original trend to the maximum extent. Conversely, if the calculated SNR is 12 dB, lower than the low threshold, it indicates that the signal is severely contaminated by noise. The third parameter group is then selected, for example, a window length M3 of 31 and a polynomial order P3 of 1. The long window and low-order linear fitting provide strong smoothing, effectively filtering out noise and extracting the core signal trend. This avoids a one-size-fits-all filtering approach for all signals, achieving personalized and optimized processing for signals of different quality.

[0038] S4. Using the processed detail coefficients and smoothed approximation coefficients, N-level wavelet inverse reconstruction is performed to obtain preliminary temperature rise data. A nonlinear lumped parameter thermal network model of the high-speed bearing is established, and the parameters are identified using the preliminary temperature rise data to obtain a dynamic heat generation physical model that matches the current operating conditions. An extended Kalman filter is constructed with the dynamic heat generation physical model as the state prediction model and the preliminary temperature rise data as the measurement value. The final temperature rise data is obtained through the state estimation and update process of the extended Kalman filter.

[0039] Specifically, this step involves establishing a nonlinear lumped-parameter thermal network model of the high-speed bearing and identifying parameters using preliminary temperature rise data. Establish a lumped parameter thermal network model that includes at least three thermal nodes from the inner ring, outer ring, rolling elements, cage, and lubricant; Based on the actual heat generation mechanism of high-speed bearings, heat source terms are set in the lumped parameter heat network model. The heat generation power of high-speed bearings is related to the speed and / or load of high-speed bearings. Using the least squares method, with the goal of minimizing the sum of squared errors between the temperature rise predicted by the lumped parameter thermal network model and the preliminary temperature rise data, the equivalent thermal resistance and equivalent heat capacity parameters in the lumped parameter thermal network model are identified.

[0040] To create a simplified yet accurate lumped-parameter thermal network that reflects the thermodynamic behavior of high-speed bearings, instead of directly solving complex partial differential equations, the key components of the high-speed bearing are abstracted into several thermal nodes. For example, the inner ring, outer ring, and rolling elements are considered as three independent nodes, each with a single temperature. The generation of heat at these nodes is determined based on physical principles. For instance, for a high-speed bearing operating at 12,000 rpm and a load of 1,000 Newtons, empirical formulas can be used to calculate that the heat power generated by the friction between the rolling elements and the raceway is 50 watts, and the heat power generated by the shearing of the lubricant is 30 watts. These heat powers are then input as heat sources to the corresponding nodes in the lumped-parameter thermal network model.

[0041] Preliminary temperature rise data obtained from preprocessing, such as the measured outer ring temperature curve over time, is used to calibrate these unknown parameters. Using the least squares method, the program repeatedly adjusts the thermal resistance and heat capacity values ​​in the lumped-parameter thermal network model, calculating the difference between the predicted outer ring temperature and the measured temperature. The program iterates until a set of parameters is found that best matches the temperature rise curve predicted by the lumped-parameter thermal network model and the actual measured temperature rise curve, minimizing the sum of squared errors between them. Through this identification process, a customized model capable of accurately simulating the thermal characteristics of high-speed bearings under specific operating conditions is obtained.

[0042] The lumped-parameter thermal network model described above is nonlinear and is used to represent heat transfer and temperature distribution in complex systems. It divides the actual continuous thermal system into several discrete regions with uniform temperatures, i.e., thermal nodes or lumped parameters, and simulates heat transfer and heat storage between nodes using thermal resistance and heat capacity. Its nonlinearity is reflected in the fact that some heat transfer coefficients, heat capacities, or heat source terms in the model are not constant but dynamically change with system state variables such as temperature, flow rate, and power, resulting in a nonlinear system of differential equations describing the system behavior. Specific lumped-parameter thermal network models are existing technology and will not be elaborated further.

[0043] In an optional embodiment, the construction process of the extended Kalman filter includes: Construct a state vector and use its components as the temperature of each thermal node in the lumped parameter thermal network model; The discretized set of nonlinear differential equations representing the temperature change of each thermal node over time is used as the state transition equation of the extended Kalman filter. The temperature sequence corresponding to the outer ring or other measurable location nodes in the preliminary temperature rise data is used as the measurement vector of the extended Kalman filter, and the measurement matrix is ​​set according to the measurement vector.

[0044] The Extended Kalman Filter (EKF) can be used to track and predict the temperatures of key components inside high-speed bearings in real time, even those whose temperatures cannot be directly measured. All node temperatures to be tracked are combined into a state vector. For example, for a four-node model including the inner ring, outer ring, rolling elements, and cage, the state vector is a four-dimensional column vector containing the temperatures of these four nodes. The nonlinear differential equations corresponding to the lumped-parameter thermal network model identified in the previous step are discretized in time to obtain a set of difference equations. These difference equations describe how to predict the temperature at the next moment given the current temperatures of all nodes, forming the core of the EKF: the state transition equations.

[0045] The actual measurable temperature data is incorporated into the Extended Kalman Filter (EKF) to correct the model's predictions. In most applications, only the temperature of the bearing's outer ring can be measured; the time series of the outer ring temperature serves as the EKF's measurement vector. To inform the EKF which component of the state vector the measured value corresponds to, a measurement matrix is ​​defined. In the example of the four-node model above, if the second component of the state vector is the outer ring temperature, then the measurement matrix is ​​a matrix with 1 rows and 4 columns. Therefore, at each time step, the EKF can use the actual measured temperature of the bearing's outer ring to correct the predicted values ​​of the four node temperatures, achieving accurate and real-time estimation of the temperatures of core components, including the inner ring and rolling elements, which cannot be directly measured.

[0046] The present invention provides a high-speed bearing lubrication temperature rise data processing system, including a processor and a memory. The memory stores a computer program, and the processor executes the computer program to implement a high-speed bearing lubrication temperature rise data processing method in the above embodiments.

[0047] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as resistive random access memory (RRAM), dynamic random access memory (DRAM), static random access memory (SRAM), enhanced dynamic random access memory (EDRAM), high-bandwidth memory (HBM), hybrid memory cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented by computer-readable / executable instructions stored or otherwise maintained on such a computer-readable medium.

[0048] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A method for processing lubrication temperature rise data of high-speed bearings, characterized in that, Includes the following steps: Obtain the original lubrication temperature rise time series data of high-speed bearings under operating conditions; use a fractional B-spline wavelet basis of a preset order to perform N-level wavelet decomposition on the data to obtain N-level approximation coefficients and N-level detail coefficients. Based on the detail coefficients of the first layer, the noise standard deviation of each layer is estimated using the median absolute deviation method; for each detail coefficient, its energy concentration is calculated; for detail coefficients with more than 1 layer, the interlayer correlation coefficient between the detail coefficient and the detail coefficients of adjacent high-frequency layers is calculated. Based on the combined energy concentration, interlayer correlation coefficient, and noise standard deviation, two different threshold values, a first threshold and a second threshold, are determined for the detail coefficient of each layer. A nonlinear threshold function is used to process the detail coefficients of each layer. Then, the global signal-to-noise ratio is estimated based on the noise standard deviation between the original data energy and the first layer detail coefficients. The parameters of the Savitzky-Golay filter are adaptively determined based on the global signal-to-noise ratio, and the Nth layer approximation coefficients are smoothed. N-level wavelet inverse reconstruction is performed using the processed detail coefficients and smoothed approximation coefficients to obtain preliminary temperature rise data. A nonlinear lumped parameter thermal network model of the high-speed bearing is established, and parameters are identified using the preliminary temperature rise data to obtain a dynamic heat generation physical model that matches the current operating conditions. An extended Kalman filter is constructed with the dynamic heat generation physical model as the state prediction model and the preliminary temperature rise data as the measurement value. The final temperature rise data is obtained through the state estimation and update process of the extended Kalman filter.

2. The method for processing high-speed bearing lubrication temperature rise data according to claim 1, characterized in that, definition For the first Layer detail factor, It is a positive integer. For the first Standard deviation of noise per layer; Standard deviation of noise per layer The methods for obtaining it include: Calculate the median absolute deviation of the first level detail coefficients. The calculation formula is: ; Based on the median absolute deviation of the obtained first-level detail coefficients Using the formula Calculate the standard deviation of the first layer of noise ; Based on the statistical law of the approximate attenuation of Gaussian white noise energy between layers in wavelet decomposition, the standard deviation of noise in subsequent layers is estimated using the following formula: ,in The preset attenuation constant is greater than 1. The number of layers, and ; This indicates the median operation.

3. The method for processing high-speed bearing lubrication temperature rise data according to claim 2, characterized in that, definition For the processed first Layer detail factor, For the first The first threshold of the layer detail factor, For the first The second threshold for the layer detail coefficients; the method of processing the layer detail coefficients using a nonlinear threshold function is as follows: when hour, ;when hour, ;otherwise, Where k, α, and β are preset parameters.

4. The method for processing high-speed bearing lubrication temperature rise data according to claim 3, characterized in that, No. Energy concentration of layer detail factor The calculation formula is: ,in, Let q be the detail coefficients of the qth level, where q = 1, 2, ..., N; Represents the L2 norm; No. Layer detail coefficient and the first Interlayer correlation coefficient of layer detail coefficient The calculation formula is: ,in For covariance, Let the standard deviation be set. ; The first threshold corresponding to the detail level of each layer Second threshold The calculation method is as follows: First combine the first Energy concentration of layer detail factor and the Layer detail coefficient and the first Interlayer correlation coefficient of layer detail coefficient Calculate the adaptive factor The calculation formula is: ; Then use the formula Calculate the first The first threshold corresponding to the layer detail factor ,in, For the first The length of the layer detail factor, It is the square root; Using formula Calculate the first The second threshold corresponding to the layer detail factor ,in, The preset multiplier is greater than 1.

5. A method for processing high-speed bearing lubrication temperature rise data according to claim 3 or 4, characterized in that, The preset parameters k, α, and β in the nonlinear threshold function satisfy the following condition: 0.5 <k<1;1<α≤3;1≤β≤2。 6. The method for processing high-speed bearing lubrication temperature rise data according to claim 1, characterized in that, The construction process of the Savitzky-Golay filter includes: Based on the noise standard deviation of the original data energy and the first layer detail coefficient. Estimate the global signal-to-noise ratio ; Preset first signal-to-noise ratio threshold Second signal-to-noise ratio threshold ,and ; Based on global signal-to-noise ratio Compared with the first signal-to-noise ratio threshold Second signal-to-noise ratio threshold The comparison results determine the window length M and polynomial order P of the Savitzky-Golay filter piecewise: when When, it is set as the first parameter group (M1, P1); when At that time, it is set as the second parameter group (M2, P2); when At that time, it is set as the third parameter group (M3, P3); M1 <M2<M3,P1> P2>P3, and the window length M is an odd number greater than the polynomial order P.

7. The method for processing high-speed bearing lubrication temperature rise data according to claim 1, characterized in that, The process of establishing a nonlinear lumped-parameter thermal network model for a high-speed bearing and identifying parameters using the preliminary temperature rise data includes: Establish a lumped parameter thermal network model that includes at least three thermal nodes from the inner ring, outer ring, rolling elements, cage, and lubricant; Based on the actual heat generation mechanism of high-speed bearings, heat source terms are set in the lumped parameter heat network model. The heat generation power of high-speed bearings is related to the speed and / or load of high-speed bearings. Using the least squares method, with the goal of minimizing the sum of squared errors between the temperature rise predicted by the lumped parameter thermal network model and the preliminary temperature rise data, the equivalent thermal resistance and equivalent heat capacity parameters in the lumped parameter thermal network model are identified.

8. The method for processing high-speed bearing lubrication temperature rise data according to claim 7, characterized in that, The construction process of the extended Kalman filter includes: Construct a state vector and use its components as the temperature of each thermal node in the lumped parameter thermal network model; The discretized set of nonlinear differential equations representing the temperature change of each thermal node over time is used as the state transition equation of the extended Kalman filter. The temperature sequence corresponding to the outer ring or other measurable location nodes in the preliminary temperature rise data is used as the measurement vector of the extended Kalman filter, and the measurement matrix is ​​set according to the measurement vector.

9. The method for processing high-speed bearing lubrication temperature rise data according to claim 1, characterized in that, The number of wavelet decomposition layers N is determined based on the signal length of the original lubrication temperature rise time series data.

10. A high-speed bearing lubrication temperature rise data processing system, characterized in that, It includes a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the high-speed bearing lubrication temperature rise data processing method according to any one of claims 1 to 9.

Images