Insulation bearing residual life prediction method and related equipment

Abstract

The invention belongs to the technical field of bearing state monitoring, and discloses an insulation bearing residual life prediction method and related equipment, and the method comprises the steps: obtaining a vibration signal of a target insulation bearing, and carrying out the preprocessing of the vibration signal; performing inner and outer ring collision process simulation based on an inner and outer ring collision dynamic model of the target insulated bearing to obtain collision physical characteristics, obtaining actual measurement signal characteristics according to the preprocessed vibration signals, and fusing the collision physical characteristics and the actual measurement signal characteristics to obtain a characteristic vector; on the basis of a principal component analysis method, dimension reduction processing is carried out on the feature vectors, and health indexes are obtained; according to the health indexes, establishing a health curve model based on a random degradation model fitting method; according to the health curve model and a preset failure threshold value, predicting the residual life of the target insulated bearing; therefore, the residual life prediction reliability and precision of the insulating bearing can be improved.

Description

Technical Field

[0001] This application relates to the field of bearing condition monitoring technology, and more specifically, to a method and related equipment for predicting the remaining life of insulated bearings. Background Technology

[0002] Bearings in electric vehicle drive motors are critical core components, accounting for approximately 30% to 40% of all drive motor failures. They directly impact vehicle performance and driving safety, and in severe cases, can lead to major accidents. Therefore, reliable condition monitoring and health assessment of motor bearings to ensure their safe and stable operation has become a key research direction in China. Currently, vibration signal-based monitoring methods have become the mainstream diagnostic approach due to their rich information and sensitive response. This method identifies the health status of bearings through signal analysis based on the specific characteristics of vibration signals exhibited during abnormal bearing wear, and can further construct health curves to predict remaining lifespan.

[0003] Existing technologies still have significant shortcomings in practical applications: Electric vehicles often face complex operating conditions such as acceleration, deceleration, and sudden braking, causing the drive system to endure uncertain road excitations and multi-physics coupled vibrations, exhibiting strong nonlinear, non-stationary, and multi-scale dynamic characteristics. This complex dynamic environment poses significant challenges to traditional vibration signal analysis methods in feature extraction, state identification, and lifespan prediction, making it difficult to achieve highly reliable and accurate monitoring and assessment. If these problems are not solved, abnormal wear and aging processes of bearings cannot be captured accurately and in a timely manner, potentially leading to component damage, system performance degradation, or even sudden failures.

[0004] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0005] The purpose of this application is to provide a method and related equipment for predicting the remaining life of insulated bearings, which can improve the reliability and accuracy of predicting the remaining life of insulated bearings and meet the monitoring needs of electric vehicles in complex dynamic environments.

[0006] In a first aspect, this application provides a method for predicting the remaining life of an insulated bearing, comprising the following steps: A1. Acquire the vibration signal of the target insulated bearing and preprocess the vibration signal; A2. Based on the collision dynamics model of the inner and outer rings of the target insulated bearing, the collision process of the inner and outer rings is simulated to obtain the collision physical characteristics, and the measured signal characteristics are obtained according to the preprocessed vibration signal. The collision physical characteristics and the measured signal characteristics are fused to obtain the feature vector. A3. Based on principal component analysis, the dimensionality of the eigenvectors is reduced to obtain health indicators; A4. Based on the aforementioned health indicators, establish a health curve model using a random degradation model fitting method; A5. Based on the health curve model and the preset failure threshold, predict the remaining life of the target insulated bearing.

[0007] Secondly, this application provides an electronic device including a processor and a memory, the memory storing a computer program executable by the processor, wherein when the processor executes the computer program, it performs the steps in the insulated bearing remaining life prediction method described above.

[0008] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the steps of the insulated bearing remaining life prediction method described above.

[0009] Beneficial effects: The method and related equipment for predicting the remaining life of insulated bearings provided in this application solve the problem of low reliability of vibration signal analysis under complex working conditions by acquiring vibration signals and preprocessing them, simulating to obtain collision physical characteristics, fusing measured signal characteristics, dimensionality reduction to obtain health indicators, establishing a health curve model and predicting the remaining life. It has the advantages of improving the reliability and accuracy of predicting the remaining life of insulated bearings. Attached Figure Description

[0010] Figure 1 A flowchart of a method for predicting the remaining life of an insulated bearing provided in this application.

[0011] Figure 2 A schematic diagram of the structure of the electronic device provided in this application.

[0012] Labeling explanations: 301, processor; 302, memory; 303, communication bus. Detailed Implementation

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0014] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0015] Please refer to Figure 1 A method for predicting the remaining life of an insulated bearing, as described in some embodiments of this application, includes the following steps: A1. Acquire the vibration signal of the target insulated bearing and preprocess the vibration signal; A2. Based on the collision dynamics model of the inner and outer rings of the target insulated bearing, the collision process of the inner and outer rings is simulated to obtain the collision physical characteristics. The measured signal characteristics are obtained based on the preprocessed vibration signal, and the collision physical characteristics and measured signal characteristics are fused to obtain the feature vector. A3. Based on principal component analysis, the dimensionality of the eigenvectors is reduced to obtain health indicators; A4. Based on health indicators, establish a health curve model using a random degradation model fitting method; A5. Based on the health curve model and preset failure threshold, predict the remaining life of the target insulated bearing.

[0016] An insulated bearing is a specially designed bearing with an insulating layer between its inner and outer rings or rollers to prevent current from passing through the bearing, thereby avoiding electro-corrosion and premature failure. It is commonly used in equipment with insulation requirements, such as electric vehicle drive motors.

[0017] The inner and outer ring collision dynamics model is a physical model that describes the collision behavior of the inner and outer rings of a bearing under specific operating conditions. This model can be used to simulate and analyze the mechanical response and energy transfer during the collision process, thereby obtaining the physical characteristics of the collision.

[0018] The proposed method for predicting the remaining life of insulated bearings constructs a more comprehensive and representative feature vector by integrating the collision physical characteristics obtained from simulations of inner and outer ring collision dynamics models with measured vibration signal characteristics. Compared to traditional methods that rely solely on measured signal characteristics, this method combines theoretical analysis with actual data, enabling it to more accurately capture bearing degradation information, especially in the complex dynamic environment of electric vehicle drive motors.

[0019] Furthermore, this application employs principal component analysis (PCA) to reduce the dimensionality of the fused feature vectors, obtaining a health indicator that effectively reduces data redundancy and extracts key information. This not only simplifies the model and improves computational efficiency but also allows the health indicator to reflect the overall health status of the bearing more concisely and intuitively. In the example above, PCA reduces the multidimensional feature vectors into a single health indicator, making the bearing's degradation trend readily apparent and facilitating subsequent health curve modeling.

[0020] Furthermore, this application establishes a health curve model based on a stochastic degradation model fitting method, taking into account the randomness of the degradation process. By determining the initial degradation point, exponentially smoothing the health indicators, and assuming that the smoothed health indicators follow a Wiener process, the maximum likelihood estimation method is used to fit the model parameters, making the established health curve model more robust and predictive.

[0021] Ultimately, this application can predict the expected and variance of the remaining life of a target insulated bearing based on a health curve model and a preset failure threshold, providing a quantitative and reliable basis for maintenance decisions. This predictive maintenance strategy can effectively avoid sudden failures, reduce maintenance costs, and improve the reliability and safety of equipment operation. Compared to traditional periodic maintenance or post-failure maintenance, the method in this application can achieve more accurate prediction of maintenance timing, thereby optimizing resource allocation and improving overall operational efficiency.

[0022] The vibration signal can be obtained by installing a vibration sensor, such as an accelerometer, on the insulated bearing.

[0023] In some implementations, preprocessing includes DC component removal, bandpass filtering, and normalization.

[0024] The DC component removal process aims to eliminate the time-invariant DC offset in the vibration signal. This offset, which may be caused by sensor bias, environmental factors, or inherent characteristics of the measurement system, will mask the true AC vibration components if not removed, affecting the dynamic range of the signal and the accuracy of subsequent analysis. This can be achieved by subtracting the average value of the original signal from its value, or by using a high-pass filter to remove extremely low-frequency components.

[0025] Bandpass filtering aims to selectively retain components within a specific frequency range of a vibration signal while suppressing noise and interference outside that range. For fault diagnosis of insulated bearings, the focus is typically on a specific frequency band related to the bearing fault frequency. For example, a bandpass frequency of 2kHz-10kHz can be set to filter out low-frequency environmental vibrations and high-frequency random noise, highlighting the impact signal caused by the bearing fault. Besides Butterworth filters, various digital filters such as Chebyshev filters or elliptic filters can also be used.

[0026] Standardization aims to adjust the amplitude of vibration signals to a uniform scale or distribution, eliminating the influence of differences in measurement conditions or sensors and ensuring the comparability of signals across different time points or bearings. This is crucial for subsequent feature extraction and model training. Common standardization methods include Z-score standardization (converting data to a distribution with a mean of 0 and a standard deviation of 1) or Min-Max standardization (scaling data to between 0 and 1). For example, vibration signals can be standardized using the following formula: ; Where t is time, The vibration data at time t are the standardized data. The data is the filtered vibration data at time t. This represents the maximum absolute value of the vibration data after filtering.

[0027] The preprocessing process begins with DC component removal, eliminating the DC offset from the original vibration signal to ensure the signal baseline is near zero. This eliminates static errors introduced by the sensor or acquisition system, allowing subsequent analysis to more accurately focus on the dynamic vibration characteristics of the bearing. The DC-removed signal then enters a bandpass filtering stage. Here, by setting a specific bandpass frequency range, such as 2kHz-10kHz, low-frequency environmental noise and high-frequency random interference unrelated to bearing failure are effectively filtered out, significantly improving the signal-to-noise ratio and highlighting the impact or periodic components related to insulated bearing failure. Finally, the filtered signal undergoes standardization, normalizing the signal amplitude to a uniform scale. This eliminates differences in signal amplitude between different measurement batches or under different operating conditions, ensuring the comparability and stability of signal characteristics. Through these preprocessing steps, the original vibration signal, which may contain noise and bias, is transformed into a clean, standardized signal with a high signal-to-noise ratio. This optimized signal, used as input for acquiring measured signal features in step A2, significantly improves the accuracy and reliability of feature extraction. This, in turn, makes the fusion of collision physical features and measured signal features more effective, laying a solid foundation for the subsequent construction of health indicators and prediction of remaining lifespan. This refined signal processing ensures the accuracy and robustness of the entire chain from raw vibration data to final lifespan prediction.

[0028] In some implementations, in step A2, the collision physical characteristics include at least the peak contact force, clearance change rate, and collision energy. The measured signal characteristics include time-domain characteristics, frequency-domain characteristics, and time-frequency-domain characteristics; Temporal features include at least the root mean square value and kurtosis; Frequency domain characteristics include at least the fault characteristic frequency amplitude; The time-frequency domain features include at least the envelope spectrum amplitude and the wavelet energy entropy.

[0029] Specifically, in step A2, the collision process between the inner and outer rings of the target insulated bearing is simulated based on the collision dynamics model of the inner and outer rings to obtain the collision physical characteristics. Collision physical characteristics refer to the physical quantities describing the internal collision dynamics of the bearing, obtained through simulation of the collision dynamics model of the inner and outer rings of the target insulated bearing. These characteristics can quantify the dynamic behavior inside the bearing at the physical mechanism level. The inner and outer ring collision dynamics model can be constructed using finite element analysis software, multibody dynamics simulation software, or theoretical mathematical models. This model can simulate the collision behavior of the inner and outer rings of the bearing under different operating conditions, thereby obtaining collision physical characteristics such as peak contact force, clearance change rate, and collision energy. These physical characteristics can theoretically reflect the damage state of the bearing. For example, the inner and outer ring collision dynamics model can be: ; in, Let be the inner ring acceleration at time t, and q be the number of rollers in the target insulated bearing. Let m be the radial deformation of the m-th roller. Let m be the position angle of the m-th roller. For contact function, For the gap, This represents the difference in displacement between the inner and outer rings. For symbolic functions, Let m be the weight of the m-th roller. Equivalent load-deformation coefficient It is the acceleration due to gravity; in, , , , The load-deformation coefficient of the inner ring. The load-deformation coefficient of the outer ring. The elastic modulus of the roller. Let be the elastic modulus of the inner raceway. Let be the elastic modulus of the outer raceway. For the Poisson's ratio of the roller, The Poisson's ratio of the inner raceway. The Poisson's ratio of the outer raceway. The coefficient of elasticity for contact deformation (e.g., for ball bearings). For cylindrical roller bearings, the ratio is 3 / 2. (10 / 9) For the raceway curvature and; in, ; in, , This represents the displacement of the outer ring at time t (referring to the displacement of the center point). This represents the displacement of the inner ring at time t (referring to the displacement of the center point). in, ; in, , For the initial clearance, The wear effect coefficient is... The wear amount can be obtained through simulation or theoretical calculation based on the historical speed and load data of the target insulated bearing.

[0030] Peak contact force refers to the maximum instantaneous value of the contact force when the inner and outer rings of a bearing collide. This value reflects the intensity of the impact inside the bearing and is an important indicator for assessing the degree of bearing damage. It can be obtained by monitoring the change of the normal force at the contact point or contact area over time in a dynamic simulation and recording its maximum instantaneous value; or by post-processing the simulation results to identify and extract local maxima in the contact force time series.

[0031] Clearance variation rate refers to the rate at which the radial or axial clearance between the inner and outer rings of a bearing changes over time. This parameter indicates the degree of bearing wear, loosening, or deformation and is a key parameter reflecting the bearing's health status. In dynamic simulations, it can be obtained by tracking the distance between the center points of the inner and outer rings or a specific reference point and calculating its derivative with respect to time; or by analyzing the relative displacements of the bearing components during the simulation to quantify the dynamic trend of clearance variation.

[0032] Collision energy refers to the energy dissipated or transferred by a system during the collision between the inner and outer rings of a bearing. This energy characterizes the severity of the collision and energy loss, and is closely related to bearing fatigue damage. In dynamic simulations, collision energy can be determined by calculating the integral of contact force and relative displacement, or by monitoring changes in the total energy of the system; alternatively, it can be determined by analyzing the energy dissipated by damping elements or contact damping in the simulation model.

[0033] Simultaneously, in step A2, measured signal features are extracted from the preprocessed vibration signal. For example, time-domain features such as root mean square (RMS) and kurtosis, reflecting the signal's energy and impact, can be extracted; frequency-domain features such as fault characteristic frequency amplitudes, correlated with specific bearing fault modes, can be extracted; and time-frequency-domain features such as envelope spectrum amplitude and wavelet energy entropy, revealing local variations in the signal over time and frequency, can be extracted. These collision physical features and measured signal features are then fused, for example, through simple concatenation, to form a comprehensive feature vector. This fusion strategy combines the advantages of theoretical simulation and actual measurement, enabling a more comprehensive and accurate description of the health status of insulated bearings, especially under complex operating conditions, compensating for the deficiencies of single features.

[0034] Measured signal characteristics refer to quantitative indicators extracted from actual vibration signals that reflect the bearing's operating state. These characteristics can comprehensively capture the nonlinear and non-stationary properties of vibration signals, providing data support for bearing condition assessment. For example, digital signal processing techniques can be used to perform time-domain, frequency-domain, or time-frequency-domain analysis on preprocessed vibration signals to extract corresponding statistics or spectral features; or machine learning or deep learning methods can be used to automatically learn and extract high-dimensional features from the original vibration signals.

[0035] The root mean square (RMS) value is the effective value of the vibration signal amplitude, reflecting the average energy level of the signal, and is often used to assess the stability of equipment operation. It can be obtained by averaging the squares of the signal time series and then taking the square root; or by directly measuring it using a hardware RMS converter.

[0036] Kurtosis is the peak-to-trend coefficient of the amplitude distribution of a vibration signal, reflecting the strength of the impact component in the signal and being sensitive to early faults. It can be obtained by calculating the ratio of the fourth-order central moment to the square of the variance of the signal; or by calculating the kurtosis of the signal using statistical analysis software.

[0037] The fault characteristic frequency amplitude refers to the amplitude of a specific frequency component associated with a fault in the bearing's inner or outer rings, rollers, or cage. This amplitude is directly related to the frequency response of a specific bearing defect, enhancing the accuracy of fault identification. It can be obtained by using Fourier transform to obtain the spectrum and then extracting the corresponding amplitude at the theoretically calculated fault characteristic frequency; or by using envelope demodulation technology to extract the low-frequency modulation component from the high-frequency carrier signal and then analyzing its spectrum.

[0038] Envelope spectrum amplitude refers to the spectral amplitude of the signal envelope obtained through envelope demodulation technology. It is used to extract fault features from the modulated signal, and is particularly suitable for early fault diagnosis of rolling bearings. It can be obtained by bandpass filtering the pre-processed vibration signal, performing a Hilbert transform to obtain an analytic signal, calculating its magnitude to obtain the envelope, and finally performing a Fourier transform on the envelope to obtain the envelope spectrum and extract the amplitude at a specific frequency; or by using synchronous averaging or order tracking techniques to process the signal at a specific rotational speed and then performing envelope analysis.

[0039] Wavelet energy entropy is a measure of the uniformity or complexity of the distribution of signal energy across different frequency bands based on wavelet decomposition. This entropy value quantifies the complexity of the signal's time-frequency distribution and has a good characterizing ability for the nonlinearity and non-stationarity of the signal. It can be obtained by performing multi-level wavelet decomposition on the signal, calculating the energy of each wavelet decomposition layer (frequency band), and then calculating the wavelet energy entropy according to the definition of Shannon entropy; or by using wavelet packet decomposition to obtain a finer frequency band division, and then calculating the entropy value of the energy in each frequency band.

[0040] This application's solution preprocesses the vibration signal of the target insulated bearing and simulates the collision process between the inner and outer races based on an inner and outer race collision dynamics model to obtain the collision physical characteristics. These collision physical characteristics, such as peak contact force, clearance change rate, and collision energy, quantify the dynamic behavior inside the bearing at the physical mechanism level, overcoming the shortcomings of traditional methods that rely solely on signal analysis. Simultaneously, measured signal characteristics are obtained from the preprocessed vibration signal. These characteristics, through multi-dimensional (time domain, frequency domain, and time-frequency domain) analysis, comprehensively capture the nonlinear, non-stationary, and multi-scale dynamic characteristics of the vibration signal. Time-domain characteristics, such as root mean square value and kurtosis, directly reflect the energy and impact of the signal; frequency-domain characteristics, such as fault characteristic frequency amplitude, are associated with specific fault frequencies; and time-frequency domain characteristics, such as envelope spectrum amplitude and wavelet energy entropy, are adept at handling transient and non-stationary signals. By fusing these physical simulation characteristics with multi-dimensional measured signal characteristics, a more comprehensive and representative feature vector is constructed. This fusion mechanism enables the feature vector to not only contain signal appearance information, but also incorporate deep physical mechanisms, thereby more accurately reflecting the actual health status changes of the bearing under complex working conditions, laying a solid foundation for the subsequent construction of health indicators and prediction of remaining life.

[0041] In some implementations, step A3 includes: A301. Standardize the feature vectors to obtain standardized feature vectors; A302. Based on principal component analysis, the dimensionality of the eigenvectors is reduced to obtain dimensionality-reduced eigenvectors; A303. Normalize the dimensionality-reduced feature vectors to obtain health indicators.

[0042] Specifically, in step A301, the standardization of the feature vector aims to eliminate differences in the dimensions and numerical ranges of the features, making them comparable. By transforming the original feature data to a uniform scale, it avoids certain features with larger values ​​dominating subsequent analysis, thus ensuring that the contribution of all features to the health indicators is fair and effective. One implementation is Z-score standardization, which involves subtracting the mean from each feature value and dividing by its standard deviation, resulting in a mean of 0 and a standard deviation of 1 for the processed data. Another implementation is Min-Max standardization, which linearly scales the feature values ​​to a preset range, such as [0,1] or [-1,1]. This is typically achieved by subtracting the minimum value and dividing by the difference between the maximum and minimum values.

[0043] In step A302, dimensionality reduction is performed on the eigenvectors using Principal Component Analysis (PCA) to obtain dimensionality-reduced eigenvectors. PCA is a commonly used multivariate statistical method used to transform a set of potentially correlated variables into a set of linearly uncorrelated variables through orthogonal transformations. These uncorrelated variables are called principal components. The purpose of dimensionality reduction is to reduce the number of features while retaining most of the information in the original data, thereby reducing model complexity, computational load, and removing noise and redundant information from the data. Through PCA, the most important directions of change in the data (i.e., principal components) can be identified, and the data can be projected onto these principal components to obtain a feature representation with lower dimensionality but minimal information loss. For example, the covariance matrix of the standardized eigenvectors can be calculated, and then its eigenvalues ​​and eigenvectors can be solved. The eigenvectors corresponding to the largest eigenvalues ​​can be selected as principal components, and the standardized eigenvectors can be projected onto these principal components to obtain dimensionality-reduced eigenvectors. Alternatively, nonlinear dimensionality reduction methods such as kernel principal component analysis (KPCA) can be used to handle nonlinear relationships in the data, thereby extracting complex features more effectively.

[0044] In step A303, the dimensionality-reduced feature vector is normalized to obtain the health index. The purpose of normalization is to map the dimensionality-reduced feature vector to a specific numerical range, making it easier to interpret and compare, and providing a unified and stable input for subsequent health curve modeling. This helps quantify the health state into an intuitive index, for example, mapping it to a range of 0 to 1, where 0 represents complete health and 1 represents complete failure. One implementation is to use the Sigmoid or Tanh function for non-linear normalization, compressing the values ​​of the dimensionality-reduced feature vector to the (0,1) or (-1,1) interval. This method has some robustness to outliers. Another implementation is to use linear normalization, such as Min-Max normalization, linearly scaling the values ​​of the dimensionality-reduced feature vector to the [0,1] interval. This makes the physical meaning of the health index clearer and facilitates the setting of failure thresholds.

[0045] This application proposes a method for predicting the remaining life of insulated bearings. After obtaining a feature vector that integrates collision physical characteristics and measured signal characteristics, this method first standardizes the feature vector to address the issue that the features may have different dimensions and scales. This step ensures that all features are numerically of the same order of magnitude, eliminating the impact of dimensional differences on subsequent analysis and laying the foundation for accurate extraction of health information. Subsequently, based on the standardized feature vector, principal component analysis is used for dimensionality reduction. This step identifies the most important directions of change in the data, mapping the high-dimensional feature space to a low-dimensional space, effectively removing redundant information and noise while retaining key information reflecting the bearing's health status. The dimensionality-reduced feature vector has a lower dimension but contains the main variation information of the original data, avoiding model overfitting or low computational efficiency caused by too many features or strong correlations. Finally, the dimensionality-reduced feature vector is normalized, mapping it to a preset numerical range with clear physical meaning, thus obtaining the final health index. This series of processing steps, from eliminating dimensional differences and extracting core information to unifying quantitative representation, progresses step by step to ensure that the constructed health indicators can accurately, stably, and effectively reflect the actual health status of the insulated bearings, providing reliable input for subsequent health curve modeling and remaining life prediction. Through this refined processing flow, this method can effectively address the nonlinear and non-stationary characteristics of vibration signals under complex operating conditions, making the health indicators extracted from multi-source features more robust and accurate, thereby significantly improving the reliability and stability of remaining life prediction for insulated bearings.

[0046] In some implementations, step A4 includes: A401. Obtain the relative root mean square value of the preprocessed vibration signal, and determine the initial decay point of the target insulated bearing based on the relative root mean square value; A402. Using exponential smoothing, the health indicators after the initial decline point are smoothed to obtain smoothed health indicators. A403. Based on the smoothed health indicators, establish a health curve model using a random degradation model fitting method.

[0047] Specifically, in step A401, obtaining the relative root mean square (RMS) value of the preprocessed vibration signal refers to calculating the RMS value of the preprocessed vibration signal and comparing or normalizing it with a reference value to quantify the change in the current vibration level relative to the healthy state. This relative RMS value, as a sensitive degradation indicator, can reflect subtle changes in the operating state of the insulated bearing and is a key basis for identifying the degradation initiation point. It can be implemented by dividing the current RMS value by the reference RMS value measured in the bearing's healthy state, or by calculating the difference between the current RMS value and the RMS value in the healthy state and then normalizing it.

[0048] Determining the initial degradation point of a target insulated bearing based on its relative root mean square (RMS) value aims to accurately identify the critical time point at which the bearing transitions from normal operation to degradation. Accurate identification of this initial degradation point is crucial for subsequent health curve modeling, preventing misjudging normal fluctuations as the onset of degradation or delaying the identification of true degradation due to noise interference. One approach is to set a statistical threshold; when the RMS value exceeds this threshold multiple times consecutively, the first time it exceeds the threshold is marked as the initial degradation point. Another approach is to use statistical process control charts (such as the Cumulative Sum (CUSUM) control chart or the Exponentially Weighted Moving Average (EWMA) control chart) to monitor changes in the RMS value; when the chart shows a significant trend change, the initial degradation point is identified.

[0049] In step A402, exponential smoothing is used to smooth the health indicators after the initial decline point, resulting in smoothed health indicators. This smoothed health indicator is obtained by using a weighted averaging method to reduce random fluctuations and noise in the health indicators, thereby revealing their potential long-term degradation trends. This processing method makes the health indicator data more stable and continuous, providing more reliable input for subsequent model fitting. Exponential smoothing can be implemented through simple exponential smoothing, where the current smoothed value is a weighted average of the current observation and the previous smoothed value, expressed by the formula: , Here is the smoothed health indicator at time t. Let t be the health indicator at time t. Let t be the health indicator of the previous time step. It can be used as a smoothing factor; or, when health indicators show a clear trend, double exponential smoothing or triple exponential smoothing methods can be used for processing.

[0050] In step A403, based on the smoothed health indicators and a stochastic degradation model fitting method, the health curve model is established. This refers to using mathematical statistical methods to construct a model that describes the evolution of the health state of the insulated bearing over time, taking into account the stochastic uncertainties in the degradation process. This model is the basis for predicting remaining life. One implementation method is to use stochastic process models such as the Wiener process or Gamma process to describe the evolution of health indicators and fit the model parameters using methods such as maximum likelihood estimation. Another implementation method is to use nonparametric models such as Gaussian process regression to fit the smoothed health indicators while considering uncertainties, thereby establishing a health curve model.

[0051] The proposed solution achieves its intended function through the following steps: First, the relative root mean square (RMS) value of the preprocessed vibration signal is obtained. This RMS value serves as a sensitive indicator of the health status of the insulated bearing, effectively reflecting changes in its operating state. Based on this RMS value, the initial degradation point of the target insulated bearing is precisely determined. This allows subsequent health curve modeling to begin from the point when the insulated bearing truly begins to degrade, avoiding misjudging fluctuations or noise during normal operation as degradation, thus ensuring the accuracy of the starting point of the health curve. Subsequently, the health indicators after the initial degradation point are processed using exponential smoothing. This step effectively filters out potential short-term fluctuations and random noise in the health indicators, making the health indicator data more stable and continuous, and more clearly showing the true degradation trend of the insulated bearing. Finally, based on these smoothed, more representative health indicators, a health curve model is established using a stochastic degradation model fitting method. The synergistic effect of this series of steps ensures that the established health curve model can accurately capture the degradation pattern of the insulated bearing and fully consider the randomness in the degradation process, providing a solid and reliable foundation for subsequent remaining life prediction. Compared with direct modeling based solely on raw health indicators, this approach significantly improves the quality of health indicators and the model's fitting accuracy by introducing initial decay point detection and exponential smoothing. This allows for fuller utilization of the health indicators constructed in the basic approach through vibration signal preprocessing, feature extraction and fusion, and principal component analysis. The resulting health curve model can more realistically and accurately describe the degradation process of insulated bearings.

[0052] Preferably, step A401 may include: Obtain the relative root mean square value of the preprocessed vibration signal; Calculate the mean and standard deviation of the relative root mean square values; The sum of the mean and three times the standard deviation is used as the judgment threshold; When N consecutive relative root mean square values ​​are greater than the judgment threshold, the time corresponding to the first relative root mean square value among the N consecutive relative root mean square values ​​is determined as the initial decay point of the target insulated bearing; N is the preset number of thresholds.

[0053] The calculation method for the relative root mean square (RMS) value is described above and will not be repeated here. Calculating the mean and standard deviation of the relative RMS value quantifies its statistical distribution characteristics over a specific time period. The mean reflects the central tendency of the data, while the standard deviation reflects its dispersion. These statistics are key parameters for constructing the judgment threshold, objectively describing the vibration characteristics of the bearing under normal or stable operating conditions. The calculation can be performed by continuously monitoring and recording the relative RMS value during normal bearing operation, then statistically analyzing this historical data to obtain its mean and standard deviation; or by using a sliding window average and standard deviation calculation, dynamically updating these statistics to adapt to environmental changes. The sum of the mean and three times the standard deviation is calculated as the judgment threshold. This step aims to establish an anomaly detection boundary based on statistical principles. Using the sum of the mean and three times the standard deviation as the judgment threshold is based on the "3σ" principle, which states that in a normal distribution, the probability of a data point falling within the range of the mean plus or minus three standard deviations is approximately 99.73%. Therefore, data points outside this range are considered abnormal, effectively distinguishing normal fluctuations from potential signs of degradation. The calculation can be implemented directly through programming, substituting the previously calculated average and standard deviation into the formula; or, based on actual working conditions and experience, the coefficient "three times" can be fine-tuned, for example, adjusted to 2.5 times or 3.5 times the standard deviation, to balance the false alarm rate and the false negative rate. When N consecutive relative root mean square values ​​exceed the judgment threshold, the time corresponding to the first relative root mean square value among these N consecutive relative root mean square values ​​is determined as the initial decay point of the target insulated bearing; N is a preset number of thresholds. This step introduces a continuity confirmation mechanism to improve the robustness and accuracy of the initial decay point determination, avoiding misjudgments caused by instantaneous noise or random events. A single data point exceeding the threshold may be due to random fluctuations, but multiple consecutive data points continuously exceeding the threshold strongly indicate that the bearing condition has undergone a substantial change. N, as the preset number of thresholds, allows for adjustment of the judgment sensitivity according to the actual application scenario. One approach is to use a counter to track the number of consecutive times the threshold is exceeded. Once the counter reaches N, the time point when the threshold is first exceeded is recorded. Alternatively, a sliding window detection method can be used to check whether there are N consecutive data points that meet the condition at each time step.

[0054] This application aims to accurately and robustly identify the onset of bearing degradation through a series of ordered steps. First, the system acquires the relative root mean square (RMS) values ​​of preprocessed vibration signals, ensuring the quality and comparability of the input data and laying the foundation for subsequent statistical analysis. After acquiring this basic data, the system calculates the mean and standard deviation of these relative RMS values. These two statistical quantities objectively depict the vibration characteristics distribution of the bearing under normal or steady-state conditions, providing a quantitative basis for establishing a scientific judgment benchmark. Based on this, by setting the sum of the calculated mean and three times the standard deviation as the judgment threshold, this method constructs an anomaly boundary based on statistical principles. This threshold effectively distinguishes normal fluctuations from potential degradation signals, filtering out most random noise and thus reducing the risk of misjudgment due to single measurement errors. More importantly, to further enhance the reliability of the judgment, this method introduces a continuity confirmation mechanism: only when N consecutive relative RMS values ​​are continuously greater than this judgment threshold is the time point corresponding to the first relative RMS value among these N values ​​determined as the initial degradation point of the target insulated bearing. This requirement for continuity significantly enhances the resistance to transient interference, ensuring that the identified decay initiation point is a real and stable state change, rather than an accidental signal fluctuation. The preset threshold N allows for flexible adjustment based on specific application scenarios and sensitivity requirements. Through the above method, in step A401, this application provides a precise starting point for establishing the health curve model, enabling subsequent health index smoothing (step A402) and random degradation model fitting (step A403) to begin from a reliable benchmark, thereby significantly improving the accuracy of the entire health curve model and the reliability of remaining life prediction. This initial decay point identification mechanism based on statistical principles and continuity verification, combined with the steps in the basic method of obtaining the relative root mean square value and determining the initial decay point, overcomes the inaccurate judgment problems caused by the lack of specific quantitative standards and insufficient anti-interference capabilities in traditional methods, providing a solid foundation for accurate life prediction of insulated bearings.

[0055] Preferably, step A403 may include: Based on the assumption that smoothed health indicators follow a Wiener process, the following stochastic degradation model is constructed: ; Where t is time, Let be the smoothed health index at time t. The drift coefficient, Where is the diffusion coefficient. For random terms that obey standard Brownian motion, To observe the noise, The standard deviation of the normal distribution. This indicates that the mean is 0 and the variance is 0. The normal distribution; Based on the smoothed health indicators, the maximum likelihood estimation method is used to fit the random degradation model. , and The fitted random degradation model was used as the health curve model.

[0056] In this application, the assumption that the smoothed health index follows a Wiener process is key to modeling the stochasticity and continuity of the degradation process of insulated bearings. The Wiener process, also known as Brownian motion, is a continuous-time stochastic process characterized by independent and stationary increments that follow a normal distribution. This characteristic makes it highly suitable for describing the continuous and stochastic deterioration of the health status of insulated bearings during operation due to factors such as wear and fatigue. By introducing the Wiener process, random fluctuations in the degradation process can be effectively captured, rather than relying solely on deterministic trends. Furthermore, other stochastic process assumptions, such as gamma processes or inverse Gaussian processes, can also be used to describe degradation behavior, but the Wiener process demonstrates excellent applicability in handling continuous degradation and random disturbances.

[0057] The constructed stochastic degradation model aims to quantify the change in the health state of insulated bearings over time. In this model, It is a quantitative value used to measure the health status of insulated bearings. It characterizes the average trend or rate of decline of health indicators over time, i.e. the deterministic decline in health status. It reflects the intensity of random fluctuations in the degradation process, that is, the degree of random dispersion of health indicators around the average trend. B(t) is a random term that follows standard Brownian motion, which introduces the randomness of the degradation process and simulates the impact of unpredictable external disturbances or internal micro-changes on health status. This model is used to describe potential errors or unmodeled random factors in the measurement of health indicators. By combining deterministic trends, random fluctuations, and observational noise, it can more comprehensively and accurately characterize the actual degradation trajectory of insulated bearings.

[0058] The maximum likelihood estimation method is used to fit the random degradation model. , and Maximizing the likelihood function of the observed smoothed health index sequence is a crucial step in determining the model parameters. The basic idea of ​​maximum likelihood estimation (MLE) is to select the parameter values ​​that maximize the probability of these data occurrences as the model estimates, given a set of observed data. For the stochastic degradation model in this application, by maximizing the likelihood function of the observed smoothed health index sequence, we can obtain... , and The optimal estimate, for example, the likelihood function can be , Let n be the likelihood function value, and n be the number of data points for the smoothed health indicator. The i-th time increment (i.e., the difference between the sampling time of the i-th smoothed health indicator and the sampling time of the previous smoothed health indicator, where the smoothed health indicator corresponding to the initial decline point can be set as the 0th smoothed health indicator). This represents the increment of the i-th smoothed health indicator (i.e., the difference between the i-th smoothed health indicator and the previous smoothed health indicator). This method can fully utilize historical data to ensure that the estimated model parameters are statistically sound, thereby improving the model's fitting accuracy and predictive ability. Besides maximum likelihood estimation, least squares or Bayesian estimation can also be used for parameter fitting, but maximum likelihood estimation theoretically has advantages such as asymptotic unbiasedness, efficiency, and normality.

[0059] Using the fitted random degradation model as the health curve model means that once the model parameters... , and Once determined, the model becomes a mathematical expression describing the evolution of the health status of insulated bearings. This health curve model not only reflects the average degradation trend of insulated bearings but also quantifies the uncertainties in the degradation process, providing a reliable basis for subsequent remaining life prediction.

[0060] This application constructs a stochastic degradation model that incorporates deterministic drift, random fluctuations, and observation noise, based on the assumption that the smoothed health indicators follow a Wiener process. This model accurately captures the complex characteristics of nonlinearity, non-stationarity, and random noise during the degradation process of insulated bearings. Subsequently, the maximum likelihood estimation method is used to efficiently and accurately estimate the key parameters of the model from historical data, ensuring the model's good adaptability to actual degradation behavior. This method cleverly combines stochastic process theory with statistical estimation techniques, overcoming the limitations of traditional deterministic models in handling the randomness of the degradation process. Through the above steps, this application can establish a highly reliable health curve model, laying a solid foundation for predicting the remaining life of insulated bearings. Compared with methods that rely solely on the original health indicators or simple trend fitting, this application smooths the health indicators in step A402, effectively reducing random noise in the data. This allows subsequent stochastic degradation model fitting to be performed on a more stable and representative data basis, further improving the accuracy and robustness of the health curve model.

[0061] In some implementations, step A5 includes: A501. Based on the health curve model, calculate the estimated health indicators at the current moment; A502. Based on the estimated health indicators and preset failure threshold at the current moment, calculate the expected and variance of the remaining life of the target insulated bearing using the following formula: ; ; in, As a result of the expected remaining lifespan, To preset the failure threshold, For the estimated health indicators at the current moment, The current time.

[0062] In step A501, calculating the estimated health index at the current moment refers to determining the health status of the insulated bearing at the current moment based on the established health curve model. This can be achieved by using the current time... Substitute the values ​​into the fitted health curve model to obtain predicted health index values. This step ensures that subsequent remaining life predictions are based on the latest understanding of the bearing degradation trajectory.

[0063] In step A502, the preset failure threshold D is a key parameter that defines the critical point at which the insulated bearing is considered to have failed or reached an unacceptable level of degradation. This threshold represents a specific health indicator value; once the bearing's health indicator falls below or exceeds this value, its performance is considered impaired. This threshold can be determined based on engineering experience, manufacturer specifications, safety regulations, or statistical analysis of historical failure data; for example, it can be set to 0.2 or dynamically adjusted according to specific operating conditions. Calculating the expected and variance of the remaining life of the target insulated bearing using the above formulas involves using specific mathematical expressions to quantify the remaining operating time of the bearing from the current moment until it reaches the preset failure threshold, while simultaneously assessing the uncertainty of this prediction. These formulas are typically derived from underlying stochastic degradation models (such as Wiener processes), providing not only a point estimate (expected value) but also a discrete measure (variance), thus making the remaining life prediction more statistically significant and robust. Since the smoothed health indicator follows a Wiener process, the probability density function of the remaining life RUL is an inverse Gaussian distribution. s is a time variable. Let s represent the probability density function of the remaining lifetime. Based on this probability density function, the formulas for calculating the expected value and variance of the remaining lifetime are as described above. E(RUL) represents the expected value of the remaining lifetime, i.e., the most likely time from the current moment until the bearing health indicators reach the preset failure threshold, providing a primary reference for maintenance planning and operation scheduling. Var(RUL) represents the variance of the remaining lifetime, which quantifies the degree of dispersion or uncertainty around the expected value of the remaining lifetime. Smaller variance indicates higher prediction accuracy, while larger variance suggests greater uncertainty, which is crucial for risk assessment and decision-making.

[0064] Through the aforementioned technical solution, this application can provide the statistical characteristics of remaining life, namely the expected value and variance of remaining life, thereby overcoming the limitations of traditional methods that only provide point estimates and lack quantification of prediction uncertainty. This comprehensive statistical characterization enables maintenance personnel not only to understand when the bearing is expected to fail, but also to assess the reliability of the prediction, thereby conducting more accurate risk assessments and formulating more optimized maintenance strategies. Especially after combining with a health curve model based on the Wiener process, these statistics directly reflect the actual random degradation pattern of the bearing, making the prediction results more practically instructive and reliable, significantly improving the accuracy and practicality of remaining life prediction for insulated bearings.

[0065] Please refer to Figure 2 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device includes a processor 301 and a memory 302. The processor 301 and the memory 302 are interconnected and communicate with each other via a communication bus 303 and / or other connection mechanisms (not shown). The memory 302 stores a computer program executable by the processor 301. When the electronic device is running, the processor 301 executes the computer program to perform the remaining life prediction method for insulated bearings in any optional implementation of the above embodiments, to achieve the following functions: acquiring the vibration signal of the target insulated bearing and preprocessing the vibration signal; simulating the collision process of the inner and outer rings based on the collision dynamics model of the target insulated bearing to obtain collision physical characteristics, and obtaining measured signal characteristics based on the preprocessed vibration signal, fusing the collision physical characteristics and measured signal characteristics to obtain a feature vector; performing dimensionality reduction processing on the feature vector based on principal component analysis to obtain a health index; establishing a health curve model based on the health index and a random degradation model fitting method; and predicting the remaining life of the target insulated bearing based on the health curve model and a preset failure threshold.

[0066] This application provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it performs the remaining life prediction method for insulated bearings in any optional implementation of the above embodiments to achieve the following functions: acquiring the vibration signal of the target insulated bearing and preprocessing the vibration signal; simulating the collision process of the inner and outer rings based on the collision dynamics model of the target insulated bearing to obtain the collision physical characteristics, and obtaining the measured signal characteristics based on the preprocessed vibration signal, fusing the collision physical characteristics and the measured signal characteristics to obtain a feature vector; performing dimensionality reduction processing on the feature vector based on principal component analysis to obtain a health index; establishing a health curve model based on the health index and a random degradation model fitting method; and predicting the remaining life of the target insulated bearing based on the health curve model and a preset failure threshold. The computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0067] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for predicting the remaining life of an insulated bearing, characterized in that, Includes the following steps: A1. Acquire the vibration signal of the target insulated bearing and preprocess the vibration signal; A2. Based on the collision dynamics model of the inner and outer rings of the target insulated bearing, the collision process of the inner and outer rings is simulated to obtain the collision physical characteristics, and the measured signal characteristics are obtained according to the preprocessed vibration signal. The collision physical characteristics and the measured signal characteristics are fused to obtain the feature vector. A3. Based on principal component analysis, the dimensionality of the eigenvectors is reduced to obtain health indicators; A4. Based on the aforementioned health indicators, establish a health curve model using a random degradation model fitting method; A5. Based on the health curve model and the preset failure threshold, predict the remaining life of the target insulated bearing.

2. The method for predicting the remaining life of an insulated bearing according to claim 1, characterized in that, The preprocessing includes DC component removal, bandpass filtering, and normalization.

3. The method for predicting the remaining life of an insulated bearing according to claim 1, characterized in that, The collision physical characteristics include at least the peak contact force, clearance change rate, and collision energy. The measured signal characteristics include time-domain characteristics, frequency-domain characteristics, and time-frequency-domain characteristics; The time-domain features include at least the root mean square value and kurtosis; The frequency domain features include at least the fault characteristic frequency amplitude; The time-frequency domain features include at least the envelope spectrum amplitude and wavelet energy entropy.

4. The method for predicting the remaining life of an insulated bearing according to claim 1, characterized in that, Step A3 includes: A301. Standardize the feature vector to obtain a standardized feature vector; A302. Based on principal component analysis, the eigenvectors are dimensionality reduced to obtain dimensionality-reduced eigenvectors; A303. Normalize the reduced-dimensional feature vector to obtain the health index.

5. The method for predicting the remaining life of an insulated bearing according to claim 1, characterized in that, Step A4 includes: A401. Obtain the relative root mean square value of the preprocessed vibration signal, and determine the initial decay point of the target insulated bearing based on the relative root mean square value; A402. The health indicators after the initial decline point are smoothed using an exponential smoothing method to obtain smoothed health indicators; A403. Based on the smoothed health indicators, establish the health curve model using a random degradation model fitting method.

6. The method for predicting the remaining life of an insulated bearing according to claim 5, characterized in that, Step A401 includes: Obtain the relative root mean square value of the preprocessed vibration signal; Calculate the mean and standard deviation of the relative root mean square values; The sum of the average value and three times the standard deviation is used as the judgment threshold; When N consecutive relative root mean square values ​​are greater than the judgment threshold, the time corresponding to the first relative root mean square value among the N consecutive relative root mean square values ​​is determined as the initial decay point of the target insulated bearing; N is a preset number threshold.

7. The method for predicting the remaining life of an insulated bearing according to claim 5, characterized in that, Step A403 includes: Based on the assumption that smoothed health indicators follow a Wiener process, the following stochastic degradation model is constructed: ; Where t is time, Let be the smoothed health index at time t. The drift coefficient, Where is the diffusion coefficient. For random terms that obey standard Brownian motion, To observe the noise, The standard deviation of the normal distribution. This indicates that the mean is 0 and the variance is 0. The normal distribution; Based on the smoothed health indicators, the maximum likelihood estimation method is used to fit the random degradation model. , and The fitted random degradation model was used as the health curve model.

8. The method for predicting the remaining life of an insulated bearing according to claim 7, characterized in that, Step A5 includes: A501. Calculate the estimated health indicators at the current moment based on the health curve model. A502. Based on the estimated health indicators and preset failure threshold at the current moment, calculate the expected and variance of the remaining life of the target insulated bearing using the following formula: ； ； in, As a result of the expected remaining lifespan, To preset the failure threshold, For the estimated health indicators at the current moment, The current time.

9. An electronic device, characterized in that, It includes a processor and a memory, the memory storing a computer program executable by the processor, which, when executed by the processor, performs the steps of the method for predicting the remaining life of an insulated bearing as described in any one of claims 1-8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it performs the steps of the method for predicting the remaining life of an insulated bearing as described in any one of claims 1-8.

Images