Train running gear bearing state prediction and early warning method and system

Abstract

The invention discloses a train running gear bearing state prediction and early warning method and system, and the method comprises the steps: synchronously collecting vibration, noise and temperature data, and building a standard vibration reference based on rotation period segmentation; extracting a component characteristic frequency and confirming fault information by adopting a frequency spectrum decomposition and intermodulation analysis technology, and determining a monitoring working frequency and a component weight table; a temperature confrontation processing mechanism is introduced, decoupling parameters are generated through a heat distribution matrix and directional heat flow analysis, and accurate separation of temperature influence and mechanical wear signals is achieved; constructing an equivalent transformation matrix based on a noise dependency relationship, and performing standardized transformation on pure vibration data to generate a comparable state index; through abrasion tracking and vibration transmission analysis, path reconstruction is carried out in combination with transfer parameters, and accurate positioning and evolution prediction of faults are achieved; and finally, dynamically adjusting the component weight based on the path control signal, establishing a four-level grading early warning system, and realizing real-time monitoring and intelligent early warning of the running state of the bearing.

Description

Technical Field

[0001] This invention relates to the field of mechanical fault diagnosis technology, and in particular to a method and system for predicting and warning the condition of bearings in train running gear. Background Technology

[0002] As a core component of rotating machinery, bearings directly affect the reliability and safety of the entire machine. Vibration monitoring is a primary method for bearing fault diagnosis. However, bearing vibration signals under actual operating conditions are subject to complex influences from various factors such as temperature, noise, and load. In particular, temperature effects significantly alter the bearing's stiffness and damping characteristics, leading to non-fault-related changes in the vibration signal and severely interfering with the accurate identification of fault characteristics. Traditional vibration analysis methods struggle to effectively separate temperature effects from mechanical wear signals, easily resulting in misdiagnosis or missed diagnosis, thus impacting preventative maintenance decisions.

[0003] Furthermore, existing technologies lack an effective standardized conversion mechanism when dealing with bearing vibration under varying noise conditions. This makes it impossible to map vibration data under different noise conditions to a unified evaluation benchmark, resulting in a lack of comparability in condition assessment results. Simultaneously, the propagation paths of multi-component coupled faults are complex, making it difficult for existing methods to accurately locate fault sources and predict fault evolution trends. This hinders the development of precise graded early warning systems and restricts the advancement of intelligent fault diagnosis technology. Summary of the Invention

[0004] This invention provides a method and system for predicting and warning the condition of bearings in train running gear. The aim is to establish a standard vibration benchmark and extract the characteristic frequencies of components. Through temperature-vibration correlation analysis and counter-vibration processing techniques, it achieves precise separation of temperature influences in vibration signals, obtaining vibration data reflecting pure mechanical wear. By constructing an equivalent transformation matrix, it maps vibrations under varying noise conditions to a standard state, achieving unified evaluation of vibration data under different operating conditions. Through fault propagation path reconstruction and multi-level vibration fusion, it accurately locates fault sources and predicts evolution trends. Finally, it establishes a four-level hierarchical early warning mechanism based on dynamic weight adjustment, providing reliable technical support for intelligent fault diagnosis and preventative maintenance of bearings.

[0005] The first aspect of this invention proposes a method for predicting and warning the condition of bearings in train running gear, comprising the following steps: Acquire bearing vibration signals, noise data, and temperature data; generate a rotation period based on the noise data; synchronously segment the vibration signal according to the rotation period to form vibration segments; and fuse the vibration segments to generate a standard vibration reference. The component characteristic frequencies are extracted using the standard vibration reference, and the component characteristic frequencies are subjected to intermodulation analysis to confirm fault information. The fault information is then subjected to fault frequency analysis to determine the monitoring operating frequency. Finally, a component weight table is determined based on the monitoring operating frequency. Based on the component weight table, the temperature data is spatially calibrated to generate a temperature distribution map. Based on the temperature distribution map and the component characteristic frequency, a correlation analysis is performed to generate decoupling parameters. Temperature countermeasures are performed according to the decoupling parameters to generate countermeasures effect. The countermeasures effect is used to identify pure mechanical wear signals and generate pure vibration data. An equivalent transformation matrix is ​​constructed based on the pure vibration data and the noise data. The pure vibration data is then standardized and transformed using the equivalent transformation matrix to generate equivalent vibration data. The equivalent vibration data is then correlated with the standard vibration benchmark to generate standard state indicators. Wear tracking and vibration transmission analysis are performed on the standard state indicators to generate transfer parameters, and path reconstruction processing is performed based on the transfer parameters to generate path control signals; Based on the path control signal, the component weight table is adjusted to generate weight adjustment data, and a graded early warning signal is generated based on the weight adjustment data.

[0006] A second aspect of the present invention provides a predictive and early warning system for the condition of a train running gear bearing, comprising: The data acquisition module is used to acquire bearing vibration signals, noise data, and temperature data; generate a rotation period based on the noise data; synchronously segment the vibration signal according to the rotation period to form vibration segments; and fuse the vibration segments to generate a standard vibration reference. The feature extraction module is used to extract component feature frequencies through the standard vibration reference, perform intermodulation analysis on the component feature frequencies to confirm fault information, perform fault frequency analysis on the fault information to determine the monitoring operating frequency, and determine the component weight table based on the monitoring operating frequency. The temperature processing module is used to spatially calibrate the temperature data based on the component weight table to generate a temperature distribution map, perform correlation analysis between the temperature distribution map and the component characteristic frequency to generate decoupling parameters, perform temperature countermeasure processing based on the decoupling parameters to generate countermeasure effects, and use the countermeasure effects to identify pure mechanical wear signals to generate pure vibration data. The working condition standardization module is used to construct an equivalent transformation matrix based on the pure vibration data and the noise data, perform standardization transformation on the pure vibration data through the equivalent transformation matrix to generate equivalent vibration data, and perform correlation analysis between the equivalent vibration data and the standard vibration benchmark to generate standard state indicators. The fault control module is used to perform wear tracking and vibration transmission analysis on the standard state indicators to generate transfer parameters, and to perform path reconstruction processing based on the transfer parameters to generate path control signals. The early warning decision module is used to adjust the component weight table based on the path control signal to generate weight adjustment data, and to generate a graded early warning signal based on the weight adjustment data.

[0007] The beneficial effects of this invention are reflected in the following points: First, by using a temperature-vibration coupled field model and a temperature-resistance processing mechanism, based on an established standard vibration benchmark that includes vibration segment fusion, it can accurately extract pure mechanical wear components from complex vibration signals, eliminating vibration characteristic drift caused by temperature changes, improving the accuracy and reliability of fault identification, and avoiding the temperature-related misdiagnosis phenomenon commonly found in traditional methods. Second, by confirming fault information through intermodulation analysis of component characteristic frequencies, determining key monitoring frequencies and component weights, and then combining an equivalent transformation matrix to map vibration data under different noise conditions to a standard state space, it enables bearing states under various operating conditions to have a unified evaluation benchmark and comparability, not only expanding the applicable scope of fault diagnosis technology but also realizing state tracking and trend analysis throughout the equipment's entire life cycle. Third, by reconstructing fault propagation paths and performing multi-level vibration fusion analysis, it accurately locates the current fault source and predicts the fault development trend. Combined with weight sensitivity jump detection and dynamic adjustment mechanisms, it achieves adaptive four-level graded early warning, enabling maintenance personnel to grasp equipment state changes in advance, rationally arrange maintenance plans, and realize the transformation from passive maintenance to preventive maintenance.

[0008] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0009] The accompanying drawings illustrate specific examples of the technical solutions described in this invention and, together with the detailed embodiments, form part of the specification, serving to explain the technical solutions, principles, and effects of this invention.

[0010] Figure 1 This is a flowchart illustrating a method for predicting and warning the condition of bearings in a train running gear according to the present invention.

[0011] Figure 2 This is a structural block diagram of a train running gear bearing condition prediction and early warning system according to the present invention. Detailed Implementation

[0012] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0013] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0014] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0015] The technical solutions of the embodiments of this application will be described below.

[0016] like Figure 1 As shown, this embodiment of the invention provides a method for predicting and warning the condition of a train running gear bearing, including the following steps S110-S160: Step S110: Obtain bearing vibration signal, noise data and temperature data; generate rotation period based on noise data; synchronously segment the vibration signal according to the rotation period to form vibration segments; and fuse the vibration segments to generate a standard vibration reference.

[0017] Specifically, bearing vibration signals, noise data, and temperature data are acquired. Vibration signals are collected using an accelerometer mounted on the bearing housing. The sensor type is a piezoelectric accelerometer with a measurement range of 0-10kHz, a sensitivity of 100mV / g, and a sampling frequency set to 25.6kHz to ensure the capture of high-frequency vibration characteristics. The vibration signal includes radial and axial vibration components, reflecting the radial load and axial thrust state of the bearing, respectively. Noise data is acquired in real time using a sound level meter or acoustic sensor, with a measurement range of 30-130dB, a frequency response of 20Hz-20kHz, and a sampling frequency of 25.6kHz. The noise data includes the periodic acoustic characteristics of bearing rotation, with the rotational speed proportional to the fundamental noise frequency. Temperature data is measured using a PT100 platinum resistance temperature sensor to measure the bearing outer ring temperature, with a measurement range of -50℃ to 200℃, an accuracy of ±0.1℃, and a sampling interval of 1 second. All three types of data are synchronized using a unified timestamp, achieving millisecond-level time synchronization accuracy to ensure accurate and reliable temporal correspondence between the vibration signal, noise data, and temperature data. Data preprocessing includes removing DC components, eliminating trend terms, and removing outliers to ensure signal quality. The data buffer uses a circular buffer structure with a capacity of 10 minutes of continuous data, supporting real-time analysis and historical backtesting.

[0018] Rotation periods are generated based on noise data. A short-time Fourier transform is performed on the noise data with a window length of 1024 points and 50% overlap, identifying the fundamental frequency within the 0.5-100Hz range. The rotation period is the reciprocal of the fundamental frequency of the noise data; its accuracy directly depends on the quality of the noise data and the precision of the fundamental frequency identification. Considering the dynamic characteristics of the fundamental frequency changes in the noise data, a real-time update mechanism is adopted, calculating the corresponding instantaneous rotation period based on the noise data at each sampling moment. The fundamental frequency stability of the noise data is determined by calculating the standard deviation of 100 adjacent sampling points; a standard deviation less than 2% of the average fundamental frequency is considered stable. For variable-speed operation, a piecewise linear interpolation method is used to divide the fundamental frequency change process of the noise data into multiple quasi-steady-state intervals, with a fundamental frequency change rate of less than 5 rpm / s within each interval. The validity of the rotation period is checked through continuity judgment; the difference between adjacent periods does not exceed 5%, and periods exceeding the threshold are marked as abnormal and discarded. A rotation period sequence array is established to record the start time, end time, and duration of each valid rotation period, providing a precise time reference for vibration signal segmentation.

[0019] Vibration signals are synchronously segmented into vibration segments based on the rotation period. The starting moment of the rotation period is used as the synchronization reference for precise segmentation of the vibration signal. Each vibration segment corresponds to the complete vibration data of one rotation of the bearing, including the characteristic frequency components of each bearing component (inner ring, outer ring, rolling elements, and cage). Segmentation is based on the rotation period to ensure a strict correspondence between the vibration segment and the rotation period. A zero-phase point alignment method is used to ensure that the starting phase of each vibration segment is consistent, facilitating data fusion. For non-integer multiple sampling cases, cubic spline interpolation is used to resample the vibration segments to a uniform length, with the standard segment length set at 2048 sampling points. Hanning window functions are used for segment boundary processing, with a window length of 5% of the segment length to avoid spectral leakage caused by segmentation. The signal-to-noise ratio (SNR) of the vibration segments is calculated to mark data quality; vibration segments with an SNR below 10dB have reduced weight in the fusion process. A vibration segment index table is established to record metadata such as the sequence number, timestamp, corresponding rotation period value, temperature value, and quality marker for each vibration segment. Multiple vibration segments are extracted continuously. The number of segments is determined based on the stability of the rotation period. Under stable conditions, 20-50 vibration segments are extracted, while under variable speed conditions, they are extracted in groups according to the rotation period interval.

[0020] In some embodiments, the step of fusing the vibration segments to generate a standard vibration reference includes: identifying abrupt peaks within the vibration segments; decomposing the vibration segments into multiple steady-state segments based on the abrupt peaks; acquiring and accumulating the vibration energy of the steady-state segments to form a total energy value; and performing waveform compensation correction on the total energy value to generate a standard vibration reference.

[0021] Identifying abrupt peaks within vibration segments: A local extremum search algorithm is employed to find local maxima within each vibration segment. Peak determination criteria are: the amplitude of the current point is greater than the amplitudes of the five sampling points before and after it, and the amplitude exceeds 1.5 times the segment mean. Abrupt peak identification is performed by calculating the relative rate of change of amplitude between adjacent peaks; when the relative change in amplitude between adjacent peaks exceeds 30%, the peak position is identified as an abrupt peak. Time interval constraints ensure that the minimum interval between abrupt peaks is greater than 0.1 times the rotation period, avoiding false abrupt changes caused by noise. Features extracted from abrupt peaks include: abrupt change amplitude, abrupt change duration, and frequency changes before and after the abrupt change. A sequence of abrupt peaks is established, recording the position index, amplitude, and abrupt change type (rising or falling) for each abrupt peak, providing accurate boundary markers for vibration segment decomposition. The accuracy of abrupt peak identification directly affects the rationality of steady-state segment division; through multiple iterations to optimize the peak determination threshold, it is ensured that abrupt peaks can accurately identify significant changes in vibration characteristics.

[0022] The vibration segment is decomposed into multiple steady-state segments based on the wave crest abrupt change point. Using the wave crest abrupt change point as the boundary, the complete vibration segment is divided into several steady-state segments. A steady-state segment is defined as a region between wave crest abrupt change points where the vibration characteristics are relatively stable, with gradual changes in vibration amplitude and frequency components. The decomposition algorithm ensures the integrity of each steady-state segment, avoiding segmentation in the middle of the vibration cycle. During steady-state segment length checking, excessively short steady-state segments (less than 0.05 times the rotation period) are merged with adjacent steady-state segments to avoid over-decomposition. Each steady-state segment undergoes independent feature analysis, extracting time-domain statistical features and the dominant frequency in the frequency domain. The steady-state performance index is determined by calculating the ratio of the amplitude standard deviation to the mean within the steady-state segment; a ratio less than 0.2 indicates good steady-state performance. A steady-state segment library is established, assigning a unique identifier to each steady-state segment and recording its position and duration within the original vibration segment. The quality of steady-state segment division is ensured by the accuracy of the wave crest abrupt change points, and the decomposed steady-state segments provide stable and reliable analysis units for energy calculations.

[0023] The vibration energy of steady-state segments is acquired and accumulated to form the total energy value. A time-domain energy integration method is used, E = ∫x²(t)dt, where x(t) is the vibration acceleration signal, and the integration interval is the time range of the steady-state segment. The trapezoidal integral method is used for discretization to ensure the accuracy and stability of the numerical calculation. Vibration energy is acquired from each steady-state segment, and energy normalization is performed to eliminate the influence of differences in steady-state segment lengths. The normalized vibration energy equals the steady-state segment energy divided by the segment duration. The vibration energy weight allocation is determined based on the characteristic importance and signal quality of the steady-state segments: steady-state segments with a signal-to-noise ratio (SNR) higher than 15dB have a weight of 1.0, those between 10-15dB have a weight of 0.8, and those below 10dB have a weight of 0.5. A weighted summation method is used in the accumulation process, with the weight coefficients dynamically adjusted according to the steady-state performance indicators and signal quality of the steady-state segments. The vibration energy of all steady-state segments is accumulated to form the total energy value. The total energy value calculation considers the contributions of radial and axial vibrations, and a vector synthesis method is used to obtain the comprehensive vibration energy. The total energy value comprehensively reflects the vibration intensity of the entire vibration segment within the rotation cycle, providing a quantitative basis for compensation and correction.

[0024] Waveform compensation correction is applied to the total energy value to generate a standard vibration reference. The compensation correction considers the structural characteristics and operating conditions of the bearing, correcting energy deviations caused by geometric asymmetry and uneven load distribution. The compensation coefficient is calculated using the bearing's geometric parameters, including the influence of structural parameters such as the number of rolling elements, contact angle, and pitch circle diameter. The load compensation factor is determined based on the ratio of radial load to axial load, and the compensation formula considers the influence of load direction on vibration energy distribution. Temperature compensation corrects the effect of temperature changes on vibration characteristics; the compensation coefficient α = 1 + β × (T - T0), where T is the measured temperature recorded in the segment, T0 is the standard temperature of 20℃, and β is the temperature coefficient 0.002 / ℃. Fundamental frequency compensation eliminates the interference of speed fluctuations on the calculation of the total energy value. Using the rotational period values ​​recorded in the segment index table, the total energy value at different speeds is normalized to the standard speed conditions; the compensation coefficient γ = (f0 / f0_std)^k, where f0 is the measured fundamental frequency, f0_std is the standard fundamental frequency of 25Hz, and k is the frequency exponent of 1.5. The correction algorithm employs a multi-parameter fusion method, sequentially compensating for temperature, fundamental frequency, and load on the total energy value, comprehensively considering the interactive effects of various compensation factors. The standard vibration benchmark is generated through statistical processing of the corrected total energy value, using a weighted average method to synthesize the correction results of multiple vibration segments. This standard vibration benchmark serves as a reference value for the vibration energy under normal bearing operating conditions, providing a comparative benchmark for fault detection and condition assessment.

[0025] Step S120: Extract component characteristic frequencies through standard vibration reference, perform intermodulation analysis on component characteristic frequencies to confirm fault information, perform fault frequency analysis on fault information to determine monitoring operating frequency, and determine component weight table based on monitoring operating frequency.

[0026] Specifically, component characteristic frequencies are extracted using a standard vibration benchmark. A Fast Fourier Transform (FFT) is performed on the standard vibration benchmark with a spectral resolution of 0.5 Hz to ensure accurate identification of each component's characteristic frequency. The rotational frequency parameter in the component characteristic frequency calculation is obtained by converting the noise fundamental frequency. The inner ring characteristic frequency is calculated based on the bearing geometry and rotational frequency, taking into account the influence of the number of rolling elements, rolling element diameter, pitch circle diameter, and contact angle. The outer ring characteristic frequency is calculated similarly to the inner ring, but with the opposite sign to reflect the impact frequency during outer ring failure. The rolling element characteristic frequency is calculated based on the ratio of the pitch circle diameter to the rolling element diameter; this frequency significantly increases when the rolling elements are damaged. The cage characteristic frequency is typically the lowest component among all characteristic frequencies. A spectral peak search algorithm searches for the actual peak value within ±2% of the calculated theoretical frequency, considering the influence of manufacturing tolerances and installation errors. Harmonic components of the component characteristic frequencies are extracted up to the 5th harmonic; the intensity of higher harmonics indicates the severity of the fault. The component characteristic frequency amplitudes are recorded and normalized, using the maximum peak value as a benchmark, to form a complete component characteristic frequency dataset.

[0027] Intermodulation analysis of component characteristic frequencies confirms fault information. Based on nonlinear system theory, intermodulation analysis analyzes the intermodulation of component characteristic frequencies. When multiple components have simultaneous defects, sum and difference frequency components are generated. Second-order intermodulation frequencies include the sum and difference frequency components of different component characteristic frequencies. Third-order intermodulation considers more complex multi-frequency combinations, reflecting the characteristics of multi-component coupled faults. For example, when the inner ring characteristic frequency is 100Hz and the outer ring characteristic frequency is 60Hz, intermodulation analysis will detect frequency components with a sum frequency of 160Hz and a difference frequency of 40Hz. When the amplitudes of these intermodulation frequencies significantly increase, it indicates that defects exist simultaneously in the inner and outer rings and influence each other. Third-order intermodulation may generate more complex combination frequencies such as 220Hz (100+60+60), reflecting the coupling depth of the fault. Harmonic intermodulation analysis considers complex combinations involving the 5th harmonic extracted from the component characteristic frequencies. When harmonics participate in intermodulation, it indicates a deeper fault severity. The amplitude of the intermodulation products is obtained through spectral analysis and compared with the normalized amplitude of the original component characteristic frequencies. The amplitude ratio reflects the coupling strength between components. The propagation contribution is determined by combining the intermodulation strength and frequency of occurrence of the component characteristic frequencies; a high contribution indicates that the component plays a key role in the fault propagation chain. The propagation rate is calculated by the growth rate of the component characteristic frequency amplitude; rapid growth (growth rate > 20% / hour) indicates rapid fault deterioration. The results of the intermodulation analysis are integrated into fault information, which includes key parameters such as the intermodulation frequency, amplitude ratio, propagation contribution, and propagation rate of each component.

[0028] Fault frequency analysis is performed on fault information to determine the monitoring operating frequency. Key frequency components are extracted from the fault information for analysis. These key frequency components include the original component characteristic frequencies, major intermodulation frequencies, and high-energy harmonic frequencies. The frequency importance score is based on three indicators: occurrence probability P, amplitude A, and propagation contribution C obtained from the fault information. The comprehensive score S = 0.4 × P + 0.3 × A + 0.3 × C, where S is the comprehensive frequency importance score, P is the occurrence probability, A is the amplitude, and C is the propagation contribution. Frequency stability analysis calculates the coefficient of variation of each frequency component in the fault information under different operating conditions. Frequencies with a coefficient of variation less than 0.1 are considered stable and reliable. The monitoring operating frequency set is determined through an optimization algorithm, aiming to cover the most fault modes in the fault information with the fewest frequency points. The final monitoring operating frequency set consists of 5-10 key frequency points, sorted from high to low according to the importance score S, covering the main fault characteristics and key intermodulation frequencies of each component, forming a complete monitoring operating frequency set.

[0029] In some embodiments, determining the component weight table based on the monitoring operating frequency includes: identifying resonance characteristics from the monitoring operating frequency; generating a secondary resonance inside the bearing using the resonance characteristics; superimposing the component characteristic frequency with the secondary resonance to form an enhanced resonance; and assigning weights to the enhanced resonance to generate a component weight table.

[0030] Resonance characteristics are identified from the monitored operating frequency. Frequency response function analysis is used to locate frequencies within the monitored operating frequency range where the response peak is significantly amplified. The criteria for identifying resonance characteristics are that the frequency response amplitude exceeds three times the average response, and the phase undergoes a 90-degree abrupt change. The natural frequency in the resonance characteristics is calculated based on the bearing's structural parameters and material properties, depending on the system's stiffness and mass. The resonance bandwidth is measured using the half-power point method; the bandwidth is the difference between two frequency points where the response drops to 0.707 times the peak value. The quality factor Q is the ratio of the natural frequency to the bandwidth, reflecting the sharpness of the resonance characteristic. A quality factor greater than 10 indicates significant resonance, and a higher quality factor enhances the superposition effect. Multi-order resonance identification considers the first three vibration modes of the bearing system, recording the resonance frequency and corresponding quality factor for each mode, which constitutes complete resonance characteristic data. The resonance amplification factor R_amp is defined as the ratio of the response at the resonance frequency to the static response, used to quantify the enhancement effect of the resonance characteristic; a larger R_amp indicates a stronger resonance enhancement capability.

[0031] Secondary resonances are generated within the bearing using resonant characteristics. Parametric resonance occurs when the excitation frequency approaches half the system's natural frequency. The secondary resonance frequencies are calculated for the first three identified resonances, each twice the natural frequency of the resonant characteristic, resulting in multiple secondary resonance frequencies. For example, if the bearing's natural frequency is 200Hz, and the component's characteristic frequency is 100Hz (half the natural frequency), a 200Hz secondary resonance will be generated. This is analogous to a child swinging on a swing; each two pushes (100Hz) complete a full swing cycle (200Hz). In a bearing, when a rolling element passes a position at 100Hz, two impacts will result in enhanced resonance at 200Hz, forming a secondary resonance. Excitation condition analysis determines the minimum excitation amplitude required to generate secondary resonance, typically 1.5 times the primary resonance threshold of the resonant characteristic. The energy transfer path traces the propagation of the secondary resonance within the bearing, from the excitation source through the rolling elements to the inner and outer rings. The coupling coefficient κ is calculated to reflect the energy conversion efficiency from the resonant characteristic to the secondary resonance. The coupling coefficient κ = A_2 / A_1, where κ is the coupling coefficient, A_2 is the amplitude of the secondary resonance, and A_1 is the amplitude of the primary resonance of the resonant characteristic. The larger the coupling coefficient, the higher the energy conversion efficiency.

[0032] The component's characteristic frequency is superimposed with the secondary resonance to form an enhanced resonance. The superposition mechanism is based on the principle of wave interference; when the component's characteristic frequency and the secondary resonance frequency are close, a beat frequency phenomenon occurs. For example, when the component's characteristic frequency is 198Hz and the secondary resonance frequency is 200Hz, their superposition produces a 2Hz beat frequency, manifested as a change in vibration amplitude every 0.5 seconds. When the vibrations of the two frequencies are superimposed in phase, the amplitude can reach twice that of the individual vibrations, resulting in a significant enhanced resonance effect. The frequency deviation Δf_i is calculated considering multiple secondary resonances: Δf_i = min(|f_component - f_2i| / f_2i), where Δf_i is the frequency deviation, f_component is the component's characteristic frequency, and f_2i is the i-th order secondary resonance frequency. The secondary resonance with the smallest deviation is selected for superposition. Phase relationship analysis determines the optimal superposition conditions; in-phase superposition produces the maximum enhancement effect. In the time domain, this is manifested as periodic modulation of the amplitude, with the modulation depth reflecting the degree of enhancement of the enhanced resonance. The energy concentration effect significantly increases the energy of specific frequency components. The energy enhancement factor E_factor = (1 + R_amp × (1 - Δf_i)) × (1 + Q / 20), where E_factor is the energy enhancement factor, R_amp is the resonance amplification factor (derived from the identification result of resonance characteristics), 1 - Δf_i is the frequency matching degree (the closer to 1, the better the matching), and Q is the quality factor. The spatial enhancement distribution reflects the difference in the degree of resonance enhancement at different locations, with the maximum enhancement usually occurring in the load region.

[0033] A component weight table is generated by weighting the enhanced resonance. The weighting comprehensively considers multiple characteristic parameters of the enhanced resonance, including the energy enhancement factor E_factor, the resonance amplification factor R_amp, and the quality factor Q. The weight calculation uses a weighted fusion method, with the component weight formula being Wi = 0.5 × E_factor + 0.3 × R_amp_norm + 0.2 × Q_norm, where Wi is the weight of the i-th component, E_factor is the energy enhancement factor, R_amp_norm is the normalized resonance amplification factor calculated by R_amp_norm = R_amp / R_amp_max (where R_amp_max is the maximum resonance amplification factor among all components), and Q_norm is the normalized quality factor calculated by Q_norm = Q / Q_max (where Q_max is the maximum quality factor among all components). The weight coefficients 0.5, 0.3, and 0.2 reflect the relative importance of the energy enhancement factor, resonance amplification, and resonance sharpness in the weighting, respectively, and the sum of the coefficients is 1 to ensure the normalization characteristic of the weights. Considering the spatial distribution differences in enhanced resonance, the component weights in the loaded and unloaded regions are fine-tuned. The enhancement coefficient for the loaded region is set to 1.2, and the enhancement coefficient for the unloaded region is set to 0.8. The adjusted weight W_i' = W_i × η, where η is the enhancement coefficient for the corresponding region. The adjusted weights of all components are normalized, and the final weight W_final(i) = W_i' / Σ(W_i'), ensuring that the sum of the weights of all components is 1. The generated component weight table contains the component number, component name, and final weight W_final. The component weight table is used for subsequent spatial calibration and monitoring resource allocation. Components with higher weights will have denser interpolation nodes.

[0034] Step S130: Spatial calibration of temperature data is performed based on component weight table to generate temperature distribution map. Correlation analysis is performed between temperature distribution map and component characteristic frequency to generate decoupling parameters. Temperature countermeasure processing is performed based on decoupling parameters to generate countermeasure effect. The countermeasure effect is used to identify pure mechanical wear signal to generate pure vibration data.

[0035] Specifically, a temperature distribution map is generated by spatially calibrating the temperature data based on a component weight table. A multi-point temperature sensor interpolation method is used based on the collected temperature data, and the importance of interpolation nodes is determined according to the component weight table. Higher-weight component regions have denser interpolation nodes, with the node density proportional to the weight value in the component weight table, reaching a maximum density of 5 nodes / cm². The interpolation algorithm uses Kriging interpolation, considering the spatial correlation of the temperature field during spatial interpolation, and a Gaussian model is selected as the correlation function. The temperature gradient of the temperature data includes radial and axial components; the radial component reflects the radial temperature change rate, and the axial component reflects the axial temperature change rate. The spatial resolution is determined based on the bearing size, with a typical value of a 2mm × 2mm grid. Boundary conditions are set considering the heat dissipation characteristics of the bearing housing; the outer boundary uses a convective heat transfer boundary, and the convection coefficient is determined based on environmental conditions. Temperature data normalization maps temperature values ​​to the 0-1 range, facilitating comparative analysis under different operating conditions. A hotspot identification algorithm finds local maxima in the temperature distribution from the temperature data; hotspot locations typically correspond to high friction or fault areas. Through the above spatial calibration process, discrete temperature data are converted into a continuous temperature distribution map, which fully describes the spatial temperature field characteristics of the bearing.

[0036] In some embodiments, the step of generating decoupling parameters based on the correlation analysis between the temperature distribution map and the component characteristic frequency includes: constructing a heat distribution matrix based on the temperature distribution map and the component characteristic frequency; extracting radial heat flow and axial heat flow from the heat distribution matrix respectively; generating a directional heat flow coefficient by comparing the difference between the radial heat flow and the axial heat flow; and determining the decoupling parameters by performing cross-analysis on the directional heat flow coefficient and the heat distribution matrix.

[0037] A heat distribution matrix is ​​constructed based on temperature distribution maps and component characteristic frequencies. Temperature gradients at various spatial locations are extracted from the temperature distribution maps, and combined with the energy density of the component's characteristic frequencies, a heat distribution matrix is ​​constructed using a coupling strength product approach. The matrix elements... Where H_ij is the thermal-vibrational coupling strength at position i and frequency j, Let E_j be the temperature gradient at location i (from the temperature distribution map), and E_j be the energy density of the component's characteristic frequency j. The matrix elements corresponding to hot spots in the temperature distribution map have their weights increased by 50%, reflecting the strengthening effect of local high temperatures on coupling strength. The heat distribution matrix has dimensions M×N, where M is the number of spatial grid points and N is the number of component characteristic frequencies. Frequency energy is calculated by integrating the power spectral density of the component's characteristic frequencies, with an integration bandwidth of ±5% of the characteristic frequency. Spatial weights are introduced to reflect the influence of the component weight table; elements in the heat distribution matrix of high-weight regions are multiplied by the corresponding weight coefficient. Time averaging eliminates instantaneous fluctuations, with the averaging time window set to 10 rotation cycles. The heat distribution matrix is ​​normalized using the Frobenius norm to ensure comparability of the matrices under different operating conditions. The construction of the heat distribution matrix completes the spatial-frequency correlation mapping between the temperature distribution map and the component's characteristic frequencies.

[0038] Radial and axial heat flows are extracted from the heat distribution matrix. Spatial temperature gradient components are extracted from the heat distribution matrix; the matrix element H_ij contains the temperature gradient at position i. Information. Based on the extracted temperature gradient, Fourier's law is applied to calculate radial and axial heat flow respectively. Heat flux density is proportional to the temperature gradient, and the thermal conductivity value is chosen considering material properties and temperature dependence, with a typical value of 45 W / (m·K). The integration path for radial heat flow is selected along the radial heat transfer path of the bearing, from the inner ring through the rolling elements to the outer ring. The integration path for axial heat flow is selected along the bearing axis, including the end seal and the bearing end heat dissipation path. The radial heat flow dominance zone identifies the main location of frictional heat generation, typically with the maximum radial heat flow in the load area. Axial heat flow analysis examines the heat dissipation of the end seal and bearing end; a large axial heat flow indicates good end heat dissipation. The radial and axial heat flow distribution data extracted from the heat distribution matrix provide a basis for analyzing the directional characteristics of heat flow.

[0039] A directional heat flux coefficient is generated by comparing the differences between radial and axial heat fluxes. Vector analysis is performed on the radial and axial heat fluxes to extract the modulus and direction information of the radial heat flux vector Q_r and the axial heat flux vector Q_a. The modulus ratio is calculated to reflect the relative intensity of the radial and axial heat fluxes; the vector angle is calculated to reflect the degree of deviation of the heat flux direction. Based on the comprehensive analysis of the modulus ratio and the vector angle, the directional heat flux coefficient D = sgn(R-1) × (1 - exp(-|R-1|)) × cos(θ) is generated, where R is the modulus ratio |Q_r| / |Q_a|, θ is the vector angle, and sgn is the sign function. The directional heat flux coefficient ranges from [-1, 1], with positive values ​​indicating radial heat flux dominance and negative values ​​indicating axial heat flux dominance; the absolute value reflects the degree of dominance. Spatial distribution characteristics are analyzed to understand the variation of the directional heat flux coefficient in different regions. In the load area, radial heat flux is usually dominant, and the directional heat flux coefficient shows positive values ​​and large values. By comparing the intensity and direction differences between radial and axial heat flow, directional heat flux coefficient distribution data describing the directional characteristics of heat flow are generated.

[0040] Decoupling parameters are determined through cross-analysis of the directional heat flux coefficient and the heat distribution matrix. For example, when radial heat flux is dominant (directional heat flux coefficient is 0.8) and high-frequency vibration energy is concentrated in the load area, the decoupling parameters extracted through cross-analysis reflect the coupling mode of "radial heat flux-high-frequency vibration". These decoupling parameters are used for temperature countermeasures to specifically eliminate the influence of radial temperature changes on high-frequency vibration. Matrix operations and singular value decomposition are used to cross-analyze the directional heat flux coefficient distribution and the heat distribution matrix to extract the main thermal-vibration coupling modes. The magnitude of the singular values ​​reflects the importance of each mode. The decoupling parameter vector consists of the eigenvectors corresponding to the first k singular values, with k chosen to achieve a cumulative contribution rate of 95%. The quantitative index of decoupling strength is defined as the ratio of the maximum singular value to the minimum singular value; a large ratio indicates that the decoupling parameters can effectively distinguish different coupling modes. The final set of decoupling parameters contains 3-5 main parameters, each corresponding to a specific thermal-vibration coupling mode, providing a quantitative basis for temperature countermeasures.

[0041] In some embodiments, the step of generating a countermeasure effect by performing temperature countermeasure processing based on the decoupling parameters includes: performing thermal diffusion path analysis on the decoupling parameters to form temperature propagation ripples; extracting the attenuation rate and propagation range from the temperature propagation ripples respectively; deducing the temperature interference start time based on the attenuation rate; and determining the countermeasure effect based on the temperature interference start time and the propagation range.

[0042] Thermal diffusion path analysis of the decoupling parameters generates temperature propagation ripples. For example, when the temperature in a certain area of ​​the bearing increases by 5°C due to friction, this temperature increase will increase the vibration signal amplitude by 0.1g (temperature-vibration coupling). Temperature countermeasures are implemented by applying reverse compensation to actively cool the area and offset the effect of the temperature increase. When the compensation effect reaches 80%, the temperature influence in the vibration signal is eliminated, retaining only the vibration characteristics generated by pure mechanical wear. The finite element method is used to solve the thermal diffusion equation, with initial conditions determined by the decoupling parameters, which are mapped to the initial temperature distribution. Boundary conditions consider the actual installation of the bearing, with the bearing housing contact surface set as an isothermal boundary. The time step is selected to satisfy the stability condition, ensuring stable convergence of the numerical solution. The morphology of the temperature propagation ripples is shown through the time evolution of isotherms, with the density of isotherms reflecting the magnitude of the temperature gradient. The propagation velocity is calculated by tracking the positional changes of specific isotherms in the temperature propagation ripples. Spatial attenuation characteristics are studied to investigate the change in the amplitude of the temperature propagation ripples with propagation distance; exponential attenuation indicates that heat conduction dominates. By analyzing the thermal diffusion path of the decoupled parameters, temperature propagation ripple data describing the spatiotemporal evolution characteristics of the temperature field are generated.

[0043] The attenuation rate and propagation range are extracted from the temperature propagation ripple. An exponential decay function is fitted based on the spatial variation of the temperature propagation ripple peak value to determine the attenuation rate. Temporal attenuation analysis is used to study the change in temperature propagation ripple intensity over time; the time constant reflects the rate of attenuation. The propagation range is defined as the distance within which the temperature propagation ripple amplitude drops to 10% of its initial value, within which the temperature effect is significant. Directional propagation characteristics consider the anisotropy of the bearing structure; the radial and axial propagation ranges are typically different. Frequency dependence studies the attenuation characteristics of different frequency components; the attenuation rate of high-frequency components is generally faster, and the attenuation rate increases with frequency. The typical basic attenuation coefficient is 0.1 / m, with a reference frequency of 1kHz. From the spatial and temporal evolution characteristics of the temperature propagation ripple, two key parameters—attenuation rate and propagation range—are extracted to describe the attenuation law and scope of the temperature effect.

[0044] The onset time of temperature disturbance is deduced by inversely calculating the decay rate. Starting from the currently observed temperature field state, a reverse-time heat conduction model is established using the decay rate. Based on the decay rate characteristics, a time-reverse evolution method is used to deduce the current temperature field backward along the time axis, gradually reconstructing the historical evolution process of the temperature field. In the reverse evolution, the decay rate controls the reverse growth rate of the temperature field; the larger the decay rate, the faster the reverse growth. Through continuous reverse evolution, the temperature field returns to the initial perturbation state; this moment is the onset time of temperature disturbance. An iterative optimization method improves the accuracy of the inverse calculation. The objective function is the mean square error between the predicted and measured temperatures, and the convergence criterion is set as the time difference between adjacent iterations being less than 0.1 seconds. The physical rationality of the onset time of temperature disturbance is checked to ensure that the inverse calculation time is within a reasonable range. Through the inverse analysis of the decay rate, the onset time of temperature disturbance is accurately determined, providing a time reference for constructing the target temperature field.

[0045] For example, determining the countermeasure effect based on the temperature interference start time and the propagation range includes: constructing a temperature field to generate a target temperature field based on the temperature interference start time; performing deviation analysis between the target temperature field and the propagation range to obtain a temperature control deviation; applying active temperature compensation based on the temperature control deviation to form a temperature countermeasure field; and performing interactive analysis between the temperature countermeasure field and the propagation range to generate a countermeasure effect.

[0046] The target temperature field is generated by constructing a temperature field based on the initial moment of temperature disturbance. The initial temperature field is reconstructed based on the initial moment of temperature disturbance, and the temperature distribution at the initial moment of temperature disturbance is recovered from historical data using an interpolation method. The radial basis function is selected as the interpolation basis function. A time evolution simulation is performed starting from the initial moment of temperature disturbance, and the heat conduction equation is solved using an explicit time-progression scheme to extrapolate to the current moment to form the target temperature field. The establishment of the target temperature field considers the steady-state temperature distribution characteristics under ideal operating conditions, and the ideal temperature gradient is determined through numerical simulation or empirical models. Boundary temperature settings consider ambient temperature changes and heat dissipation conditions. The heat source distribution is determined based on the component weight table and the characteristic frequency energy of the components, with stronger heat sources set in high-weight, high-energy regions. The target temperature field, as the desired temperature distribution state, provides a benchmark for deviation analysis.

[0047] The temperature control deviation is obtained by performing deviation analysis between the target temperature field and the propagation range. Deviation analysis is performed point-by-point within the propagation range, calculating the difference between the target temperature field value T_target and the measured temperature field value T_actual at each spatial grid point, where the deviation ΔT = T_target - T_actual. The deviation values ​​of each grid point are combined to obtain the spatial distribution field of the temperature control deviation. Statistical analysis is performed to calculate the mean, standard deviation, and maximum value of the temperature control deviation. The root mean square deviation σ_rms = sqrt(mean(ΔT²)) reflects the overall control accuracy. The deviation gradient is calculated using the finite difference method. Identify areas of drastic temperature control deviation changes; regions with large gradients require finer control. Track the temporal evolution of the temperature control deviation over time; an upward trend indicates deteriorating control effectiveness. By comparing the target temperature field with the actual temperature distribution within the propagation range, obtain temperature control deviation data to determine the magnitude and location of the temperature deviation requiring compensation.

[0048] Active temperature compensation is applied to the temperature control deviation to form a temperature countermeasure field. Based on the principle of applying compensation in the opposite direction to the temperature control deviation, the temperature control deviation is determined by sign: cooling compensation is applied to regions with positive deviation (actual temperature higher than the target temperature), and heating compensation is applied to regions with negative deviation (actual temperature lower than the target temperature). The calculation of the compensation intensity comprehensively considers the magnitude and frequency characteristics of the temperature control deviation. The compensation power P(x,f) = C_base × |ΔT(x)| × H(f), where P is the compensation power, C_base is the basic compensation coefficient (typically 20W / K), ΔT(x) is the temperature control deviation at position x, and H(f) is the frequency response function (the coefficient decreases with higher frequencies, with a cutoff frequency of 500Hz). For each spatial location within the propagation range, the corresponding compensation power is calculated based on its temperature control deviation value, and the application direction is opposite to the deviation. By applying active temperature compensation to the temperature control deviation, a temperature countermeasure field opposite to the temperature interference is formed within the propagation range. The intensity distribution of the temperature countermeasure field is determined by the magnitude of the temperature control deviation; locations with larger deviations have higher compensation intensity, forming a countermeasure field with a spatial distribution opposite to the interference field. Temperature-resistant field counteracts the effects of temperature disturbances on vibration signals through an active compensation mechanism.

[0049] Interactive analysis of the temperature countermeasure field and propagation range is performed to generate the countermeasure effect. The effect of the temperature countermeasure field within the propagation range is analyzed, and the countermeasure efficiency is calculated by comparing the temperature change amplitude before and after applying the temperature countermeasure field. The countermeasure efficiency is defined as the ratio of the temperature reduction within the propagation range to the initial temperature interference. A ratio greater than 0.8 indicates excellent countermeasure effect, 0.6-0.8 is good, and a ratio below 0.6 requires optimization of the countermeasure parameters. The coupling relationship between temperature change and vibration signal within the propagation range is analyzed. By comparing the vibration amplitude changes under different temperature conditions, the temperature influence coefficient is extracted. The temperature influence coefficient reflects the change in vibration amplitude caused by a 1°C temperature change. It is obtained by sampling temperature-vibration data pairs at multiple locations within the propagation range and performing linear regression fitting to obtain the slope, with a typical value of 0.02 g / °C. The stability of the countermeasure effect is verified by repeated measurements in different areas of the propagation range, ensuring that the coefficient of variation of the temperature influence coefficient is less than 0.15 and the fluctuation of the countermeasure efficiency is less than 10%. The countermeasure effect output includes two core parameters: the temperature influence coefficient and the countermeasure efficiency, serving as the decoupling basis for pure vibration signal extraction.

[0050] Pure mechanical wear signals are identified and pure vibration data is generated using the adversarial effect. The temperature influence coefficient and adversarial efficiency obtained from adversarial effect analysis are used as decoupling criteria to separate the temperature influence from the original vibration signal. The original vibration signal comes from the bearing vibration signal collected by S110. The vibration signal is decomposed using the empirical mode decomposition method, breaking down the original signal into multiple intrinsic mode functions. Temperature-related vibration components are identified by calculating the cross-correlation coefficient between each mode function and temperature change; modes with a correlation coefficient greater than 0.7 are identified as temperature-related vibration components. The formula for extracting the pure mechanical wear signal is V_pure = V_total - T_vib, where V_pure is the pure mechanical wear signal, V_total is the total vibration signal, and T_vib is the temperature-related vibration component. The extraction effect is evaluated by the adversarial efficiency; when the adversarial efficiency is greater than 0.8, temperature decoupling is considered successful, and the pure mechanical wear signal is considered reliable. Residual influence is evaluated by calculating the proportion of temperature-related components in the decoupled pure mechanical wear signal; when the residual influence is less than 5%, the pure vibration standard is met. The extracted pure mechanical wear signals form pure vibration data, which eliminates temperature interference and truly reflects the mechanical wear state of the bearing.

[0051] Step S140: Construct an equivalent transformation matrix based on pure vibration data and noise data; standardize and transform the pure vibration data using the equivalent transformation matrix to generate equivalent vibration data; and perform correlation analysis between the equivalent vibration data and the standard vibration benchmark to generate standard state indicators.

[0052] Specifically, an equivalent transformation matrix is ​​constructed based on pure vibration data and noise data. Using a state-space mapping method, the pure vibration data under varying noise conditions is mapped to the standard noise conditions. The equivalent transformation matrix has an N×N dimension, where N is the characteristic dimension of the pure vibration data, typically 64 or 128. The dimension selection is determined through principal component analysis, retaining 99% of the information. Noise data normalization maps the actual noise level to the standard noise level, which is set to 80% of the rated noise. The noise dependence of vibration amplitude is described by a power-law model: A = A0 × (N / N0)^α, where A is the vibration amplitude, A0 is the baseline vibration amplitude, N is the actual noise level, N0 is the standard noise level, and α is the noise index, obtained through fitting experimental data, with a typical value between 1.5 and 2.5. The equivalent transformation matrix is ​​constructed using singular value decomposition, containing an orthogonal matrix and a diagonal matrix. The diagonal elements are determined based on the power relationship of the noise data ratios. Phase compensation takes into account the phase shift caused by changes in noise data. The compensation angle is calculated using the noise difference and time, achieving a phase compensation accuracy of ±1 degree. Through the above construction process, a complete equivalent transformation matrix is ​​formed, which can uniformly map pure vibration data under different noise conditions to standard conditions.

[0053] In some embodiments, the step of standardizing the pure vibration data using the equivalent transformation matrix to generate equivalent vibration data includes: decomposing the pure vibration data according to the load dimension to obtain load vibration components; generating a noise correction factor based on the noise data; applying the equivalent transformation matrix to the load vibration components to generate intermediate transformation results; and standardizing the intermediate transformation results and the noise correction factor to obtain equivalent vibration data.

[0054] The pure vibration data is decomposed according to the load dimension to obtain load vibration components. Radial load vibration components are extracted from the pure vibration data using radial vibration sensor data, reflecting the radial force acting on the bearing. Components within the 0.5-5 times fundamental frequency range are retained during decomposition. Axial load vibration components are extracted from the axial vibration signals of the pure vibration data, including the effects of axial thrust and preload. Axial components are typically an order of magnitude smaller than radial components. The composite load decomposition considers the coupling effect of radial and axial load vibration components, employing an independent component analysis method with 100 iterations to ensure convergence. The non-uniformity of the load vibration component distribution is assessed through vibration differences in the circumferential direction. The load non-uniformity coefficient U, the ratio of the maximum to minimum vibration value in the circumferential direction, is used for subsequent weight adjustment. Through this decomposition process, complete radial and axial load vibration component data are obtained from the pure vibration data, providing a basis for equivalent transformation.

[0055] Noise correction factors are generated based on noise data. First, the actual noise level N is extracted from the noise data. Power spectrum analysis is performed on the noise data to calculate the average energy within a specific frequency range as the noise level. The noise correction factor adopts a negative power law relationship with the noise ratio: K_N = (N / N0)^(-α), where K_N is the noise correction factor, N is the actual noise level, N0 is the standard noise level (80% of the rated noise), and α is the noise index (derived from the power-law model used in constructing the equivalent transformation matrix, typically 1.5-2.5). The noise correction factor exhibits the opposite noise dependence to the vibration amplitude. When the actual noise is higher than the standard noise, the noise correction factor is less than 1, used to suppress the increase in spurious vibration caused by noise. When the actual noise is lower than the standard noise, the noise correction factor is greater than 1, used to compensate for the attenuation of vibration amplitude caused by the reduction in noise. The physical meaning of the noise correction factor is to compensate for the impact of changes in noise data on the vibration amplitude, making vibrations under different noise conditions comparable. The noise correction factor is calculated in real time based on the current noise data to ensure synchronization with actual noise conditions.

[0056] An equivalent transformation matrix is ​​applied to the load vibration components to generate intermediate transformation results. The equivalent transformation matrix is ​​used to transform the load vibration components, mapping them to standard noise conditions through matrix multiplication. Block matrix multiplication is employed to improve computational efficiency, with a block size of 32×32. The weights of the load vibration components are dynamically set according to the load type and magnitude; the typical weight for radial load vibration components is 0.7, and for axial load vibration components it is 0.3. Non-uniform loads are compensated for using a scaling factor (2-U) / (1+U), where U is the load non-uniformity coefficient. An equivalent transformation matrix operation is then applied to the weighted load vibration components to complete the mapping from actual noise conditions to standard noise conditions. Phase compensation, combined with the aforementioned phase offset information, applies corresponding phase correction to the transformed load vibration components to ensure time synchronization. The intermediate transformation results generated by transforming the load vibration components using the equivalent transformation matrix retain the fault characteristics of the original vibration while eliminating the influence of noise variations.

[0057] The intermediate transformation result and the noise correction factor are subjected to a normalization transformation to obtain equivalent vibration data. The normalization transformation of the intermediate transformation result and the noise correction factor is divided into three steps. In the first step, the L2 norm of the intermediate transformation result is calculated, ||V_mid||_2 = sqrt(Σ(V_mid[i]²)), where the squares of all sampling points of the intermediate transformation result are summed and then square-rooted to obtain the total energy value. In the second step, the energy normalization of the intermediate transformation result is performed, V_norm = V_mid / ||V_mid||_2, where the intermediate transformation result is divided by its norm to make the energy of the normalized signal equal to 1 and the dynamic range unified. In the third step, the normalized intermediate transformation result is multiplied by the noise correction factor, V_eq = K_N × V_norm, to complete the final normalization transformation. The complete formula for the normalization transformation is V_eq = K_N × V_mid / ||V_mid||_2, where V_eq is the equivalent vibration data, K_N is the noise correction factor, and V_mid is the intermediate transformation result. The role of the noise correction factor is to further compensate for the influence of noise changes on the basis of energy normalization, so that the equivalent vibration data corresponds to the standard noise condition. Through the normalization transformation, equivalent vibration data is obtained, and the equivalent vibration data eliminates the influence of noise changes and energy fluctuations, and can be used for condition assessment under unified standard conditions.

[0058] The equivalent vibration data and the standard vibration reference are subjected to a correlation analysis to generate a standard condition index. The equivalent vibration data and the standard vibration reference are subjected to a correlation analysis. The cross-correlation function is used to calculate the similarity between the equivalent vibration data and the standard vibration reference, and normalization is performed to ensure that the result is in the range of [0, 1]. The correlation coefficient R_max between the equivalent vibration data and the standard vibration reference reflects the degree of state similarity. A correlation coefficient greater than 0.8 indicates a normal state, 0.6 - 0.8 indicates a state of concern, and less than 0.6 indicates an abnormal state. The spectral coherence analysis calculates the correlation degree between the equivalent vibration data and the standard vibration reference in different frequency components, and the average coherence value γ²_avg quantifies the similarity in the frequency domain. The feature matching analysis compares the characteristic frequency distributions of the equivalent vibration data and the standard vibration reference, and the matching degree F_match reflects the consistency of the fault characteristics. The standard condition index S = 100 × (0.5 × R_max + 0.3 × γ²_avg + 0.2 × F_match), where S is the standard condition index, comprehensively reflecting the correlation degree between the equivalent vibration data and the standard vibration reference. The classification of the standard condition index: S > 70 is healthy, 40 < S ≤ 70 is slightly abnormal, 20 < S ≤ 40 is moderately faulty, and S ≤ 20 is severely faulty, providing a quantitative basis for the bearing condition assessment.

[0059] Step S150, perform wear tracking and vibration transfer analysis on the standard condition index to generate transfer parameters, and perform path reconstruction processing according to the transfer parameters to generate a path control signal.

[0060] Specifically, wear tracking and vibration transmission analysis are performed on standard state indicators to generate transfer parameters. An incremental accumulation algorithm is used to accumulate the decrease in the standard state indicator at each time step. The accumulation calculation formula is C(t) = C(t-1) + max(0, I(t-1) - I(t)), where C(t) is the cumulative wear value at time t, and I(t) is the standard state indicator at time t. Wear stage identification is based on the slope change of the accumulation curve. The slope is larger in the initial wear stage, gentler in the stable wear stage, and sharply increases in the accelerated wear stage. The accumulation curve is matched with typical failure modes, and the failure type and severity are determined by combining the standard state indicator values. When the standard state indicator is below 40, a severe failure mode is prioritized; when the standard state indicator is between 40 and 70, a moderate failure mode is matched; and when the standard state indicator is above 70, an early failure mode is matched. Spatial location is determined by vibration energy distribution. The bearing circumference is divided into 12 sectors, and the sector with the highest energy is the failure location. Depth location assesses the radial depth of the failure, and the failure size is estimated based on the harmonic components of the characteristic frequency. Vibration transmission analysis was conducted, and the vibration intensity index of the fault area was calculated based on the fault location, depth, and size information to form a comprehensive intensity assessment. The vibration intensity level was divided into four levels: slight, moderate, severe, and dangerous, each corresponding to different transmission characteristics and attenuation patterns. The vibration generated at the fault location was transmitted to the measuring point through the bearing structure. The transfer function model considered the effects of material damping and structural damping. Transmission characteristic analysis revealed that the transmission path was clear and the attenuation coefficient was small in the initial wear stage; the attenuation coefficient increased by 20-30% in the accelerated wear stage. The attenuation coefficient calculation considered distance and medium characteristics; the low-frequency components generated by large-size faults showed less attenuation. Frequency selectivity analysis established a frequency-attenuation curve. The transfer function, attenuation coefficient, path weight, frequency selectivity characteristics, and transfer delay were integrated into transfer parameters, which describe the transmission characteristics of vibration from the fault source to the measuring point.

[0061] In some embodiments, the step of generating a path control signal by performing path reconstruction processing based on the transfer parameters includes: adjusting the vibration path according to the transfer parameters to obtain an initial vibration layer; performing a first-level amplification analysis based on the initial vibration layer to obtain an enhanced vibration layer; performing a second-level screening analysis on the enhanced vibration layer to obtain a core vibration layer, wherein the core vibration layer includes wear-related vibration, lubrication characteristic vibration, and structural stability vibration; and performing multi-level vibration fusion processing on the core vibration layer to determine the path control signal.

[0062] The initial vibration layer is obtained by adjusting the vibration path based on the transfer parameters. Inverse reconstruction is performed based on the transfer function in the transfer parameters, and the source signal is recovered through deconvolution. For example, if the sensor measures a mixed vibration signal amplitude of 6 mm / s, after deconvolution processing using the transfer function H(ω), the vibration amplitude at the fault source is recovered to be 10 mm / s. A blind source separation algorithm is used to separate multiple paths, decomposing the measured signal into direct path, single reflection, and multiple reflection components. The separation matrix is ​​determined using the FastICA algorithm. Weight allocation is set according to the path weights in the transfer parameters. For example, the recovered 10 mm / s vibration is decomposed by path as follows: direct path 6 mm / s (weight 0.6), single reflection path 3 mm / s (weight 0.3), and multiple reflection path 1 mm / s (weight 0.1), ensuring that the influence of the main path is fully reflected. Phase correction compensates for the propagation delay of different paths. The phase correction amount for each frequency component is calculated based on the transfer delay in the transfer parameters. For example, the arrival time of the direct path is 0 ms, the single reflection path has a delay of 2 ms, and the multiple reflection path has a delay of 5 ms. Phase correction aligns the signals of the three paths in time. Amplitude recovery is performed using inverse compensation based on the attenuation coefficient in the transfer parameters to compensate for energy loss during propagation along the three paths, thus restoring the true amplitude of the fault source. Frequency equalization compensates for frequency-selective attenuation in the transfer parameters by designing inverse filters based on the attenuation differences of different frequencies along each path, and employing the minimum phase method to ensure system stability. Through the above path adjustment processing based on transfer parameters, initial vibration layer data is obtained, and the initial vibration layer completes the preliminary reconstruction from the vibration at the measuring point to the vibration of the fault source.

[0063] The enhanced vibration layer is obtained through first-stage amplification analysis based on the initial vibration layer. This first-stage amplification employs the principle of resonant amplification, identifying components in the initial vibration layer that are close to the system's natural frequency. The resonant quality factor determines the amplification factor. The frequency-selective amplifier design uses a second-order bandpass filter bank, with the center frequency covering the main fault characteristic frequencies. The Q value of each filter is set to 10-20. Gain scheduling is adaptively adjusted according to the signal strength of the initial vibration layer: 20dB for weak signals, 10dB for medium signals, and 0dB for strong signals to prevent over-amplification. Nonlinear amplification considers the impulse characteristics of the initial vibration layer signal, employing square-law amplification after envelope detection: A_enhanced = A_input × (1 + k × A_envelope²), where A_enhanced is the enhanced amplitude, A_input is the initial vibration layer input amplitude, A_envelope is the envelope amplitude, and k is a dimensionless nonlinear coefficient (typically 0.5 after amplitude normalization). Through the above amplification analysis, the weak fault features in the initial vibration layer are amplified to generate an enhanced vibration layer. The enhanced vibration layer has a higher signal-to-noise ratio, which facilitates subsequent fault feature identification.

[0064] A secondary screening analysis was performed on the enhanced vibration layer to obtain the core vibration layer. This secondary screening employed a multi-criteria decision-making method, comprehensively considering frequency characteristics, energy distribution, and time-domain modes. Wear-related vibrations were identified from the enhanced vibration layer through characteristic frequency matching, with a matching tolerance set at ±2% of the theoretical frequency. The matching score was calculated using an exponential function. Lubrication-related vibrations were extracted from the enhanced vibration layer based on modulation analysis. The carrier frequency was the bearing's characteristic frequency, and the modulation frequency was an integer multiple of the fundamental frequency. A modulation depth greater than 0.1 indicated poor lubrication. Structural stability vibrations were identified from the enhanced vibration layer through modal analysis. A deviation of less than 5% compared to the modal frequencies calculated by finite element analysis was considered structural vibration. Energy threshold screening removed low-energy components from the enhanced vibration layer, with a threshold set at 1% of the total energy, retaining the main vibration components. Classification confidence was calculated based on a comprehensive score of multiple features. Components with a confidence score greater than 0.8 were selected from the enhanced vibration layer and included in the core vibration layer. The core vibration layer contains three core components: wear-related vibration, lubrication characteristic vibration, and structural stability vibration, providing accurate vibration characteristics for the generation of path control signals.

[0065] Multi-level vibration fusion processing is performed on the core vibration layer to determine the path control signal. A layered weighted strategy is adopted for multi-level vibration fusion processing of the core vibration layer. The first layer is the vibration type layer, with wear-related vibrations having a weight of 0.5, lubrication characteristic vibrations having a weight of 0.3, and structural stability vibrations having a weight of 0.2. The weights are dynamically adjusted according to the severity of the fault. The second layer is the frequency layer, where various vibrations in the core vibration layer are weighted according to frequency bands: low frequency band (0-2kHz) has a weight of 0.4, mid frequency band (2-5kHz) has a weight of 0.4, and high frequency band (5-10kHz) has a weight of 0.2, reflecting the diagnostic value of different frequency bands. The third layer is the time layer, considering the time-varying characteristics of the core vibration layer vibration, with recent data having a higher weight than historical data, and the weights are allocated according to an exponential decay law. The fusion algorithm adopts the evidence theory framework, S_control=Σ_iΣ_jΣ_k(w_i×w_j×w_k×V_ijk), where S_control is the path control signal, i is the vibration type index (i=1,2,3, corresponding to wear correlation, lubrication characteristics, and structural stability, respectively), j is the frequency band index (j=1,2,3, corresponding to low frequency, mid frequency, and high frequency, respectively), k is the time index, w_i, w_j, and w_k are the three-layer weights, and V_ijk is the corresponding vibration component. Signal normalization ensures that the amplitude range of the path control signal is within the [-1, 1] interval, facilitating its use in subsequent control systems. Time-series coding arranges the fusion results in chronological order, forming a continuous path control signal sequence, with the sampling rate consistent with the original vibration signal.

[0066] Step S160: Generate weight adjustment data based on the weight table of the path control signal adjustment components, and generate a graded early warning signal based on the weight adjustment data.

[0067] In some embodiments, the step of adjusting the component weight table based on the path control signal to generate weight adjustment data includes: identifying weight adjustment requirements based on the path control signal; applying weight changes within the range of the weight adjustment requirements to obtain component response data; identifying component points with weight sensitivity jumps from the component response data; and adjusting the weight values ​​of the component points to generate weight adjustment data.

[0068] Weight adjustment requirements are identified based on path control signals. This is achieved by analyzing the spectral characteristics and time-domain patterns of the path control signals. The dominant frequency components of the path control signals are extracted and compared with the characteristic frequencies of the components. A peak search is performed in the power spectrum of the path control signals to find significant peaks. The peak detection threshold is set to three times the average power, and the detection bandwidth is ±5% of the characteristic frequency. The matching degree between the peak and the component is M = exp(-|f_peak-f_component| / f_component), where M is the matching degree, f_peak is the peak frequency of the path control signal, and f_component is the characteristic frequency of the component. A matching degree greater than 0.8 indicates that the component requires weight adjustment. The adjustment direction is determined based on the phase information of the path control signals. A leading phase indicates that the weight needs to be increased, while a lagging phase indicates that the weight needs to be decreased. The adjustment amplitude is estimated by the ratio of the amplitude of the path control signal to the baseline value. In the case of multi-component competition, priority ranking is based on the product of the matching degree and the amplitude, prioritizing the adjustment of components with higher scores. The weight adjustment requirement identification results include a list of components to be adjusted, the direction of adjustment, the suggested magnitude, and the urgency level, forming a structured description of the weight adjustment requirement.

[0069] Weight changes were applied within the weight adjustment requirement range to acquire component response data. A controlled experimental method was used to gradually change the weight values ​​while maintaining system stability. The initial disturbance amplitude was determined based on the suggested amplitude and urgency level of the weight adjustment requirement; the suggested amplitude was used directly for urgent needs, 50% for routine needs, and 30% for non-urgent needs. The disturbance sequence was a pseudo-random binary sequence with a length of 127. The data acquisition window was 10 rotation cycles after the disturbance was applied, with a sampling rate of 25.6 kHz to ensure complete capture of the response process. Response feature extraction included rise time, overshoot, settling time, and steady-state error, forming a response feature vector. The acquired response features, time-domain waveforms, and frequency-domain characteristics were integrated into component response data, which comprehensively describes the dynamic response characteristics of the component to weight changes.

[0070] Identify component points with abrupt changes in weight sensitivity from component response data. Sensitivity is defined as... Where S represents sensitivity, R represents the vibration response amplitude of the component response data, and W represents the component weight value, calculated approximately using finite difference. Sensitivity calculation comprehensively considers the transfer function gain and time-domain response characteristics in the component response data. The sensitivity curve is plotted using cubic spline interpolation, with a weight range of 0.05-0.5 and a step size of 0.05, obtaining a continuous sensitivity distribution. The jump detection algorithm is based on derivative analysis, calculating the first derivative of sensitivity; points with an absolute derivative value greater than the threshold of 10 are marked as jump points. Jump types are classified into positive jumps (sudden increase in sensitivity) and negative jumps (sudden decrease in sensitivity), with different processing strategies employed for different types. Jump amplitude quantification is calculated by the difference in sensitivity before and after the jump; an amplitude greater than 5 is considered a significant jump. Components with significantly changed weight sensitivity are defined as component points, which are the key targets for weight adjustment.

[0071] Weight adjustment data is generated by adjusting the weight values ​​of component points. For identified component points, weight adjustments are made based on jump characteristics. Component points with positive jumps are shifted forward to keep the system operating in a low-sensitivity region, while component points with negative jumps maintain their current weights to avoid entering a high-sensitivity region. The larger the jump amplitude, the larger the adjustment amount. The adjustment step size is determined comprehensively based on the sensitivity and jump amplitude of the component point. A small step size of 0.005 is used for component points in the high-sensitivity region (sensitivity greater than 10) with large jumps (jump value greater than 8), 0.02 for component points in the medium-sensitivity region (sensitivity 5-10), and 0.05 for component points in the low-sensitivity region (sensitivity less than 5). Iterative optimization uses the golden section search to find the optimal weight value of the component points within the feasible region, with a convergence accuracy of 0.001. The objective function is defined as minimizing vibration energy while avoiding the jump regions of component points. Constraints include weight range constraints (0.05-0.5), weight sum constraints (summing up to 1), and rate of change constraints. The weight adjustment data is fully recorded to form a structured data table containing five key pieces of information: the initial weight records the weight configuration before adjustment, the target weight reflects the ideal value determined by the optimization algorithm, the actual adjustment amount quantifies the magnitude of weight change, the expected effect is the percentage reduction of theoretical vibration energy calculated based on the objective function, and the verification result is the verification pass rate obtained by comparing measured data.

[0072] A tiered early warning signal is generated based on the weight adjustment data. The tiered early warning signal system adopts a four-level system: Normal (green), Attention (yellow), Warning (orange), and Danger (red), with each level corresponding to a different handling strategy. The early warning threshold is determined based on the actual adjustment amount in the weight adjustment data. The actual adjustment amount of all component points in the weight adjustment data is extracted, and the weight change rate is calculated as the ratio of the adjustment amount to the initial weight. The change rate thresholds are set at 5%, 10%, and 20% respectively, corresponding to the Attention, Warning, and Danger levels. The comprehensive scoring mechanism considers both the actual adjustment amount in the weighted adjustment data and the verification results. The change rate score is normalized to a value of 0-100 based on the weighted change rate; a change rate exceeding 20% ​​is awarded a full score of 100. The verification score is determined based on the verification pass rate in the weighted adjustment data: a pass rate above 90% is awarded 100 points, 80%-90% 80 points, 60%-80% 60 points, and below 60% 40 points. The comprehensive score is calculated by weighting the change rate score at 60% and the verification score at 40%. The reliability assessment of the early warning system is determined by comparing the deviation between the expected effect in the weighted adjustment data and the verification results. The deviation is calculated as the percentage difference between the two results and the expected effect. If the deviation is less than 10%, the early warning level remains unchanged; if the deviation is between 10% and 30%, the early warning level is reduced by one level; and if the deviation is greater than 30%, the early warning level is reduced by two levels, ensuring the accuracy of the early warning. Warning level determination rules: A comprehensive score below 30 is normal, 30-50 is attention, 50-70 is warning, and above 70 is danger. The graded warning signal coding adopts a standard format including level code, component number, timestamp, severity value, and reliability. The graded warning signal output provides a decision-making basis for the bearing health management system.

[0073] To implement the train running gear bearing condition prediction and early warning method corresponding to the above method embodiments, in order to achieve the corresponding functions and technical effects. See also Figure 2 , Figure 2 This diagram illustrates a structural block diagram of a train running gear bearing condition prediction and early warning system 200 according to an embodiment of this application. For ease of explanation, only the parts relevant to this embodiment are shown. The train running gear bearing condition prediction and early warning system 200 provided in this embodiment includes: Data acquisition module 201 is used to acquire bearing vibration signal, noise data and temperature data, generate a rotation period based on the noise data, synchronously segment the vibration signal according to the rotation period to form vibration segments, and fuse the vibration segments to generate a standard vibration reference. Feature extraction module 202 is used to extract component characteristic frequencies through the standard vibration reference, perform intermodulation analysis on the component characteristic frequencies to confirm fault information, perform fault frequency analysis on the fault information to determine the monitoring working frequency, and determine the component weight table based on the monitoring working frequency. Temperature processing module 203 is used to spatially calibrate the temperature data based on the component weight table to generate a temperature distribution map, perform correlation analysis between the temperature distribution map and the component characteristic frequency to generate decoupling parameters, perform temperature countermeasure processing based on the decoupling parameters to generate countermeasure effect, and use the countermeasure effect to identify pure mechanical wear signals to generate pure vibration data. The working condition standardization module 204 is used to construct an equivalent transformation matrix based on the pure vibration data and the noise data, perform standardization transformation on the pure vibration data through the equivalent transformation matrix to generate equivalent vibration data, and perform correlation analysis between the equivalent vibration data and the standard vibration benchmark to generate standard state indicators. The fault control module 205 is used to perform wear tracking and vibration transmission analysis on the standard state indicators to generate transfer parameters, and to perform path reconstruction processing based on the transfer parameters to generate path control signals. The early warning decision module 206 is used to adjust the component weight table based on the path control signal to generate weight adjustment data, and generate a graded early warning signal based on the weight adjustment data.

[0074] The train running gear bearing condition prediction and early warning system 200 described above can implement the train running gear bearing condition prediction and early warning method of the above method embodiments. The options in the above method embodiments are also applicable to this embodiment, and will not be detailed here. The remaining contents of this application embodiment can be referred to the contents of the above method embodiments, and will not be repeated in this embodiment.

[0075] The purpose of the above embodiments is to reproduce and derive the technical solution of the present invention by way of example, and to fully describe the technical solution, purpose and effect of the present invention. The purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosure of the present invention, and not to limit the scope of protection of the present invention.

[0076] The above embodiments are not an exhaustive list based on the present invention, and there may be many other embodiments not listed. Any substitutions and improvements made without departing from the concept of the present invention are within the protection scope of the present invention.

Claims

1. A method for predicting and warning the condition of bearings in train running gear, characterized in that, include: Acquire bearing vibration signals, noise data, and temperature data; generate a rotation period based on the noise data; synchronously segment the vibration signal according to the rotation period to form vibration segments; and fuse the vibration segments to generate a standard vibration reference. The component characteristic frequencies are extracted using the standard vibration reference, and the component characteristic frequencies are subjected to intermodulation analysis to confirm fault information. The fault information is then subjected to fault frequency analysis to determine the monitoring operating frequency. Finally, a component weight table is determined based on the monitoring operating frequency. Based on the component weight table, the temperature data is spatially calibrated to generate a temperature distribution map. Based on the temperature distribution map and the component characteristic frequency, a correlation analysis is performed to generate decoupling parameters. Temperature countermeasures are performed according to the decoupling parameters to generate countermeasures effect. The countermeasures effect is used to identify pure mechanical wear signals and generate pure vibration data. An equivalent transformation matrix is ​​constructed based on the pure vibration data and the noise data. The pure vibration data is then standardized and transformed using the equivalent transformation matrix to generate equivalent vibration data. The equivalent vibration data is then correlated with the standard vibration benchmark to generate standard state indicators. Wear tracking and vibration transmission analysis are performed on the standard state indicators to generate transfer parameters, and path reconstruction processing is performed based on the transfer parameters to generate path control signals; Based on the path control signal, the component weight table is adjusted to generate weight adjustment data, and a graded early warning signal is generated based on the weight adjustment data.

2. The method according to claim 1, characterized in that, The step of fusing the vibration segments to generate a standard vibration reference includes: Identify the abrupt change points of wave peaks within the vibration segment; The vibration segment is decomposed into multiple steady-state segments based on the wave peak abrupt change point; The vibrational energy of the steady-state segments is obtained and accumulated to form a total energy value; The total energy value is then subjected to waveform compensation correction to generate a standard vibration reference.

3. The method according to claim 1, characterized in that, The step of determining the component weight table based on the monitoring operating frequency includes: Identify resonance characteristics from the monitored operating frequency; The aforementioned resonance characteristics are used to generate a secondary resonance inside the bearing; The characteristic frequency of the component is superimposed with the secondary resonance to form an enhanced resonance; The enhanced resonance is weighted to generate a component weight table.

4. The method according to claim 1, characterized in that, The step of generating decoupling parameters based on the correlation analysis between the temperature distribution map and the component characteristic frequency includes: A heat distribution matrix is ​​constructed based on the temperature distribution map and the characteristic frequency of the component. Radial heat flow and axial heat flow are extracted from the heat distribution matrix, respectively; The difference between the radial heat flow and the axial heat flow is used to generate a directional heat flow coefficient; The decoupling parameters are determined by cross-analysis of the directional heat flux coefficient and the heat distribution matrix.

5. The method according to claim 1, characterized in that, The step of generating a countermeasure effect by performing temperature countermeasure processing based on the decoupling parameters includes: Thermal diffusion path analysis is performed on the decoupling parameters to generate temperature propagation ripples; The attenuation rate and propagation range are extracted from the temperature propagation ripples, respectively. The onset time of the temperature disturbance is deduced based on the attenuation rate. The effectiveness of the countermeasures is determined based on the start time of the temperature interference and the propagation range.

6. The method according to claim 1, characterized in that, The process of standardizing and transforming the pure vibration data using the equivalent transformation matrix to generate equivalent vibration data includes: The pure vibration data is decomposed according to the load dimension to obtain the load vibration components; A noise correction factor is generated based on the noise data; The load vibration components are transformed using the equivalent transformation matrix to generate intermediate transformation results; The intermediate transformation result is standardized and converted with the noise correction factor to obtain equivalent vibration data.

7. The method according to claim 1, characterized in that, The step of generating a path control signal by performing path reconstruction processing based on the transfer parameters includes: The initial vibration layer is obtained by adjusting the vibration path according to the transfer parameters. Based on the initial vibration layer, a first-level amplification analysis is performed to obtain the enhanced vibration layer; A secondary screening analysis is performed on the enhanced vibration layer to obtain the core vibration layer, which includes wear-related vibration, lubrication characteristic vibration and structural stability vibration. The core vibration layer is subjected to multi-level vibration fusion processing to determine the path control signal.

8. The method according to claim 1, characterized in that, The step of adjusting the component weight table based on the path control signal to generate weight adjustment data includes: The weight adjustment requirements are identified based on the path control signals. Within the range of weight adjustment requirements, weight changes are applied to obtain component response data; Identify component points with weight sensitivity jumps from the component response data; The weight values ​​of the component points are adjusted to generate weight adjustment data.

9. The method according to claim 5, characterized in that, The determination of the countermeasure effect based on the onset time of the temperature interference and the propagation range includes: The target temperature field is generated by constructing a temperature field based on the initial time of the temperature disturbance. The temperature control deviation is obtained by performing a deviation analysis between the target temperature field and the propagation range. Active temperature compensation is applied to form a temperature resistance field based on the temperature control deviation; The interaction between the temperature resistance field and the propagation range generates a resistance effect.

10. A predictive and early warning system for the condition of bearings in a train running gear, characterized in that, include: The data acquisition module is used to acquire bearing vibration signals, noise data, and temperature data; generate a rotation period based on the noise data; synchronously segment the vibration signal according to the rotation period to form vibration segments; and fuse the vibration segments to generate a standard vibration reference. The feature extraction module is used to extract component feature frequencies through the standard vibration reference, perform intermodulation analysis on the component feature frequencies to confirm fault information, perform fault frequency analysis on the fault information to determine the monitoring operating frequency, and determine the component weight table based on the monitoring operating frequency. The temperature processing module is used to spatially calibrate the temperature data based on the component weight table to generate a temperature distribution map, perform correlation analysis between the temperature distribution map and the component characteristic frequency to generate decoupling parameters, perform temperature countermeasure processing based on the decoupling parameters to generate countermeasure effects, and use the countermeasure effects to identify pure mechanical wear signals to generate pure vibration data. The working condition standardization module is used to construct an equivalent transformation matrix based on the pure vibration data and the noise data, perform standardization transformation on the pure vibration data through the equivalent transformation matrix to generate equivalent vibration data, and perform correlation analysis between the equivalent vibration data and the standard vibration benchmark to generate standard state indicators. The fault control module is used to perform wear tracking and vibration transmission analysis on the standard state indicators to generate transfer parameters, and to perform path reconstruction processing based on the transfer parameters to generate path control signals. The early warning decision module is used to adjust the component weight table based on the path control signal to generate weight adjustment data, and to generate a graded early warning signal based on the weight adjustment data.

Images