A bearing wear state real-time monitoring system

Abstract

This invention discloses a real-time bearing wear condition monitoring system, belonging to the field of bearing detection technology. It solves the problems encountered during bearing processing, such as progressive wear of machine tool spindle bearings leading to decreased rotational accuracy, difficulty in capturing attenuated vibration signals, and temperature rise in the grinding zone caused by fluctuations in grinding wheel cutting depth, resulting in microscopic burns on the workpiece surface. These burn layers exacerbate grinding wheel blockage, leading operators to misdiagnose the problem as wheel passivation and repeatedly overhaul the machine. The system includes a multi-source sensor unit for real-time acquisition of multi-source sensor signals during machine tool processing; and a signal conditioning and data acquisition module. This invention obtains multi-source sensor signals, preprocesses them to extract instantaneous rotational speed curves, performs angular domain resampling and signal conversion, extracts bearing fault features, grinding wheel state features, and thermal effect feature vectors, fuses and standardizes them, inputs them into a pre-trained classification model, calculates the probability of spindle wear and grinding wheel passivation, and outputs a diagnostic category based on threshold comparison.

Description

Technical Field

[0001] This invention relates to the field of bearing testing technology, and in particular to a real-time bearing wear condition monitoring system. Background Technology

[0002] Real-time bearing wear monitoring is a crucial technology in modern intelligent manufacturing, ensuring stable production and product quality. It primarily assesses bearing operating conditions by collecting various physical signals. Monitoring includes vibration and acoustic emission, temperature, displacement, surface morphology, and lubricating oil status, reflecting the degree and trend of wear from different perspectives. Related technologies fall into three categories: physical parameter monitoring, geometric measurement, and intelligent embedded sensing, utilizing methods such as vibration analysis, capacitance and laser measurement, and fiber optic and MEMS sensing, respectively. Current technologies are evolving towards multi-sensor information fusion, AI-based fault diagnosis, and cloud-edge-device collaborative data processing. Adaptive solutions can be selected based on on-site accuracy, cost, and environmental conditions, helping companies shift from reactive maintenance to predictive maintenance, thereby improving equipment operating efficiency and production safety.

[0003] During the bearing manufacturing process, the high-precision bearings inside the machine tool spindle that support the grinding wheel undergo progressive micro-wear due to long-term service, resulting in a slight decrease in spindle rotation accuracy. This minute change is caused by the wear source being deep inside the machine tool, and its vibration signal must be transmitted through a complex mechanical structure before it can be picked up by external sensors, resulting in significant signal attenuation and difficulty in early detection. This causes spindle runout, leading to periodic fluctuations in the grinding wheel's cutting depth, which causes a sharp increase in instantaneous temperature in the grinding zone. This results in micro-scale burns on the workpiece surface that are invisible to the naked eye. The burn layer, due to changes in its material properties, in turn exacerbates the clogging and abnormal wear of the grinding wheel. As observed by on-site operators, the typical appearance of the grinding wheel is dull and the vibration is aggravated, leading to a cycle of misjudgment where "repeated grinding wheel dressing does not solve the problem." Ultimately, all bearing workpieces produced during this fault investigation period are shipped off the production line with irreversible burn layers. These workpieces, which appear to be of acceptable quality but actually have hidden damage, only reveal their abnormalities when they are transferred to the finished product noise testing or durability testing stage, causing batch scrapping and serious quality traceability issues.

[0004] Therefore, a real-time bearing wear condition monitoring system is proposed to solve or alleviate the above problems. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a real-time bearing wear condition monitoring system.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A real-time bearing wear condition monitoring system, including A multi-source sensor unit is used to acquire multi-source sensor signals in real time during machine tool processing. The signal conditioning and data acquisition module is used to perform signal conditioning and analog-to-digital conversion on the received multi-source sensor signals to obtain digitized multi-channel raw signals. The data preprocessing module is used to resample, bandpass filter, and extract instantaneous speed curves from the multi-channel raw signals to obtain the preprocessed signals and instantaneous speed curves. The angular domain resampling module is used to convert the time-domain signal of the preprocessed signal into an angular domain signal based on the instantaneous rotational speed curve. The bearing fault feature extraction module is used to extract bearing fault feature vectors including the order energy of various fault types based on the angular domain signal and preset bearing geometric parameters. The grinding wheel state feature extraction module is used to extract the grinding wheel passivation index, which characterizes the degree of grinding wheel passivation, based on the acoustic emission signal in the preprocessed signal through wavelet packet decomposition. The thermal effect feature extraction module is used to extract thermal effect feature vectors that reflect grinding force fluctuations and thermal damage risks based on the grinding force signal, instantaneous speed curve, and bearing geometric parameters in the preprocessed signal. The feature fusion and standardization module is used to fuse the bearing fault feature vector, grinding wheel passivation index and thermal effect feature vector into a total feature vector, and then perform standardization processing to generate a standardized feature vector. The diagnostic model module is used to calculate the spindle wear probability and grinding wheel passivation probability based on the input standardized feature vector through a multi-output logistic regression classification model. The diagnostic output and display module is used to determine the diagnostic category based on the comparison results of the spindle wear probability and grinding wheel passivation probability with preset thresholds, and to generate a diagnostic report for display. The data storage and traceability module is used to bind and store diagnostic results and their associated feature data with the currently processed workpiece ID, and provides a historical data traceability interface.

[0007] Preferably, the output terminal of the multi-source sensor unit is connected to the input terminal of the signal conditioning and data acquisition module; the output terminal of the signal conditioning and data acquisition module is connected to the input terminal of the data preprocessing module; the first output terminal of the data preprocessing module is connected to the first input terminal of the angular domain resampling module; the second output terminal of the data preprocessing module is connected to the second input terminals of both the angular domain resampling module and the thermal effect feature extraction module; the output terminal of the angular domain resampling module is connected to the input terminals of both the bearing fault feature extraction module and the thermal effect feature extraction module; and the output terminal of the bearing fault feature extraction module is connected to the feature fusion module. The first input terminal of the feature fusion and standardization module is connected to the first input terminal of the grinding wheel state feature extraction module. The input terminal of the grinding wheel state feature extraction module is connected to the second input terminal of the feature fusion and standardization module. The output terminal of the thermal effect feature extraction module is connected to the third input terminal of the feature fusion and standardization module. The output terminal of the feature fusion and standardization module is connected to the input terminal of the diagnostic model module. The output terminal of the diagnostic model module is connected to the input terminal of the diagnostic output and display module. The control output terminal of the diagnostic output and display module is connected to the input terminal of the data storage and traceability module.

[0008] Preferably, the steps for resampling, bandpass filtering, and extracting instantaneous speed curves from the multi-channel raw signal to obtain the preprocessed signal and instantaneous speed curves specifically include the following steps: It receives multi-channel digital signals and resamples all signals to a uniform maximum sampling rate through anti-aliasing filtering and cubic spline interpolation to ensure time alignment. Bandpass filtering was performed according to the signal type. Specifically, vibration signals were filtered using a 0.5 Hz to 10 kHz bandpass filter to retain bearing fault characteristic frequencies; acoustic emission signals were filtered using a 10 kHz high-pass filter to remove low-frequency mechanical noise; current signals were filtered using a 0.1 Hz to 2 kHz bandpass filter to retain load fluctuation information; and grinding force signals were filtered using a 0.1 Hz to 5 kHz bandpass filter. Instantaneous speed extraction: If a speed pulse signal exists, calculate the reciprocal of the time interval between adjacent pulses and multiply it by the number of pulses per revolution to obtain the instantaneous speed. Then, obtain a continuous instantaneous speed curve through linear interpolation. If there is no speed pulse, perform a short-time Fourier transform on the spindle motor current signal, extract the frequency component with the largest amplitude in the time spectrum as the instantaneous rotation frequency, and multiply it by 60 to obtain the instantaneous speed curve. The preprocessed signal and instantaneous speed curve are output to subsequent modules.

[0009] Preferably, the step of converting the time-domain signal of the preprocessed signal into an angular-domain signal based on the instantaneous rotational speed curve specifically includes the following steps: Receive the preprocessed signal and instantaneous speed curve, and calculate the cumulative spindle rotation angle based on the instantaneous speed curve; Set the number of sampling points per revolution, calculate the angular domain sampling interval, and generate an angle sequence with equal angular intervals; By using cubic spline interpolation, the time-domain amplitude of the preprocessed signal is mapped onto the angle sequence to obtain the angular domain signal, and the angular domain signal is output to the bearing fault feature extraction module and the thermal effect feature extraction module.

[0010] Preferably, the step of extracting a bearing fault feature vector including specific energies of multiple fault types based on the angular domain signal and preset bearing geometric parameters specifically includes the following steps: It receives vibration and current signals from the angular domain signal, as well as preset bearing geometric parameters; The theoretical failure order ratios of the inner ring, outer ring, rolling elements, and cage are calculated based on geometric parameters. Bandpass filtering is performed on the angular domain vibration signal and current signal within the preset high-frequency resonance band; Perform a Hilbert transform on the filtered signal and take the modulus to obtain the envelope signal; Perform a Fourier transform on the envelope signal to obtain the order spectrum; Within a preset bandwidth range near each theoretical fault order, the peak amplitude of the order spectrum is extracted; The peak amplitudes of the four fault types extracted from the vibration and current signals are combined to form an 8-dimensional bearing fault feature vector, and this feature vector is output to the feature fusion and standardization module.

[0011] Preferably, the step of extracting the grinding wheel passivation index, which characterizes the degree of grinding wheel passivation, from the acoustic emission signal in the preprocessed signal via wavelet packet decomposition specifically includes the following steps: Receive the acoustic emission signal from the preprocessed signal; The acoustic emission signal is decomposed into wavelet packet coefficients of each node by performing wavelet packet decomposition at a preset number of levels. Calculate the energy of each node in the last layer within a preset time window, and normalize it to obtain the energy distribution vector; The actual frequency band range of each node is determined based on the sampling rate and the number of wavelet packet decomposition layers. Based on the preset boundaries of the low-frequency and high-frequency regions, the normalized energy of the corresponding nodes is accumulated to obtain the total energy of the low-frequency region and the total energy of the high-frequency region. The ratio of total energy in the low-frequency region to total energy in the high-frequency region is calculated and used as the grinding wheel passivation index. This index is then output to the feature fusion and normalization module.

[0012] Preferably, the step of extracting a thermal effect feature vector reflecting grinding force fluctuations and thermal damage risk based on the grinding force signal, instantaneous speed curve, and bearing geometric parameters in the preprocessed signal specifically includes the following steps: It receives the normal force signal, instantaneous speed curve and angular domain signal from the grinding force signal, as well as the theoretical fault order ratio from the bearing fault feature extraction module; A continuous wavelet transform is performed on the normal force signal to obtain its time-frequency representation. Based on the instantaneous speed curve and the theoretical fault order ratio, the instantaneous fault frequency corresponding to the four types of faults—inner ring, outer ring, rolling element, and cage—is calculated in real time at the current moment. The instantaneous fault frequency is converted into the corresponding wavelet scale, and the wavelet coefficient amplitude at this scale is extracted from the time-frequency representation to obtain the amplitude curve that changes with time. Within a preset time window, the mean and standard deviation of the amplitude curve corresponding to each fault type are calculated, and then the coefficient of variation is obtained, and the peak factor is calculated. The coefficients of variation and peak factors of the four fault types are combined to form an 8-dimensional thermal effect feature vector, and this feature vector is output to the feature fusion and standardization module.

[0013] Preferably, the process of fusing the bearing fault feature vector, the grinding wheel passivation index, and the thermal effect feature vector into a total feature vector, and then performing standardization processing to generate a standardized feature vector, specifically includes the following steps: The bearing fault feature vector, grinding wheel passivation index and thermal effect feature vector are received and concatenated to form a 17-dimensional total feature vector. Retrieve the mean and standard deviation of each feature dimension of pre-stored historical normal operating condition data; Z-score standardization is performed on each dimension of the current total feature vector, which is obtained by subtracting the historical mean and dividing by the historical standard deviation. The standardized feature vector is then output to the diagnostic model module.

[0014] Preferably, the step of calculating the spindle wear probability and grinding wheel passivation probability using a multi-output logistic regression classification model based on the input standardized feature vector specifically includes the following steps: Internally, there is a pre-trained multi-output logistic regression classification model, which includes two independent logistic regression classifiers, corresponding to the spindle wear probability and the grinding wheel passivation probability, respectively. The system receives standardized feature vectors, substitutes them into the calculation formulas of two classifiers, calculates the spindle wear probability and grinding wheel passivation probability, and outputs the two probability values ​​to the diagnostic output and display module.

[0015] Preferably, the step of determining the diagnostic category based on the comparison results of the spindle wear probability and the grinding wheel passivation probability with a preset threshold, and generating a diagnostic report for display, specifically includes the following steps: Receive the spindle wear probability and the grinding wheel passivation probability, and compare them with a preset first threshold and a second threshold; The diagnostic category is determined based on the comparison results: if the spindle wear probability is ≥ the first threshold and the grinding wheel passivation probability is < the second threshold, the diagnosis is spindle bearing wear dominant; if the grinding wheel passivation probability is ≥ the second threshold and the spindle wear probability is < the first threshold, the diagnosis is grinding wheel passivation dominant; if both are ≥ their respective thresholds, the diagnosis is coupling abnormality; if both are < their respective thresholds, the diagnosis is normal. For the fault state, calculate the contribution value of each feature to the diagnostic result, and select the top few features with the largest absolute values ​​as key evidence. Generate a diagnostic report that includes diagnosis time, equipment information, diagnosis category, probability value, key evidence, and targeted maintenance recommendations; The diagnostic report is displayed in real time on the local display terminal and sent to the remote monitoring center via the network. At the same time, the diagnostic results, key evidence, feature vectors and corresponding original signal fragments are sent to the data storage and traceability module.

[0016] The present invention has the following beneficial effects: This invention acquires multi-source sensor signals during the machining process and preprocesses them to extract instantaneous rotational speed curves. Then, based on the preprocessed signals and instantaneous rotational speed curves, angular domain resampling is performed to convert the time-domain signals into angular domain signals. Based on the angular domain signals and preset bearing geometric parameters, bearing fault feature vectors are extracted. Simultaneously, grinding wheel state features are extracted based on acoustic emission signals in the preprocessed signals, and thermal effect feature vectors are extracted based on grinding force signals, instantaneous rotational speed curves, and bearing geometric parameters in the preprocessed signals. The extracted bearing fault feature vectors, grinding wheel state features, and thermal effect feature vectors are then fused and standardized to generate standardized feature vectors. These standardized feature vectors are then input into a pre-trained classification model to calculate the spindle wear probability and grinding wheel passivation probability. Finally, the diagnostic category is output based on the comparison results of the spindle wear probability and grinding wheel passivation probability with preset thresholds. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1This is a structural block diagram of the present invention.

[0019] 1. Multi-source sensor unit; 2. Signal conditioning and data acquisition module; 3. Data preprocessing module; 4. Angular domain resampling module; 5. Bearing fault feature extraction module; 6. Grinding wheel condition feature extraction module; 7. Thermal effect feature extraction module; 8. Feature fusion and standardization module; 9. Diagnostic model module; 10. Diagnostic output and display module; 11. Data storage and traceability module. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0021] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0022] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0023] In the description of this invention, it should be understood that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of this invention is in use, or the orientation or positional relationship commonly understood by those skilled in the art. They are only used to facilitate the description of this invention and to simplify the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0024] Furthermore, the terms "first," "second," and "third" are used only to distinguish descriptions and should not be interpreted as indicating or implying relative importance.

[0025] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0026] A real-time bearing wear condition monitoring system, such as Figure 1 As shown, the system includes a multi-source sensor unit 1, a signal conditioning and data acquisition module 2, a data preprocessing module 3, an angular domain resampling module 4, a bearing fault feature extraction module 5, a grinding wheel state feature extraction module 6, a thermal effect feature extraction module 7, a feature fusion and standardization module 8, a diagnostic model module 9, a diagnostic output and display module 10, and a data storage and traceability module 11. The output of the multi-source sensor unit 1 is connected to the input of the signal conditioning and data acquisition module 2. The output of the signal conditioning and data acquisition module 2 is connected to the input of the data preprocessing module 3. The first output of the data preprocessing module 3 is connected to the first input of the angular domain resampling module 4. The second output of the data preprocessing module 3 is connected to the second inputs of both the angular domain resampling module 4 and the thermal effect feature extraction module 7. The output of the angular domain resampling module 4... The output terminals are respectively connected to the input terminals of the bearing fault feature extraction module 5 and the third input terminal of the thermal effect feature extraction module 7. The output terminal of the bearing fault feature extraction module 5 is connected to the first input terminal of the feature fusion and standardization module 8. The input terminal of the grinding wheel state feature extraction module 6 is connected to the first output terminal of the data preprocessing module 3. The output terminal of the grinding wheel state feature extraction module 6 is connected to the second input terminal of the feature fusion and standardization module 8. The output terminal of the thermal effect feature extraction module 7 is connected to the third input terminal of the feature fusion and standardization module 8. The output terminal of the feature fusion and standardization module 8 is connected to the input terminal of the diagnostic model module 9. The output terminal of the diagnostic model module 9 is connected to the input terminal of the diagnostic output and display module 10. The control output terminal of the diagnostic output and display module 10 is connected to the input terminal of the data storage and traceability module 11. Multi-source sensor unit 1 is used to acquire multi-source sensor signals in real time during machine tool processing, specifically including the following steps: An accelerometer mounted near the spindle bearing housing picks up vibration signals; an acoustic emission sensor mounted near the workpiece or on the spindle picks up acoustic emission signals; a Hall current sensor or clamp meter picks up spindle motor current signals; a piezoelectric force sensor or an indirect force measurement unit based on spindle motor power picks up grinding force signals; a speed sensor or speed pulse signal obtained from a servo driver; this unit continuously outputs raw analog signals to the next module; Signal conditioning and data acquisition module 2 is used to perform signal conditioning and analog-to-digital conversion on the received multi-source sensor signals to obtain digitized multi-channel raw signals. Specifically, it includes the following steps: The received raw analog signals are conditioned, including amplification, filtering, and isolation, to meet the requirements of analog-to-digital conversion; the signals of each channel are synchronously converted to digital at a preset sampling rate to obtain digitized multi-channel raw signals; the digital signals are packaged and transmitted to the data preprocessing module 3 via a high-speed bus; Data preprocessing module 3 is used to resample, bandpass filter, and extract instantaneous speed curves from the multi-channel raw signals to obtain preprocessed signals and instantaneous speed curves. Specifically, it includes the following steps: The system receives multi-channel digital signals and resamples all signals to a uniform maximum sampling rate using anti-aliasing filtering and cubic spline interpolation to ensure time alignment. Bandpass filtering is applied according to signal type: vibration signals are filtered from 0.5Hz to 10kHz to preserve bearing fault characteristic frequencies; acoustic emission signals are filtered from 10kHz to remove low-frequency mechanical noise; current signals are filtered from 0.1Hz to 2kHz to preserve load fluctuation information; and grinding force signals are filtered from 0.1Hz to 5kHz. Instantaneous speed is extracted: if speed pulse signals exist, the reciprocal of the time interval between adjacent pulses is calculated and multiplied by the number of pulses per revolution to obtain the instantaneous speed, which is then obtained through linear interpolation to obtain a continuous instantaneous speed curve. If no speed pulses exist, a short-time Fourier transform is performed on the spindle motor current signal, and the frequency component with the largest amplitude in the time spectrum is extracted as the instantaneous rotational frequency, multiplied by 60 to obtain the instantaneous speed curve. The preprocessed signal and the instantaneous speed curve are output to subsequent modules. The angular domain resampling module 4 is used to convert the time-domain signal of the preprocessed signal into an angular domain signal based on the instantaneous rotational speed curve. Specifically, it includes the following steps: The system receives the preprocessed signal and the instantaneous rotation speed curve, calculates the cumulative rotation angle of the spindle based on the instantaneous rotation speed curve, sets the number of sampling points per revolution, calculates the angular domain sampling interval, and generates an angle sequence with equal angular intervals. Through cubic spline interpolation, the time domain amplitude of the preprocessed signal is mapped onto the angle sequence to obtain the angular domain signal, and the angular domain signal is output to the bearing fault feature extraction module 5 and the thermal effect feature extraction module 7. The bearing fault feature extraction module 5 is used to extract bearing fault feature vectors including specific energies of various fault types based on the angular domain signal and preset bearing geometric parameters. Specifically, it includes the following steps: The system receives vibration and current signals from the angular domain signal, along with preset bearing geometric parameters. Based on these parameters, it calculates the theoretical fault order ratios of the inner ring, outer ring, rolling elements, and cage. It performs bandpass filtering on the angular domain vibration and current signals within a preset high-frequency resonance band. It then performs a Hilbert transform on the filtered signals and takes the modulus to obtain the envelope signal. Finally, it performs a Fourier transform on the envelope signal to obtain the order ratio spectrum. Within a preset bandwidth near each theoretical fault order ratio, it extracts the peak amplitude of the order ratio spectrum. The system combines the peak amplitudes of the four fault types extracted from the vibration and current signals to form an 8-dimensional bearing fault feature vector, and outputs this feature vector to the feature fusion and standardization module 8. The grinding wheel state feature extraction module 6 is used to extract the grinding wheel passivation index, which characterizes the degree of grinding wheel passivation, based on the acoustic emission signal in the preprocessed signal through wavelet packet decomposition. Specifically, it includes the following steps: The system receives the acoustic emission signal from the preprocessed signal; performs wavelet packet decomposition on the acoustic emission signal at a preset number of layers to obtain the wavelet packet coefficients of each node; calculates the energy of each node in the last layer within a preset time window and normalizes it to obtain the energy distribution vector; determines the actual frequency band range of each node based on the sampling rate and the number of wavelet packet decomposition layers; accumulates the normalized energy of the corresponding nodes according to the preset low-frequency and high-frequency boundaries to obtain the total energy in the low-frequency region and the total energy in the high-frequency region; calculates the ratio of the total energy in the low-frequency region to the total energy in the high-frequency region as the grinding wheel passivation index, and outputs the index to the feature fusion and normalization module 8. The thermal effect feature extraction module 7 is used to extract thermal effect feature vectors reflecting grinding force fluctuations and thermal damage risks based on the grinding force signal, instantaneous speed curve, and bearing geometric parameters in the preprocessed signal. Specifically, it includes the following steps: The system receives the normal force signal, instantaneous rotational speed curve, and angular domain signal from the grinding force signal, as well as the theoretical fault order ratio from the bearing fault feature extraction module 5. It performs continuous wavelet transform on the normal force signal to obtain a time-frequency representation. Based on the instantaneous rotational speed curve and the theoretical fault order ratio, it calculates the instantaneous fault frequencies corresponding to four types of faults—inner ring, outer ring, rolling element, and cage—in real time. It converts the instantaneous fault frequencies into corresponding wavelet scales, extracts the wavelet coefficient amplitudes at that scale from the time-frequency representation, and obtains the amplitude curves that change over time. Within a preset time window, it calculates the mean and standard deviation of the amplitude curves corresponding to each fault type, thereby obtaining the coefficient of variation and calculating the peak factor. It combines the coefficients of variation and peak factors of the four fault types to form an 8-dimensional thermal effect feature vector and outputs this feature vector to the feature fusion and standardization module 8. The feature fusion and standardization module 8 is used to fuse the bearing fault feature vector, grinding wheel passivation index, and thermal effect feature vector into a total feature vector, and then perform standardization processing to generate a standardized feature vector. Specifically, it includes the following steps: Receive the bearing fault feature vector, grinding wheel passivation index, and thermal effect feature vector, and concatenate the three to form a 17-dimensional total feature vector; call the mean and standard deviation of each feature dimension of the pre-stored historical normal working condition data; perform Z-score standardization on each dimension of the current total feature vector, that is, subtract the historical mean and divide by the historical standard deviation to obtain the standardized feature vector, and output the standardized feature vector to the diagnostic model module 9; The diagnostic model module 9 is used to calculate the spindle wear probability and grinding wheel passivation probability based on the input standardized feature vector using a multi-output logistic regression classification model. Specifically, it includes the following steps: Internally, a pre-trained multi-output logistic regression classification model is deployed. This model includes two independent logistic regression classifiers, corresponding to the spindle wear probability and the grinding wheel passivation probability, respectively. It receives the standardized feature vector, substitutes it into the calculation formula of the two classifiers, calculates the spindle wear probability and the grinding wheel passivation probability, and outputs the two probability values ​​to the diagnostic output and display module 10. The diagnostic output and display module 10 is used to determine the diagnostic category based on the comparison results of the spindle wear probability and the grinding wheel passivation probability with preset thresholds, and to generate a diagnostic report for display. Specifically, it includes the following steps: The system receives the spindle wear probability and grinding wheel passivation probability and compares them with a preset first threshold and a second threshold. Based on the comparison results, a diagnostic category is determined: if the spindle wear probability ≥ the first threshold and the grinding wheel passivation probability < the second threshold, the diagnosis is spindle bearing wear dominant; if the grinding wheel passivation probability ≥ the second threshold and the spindle wear probability < the first threshold, the diagnosis is grinding wheel passivation dominant; if both are ≥ their respective thresholds, the diagnosis is coupling abnormality; if both are < their respective thresholds, the diagnosis is normal. For fault states, the system calculates the contribution of each feature to the diagnostic result and selects the top few features with the largest absolute values ​​as key evidence. A diagnostic report is generated, including diagnostic time, equipment information, diagnostic category, probability value, key evidence, and targeted maintenance recommendations. The diagnostic report is displayed in real-time on a local display terminal and sent to a remote monitoring center via the network. Simultaneously, the diagnostic results, key evidence, feature vectors, and corresponding original signal segments are sent to the data storage and traceability module 11. The data storage and traceability module 11 is used to bind and store the diagnostic results and their associated feature data with the currently processed workpiece ID, and provides a historical data traceability interface, specifically including the following steps: It receives and stores diagnostic results, key evidence, feature vectors, and corresponding raw signal fragments from the diagnostic output module; it binds the above data with the currently processed workpiece ID to establish a correlation between the processing batch and the diagnostic record; it provides a data query interface to support tracing historical diagnostic information by time, workpiece ID, fault type, and other conditions; and it supports exporting data for offline analysis or model optimization.

[0027] This real-time bearing wear condition monitoring system performs the following steps, including: Acquire multi-source sensor signals during the grinding process; The multi-source sensor signals are preprocessed to obtain preprocessed signals, and instantaneous speed curves are extracted based on these signals. Furthermore, the multi-source sensor signals include at least vibration signals, acoustic emission signals, spindle motor current signals, grinding force signals, and speed pulse signals or encoder signals. All sensor signals are then resampled to the same sampling frequency using anti-aliasing filtering and cubic spline interpolation. The maximum value of the original sampling rate for each sensor is taken as the same sampling frequency. Bandpass filtering is applied to the resampled signals according to their signal type. For vibration signals, a bandpass filter with a first preset passband frequency range is used to preserve the bearing fault characteristic frequencies and their harmonics. A high-pass filter is used to remove low-frequency mechanical noise for the acoustic emission signal; a band-pass filter with a second preset passband frequency range is used to retain load fluctuation information for the spindle motor current signal; and a band-pass filter with a third preset passband frequency range is used for the grinding force signal. If a speed pulse signal exists, the instantaneous speed is obtained by calculating the reciprocal of the time interval between adjacent pulses and multiplying it by the number of pulses per revolution. Then, a continuous instantaneous speed curve is obtained through linear interpolation. If there is no speed pulse signal, a short-time Fourier transform is performed on the spindle motor current signal, and the frequency component with the largest amplitude in its time spectrum is extracted as the instantaneous speed frequency. After converting it into instantaneous speed, a continuous instantaneous speed curve is obtained. Based on the preprocessed signal and instantaneous rotational speed curve, angular domain resampling is performed to convert the time-domain signal into an angular domain signal. Furthermore, based on the instantaneous rotational speed curve, the cumulative spindle rotation angle is calculated by dividing the instantaneous rotational speed by 60, multiplying by 2π, and then integrating over time. The number of sampling points per revolution is set, and the angular domain sampling interval is calculated by dividing 2π by the number of sampling points per revolution. Based on the cumulative spindle rotation angle and the angular domain sampling interval, an angle sequence with equal angular intervals is generated. Through cubic spline interpolation, the time-domain amplitude of the preprocessed signal is mapped onto the angle sequence with equal angular intervals to obtain the angular domain signal. Based on the angular domain signal and preset bearing geometric parameters, a bearing fault feature vector is extracted. Furthermore, based on the number of rolling elements, rolling element diameter, pitch circle diameter, and contact angle in the bearing geometric parameters, the fault order ratios of the inner ring, outer ring, rolling elements, and cage are calculated respectively. The vibration and current signals in the angular domain signal are bandpass filtered within a preset high-frequency resonant band. A Hilbert transform is performed on the filtered angular domain signal, and its magnitude is taken to obtain the envelope signal. A Fourier transform is performed on the envelope signal to obtain the order spectrum. The peak amplitude of the order spectrum is extracted within a preset bandwidth near each theoretical fault order. The peak amplitudes of the four fault types extracted from the vibration and current signals are combined to form the bearing fault feature vector. Based on the acoustic emission signal in the preprocessed signal, the grinding wheel state features are extracted. Furthermore, the acoustic emission signal is decomposed into wavelet packets at a preset number of layers to obtain the decomposition coefficients of each node in each layer. Within a preset time window, the wavelet packet energy of each node in the last layer is calculated and normalized to obtain the energy distribution vector. Based on the sampling rate and the number of wavelet packet decomposition layers, the actual frequency band range corresponding to each node in the last layer is determined. According to the preset low-frequency and high-frequency boundaries, the corresponding node energies in the energy distribution vector are accumulated to obtain the total energy in the low-frequency region and the total energy in the high-frequency region. The ratio of the total energy in the low-frequency region to the total energy in the high-frequency region is calculated as the grinding wheel passivation index, used to characterize the degree of grinding wheel passivation. Based on the grinding force signal, instantaneous rotational speed curve, and bearing geometric parameters in the preprocessed signal, a thermal effect feature vector is extracted. Furthermore, a continuous wavelet transform is performed on the normal force signal in the grinding force signal to obtain a time-frequency representation. The continuous wavelet transform uses the Morlet wavelet basis function. Based on the theoretical fault order ratio calculated from the instantaneous rotational speed curve and bearing geometric parameters, the instantaneous fault frequencies corresponding to the four faults (inner ring, outer ring, rolling element, and cage) at the current moment are calculated in real time. The instantaneous fault frequencies are converted into corresponding wavelet scales, and the wavelet coefficient amplitudes at this scale are extracted from the time-frequency representation to obtain amplitude curves that vary with time. Within a preset time window, the mean and standard deviation of the amplitude curves corresponding to each fault type are calculated to obtain the coefficient of variation reflecting the relative intensity of the fluctuation, and the peak factor reflecting the impact intensity is also calculated. The coefficient of variation and peak factor calculated for the four fault types are combined to form a thermal effect feature vector. The bearing fault feature vector, grinding wheel condition feature vector, and thermal effect feature vector are fused and standardized to generate a standardized feature vector. Furthermore, the bearing fault feature vector, grinding wheel condition feature vector, and thermal effect feature vector are concatenated to obtain a total feature vector. Based on historical normal operating condition data, the mean and standard deviation of each feature dimension are calculated. For each dimension in the total feature vector at the current moment, its corresponding historical mean is subtracted, and then divided by its corresponding historical standard deviation to obtain a standardized feature vector. The standardized feature vectors are input into a pre-trained classification model to calculate the spindle wear probability and the grinding wheel passivation probability. Furthermore, historical data samples are collected, each sample including a standardized feature vector and a corresponding label. The label includes two categories: whether it is the root cause of spindle bearing wear and whether it is the root cause of grinding wheel passivation. Two independent logistic regression classifiers are constructed to output the spindle wear probability and the grinding wheel passivation probability, respectively. The expression for each classifier is a composite function of an exponential function with the natural constant e as the base, where the exponential part is the weighted sum of the bias term and each feature. A cross-entropy loss function is used with an L2 regularization term, and the weight coefficients of the two classifiers are optimized using gradient descent. Based on the comparison results of the spindle wear probability and grinding wheel passivation probability with preset thresholds, a diagnostic category is output. Furthermore, a first threshold and a second threshold are preset. If the spindle wear probability is greater than or equal to the first preset threshold and the grinding wheel passivation probability is less than the second preset threshold, the diagnosis is spindle bearing wear dominant. If the grinding wheel passivation probability is greater than or equal to the second preset threshold and the spindle wear probability is less than the first preset threshold, the diagnosis is grinding wheel passivation dominant. If both the spindle wear probability and the grinding wheel passivation probability are greater than or equal to their respective preset thresholds, the diagnosis is coupling anomaly. If both the spindle wear probability and the grinding wheel passivation probability are less than their respective preset thresholds, the diagnosis is normal. For states diagnosed as faults, the contribution value of each feature in the classification model to the diagnostic result is calculated. The contribution value is the product of the weight coefficient corresponding to the feature and the standardized feature value. A preset number of features with the largest absolute contribution value are selected as key evidence output. A diagnostic report including fault type, probability value, key evidence, and maintenance suggestions is generated based on the diagnostic category. The diagnostic results of the current processing batch are bound and stored with the workpiece ID for quality traceability.

[0028] The multi-source sensor unit 1 deployed on the machine tool acquires vibration signals, acoustic emission signals, spindle motor current signals, grinding force signals, and speed pulse signals in real time during the machining process. These signals reflect the state of the spindle and the grinding zone from different physical dimensions. Among them, the vibration signal is sensitive to mechanical impact, the acoustic emission signal is sensitive to material micro-deformation and friction changes, the current signal can indirectly reflect load fluctuations, the grinding force signal is directly related to the cutting process, and the speed signal provides a benchmark for subsequent order analysis.

[0029] Subsequently, the signal conditioning and data acquisition module 2 amplifies the original analog signal, performs anti-aliasing filtering, and synchronously converts it into a digital signal to ensure that the multi-channel data is strictly aligned in time, laying the foundation for subsequent fusion analysis.

[0030] The data preprocessing module 3 first resamples all signals to a uniform maximum sampling rate to eliminate time misalignment caused by inconsistent sampling rates. Then, it performs targeted bandpass filtering based on the characteristics of different signals. For example, vibration signals retain the 0.5 Hz to 10 kHz frequency band to cover bearing fault characteristic frequencies and their harmonics, acoustic emission signals use a 10 kHz high-pass filter to remove low-frequency mechanical vibration interference, current signals retain the 0.1 Hz to 2 kHz frequency band to focus on load fluctuations, and grinding force signals retain the 0.1 Hz to 5 kHz frequency band to capture changes in cutting force. This filtering process effectively removes noise components unrelated to the fault and significantly improves the signal-to-noise ratio.

[0031] Meanwhile, the data preprocessing module 3 extracts the instantaneous speed curve based on the speed pulse or current signal. This curve reflects the small fluctuations in the spindle speed and provides an accurate reference for subsequent angular domain resampling, thereby overcoming the problem of fault characteristic frequency diffusion caused by speed fluctuations in traditional spectrum analysis.

[0032] The angular domain resampling module 4 calculates the cumulative spindle rotation angle using the instantaneous rotation speed curve and maps the time-domain signal to the angular domain with equal angular intervals through cubic spline interpolation, generating angular domain vibration signal, angular domain current signal, and angular domain grinding force signal. This conversion completely eliminates the influence of rotation speed fluctuation on the periodicity of the signal, so that the fault impact related to the rotation angle can be stably expressed as a peak energy of a specific order ratio in the angular domain. Even if the vibration signal caused by bearing wear is greatly attenuated after being transmitted through the mechanical structure, its periodic characteristics can still be effectively enhanced and identified in the angular domain order ratio spectrum.

[0033] The bearing fault feature extraction module 5 pre-calculates the theoretical fault order ratios of the inner ring, outer ring, rolling elements, and cage based on the bearing's geometric parameters. Then, it performs bandpass filtering on the angular domain vibration signal and current signal within a preset high-frequency resonance band. This resonance demodulation technique utilizes the principle that bearing fault impacts can excite high-frequency resonances in the structure, further amplifying the fault features. The filtered signal is then subjected to a Hilbert transform to obtain the envelope signal, and a Fourier transform is used to obtain the order ratio spectrum. Peak amplitudes are extracted near each theoretical fault order ratio and combined to form a bearing fault feature vector with eight dimensions. This vector directly quantifies the degree of wear of each component of the main shaft bearing. Even if the original vibration is weak, it can still be reliably captured after order ratio tracking and resonance demodulation.

[0034] The grinding wheel condition feature extraction module 6 focuses on acoustic emission signals, performing multi-layer wavelet packet decomposition to decompose the signals into different frequency bands. It calculates the normalized energy of each node in the last layer, determines the actual frequency band range corresponding to each node based on the sampling rate and the number of wavelet packet layers, and divides the frequency band into low-frequency and high-frequency regions based on grinding process experience. When the grinding wheel is sharp, high-frequency grinding energy dominates, while low-frequency friction energy increases significantly when the grinding wheel is dull or clogged. By calculating the ratio of the total energy in the low-frequency region to the total energy in the high-frequency region, the grinding wheel dulling index is obtained. This index objectively reflects the true state of the grinding wheel, avoiding the error of operators making subjective judgments based solely on grinding sound and vibration. Thus, in the secondary dulling phenomenon of the grinding wheel caused by faults, it can accurately identify that the grinding wheel dulling is not an independent problem but a consequence of spindle wear.

[0035] The thermal effect feature extraction module 7 performs continuous wavelet transform on the normal force in the grinding force signal, and obtains the time-frequency representation of the signal using the Morlet wavelet basis. Simultaneously, based on the instantaneous rotational speed curve and the theoretical bearing fault order ratio, it calculates the instantaneous fault frequencies corresponding to four types of faults: inner ring, outer ring, rolling elements, and cage. These frequencies are then converted to wavelet scales, and the wavelet coefficient amplitudes of the corresponding scales are extracted from the time-frequency representation to form an amplitude curve that varies with time. This curve reflects the synchronization between grinding force fluctuations and bearing fault frequencies. If spindle runout causes periodic changes in the cutting depth, the force fluctuation frequency will inevitably be highly correlated with the bearing fault order ratio. Subsequently, within the time window, the coefficient of variation and peak factor of each amplitude curve are calculated. The coefficient of variation characterizes the relative intensity of the fluctuation, and the peak factor reflects the severity of the impact. These are combined to obtain an eight-dimensional thermal effect feature vector. This vector directly captures the transient fluctuations in grinding force caused by spindle wear. These fluctuations are the direct cause of flash high temperatures and microstructural burns on the workpiece surface. Therefore, the thermal effect feature vector becomes crucial evidence connecting the root cause of the fault with the final burn consequence.

[0036] The feature fusion and standardization module 8 concatenates the bearing fault feature vector, grinding wheel passivation index, and thermal effect feature vector into a seventeen-dimensional total feature vector. Based on historical normal working condition data, it calculates the mean and standard deviation of each dimension and performs Z-score standardization on the current features to eliminate differences in the dimensions and magnitudes of different features. This allows the subsequent classification model to fairly utilize all feature information. The standardized feature vector integrates information from three levels: spindle mechanical state, grinding wheel state, and thermal effects of the grinding process, providing a complete data foundation for causal decoupling.

[0037] The diagnostic model module 9 internally deploys a pre-trained multi-output logistic regression classification model. This model includes two independent logistic regression classifiers, which output the probability of spindle wear and the probability of grinding wheel passivation, respectively. The model is trained using historical data samples, each of which includes a standardized feature vector and a corresponding label. The label clearly identifies whether the sample is dominated by spindle bearing wear, grinding wheel passivation, the two coupled, or a normal state. The weight coefficients are optimized through the cross-entropy loss function, enabling the model to learn the mapping relationship between different features and the root cause of the fault. When the real-time standardized feature vector is input, the model automatically calculates two probability values. This process realizes intelligent fusion and decision-making of multi-source information, and can distinguish whether the subsequent series of problems are caused by spindle bearing wear or the independent fault caused by the natural passivation of the grinding wheel.

[0038] The diagnostic output and display module 10 compares the spindle wear probability and the grinding wheel passivation probability based on a preset first threshold and a second threshold. If the spindle wear probability exceeds the first threshold while the grinding wheel passivation probability is lower than the second threshold, the diagnosis is spindle bearing wear as the dominant factor. If the grinding wheel passivation probability exceeds the second threshold while the spindle wear probability is lower than the first threshold, the diagnosis is grinding wheel passivation as the dominant factor. If both exceed the threshold, the diagnosis is coupling abnormality, and spindle problems are given priority. If both are lower than the threshold, the diagnosis is normal. For fault states, the module also calculates the contribution value of each feature to the diagnostic result, i.e., the product of the feature weight and the standardized feature value, and selects several features with the largest absolute contribution value as key evidence, such as a significant increase in the fault order energy of the inner ring of vibration and an increase in the coefficient of variation of the inner ring of thermal effect, thereby revealing the physical basis of the diagnostic conclusion.

[0039] Finally, a diagnostic report is generated, including diagnostic time, equipment information, diagnostic category, probability value, key evidence, and targeted maintenance recommendations. This report is presented to the operator in real time via a display terminal, and the diagnostic results are bound to the current workpiece ID and sent to the data storage and traceability module 11.

[0040] The data storage and traceability module 11 stores all data from each diagnosis, including original signal segments, feature vectors, probability values, and diagnostic conclusions, in association with the workpiece ID, forming a complete quality archive. When an anomaly is found in subsequent finished product inspections, it can quickly trace back to the specific processing time and machine tool component to determine whether there is a hidden burn in the batch of workpieces. At the same time, the stored historical data can also be used for continuous optimization and updating of the model, extracting reliable features of bearing wear from weak vibration signals, and combining thermal effect evidence to confirm its impact on workpiece quality. This avoids the ineffective operation of repeatedly dressing the grinding wheel and ensures that bearing workpieces produced continuously during this monitoring period will not be taken off the production line with hidden damage.

[0041] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A real-time bearing wear condition monitoring system, characterized in that, include Multi-source sensor unit (1) is used to collect multi-source sensor signals in real time during machine tool processing; The signal conditioning and data acquisition module (2) is used to perform signal conditioning and analog-to-digital conversion on the received multi-source sensor signals to obtain digitized multi-channel raw signals. The data preprocessing module (3) is used to resample, bandpass filter and extract instantaneous speed curves of the multi-channel raw signals to obtain the preprocessed signals and instantaneous speed curves. The angular domain resampling module (4) is used to convert the time domain signal of the preprocessed signal into the angular domain signal according to the instantaneous rotation speed curve; The bearing fault feature extraction module (5) is used to extract bearing fault feature vectors including the order energy of multiple fault types based on the angular domain signal and the preset bearing geometric parameters. The grinding wheel state feature extraction module (6) is used to extract the grinding wheel passivation index, which characterizes the degree of grinding wheel passivation, by wavelet packet decomposition based on the acoustic emission signal in the preprocessed signal. The thermal effect feature extraction module (7) is used to extract thermal effect feature vectors reflecting grinding force fluctuations and thermal damage risks based on the grinding force signal, instantaneous speed curve and bearing geometric parameters in the preprocessed signal. The feature fusion and standardization module (8) is used to fuse the bearing fault feature vector, grinding wheel passivation index and thermal effect feature vector into a total feature vector, and perform standardization processing to generate a standardized feature vector; The diagnostic model module (9) is used to calculate the spindle wear probability and grinding wheel passivation probability based on the input standardized feature vector through a multi-output logistic regression classification model. The diagnostic output and display module (10) is used to determine the diagnostic category based on the comparison results of the spindle wear probability and the grinding wheel passivation probability with the preset threshold, and to generate a diagnostic report for display. The data storage and traceability module (11) is used to bind and store the diagnostic results and their associated feature data with the workpiece ID currently being processed, and to provide a historical data traceability interface.

2. The bearing wear condition real-time monitoring system according to claim 1, characterized in that, The output of the multi-source sensor unit (1) is connected to the input of the signal conditioning and data acquisition module (2). The output of the signal conditioning and data acquisition module (2) is connected to the input of the data preprocessing module (3). The first output of the data preprocessing module (3) is connected to the first input of the angular domain resampling module (4). The second output of the data preprocessing module (3) is connected to the second inputs of the angular domain resampling module (4) and the thermal effect feature extraction module (7), respectively. The output of the angular domain resampling module (4) is connected to the input of the bearing fault feature extraction module (5) and the third input of the thermal effect feature extraction module (7), respectively. The output of the bearing fault feature extraction module (5) is connected to the feature fusion and labeling module (7). The first input terminal of the standardization module (8) is connected to the first output terminal of the data preprocessing module (3), the output terminal of the grinding wheel state feature extraction module (6) is connected to the second input terminal of the feature fusion and standardization module (8), the output terminal of the thermal effect feature extraction module (7) is connected to the third input terminal of the feature fusion and standardization module (8), the output terminal of the feature fusion and standardization module (8) is connected to the input terminal of the diagnostic model module (9), the output terminal of the diagnostic model module (9) is connected to the input terminal of the diagnostic output and display module (10), and the control output terminal of the diagnostic output and display module (10) is connected to the input terminal of the data storage and traceability module (11).

3. The bearing wear condition real-time monitoring system according to claim 1, characterized in that, The process for resampling, bandpass filtering, and extracting instantaneous speed curves from multi-channel raw signals to obtain preprocessed signals and instantaneous speed curves specifically includes the following steps: It receives multi-channel digital signals and resamples all signals to a uniform maximum sampling rate through anti-aliasing filtering and cubic spline interpolation to ensure time alignment. Bandpass filtering was performed according to the signal type. Specifically, vibration signals were filtered using a 0.5 Hz to 10 kHz bandpass filter to retain bearing fault characteristic frequencies; acoustic emission signals were filtered using a 10 kHz high-pass filter to remove low-frequency mechanical noise; current signals were filtered using a 0.1 Hz to 2 kHz bandpass filter to retain load fluctuation information; and grinding force signals were filtered using a 0.1 Hz to 5 kHz bandpass filter. Instantaneous speed extraction: If a speed pulse signal exists, calculate the reciprocal of the time interval between adjacent pulses and multiply it by the number of pulses per revolution to obtain the instantaneous speed. Then, obtain a continuous instantaneous speed curve through linear interpolation. If there is no speed pulse, perform a short-time Fourier transform on the spindle motor current signal, extract the frequency component with the largest amplitude in the time spectrum as the instantaneous rotation frequency, and multiply it by 60 to obtain the instantaneous speed curve. The preprocessed signal and instantaneous speed curve are output to subsequent modules.

4. The bearing wear condition real-time monitoring system according to claim 1, characterized in that, The process of converting the time-domain signal of the preprocessed signal into an angular-domain signal based on the instantaneous rotational speed curve specifically includes the following steps: Receive the preprocessed signal and instantaneous speed curve, and calculate the cumulative spindle rotation angle based on the instantaneous speed curve; Set the number of sampling points per revolution, calculate the angular domain sampling interval, and generate an angle sequence with equal angular intervals; By using cubic spline interpolation, the time-domain amplitude of the preprocessed signal is mapped onto the angle sequence to obtain the angular domain signal, and the angular domain signal is output to the bearing fault feature extraction module (5) and the thermal effect feature extraction module (7).

5. The bearing wear condition real-time monitoring system according to claim 1, characterized in that, The step of extracting bearing fault feature vectors including specific energies of various fault types based on angular domain signals and preset bearing geometric parameters specifically includes the following steps: It receives vibration and current signals from the angular domain signal, as well as preset bearing geometric parameters; The theoretical failure order ratios of the inner ring, outer ring, rolling elements, and cage are calculated based on geometric parameters. Bandpass filtering is performed on the angular domain vibration signal and current signal within the preset high-frequency resonance band; Perform a Hilbert transform on the filtered signal and take the modulus to obtain the envelope signal; Perform a Fourier transform on the envelope signal to obtain the order spectrum; Within a preset bandwidth range near each theoretical fault order, the peak amplitude of the order spectrum is extracted; The peak amplitudes of the four fault types extracted from the vibration signal and current signal are combined to form an 8-dimensional bearing fault feature vector, and the feature vector is output to the feature fusion and standardization module (8).

6. The bearing wear condition real-time monitoring system according to claim 1, characterized in that, The step of extracting the grinding wheel passivation index, which characterizes the degree of grinding wheel passivation, from the acoustic emission signal in the preprocessed signal via wavelet packet decomposition specifically includes the following steps: Receive the acoustic emission signal from the preprocessed signal; The acoustic emission signal is decomposed into wavelet packet coefficients of each node by performing wavelet packet decomposition at a preset number of levels. Calculate the energy of each node in the last layer within a preset time window, and normalize it to obtain the energy distribution vector; The actual frequency band range of each node is determined based on the sampling rate and the number of wavelet packet decomposition layers. Based on the preset boundaries of the low-frequency and high-frequency regions, the normalized energy of the corresponding nodes is accumulated to obtain the total energy of the low-frequency region and the total energy of the high-frequency region. Calculate the ratio of total energy in the low-frequency region to total energy in the high-frequency region as the grinding wheel passivation index, and output the index to the feature fusion and normalization module (8).

7. The real-time bearing wear condition monitoring system according to claim 1, characterized in that, The step of extracting a thermal effect feature vector reflecting grinding force fluctuations and thermal damage risk based on the grinding force signal, instantaneous speed curve, and bearing geometric parameters in the preprocessed signal specifically includes the following steps: Receive the normal force signal, instantaneous speed curve and angular domain signal from the grinding force signal, as well as the theoretical fault order ratio from the bearing fault feature extraction module (5); A continuous wavelet transform is performed on the normal force signal to obtain its time-frequency representation. Based on the instantaneous speed curve and the theoretical fault order ratio, the instantaneous fault frequency corresponding to the four types of faults—inner ring, outer ring, rolling element, and cage—is calculated in real time at the current moment. The instantaneous fault frequency is converted into the corresponding wavelet scale, and the wavelet coefficient amplitude at this scale is extracted from the time-frequency representation to obtain the amplitude curve that changes with time. Within a preset time window, the mean and standard deviation of the amplitude curve corresponding to each fault type are calculated, and then the coefficient of variation is obtained, and the peak factor is calculated. The coefficients of variation and peak factors of the four fault types are combined to form an 8-dimensional thermal effect feature vector, and the feature vector is output to the feature fusion and standardization module (8).

8. The bearing wear condition real-time monitoring system according to claim 1, characterized in that, The process of fusing the bearing fault feature vector, grinding wheel passivation index, and thermal effect feature vector into a total feature vector, and then standardizing it to generate a standardized feature vector, specifically includes the following steps: The bearing fault feature vector, grinding wheel passivation index and thermal effect feature vector are received and concatenated to form a 17-dimensional total feature vector. Retrieve the mean and standard deviation of each feature dimension of pre-stored historical normal operating condition data; Z-score standardization is performed on each dimension of the current total feature vector, that is, the historical mean is subtracted and then divided by the historical standard deviation to obtain the standardized feature vector, and the standardized feature vector is output to the diagnostic model module (9).

9. The bearing wear condition real-time monitoring system according to claim 1, characterized in that, The method for calculating the spindle wear probability and grinding wheel passivation probability based on the input standardized feature vector using a multi-output logistic regression classification model specifically includes the following steps: Internally, there is a pre-trained multi-output logistic regression classification model, which includes two independent logistic regression classifiers, corresponding to the spindle wear probability and the grinding wheel passivation probability, respectively. Receive the standardized feature vector, substitute it into the calculation formula of the two classifiers, calculate the spindle wear probability and grinding wheel passivation probability, and output the two probability values ​​to the diagnostic output and display module (10).

10. A real-time bearing wear condition monitoring system according to claim 1, characterized in that, The process of determining the diagnostic category based on the comparison results of the spindle wear probability and the grinding wheel passivation probability with preset thresholds, and generating a diagnostic report for display, specifically includes the following steps: Receive the spindle wear probability and the grinding wheel passivation probability, and compare them with a preset first threshold and a second threshold; The diagnostic category is determined based on the comparison results: if the spindle wear probability is ≥ the first threshold and the grinding wheel passivation probability is < the second threshold, the diagnosis is spindle bearing wear dominant; if the grinding wheel passivation probability is ≥ the second threshold and the spindle wear probability is < the first threshold, the diagnosis is grinding wheel passivation dominant; if both are ≥ their respective thresholds, the diagnosis is coupling abnormality; if both are < their respective thresholds, the diagnosis is normal. For the fault state, calculate the contribution value of each feature to the diagnostic result, and select the top few features with the largest absolute values ​​as key evidence. Generate a diagnostic report that includes diagnosis time, equipment information, diagnosis category, probability value, key evidence, and targeted maintenance recommendations; The diagnostic report is displayed in real time on the local display terminal and sent to the remote monitoring center via the network. At the same time, the diagnostic results, key evidence, feature vectors and corresponding original signal fragments are sent to the data storage and traceability module (11).

Images