Method, device, equipment and medium for evaluating wear state of planetary roller screw pair

Abstract

The application relates to a planetary roller screw pair wear state evaluation method, device, equipment and medium. The method comprises the following steps: collecting a vibration signal in a high-speed running process of a planetary roller screw, performing frequency domain conversion processing on the vibration signal, extracting a frequency spectrum characteristic parameter, and determining frequency spectrum characteristic data; judging the frequency spectrum characteristic data based on a preset threshold value, determining that the vibration is abnormal if the threshold value is exceeded, extracting an abnormal frequency band parameter, and generating a vibration abnormality characteristic vector; inputting the characteristic vector into a trained convolutional neural network, combining a U-Net network segmentation technology, determining a damage position of a thread engagement area, calculating a pixel density difference of a damage area, and obtaining a quantitative value of a wear degree; based on the quantitative value, combining a standard three-dimensional model, accurately evaluating a wear state, and generating a state data set containing wear information. The method can realize accurate evaluation of the wear state of the planetary roller screw, and adapt to the actual needs of high-end equipment for transmission component state monitoring.

Description

Technical Field

[0001] This invention belongs to the field of component condition monitoring technology, and in particular relates to methods, devices, equipment and media for assessing the wear condition of planetary roller screw pairs. Background Technology

[0002] Planetary roller screw pairs, as core components of high-precision, high-load-bearing, and long-life linear transmissions, are widely used in high-end equipment fields such as aerospace, precision equipment, and heavy machinery. The wear state of their threaded meshing pairs directly determines the overall machine's operational accuracy, reliability, and service life. Currently, wear detection for planetary roller screw pairs largely relies on manual periodic inspections, disassembly and measurement, or simple time-domain vibration amplitude determination. This results in low detection efficiency, strong non-real-time performance, and difficulty in early identification of micro-wear. While existing technologies employ vibration signals for fault analysis, they often use fixed threshold judgments and single characteristic parameter analysis, making them susceptible to interference from fluctuations in operating conditions such as speed, load, and ambient temperature. Furthermore, they lack sufficient accuracy in locating damage in the threaded meshing area and a complete technical chain from damage area quantification to three-dimensional visualization assessment and equipment reliability data support, making it difficult to meet the actual needs of high-end equipment for online, accurate, and intelligent wear state assessment of planetary roller screw pairs. Summary of the Invention

[0003] Therefore, it is necessary to provide a method, device, equipment, and medium for assessing the wear condition of planetary roller screw pairs that can quickly identify abnormal vibrations, accurately locate the damaged position of the thread engagement area, and quantify the degree of wear, thereby ensuring the stable operation of the planetary roller screw pair and the entire equipment and reducing maintenance costs.

[0004] Firstly, this application provides a method for assessing the wear condition of planetary roller screw pairs, including:

[0005] Real-time vibration signals during the high-speed operation of the planetary roller screw pair are collected, and the frequency domain transformation of the real-time vibration signals is performed using the fast Fourier transform method to determine the spectral characteristics of the processed vibration signals.

[0006] The amplitude of low-frequency components in the spectral feature data is judged based on a preset threshold. If the amplitude exceeds the preset threshold, it is determined that there is a vibration anomaly and the abnormal frequency band parameters are extracted to generate a vibration anomaly feature vector.

[0007] The vibration anomaly feature vector is input into a trained convolutional neural network model for classification and identification, generating a damage location distribution map of the thread engagement area.

[0008] Damaged regions are segmented using the U-Net network algorithm based on the damage location distribution map. The pixel density difference within the damaged regions is calculated to obtain the damage quantization value.

[0009] Damage quantification values ​​are mapped to a standard 3D model of the thread engagement area of ​​the planetary roller screw pair to generate a wear state dataset.

[0010] In one embodiment, a fast Fourier transform method is used to perform frequency domain transformation processing on the real-time vibration signal to determine the spectral characteristic data of the processed vibration signal, including:

[0011] The wavelet packet threshold denoising algorithm is used to dynamically adjust the threshold parameters according to the amplitude fluctuation characteristics of the real-time vibration signal, filter out environmental interference noise, sensor inherent error and equipment operation noise, and obtain the denoised vibration signal sequence.

[0012] The denoised vibration signal sequence is processed by using Fast Fourier Transform combined with Hanning window function to convert the vibration signal sequence in the time domain into a frequency domain signal.

[0013] Based on the inherent characteristic frequency range of the thread engagement of the planetary roller screw pair, the low-frequency and high-frequency components directly related to the thread engagement behavior are separated from the vibration signal sequence converted into a frequency domain signal by an adaptive frequency band division algorithm.

[0014] Key parameters of low-frequency and high-frequency components are extracted and integrated to obtain spectral feature data of the vibration signal sequence.

[0015] Key parameters include amplitude, peak frequency, phase difference, spectral bandwidth, kurtosis, peak factor, waveform factor, and power spectral density.

[0016] In one embodiment, the amplitude of low-frequency components in the spectral feature data is judged based on a preset threshold. If the amplitude exceeds the preset threshold, it is determined that there is a vibration anomaly, and the abnormal frequency band parameters are extracted to generate a vibration anomaly feature vector, including:

[0017] An adaptive threshold model is constructed based on the normal vibration spectrum data of a standard planetary roller screw pair under different speeds and load conditions.

[0018] The adaptive threshold model can dynamically adjust the threshold range according to the current real-time operating conditions of the lead screw.

[0019] The amplitude of low-frequency components is extracted from the spectral feature data, and the amplitude of low-frequency components is compared in real time with the threshold corresponding to the current operating condition output by the adaptive threshold model to determine whether it exceeds the preset threshold.

[0020] If the vibration exceeds the preset threshold, it is determined that there is an abnormality in the planetary roller screw pair. The frequency range of the abnormal frequency band is located simultaneously, and the multi-dimensional parameters of the abnormal frequency band are extracted.

[0021] Multidimensional parameters include amplitude, root mean square amplitude, kurtosis, peak factor, spectral distortion rate, frequency offset, and amplitude fluctuation coefficient.

[0022] The extracted multi-dimensional parameters are normalized to generate vibration anomaly feature vectors.

[0023] In one embodiment, the adaptive threshold model is constructed using the following formula:

[0024]

[0025] in, This indicates the adaptive preset threshold under the current operating conditions. The reference threshold for a standard planetary roller screw pair under rated speed and rated load conditions is calibrated using standard vibration spectrum data. This indicates the current real-time rotational speed of the planetary roller screw pair. This indicates the rated speed of a standard planetary roller screw pair. This indicates the current real-time load of the planetary roller screw pair. This indicates the rated load of a standard planetary roller screw pair. This represents the speed correction factor. This represents the load correction factor, calibrated using standard operating condition tests.

[0026] In one embodiment, the vibration anomaly feature vector is input into a trained convolutional neural network model for classification and identification, generating a damage location distribution map of the thread engagement area, including:

[0027] The deep features of abnormal vibration are extracted from the vibration anomaly feature vector using the wavelet packet decomposition algorithm, resulting in an optimized vibration anomaly feature vector.

[0028] The optimized vibration anomaly feature vector is input into a pre-defined convolutional neural network model for classification and identification, and the preliminary damage location information of the thread engagement area is output.

[0029] The convolutional neural network model incorporates a channel attention mechanism, and the model is trained based on sample data of different wear types of planetary roller screw pairs.

[0030] The initial damage location information is corrected for accuracy, and a damage location distribution map of the thread engagement area is generated by combining the geometric features of the thread engagement.

[0031] In one embodiment, the damaged region is segmented using the U-Net network algorithm based on the damaged location distribution map, and the pixel density difference within the damaged region is calculated to obtain the damaged quantization value, including:

[0032] The complete damage area was obtained by segmenting the damage location distribution map using the U-Net network algorithm.

[0033] The U-Net network algorithm employs an encoder-decoder network structure, which uses skip connections to fuse feature maps of different scales to segment the damaged region from the background region.

[0034] The pixel density difference and grayscale gradient change within the damaged area are calculated. Combined with the preset mapping relationship between pixel size and the actual physical size of the lead screw, a preliminary quantitative index of the internal wear degree is obtained.

[0035] The support vector machine algorithm is used to perform regression fitting on the preliminary quantification index, and a nonlinear mapping model between the preliminary quantification value and the actual wear degree of the lead screw is constructed.

[0036] The calculation formula for the nonlinear mapping model is as follows:

[0037]

[0038] in, This indicates the actual wear level of the leadscrew. A preliminary quantitative indicator of the degree of internal wear. Represents the radial basis kernel function. , Represents kernel function parameters. This indicates the preliminary quantitative indicators of the sample. Represents the regression coefficients of the SVM model. The bias terms are all trained and calibrated using planetary roller screw pair wear sample data.

[0039] Real-time speed and load data of the planetary roller screw pair were collected, and ambient temperature data were incorporated to construct a multi-factor deviation correction model.

[0040] The preliminary quantitative indicators are input into the nonlinear mapping model to obtain the regression fitting results. The regression fitting results are then dynamically corrected by combining the multi-factor deviation correction model, and the damage quantification value after compensation for working conditions and environment is output.

[0041] In one embodiment, damage quantification values ​​are mapped to a standard three-dimensional model of the thread engagement region of the planetary roller screw pair to generate a wear state dataset, including:

[0042] The wear level is determined based on the damage quantification value and the preset wear level classification standard for planetary roller screw pairs.

[0043] Based on the damage quantification value and wear level, combined with the preset wear evolution benchmark data, the wear development trend is obtained by fitting.

[0044] A visual assessment report is generated based on the wear level and wear development trend.

[0045] Establish the correspondence between damage quantification values ​​and thread engagement area coordinates, map the damage quantification values ​​to the standard three-dimensional model of the thread engagement area of ​​the planetary roller screw pair according to coordinates, and mark the wear quantification degree of each part.

[0046] By integrating visual assessment reports, wear quantification, and damage quantification values, a standardized wear status dataset is generated.

[0047] Secondly, this application also provides a device for assessing the wear condition of planetary roller screw pairs, the device comprising:

[0048] The signal spectrum extraction module is used to collect real-time vibration signals during the high-speed operation of the planetary roller screw pair. The fast Fourier transform method is used to perform frequency domain transformation processing on the real-time vibration signals to determine the spectral characteristic data of the processed vibration signals.

[0049] The vibration anomaly detection module is used to determine the amplitude of low-frequency components in the spectral feature data based on a preset threshold. If the amplitude exceeds the preset threshold, it is determined that there is a vibration anomaly and the abnormal frequency band parameters are extracted to generate a vibration anomaly feature vector.

[0050] The damage location identification module is used to input the vibration abnormality feature vector into the trained convolutional neural network model for classification and identification, and generate a damage location distribution map of the thread engagement area.

[0051] The damage region quantization module is used to segment the damage region based on the damage location distribution map using the U-Net network algorithm, calculate the pixel density difference within the damage region, and obtain the damage quantization value.

[0052] The wear condition assessment module is used to map the damage quantification value to a standard 3D model of the thread engagement area of ​​the planetary roller screw pair, generating a wear condition dataset.

[0053] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0054] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned method.

[0055] The aforementioned method, apparatus, computer equipment, and storage medium for assessing the wear condition of planetary roller screw pairs first acquire real-time vibration signals during the high-speed operation of the planetary roller screw pair. The acquired real-time vibration signals are then processed using a Fast Fourier Transform (FFT) method to convert the time-domain vibration signal into a frequency-domain signal, thereby determining the spectral characteristic data of the processed vibration signal. Subsequently, based on a preset threshold, the amplitude of the low-frequency components in the spectral characteristic data is judged in real time. If the amplitude of the low-frequency components exceeds the preset threshold, it is determined that the planetary roller screw pair has a vibration anomaly. Simultaneously, the range of the abnormal frequency band is located, relevant parameters of the abnormal frequency band are extracted, and a vibration anomaly feature vector is generated based on the extracted abnormal frequency band parameters. Next, the vibration anomaly feature vector is input into a pre-trained convolutional neural network model for classification and recognition. Through the model's analysis of the anomaly features, a damage location distribution map of the thread engagement area is generated. Then, based on this damage location distribution map, the U-Net network algorithm is used to segment the image, separating the complete damage area from the background area. The pixel density difference within the damage area is calculated, and combined with the correspondence between pixel size and the actual physical size of the lead screw, a damage quantification value representing the degree of wear is obtained. Finally, the damage quantification value is mapped to a standard three-dimensional model of the thread engagement area of ​​the planetary roller screw pair, marking the wear degree of each part, and integrating relevant data to generate a wear state dataset. This method solves the problems of poor real-time performance, insufficient damage location accuracy, and inaccurate wear quantification in traditional wear detection. Through systematic signal processing and model analysis, it can quickly identify vibration anomalies, accurately locate the damage location in the thread engagement area, and quantify the degree of wear. The generated wear state dataset can directly provide data support for equipment reliability assessment and maintenance decisions, ensuring the stable operation of the planetary roller screw pair and the entire equipment, and reducing maintenance costs. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 A flowchart of a method for assessing the wear condition of a planetary roller screw pair provided in an embodiment of the present invention;

[0058] Figure 2 This is a structural block diagram of the planetary roller screw pair wear condition assessment device provided in an embodiment of the present invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0060] In one embodiment, such as Figure 1 As shown, this application provides a method for evaluating the wear condition of planetary roller screw pairs, which may include the following steps:

[0061] Step S101: Collect the real-time vibration signal during the high-speed operation of the planetary roller screw pair, and use the fast Fourier transform method to perform frequency domain transformation processing on the real-time vibration signal to determine the spectral characteristic data of the processed vibration signal.

[0062] Specifically, the acquired signal is the time-domain vibration signal generated by the planetary roller screw pair during operation. The acquisition process must ensure the real-time performance and integrity of the signal to avoid signal loss or distortion affecting subsequent processing. After acquisition, the Fast Fourier Transform (FFT) method is used to perform frequency domain transformation processing on the acquired real-time vibration signal, converting the original time-domain vibration signal into a frequency-domain signal, eliminating the influence of irrelevant interference in the time-domain signal, and making the frequency characteristics of the signal clearer and more distinguishable. After the above frequency domain transformation processing, the spectral characteristic data of the processed vibration signal is determined. This spectral characteristic data mainly includes the frequency distribution of the signal, the amplitude corresponding to each frequency, and other core information.

[0063] Step S102: Based on a preset threshold, the amplitude of the low-frequency component in the spectral feature data is judged. If it exceeds the preset threshold, it is determined that there is a vibration anomaly and the abnormal frequency band parameters are extracted to generate a vibration anomaly feature vector.

[0064] The preset threshold is obtained based on the normal vibration spectrum data of the standard planetary roller screw pair under different working conditions, and is used to distinguish between normal vibration and abnormal vibration. The low-frequency component is directly related to the thread engagement behavior of the planetary roller screw pair, and its amplitude change can reflect the wear state of the thread engagement area. When the amplitude of the low-frequency component exceeds the preset threshold, the frequency range of the corresponding abnormal frequency band is simultaneously located, and multi-dimensional parameters of the abnormal frequency band are extracted. The multi-dimensional parameters include amplitude, root mean square of amplitude, kurtosis, peak factor, spectral distortion rate, frequency offset, and amplitude fluctuation coefficient. After normalizing the extracted multi-dimensional parameters, a vibration anomaly feature vector for subsequent identification is generated.

[0065] Step S103: Input the vibration anomaly feature vector into the trained convolutional neural network model for classification and recognition, and generate a damage location distribution map of the thread engagement area.

[0066] The convolutional neural network model is pre-trained based on sample data of different wear types and damage degrees of planetary roller screw pairs. The model incorporates a channel attention mechanism, which can enhance the ability to extract abnormal features and suppress redundant information interference. After the generated vibration abnormal feature vector is input into the model, the model identifies the approximate location of damage in the thread engagement area by classifying, matching and analyzing the feature vector. Combined with the geometric features of the thread engagement of the planetary roller screw pair, a damage location distribution map containing damage location information is generated, clarifying the preliminary distribution range of damage in the thread engagement area.

[0067] Step S104: Based on the damage location distribution map, the U-Net network algorithm is used to segment the damage area, calculate the pixel density difference within the damage area, and obtain the damage quantization value.

[0068] The U-Net network algorithm employs an encoder-decoder network structure. By fusing feature maps of different scales through skip connections, it can accurately segment the damaged area and background area in the damage location distribution map, eliminating background interference pixels and artifact pixels to obtain the complete damaged area. For the segmented damaged area, its pixel-level data is extracted, and the pixel density difference and gray-level gradient change within the area are calculated. Combined with the preset mapping relationship between pixel size and the actual physical size of the planetary roller screw pair, the pixel-level features are transformed into actual wear-related parameters, thereby obtaining a damage quantification value that can accurately characterize the wear degree of the thread engagement area.

[0069] Step S105: Map the damage quantification value to a standard three-dimensional model of the thread engagement area of ​​the planetary roller screw pair to generate a wear state dataset.

[0070] First, a correspondence is established between damage quantification values ​​and the coordinates of the thread engagement area, ensuring that each damage quantification value can accurately correspond to the specific location of the thread engagement area of ​​the planetary roller screw pair. Then, the damage quantification values ​​at each location are mapped to a pre-established standard 3D model of the thread engagement area of ​​the planetary roller screw pair, marking the degree of wear quantification at each location and achieving 3D visualization of the wear state. Finally, the damage quantification values, 3D model annotation information, abnormal feature parameters, and relevant operating condition data are integrated to generate a standardized wear state dataset. This dataset can directly provide data support for equipment reliability assessment and maintenance decisions, helping to solve the problems of poor real-time performance, insufficient damage location accuracy, and inaccurate wear quantification in traditional wear detection, ensuring the stable operation of the planetary roller screw pair and the entire equipment, and reducing maintenance costs.

[0071] The aforementioned method for assessing the wear state of planetary roller screw pairs involves collecting vibration signals during the high-speed operation of the planetary roller screw, performing frequency domain transformation on these signals, extracting spectral feature parameters, and determining the spectral feature data of the processed vibration signal. The spectral feature data is then judged based on a preset threshold; if the threshold is exceeded, it is determined to be an abnormal vibration, and abnormal frequency band parameters are extracted to generate an abnormal vibration feature vector. This feature vector is input into a trained convolutional neural network, and combined with U-Net network segmentation technology, the damage location in the thread engagement area is determined, the pixel density difference in the damaged area is calculated, and a quantified value of the wear degree is obtained. Based on this quantified value, combined with a standard 3D model, a precise assessment of the wear state is completed, generating a state dataset containing wear information. This method, through systematic signal processing, model analysis, and data integration, solves the problems of poor real-time performance, low positioning accuracy, and inaccurate quantification in traditional wear detection, achieving precise assessment of the wear state of planetary roller screws. It provides efficient and feasible technical support for the reliable operation of equipment and meets the actual needs of high-end equipment for monitoring the condition of transmission components.

[0072] In one embodiment, the fast Fourier transform method is used to perform frequency domain transformation processing on the real-time vibration signal to determine the spectral characteristic data of the processed vibration signal, which may include the following steps:

[0073] Step S201: The wavelet packet threshold denoising algorithm is used to dynamically adjust the threshold parameters according to the amplitude fluctuation characteristics of the real-time vibration signal, filter out environmental interference noise, sensor inherent error and equipment operation noise, and obtain the denoised vibration signal sequence.

[0074] Step S202: The denoised vibration signal sequence is processed by using Fast Fourier Transform combined with Hanning window function to convert the vibration signal sequence in the time domain into a frequency domain signal.

[0075] Furthermore, the denoised vibration signal sequence is a time-domain signal, which can only reflect the variation of vibration amplitude over time and cannot directly reflect the frequency characteristics related to the wear of the planetary roller screw pair. In the processing, the Hanning window function is first used to perform weighted preprocessing on the denoised vibration signal sequence. The weighting function is... (in The signal sampling point number. The total number of signal sampling points (representing the total number of sampling points) is used to suppress spectral leakage during the Fast Fourier Transform (FFT) process, preventing frequency overlap and amplitude distortion in the frequency domain signal. Subsequently, the signal sequence, after Hanning window weighting, undergoes a Fast Fourier Transform, converting the two-dimensional "time-amplitude" signal in the time domain into a two-dimensional "frequency-amplitude" signal in the frequency domain, outputting a frequency domain signal. This frequency domain signal clearly presents the frequency distribution characteristics of the vibration signal.

[0076] If the sampling frequency of the denoised vibration signal sequence is 10kHz and the number of sampling points is 1024, after weighting with a Hanning window, the time-domain signal is converted into a frequency-domain signal by a fast Fourier transform. The frequency range of the frequency-domain signal is 0-5kHz, and the vibration amplitude corresponding to different frequencies can be clearly identified.

[0077] Step S203: Based on the inherent characteristic frequency range of the thread engagement of the planetary roller screw pair, the low-frequency and high-frequency components directly related to the thread engagement behavior are separated from the vibration signal sequence converted into a frequency domain signal by an adaptive frequency band division algorithm.

[0078] The inherent characteristic frequency range of the planetary roller screw pair's thread engagement, determined through standard planetary roller screw pair testing, is 10Hz-500Hz. Vibration signals within this range are directly related to the thread engagement behavior. During processing, an adaptive frequency band division algorithm first scans the frequency distribution of the frequency domain signal, identifying all signal components falling within the aforementioned inherent characteristic frequency range. Then, based on the frequency distribution characteristics, it automatically divides the low-frequency and high-frequency components. The low-frequency component (10Hz-100Hz) corresponds to the fundamental vibration during thread engagement, while the high-frequency component (100Hz-500Hz) corresponds to the subtle vibrations during thread engagement. Both directly reflect the thread engagement state and wear condition. This algorithm can dynamically adjust the division threshold according to the actual frequency distribution of the frequency domain signal, avoiding signal omissions or irrelevant signal mixing caused by fixed frequency band division.

[0079] If the amplitude of a frequency domain signal is concentrated in the range of 0.5-2.0V in the 10Hz-100Hz range and in the range of 0.1-0.8V in the 100Hz-500Hz range, the adaptive frequency band division algorithm will automatically identify the frequency distribution pattern and accurately separate the signal components of these two frequency bands as low-frequency and high-frequency components, respectively.

[0080] Step S204: Extract key parameters of low-frequency and high-frequency components and integrate them to obtain spectral feature data of the vibration signal sequence.

[0081] Preferably, the key parameters include amplitude, peak frequency, phase difference, spectral bandwidth, kurtosis, peak factor, waveform factor, and power spectral density.

[0082] Specifically, after acquiring real-time vibration signals during the high-speed operation of the planetary roller screw pair, a wavelet packet threshold denoising algorithm is used to preprocess the real-time vibration signals. The algorithm dynamically adjusts the threshold parameters according to the amplitude fluctuation characteristics of the real-time vibration signals. By matching the change law of the signal amplitude in real time, it accurately filters out environmental interference noise, inherent errors of the sensor itself, and irrelevant noise generated during equipment operation. While filtering out interference, it completely retains the effective vibration signal components related to the thread engagement and wear of the planetary roller screw pair, finally obtaining a denoised and clean vibration signal sequence. The denoised vibration signal sequence is then subjected to frequency domain transformation processing using a combination of Fast Fourier Transform and Hanning window function. The Hanning window function is used to suppress spectral leakage during the frequency domain transformation process, avoiding frequency domain signal distortion. This processing transforms the vibration signal sequence in the time domain into... Converting the signal to the frequency domain clearly reveals the frequency distribution characteristics of the vibration signal. Based on the inherent characteristic frequency range of the planetary roller screw pair's thread engagement, an adaptive frequency band division algorithm is used to process the converted frequency domain signal. This algorithm can automatically match the inherent characteristic frequency range of the thread engagement according to the frequency distribution characteristics of the frequency domain signal, accurately separating the low-frequency and high-frequency components directly related to the thread engagement behavior. The low-frequency components mainly correspond to the basic operating state of the thread engagement, while the high-frequency components correspond to the subtle vibration changes during the thread engagement process, both of which are closely related to the wear state of the screw. Subsequently, the key parameters of the separated low-frequency and high-frequency components are extracted, and all key parameters are classified and integrated to form spectral feature data of the vibration signal sequence that can comprehensively reflect the characteristics of the vibration signal and is directly related to the thread engagement state of the planetary roller screw pair.

[0083] This embodiment effectively solves the problems of fixed thresholds and incomplete interference filtering in traditional denoising methods by combining wavelet packet thresholding with dynamic threshold adjustment, ensuring the purity and effectiveness of the vibration signal. The application of fast Fourier transform combined with Hanning window function improves the accuracy of frequency domain transformation and avoids the impact of spectral leakage on signal feature extraction. The adaptive frequency band division algorithm can accurately separate signal components related to thread engagement, eliminate irrelevant frequency band interference, and focus on wear-related signal features. The extraction and integration of multi-dimensional key parameters comprehensively capture the frequency, amplitude, and waveform characteristics of the vibration signal, making up for the deficiency of single parameters reflecting incomplete information and effectively improving the accuracy and reliability of wear assessment.

[0084] In one embodiment, the amplitude of low-frequency components in the spectral feature data is judged based on a preset threshold. If the amplitude exceeds the preset threshold, it is determined that there is a vibration anomaly, and the abnormal frequency band parameters are extracted to generate a vibration anomaly feature vector. This may include the following steps:

[0085] Step S301: Based on the normal vibration spectrum data of the standard planetary roller screw pair under different speed and load conditions, construct an adaptive threshold model.

[0086] Preferably, the adaptive threshold model can dynamically adjust the threshold range according to the current real-time operating conditions of the lead screw.

[0087] Step S302: Extract the amplitude of low-frequency components from the spectral feature data, and compare the amplitude of low-frequency components with the threshold corresponding to the current operating condition output by the adaptive threshold model in real time to determine whether it exceeds the preset threshold.

[0088] Step S303: If the vibration exceeds the preset threshold, it is determined that there is an abnormal vibration in the planetary roller screw pair. The frequency range of the abnormal frequency band is located simultaneously, and the multi-dimensional parameters of the abnormal frequency band are extracted.

[0089] Furthermore, the multi-dimensional parameters include amplitude, root mean square amplitude, kurtosis, peak factor, spectral distortion rate, frequency offset, and amplitude fluctuation coefficient.

[0090] Step S304: Normalize the extracted multi-dimensional parameters to generate a vibration anomaly feature vector.

[0091] First, based on the normal vibration spectrum data of a standard planetary roller screw pair under different speeds and load conditions, an adaptive threshold model is constructed. This model can dynamically adjust the threshold range according to the real-time operating speed and load conditions of the screw, ensuring that the threshold accurately matches the actual operating state and providing a reliable standard for subsequent anomaly judgment. Then, the denoised vibration signal is processed using a Fast Fourier Transform, and a Hanning window is introduced to suppress spectral leakage, converting the time-domain vibration signal into a frequency-domain signal and clearly presenting the frequency distribution characteristics of the signal. Based on this, an adaptive frequency band division algorithm is used to filter the frequency-domain signal based on the inherent characteristic frequency range of the planetary roller screw pair's thread engagement, separating the low-frequency and high-frequency components directly related to the thread engagement behavior. The low-frequency component corresponds to the basic vibration of the thread engagement, while the high-frequency component corresponds to the subtle vibration changes during the thread engagement process. Together, they constitute the core signal related to screw wear. Next, multi-dimensional parameters of these two components are extracted, including amplitude, peak frequency, phase difference, spectral bandwidth, kurtosis, and peak factor. Environmental temperature, real-time speed, and load parameters are also introduced to integrate and form complete spectral feature data.

[0092] This embodiment avoids the drawbacks of fixed thresholds failing to adapt to different operating conditions by constructing an adaptive threshold model based on standard operating conditions, ensuring the accuracy of anomaly detection and reducing false positives and false negatives. The use of Fast Fourier Transform combined with Hanning window processing effectively avoids signal distortion caused by spectral leakage, ensuring the integrity of the frequency domain signal. Through adaptive frequency band division, it accurately focuses on signal components related to thread engagement, eliminating irrelevant interference and ensuring that the extracted feature parameters are directly related to the wear state of the leadscrew. The extraction and integration of multi-dimensional parameters avoids the problem of incomplete information reflected by a single parameter, significantly improving the accuracy and reliability of planetary roller leadscrew pair wear state assessment, providing scientific and effective data for equipment maintenance decisions, and ensuring stable equipment operation.

[0093] In one embodiment, the adaptive threshold model can be constructed using the following formula:

[0094]

[0095] in, This indicates the adaptive preset threshold under the current operating conditions. The reference threshold for a standard planetary roller screw pair under rated speed and rated load conditions is calibrated using standard vibration spectrum data. This indicates the current real-time rotational speed of the planetary roller screw pair. This indicates the rated speed of a standard planetary roller screw pair. This indicates the current real-time load of the planetary roller screw pair. This indicates the rated load of a standard planetary roller screw pair. This represents the speed correction factor. This represents the load correction factor, calibrated using standard operating condition tests.

[0096] This embodiment effectively solves the problem that fixed thresholds cannot adapt to different working conditions. By dynamically adjusting the threshold, it ensures the accuracy of vibration anomaly judgment and avoids misjudgment caused by changes in working conditions. At the same time, the model parameters are all calibrated through standard tests, ensuring the reliability of threshold judgment. This provides accurate judgment basis for subsequent damage location and wear quantification, thereby improving the accuracy of overall wear assessment, reducing maintenance costs, and adapting to the stringent requirements of high-end equipment for condition monitoring.

[0097] In one embodiment, inputting the vibration anomaly feature vector into a trained convolutional neural network model for classification and identification to generate a damage location distribution map of the thread engagement area may include the following steps:

[0098] Step S401: Extract the deep features of abnormal vibration from the vibration anomaly feature vector using the wavelet packet decomposition algorithm to obtain the optimized vibration anomaly feature vector.

[0099] Step S402: Input the optimized vibration anomaly feature vector into the preset convolutional neural network model for classification and recognition, and output the preliminary damage location information of the thread engagement area.

[0100] Preferably, the convolutional neural network model incorporates a channel attention mechanism, and the model is trained based on sample data of different wear types of planetary roller screw pairs.

[0101] The pre-defined convolutional neural network model incorporates a channel attention mechanism. This mechanism assigns weights to the input vibration anomaly feature vector, strengthening the weights of features related to screw wear, suppressing interference from irrelevant features, and improving the accuracy of model classification and recognition. Simultaneously, this convolutional neural network model is pre-trained based on sample data of different wear types and degrees of wear on planetary roller screw pairs. During training, model parameters are iteratively optimized to enable the model to accurately match abnormal features corresponding to different wear states, demonstrating mature classification and recognition capabilities. In processing, the model extracts, analyzes, and matches features layer by layer from the optimized vibration anomaly feature vector. Combining the correspondence between wear features and damage locations learned during training, it classifies and identifies the damage locations corresponding to vibration anomalies, clearly defining the approximate range of the thread engagement area where the damage is located. Finally, it outputs preliminary damage location information for the thread engagement area, including the approximate coordinate range of the damage and a preliminary judgment of the wear type corresponding to the damage.

[0102] Step S403: The accuracy of the preliminary damage location information is corrected, and a damage location distribution map of the thread engagement area is generated by combining the geometric features of the thread engagement.

[0103] Specifically, the generated vibration anomaly feature vector is used as input, and a wavelet packet decomposition algorithm is employed for deep feature extraction. Wavelet packet decomposition breaks down the vibration anomaly feature vector into different frequency bands, filtering out effective deep features related to the wear of the planetary roller screw pair, and eliminating redundant feature information to obtain an optimized vibration anomaly feature vector, thus improving the targeting and effectiveness of the feature vector. The optimized vibration anomaly feature vector is then input into a pre-defined convolutional neural network model for classification and recognition. This convolutional neural network model incorporates a channel attention mechanism, which enhances the extraction capability of wear-related features and suppresses interference from irrelevant features. Furthermore, this model is pre-defined based on planetary roller screws. The sample data of different wear types and wear degrees of the ball screw pair have been trained and have mature classification and recognition capabilities. The optimized vibration anomaly feature vector is analyzed, matched and classified by the convolutional neural network model to output the preliminary damage location information of the thread engagement area. The output preliminary damage location information is corrected for accuracy. Combined with the geometric features of the thread engagement of the planetary roller ball screw pair (including thread pitch, tooth profile, engagement contact area size, etc.), the deviation of the preliminary damage location is corrected, the correspondence between the damage location and the geometric structure of the thread engagement area is clarified, and finally a damage location distribution map of the thread engagement area that can accurately reflect the damage location distribution is generated.

[0104] The application of wavelet packet decomposition algorithm in this embodiment can effectively extract deep features of vibration anomalies, eliminate redundant information, solve the problem of insufficient targeting of the original feature vector, and improve the effectiveness of the feature vector; the convolutional neural network model embedding channel attention mechanism, combined with training of sample data of different wear types, greatly improves the accuracy and efficiency of damage location identification and avoids interference from irrelevant features on the identification results; the accuracy correction based on the geometric features of thread engagement corrects the deviation of the initial damage location, improves the accuracy of damage localization, and ensures that the damage location distribution map can truly and accurately reflect the damage situation of the thread engagement area; the whole process provides accurate and reliable basic support for subsequent damage area segmentation and wear quantification, effectively improves the overall accuracy and reliability of planetary roller screw pair wear condition assessment, and provides scientific data basis for equipment maintenance decisions.

[0105] In one embodiment, the damaged region is segmented using the U-Net network algorithm based on the damaged location distribution map, and the pixel density difference within the damaged region is calculated to obtain the damaged quantization value. This may include the following steps:

[0106] Step S501: The U-Net network algorithm is used to segment the damage location distribution map to obtain the complete damage area.

[0107] Preferably, the U-Net network algorithm adopts an encoder-decoder network structure, which fuses feature maps of different scales through skip connections to segment the damaged region and the background region.

[0108] Furthermore, the damage location distribution map of the thread engagement area has initially marked the approximate location of the damage, but it suffers from problems such as blurred damage area boundaries, incomplete separation from the background area, and the potential inclusion of irrelevant background pixels, making it unsuitable for direct use in subsequent wear quantification analysis. The U-Net network algorithm is employed to segment the damage location distribution map. This algorithm uses an encoder-decoder bidirectional network structure. The encoding part gradually extracts feature information from the damage location distribution map through convolution and pooling operations, achieving dimensionality reduction of image features and capturing damage features at different scales. The decoding part gradually restores the image size through deconvolution operations, and simultaneously fuses feature maps of different scales from the encoding part with feature maps of corresponding scales from the decoding part through skip connections, compensating for lost detail features during decoding and strengthening the boundary information of the damage area.

[0109] Step S502: Calculate the pixel density difference and grayscale gradient change in the damaged area, and combine the preset mapping relationship between pixel size and the actual physical size of the lead screw to obtain a preliminary quantitative index of the internal wear degree.

[0110] Step S503: The support vector machine algorithm is used to perform regression fitting on the preliminary quantification index to construct a nonlinear mapping model between the preliminary quantification value and the actual wear degree of the lead screw.

[0111] The calculation formula for the nonlinear mapping model is as follows:

[0112]

[0113] in, This indicates the actual wear level of the leadscrew. A preliminary quantitative indicator of the degree of internal wear. Represents the radial basis kernel function. , Represents kernel function parameters. This indicates the preliminary quantitative indicators of the sample. Represents the regression coefficients of the SVM model. The bias terms are all trained and calibrated using planetary roller screw pair wear sample data.

[0114] Step S504: Collect real-time speed and load data of the planetary roller screw pair, introduce ambient temperature data, and construct a multi-factor deviation correction model.

[0115] Step S505: Input the preliminary quantitative indicators into the nonlinear mapping model to obtain the regression fitting results, combine the multi-factor deviation correction model to dynamically correct the regression fitting results, and output the damage quantification value after compensation by working conditions and environment.

[0116] Specifically, the U-Net network algorithm is used to process the damage location distribution map. This algorithm, based on an encoder-decoder network structure, fuses feature maps of different scales through skip connections to accurately segment the complete damage area from the background area, effectively eliminating irrelevant background interference and clarifying the specific range of the damage area. Based on this, pixel-level analysis is performed on the segmented damage area, calculating pixel density differences and gray-level gradient changes within the area. Simultaneously, combined with a pre-defined mapping relationship between pixel size and the actual physical size of the leadscrew, the image-level features are transformed into quantifiable indicators, obtaining a preliminary quantitative value of the damage degree. Subsequently, a support vector machine algorithm is used to regress and fit this preliminary quantitative indicator, constructing a nonlinear mapping model between the preliminary quantitative value and the actual wear degree of the leadscrew. To further improve the quantification accuracy, real-time speed and load data of the planetary roller leadscrew pair are collected, and ambient temperature data is introduced to construct a multi-factor deviation correction model. After inputting the preliminary quantitative indicator into the nonlinear mapping model to obtain the regression fitting result, the deviation correction model is used for dynamic adjustment, finally outputting an accurate damage quantification value after compensation for operating conditions and the environment.

[0117] This embodiment achieves precise segmentation of the damaged area through the U-Net network algorithm, effectively avoiding background interference and ensuring the accuracy of damaged area location. By mapping pixel features to actual physical dimensions, image features are transformed into quantifiable wear indicators, solving the problem of "qualitative but not quantitative" wear assessment in traditional wear assessment. The application of the support vector machine algorithm for regression fitting establishes a precise correlation between the preliminary quantitative indicators and the actual wear degree, improving the scientific nature of the quantitative results. The introduction of a multi-factor deviation correction model compensates for the influence of operating conditions and environmental factors on the quantitative results, reducing errors. A complete link is formed from damaged area segmentation, preliminary quantification, regression fitting to deviation correction, ensuring that the output damage quantification value is accurate and reliable. This provides objective and quantifiable data support for subsequent equipment maintenance and fault early warning, improving the professionalism and practicality of wear assessment and meeting the needs of high-end equipment for accurate assessment of lead screw wear status.

[0118] In one embodiment, mapping the damage quantification values ​​to a standard three-dimensional model of the thread engagement region of the planetary roller screw pair to generate a wear state dataset may include the following steps:

[0119] Step S601: Based on the damage quantification value and the preset planetary roller screw pair wear level classification standard, determine the corresponding wear level.

[0120] Step S602: Based on the damage quantification value and wear level, and combined with the preset wear evolution benchmark data, the wear development trend is fitted to obtain the wear development trend.

[0121] Step S603: Generate a visual assessment report based on the wear level and wear development trend.

[0122] Step S604: Establish the correspondence between the damage quantification value and the coordinates of the thread engagement area, map the damage quantification value to the standard three-dimensional model of the thread engagement area of ​​the planetary roller screw pair according to the coordinates, and mark the wear quantification degree of each part.

[0123] Preferably, a pre-defined standard 3D model of the thread engagement area of ​​the planetary roller screw pair is invoked. This model is pre-constructed based on the actual dimensions of the screw and the thread engagement structure parameters, and perfectly matches the geometry and coordinate position of the actual screw thread engagement area to ensure accurate mapping. During processing, firstly, a one-to-one correspondence is established between the damage quantification value and the corresponding thread engagement area coordinates, clarifying the specific coordinate position of the thread engagement area corresponding to each damage quantification value. The coordinate position must accurately correspond to the specific tooth surface, tooth root, and other key parts of the thread engagement. Subsequently, according to the established correspondence, the damage quantification value corresponding to each coordinate is accurately mapped to the corresponding coordinate position of the standard 3D model according to the pre-defined coordinate mapping rules. Finally, the degree of wear quantification at each coordinate position is marked on the 3D model. The marking method adopts a form that can intuitively distinguish the quantification differences, ensuring that the staff can clearly identify the degree of wear in different parts of the thread engagement area. The output of this step is a standard 3D model of the thread engagement area of ​​the planetary roller screw pair, labeled with the quantified wear degree of each part. Its function is to transform the abstract damage quantification value into an intuitive 3D visualization, clarify the spatial distribution of the wear quantification value in the thread engagement area, and provide intuitive and accurate spatial location and quantification degree correlation data for the subsequent generation of visualization assessment reports and the fusion of standardized wear state datasets. At the same time, it facilitates staff to quickly locate severely worn parts.

[0124] Step S605: Integrate the visual assessment report, wear quantification degree, and damage quantification value to generate a standardized wear status dataset.

[0125] Furthermore, using the obtained damage quantification value after compensation for working conditions and environment as the core input, combined with the preset planetary roller screw pair wear level classification standard, the wear level corresponding to the planetary roller screw pair is determined by comparing the damage quantification value with the quantification threshold corresponding to each level in the standard, thus clarifying the severity of the current screw wear. On this basis, using the damage quantification value and the determined wear level as the basic data, combined with the preset wear evolution benchmark data (this benchmark data is based on the screw wear evolution law under different wear levels and different working conditions, and is calibrated through a large number of standard sample tests), a fitting algorithm is used to fit multiple sets of data to obtain the screw wear development trend, clarifying the change law of wear degree over time and subsequent evolution. The process involves several steps: First, based on the determined wear level and the fitted wear development trend, a visual assessment report is generated. This report clearly presents the current wear level, wear development rate, and potential future wear state. Second, a one-to-one correspondence is established between the damage quantification value and the coordinates of the thread engagement area. The damage quantification value corresponding to each coordinate is mapped to a standard 3D model of the thread engagement area of ​​the planetary roller screw pair according to a preset coordinate mapping rule. This accurately marks the wear quantification degree of each part of the thread engagement area, achieving an intuitive correspondence between wear location and quantification value. Finally, the generated visual assessment report, the wear quantification degree marked on the 3D model, and the original damage quantification value are integrated and summarized according to preset data specifications to generate a standardized wear state dataset.

[0126] This embodiment determines the wear level by comparing the damage quantification value with a preset standard, ensuring the objectivity and accuracy of the wear level determination and avoiding subjective judgment bias. By combining the wear evolution benchmark data with the development trend, the wear trend of the lead screw can be predicted in advance, providing a data basis for preventive maintenance. The visualized evaluation report and 3D model annotation realize the intuitive presentation of the wear state, making it easy for staff to quickly grasp the wear distribution and severity of the lead screw. The standardized wear state dataset integrates multi-dimensional evaluation data and unifies the data format, solving the problem of scattered and non-standard evaluation data at different stages. It provides comprehensive, standardized and reliable data support for subsequent equipment condition monitoring, wear model iterative optimization and maintenance strategy formulation, effectively improving the scientificity and efficiency of planetary roller lead screw pair operation and maintenance, reducing the risk of equipment failure caused by wear and extending the service life of the equipment.

[0127] In one embodiment, such as Figure 2 As shown, this application also provides a device for assessing the wear condition of planetary roller screw pairs, the device may include:

[0128] The signal spectrum extraction module 701 is used to collect real-time vibration signals during the high-speed operation of the planetary roller screw pair. The fast Fourier transform method is used to perform frequency domain transformation processing on the real-time vibration signals to determine the spectral characteristic data of the processed vibration signals.

[0129] The vibration anomaly judgment module 702 is used to judge the amplitude of low-frequency components in the spectral feature data based on a preset threshold. If the amplitude exceeds the preset threshold, it is determined that there is a vibration anomaly and the abnormal frequency band parameters are extracted to generate a vibration anomaly feature vector.

[0130] The damage location identification module 703 is used to input the vibration abnormality feature vector into the trained convolutional neural network model for classification and identification, and generate a damage location distribution map of the thread engagement area.

[0131] The damage region quantization module 704 is used to segment the damage region based on the damage location distribution map using the U-Net network algorithm, calculate the pixel density difference within the damage region, and obtain the damage quantization value.

[0132] Wear condition assessment module 705 is used to map the damage quantification value to a standard three-dimensional model of the thread engagement area of ​​the planetary roller screw pair to generate a wear condition dataset.

[0133] The aforementioned planetary roller screw pair wear condition assessment device comprises a signal spectrum extraction module responsible for collecting real-time vibration signals during the high-speed operation of the planetary roller screw pair. After acquisition, the module uses a fast Fourier transform (FFT) method to perform frequency domain conversion processing on the real-time vibration signals, transforming the time-domain vibration signals into frequency-domain signals, thereby determining the spectral characteristic data of the processed vibration signals and providing basic data support for subsequent anomaly judgment and damage identification. The vibration anomaly judgment module takes the spectral characteristic data output by the signal spectrum extraction module as input and judges the amplitude of low-frequency components in the spectral characteristic data one by one based on a preset threshold. If the amplitude of the low-frequency components exceeds the preset threshold, it determines that there is a vibration anomaly in the planetary roller screw pair. At the same time, it extracts relevant parameters of the abnormal frequency band, integrates them to generate a vibration anomaly feature vector, and provides data basis for damage location identification. The damage location identification module receives the vibration anomaly feature vector output by the vibration anomaly judgment module and inputs it into a pre-trained convolutional neural network model for classification and identification. Through the model's matching and analysis of abnormal features, it generates a damage location distribution map of the thread engagement area, clarifying the approximate range of damage. The damage area quantification module uses... The damage location distribution map is used as input. The U-Net network algorithm is used to accurately segment the damaged area, separating the complete damaged area from the background area. Then, the pixel density difference within the damaged area is calculated, and the pixel features are converted into quantifiable damage indicators, finally obtaining the damage quantification value. The wear state assessment module calls the preset standard 3D model of the thread engagement area of ​​the planetary roller screw pair. This model is pre-built based on the actual size of the screw and the thread engagement structure parameters, and it perfectly matches the geometry and coordinate position of the actual screw thread engagement area, ensuring the accuracy of the mapping. In the processing, first, a one-to-one correspondence between the damage quantification value and the corresponding thread engagement area coordinates is established, clarifying the specific coordinate position of the thread engagement area corresponding to each damage quantification value. The coordinate position is accurately mapped to the specific tooth surface, tooth root and other key parts of the thread engagement. Then, according to the preset coordinate mapping rules, the damage quantification value corresponding to each coordinate is accurately mapped to the corresponding coordinate position of the standard 3D model. Finally, the wear quantification degree of each part is marked on the 3D model in a way that can intuitively distinguish the quantification differences. Finally, all relevant data are integrated to generate a standardized wear state dataset.

[0134] The aforementioned modules work collaboratively, with clearly defined roles. The signal spectrum extraction module ensures effective conversion and feature extraction of vibration signals, providing a precise data source for subsequent assessments. The vibration anomaly judgment module can quickly identify vibration anomalies, avoiding missed or false judgments and providing accurate guidance for damage identification. The damage location identification module uses a convolutional neural network model to accurately locate the damage position and clarify the damage range. The damage area quantification module transforms image features into quantifiable indicators, achieving precise quantification of wear degree. The wear state assessment module uses standard 3D model mapping and annotation to transform abstract quantitative data into an intuitive visualization, while generating standardized datasets to provide a scientific basis for equipment maintenance. The entire module system effectively solves the problems of non-standardized data, inaccurate positioning, and unclear quantification in planetary roller screw pair wear assessment, improving the efficiency and accuracy of wear assessment, providing reliable data support for planetary roller screw pair operation and maintenance decisions, and ensuring stable equipment operation.

[0135] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0136] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the planetary roller screw pair wear condition assessment method as described above.

[0137] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0138] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0139] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for assessing the wear condition of planetary roller screw pairs, characterized in that, The method includes: Real-time vibration signals during the high-speed operation of the planetary roller screw pair are collected, and the real-time vibration signals are frequency domain transformed using the fast Fourier transform method to determine the spectral characteristic data of the processed vibration signals. The amplitude of the low-frequency component in the spectral feature data is judged based on a preset threshold. If it exceeds the preset threshold, it is determined that there is a vibration anomaly and the abnormal frequency band parameters are extracted to generate a vibration anomaly feature vector. The vibration anomaly feature vector is input into a trained convolutional neural network model for classification and identification, generating a damage location distribution map of the thread engagement area. Based on the damage location distribution map, the U-Net network algorithm is used to segment the damage region, and the pixel density difference within the damage region is calculated to obtain the damage quantization value. The damage quantification values ​​are mapped to a standard three-dimensional model of the thread engagement area of ​​the planetary roller screw pair to generate a wear state dataset.

2. The method according to claim 1, characterized in that, The step of performing frequency domain transformation processing on the real-time vibration signal using the Fast Fourier Transform method to determine the spectral characteristic data of the processed vibration signal includes: The wavelet packet threshold denoising algorithm is used to dynamically adjust the threshold parameter according to the amplitude fluctuation characteristics of the real-time vibration signal, filter out environmental interference noise, sensor inherent error and equipment operation noise, and obtain the denoised vibration signal sequence. The denoised vibration signal sequence is processed by using a fast Fourier transform combined with a Hanning window function to convert the vibration signal sequence in the time domain into a frequency domain signal. Based on the inherent characteristic frequency range of the thread engagement of the planetary roller screw pair, the vibration signal sequence converted into a frequency domain signal is separated into low-frequency and high-frequency components directly related to the thread engagement behavior through an adaptive frequency band division algorithm. Key parameters of the low-frequency and high-frequency components are extracted and integrated to obtain the spectral feature data of the vibration signal sequence; The key parameters include amplitude, peak frequency, phase difference, spectral bandwidth, kurtosis, peak factor, waveform factor, and power spectral density.

3. The method according to claim 1, characterized in that, The step of judging the amplitude of low-frequency components in the spectral feature data based on a preset threshold, and determining the presence of vibration anomalies if the amplitude exceeds the preset threshold, and extracting abnormal frequency band parameters to generate a vibration anomaly feature vector, includes: An adaptive threshold model is constructed based on the normal vibration spectrum data of a standard planetary roller screw pair under different speed and load conditions. The adaptive threshold model can dynamically adjust the threshold range according to the current real-time operating conditions of the lead screw. The amplitude of low-frequency components is extracted from the spectral feature data, and the amplitude of low-frequency components is compared in real time with the threshold corresponding to the current operating condition output by the adaptive threshold model to determine whether it exceeds the preset threshold. If the vibration exceeds the preset threshold, it is determined that there is an abnormal vibration in the planetary roller screw pair. The frequency range of the abnormal frequency band is located simultaneously, and the multi-dimensional parameters of the abnormal frequency band are extracted. The multi-dimensional parameters include amplitude, root mean square amplitude, kurtosis, peak factor, spectral distortion rate, frequency offset, and amplitude fluctuation coefficient. The extracted multi-dimensional parameters are normalized to generate a vibration anomaly feature vector.

4. The method according to claim 3, characterized in that, The adaptive threshold model is constructed using the following formula: in, This indicates the adaptive preset threshold under the current operating conditions. The reference threshold for a standard planetary roller screw pair under rated speed and rated load conditions is calibrated using standard vibration spectrum data. This indicates the current real-time rotational speed of the planetary roller screw pair. This indicates the rated speed of a standard planetary roller screw pair. This indicates the current real-time load of the planetary roller screw pair. This indicates the rated load of a standard planetary roller screw pair. This represents the speed correction factor. This represents the load correction factor, calibrated using standard operating condition tests.

5. The method according to claim 1, characterized in that, The step of inputting the vibration anomaly feature vector into a trained convolutional neural network model for classification and recognition, and generating a damage location distribution map of the thread engagement area, includes: The deep features of the abnormal vibration are extracted from the vibration anomaly feature vector using a wavelet packet decomposition algorithm to obtain an optimized vibration anomaly feature vector. The optimized vibration anomaly feature vector is input into a preset convolutional neural network model for classification and recognition, and the preliminary damage location information of the thread engagement area is output. The convolutional neural network model incorporates a channel attention mechanism, and the model is trained based on sample data of different wear types of planetary roller screw pairs. The initial damage location information is corrected for accuracy, and a damage location distribution map of the thread engagement area is generated by combining the geometric features of the thread engagement.

6. The method according to claim 1, characterized in that, The damage region is segmented using the U-Net network algorithm based on the damage location distribution map, and the pixel density difference within the damage region is calculated to obtain the damage quantization value, including: The complete damage region was obtained by segmenting the damage location distribution map using the U-Net network algorithm; The U-Net network algorithm adopts an encoder-decoder network structure and fuses feature maps of different scales through skip connections to segment the damaged region and the background region. Calculate the pixel density difference and grayscale gradient change within the damaged area, and combine the preset mapping relationship between pixel size and the actual physical size of the lead screw to obtain a preliminary quantitative index of the internal wear degree. The support vector machine algorithm is used to perform regression fitting on the preliminary quantification index to construct a nonlinear mapping model between the preliminary quantification value and the actual wear degree of the lead screw; The calculation formula for the nonlinear mapping model is as follows: in, This indicates the actual wear level of the leadscrew. A preliminary quantitative indicator of the degree of internal wear. Represents the radial basis kernel function. , Represents kernel function parameters. This indicates the preliminary quantitative indicators of the sample. Represents the regression coefficients of the SVM model. The bias terms are all trained and calibrated using planetary roller screw pair wear sample data. Real-time speed and load data of planetary roller screw pairs are collected, and ambient temperature data is incorporated to construct a multi-factor deviation correction model. The preliminary quantitative indicators are input into the nonlinear mapping model to obtain the regression fitting result. The regression fitting result is then dynamically corrected using the multi-factor deviation correction model, and the damage quantification value after compensation for working conditions and environment is output.

7. The method according to claim 1, characterized in that, The step of mapping the damage quantification value to a standard three-dimensional model of the thread engagement area of ​​the planetary roller screw pair to generate a wear state dataset includes: Based on the damage quantification value and the preset planetary roller screw pair wear level classification standard, the corresponding wear level is determined. Based on the damage quantification value and the wear level, and combined with the preset wear evolution benchmark data, the wear development trend is obtained by fitting. A visual assessment report is generated based on the wear level and the wear development trend; Establish the correspondence between the damage quantification value and the coordinates of the thread engagement area, map the damage quantification value to the standard three-dimensional model of the thread engagement area of ​​the planetary roller screw pair according to the coordinates, and mark the wear quantification degree of each part. By integrating the visual assessment report, the wear quantification level, and the damage quantification value, a standardized wear status dataset is generated.

8. A device for assessing the wear condition of planetary roller screw pairs, characterized in that, The device includes: The signal spectrum extraction module is used to collect real-time vibration signals during the high-speed operation of the planetary roller screw pair, and to perform frequency domain transformation processing on the real-time vibration signals using the fast Fourier transform method to determine the spectral characteristic data of the processed vibration signals. The vibration anomaly detection module is used to determine the amplitude of the low-frequency component in the spectral feature data based on a preset threshold. If the amplitude exceeds the preset threshold, the module determines that there is a vibration anomaly and extracts the abnormal frequency band parameters to generate a vibration anomaly feature vector. The damage location identification module is used to input the vibration abnormality feature vector into a trained convolutional neural network model for classification and identification, and generate a damage location distribution map of the thread engagement area. The damage region quantization module is used to segment the damage region based on the damage location distribution map using the U-Net network algorithm, calculate the pixel density difference within the damage region, and obtain the damage quantization value. The wear condition assessment module is used to map the damage quantification value to a standard three-dimensional model of the thread engagement area of ​​the planetary roller screw pair, generating a wear condition dataset.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images