Method and system for monitoring wear of a hydraulic motor

Abstract

This invention discloses a method and system for monitoring wear of hydraulic motors, belonging to the field of wear fault prediction technology. The method specifically includes: real-time acquisition of multi-dimensional data signals from the hydraulic motor and preprocessing them; processing the preprocessed multi-dimensional data to extract early wear characteristics of the hydraulic motor; the processing includes decomposing the preprocessed multi-dimensional data, constructing a decomposition tree, retaining nodes with preset frequencies in the decomposition tree, and reconstructing, differentially calculating, and suppressing the retained nodes with window moving average; fusing the early wear characteristics of the hydraulic motor and generating a comprehensive health index through a dimensionality reduction algorithm; judging whether the hydraulic motor has early minor wear based on a comparison of the comprehensive health index and a dynamic threshold, and issuing an alarm based on the judgment result; this application can detect early minor wear signs of hydraulic motors, allowing for advance maintenance planning and avoiding production interruptions due to sudden equipment failures.

Description

Technical Field

[0001] This invention belongs to the field of wear failure prediction technology, specifically a method and system for monitoring wear of hydraulic motors. Background Technology

[0002] Hydraulic motors are widely used in many fields such as engineering machinery, shipbuilding, and metallurgy, playing a crucial role in power output and braking during equipment operation. Due to prolonged operation under high loads and complex conditions, internal components of hydraulic motors, such as pistons, seals, and brake pads, are prone to wear. Wear leads to a decline in equipment performance, such as reduced output torque and unstable speed, and in severe cases, can cause equipment failure, production interruptions, and safety accidents. Traditional wear monitoring relies mainly on periodic manual disassembly and inspection. This method is labor-intensive, costly, and cannot provide real-time updates on equipment wear, often only discovering it when it has become quite severe.

[0003] Existing methods for monitoring wear in hydraulic motors are inefficient and untimely, failing to meet the equipment's need for real-time and accurate monitoring. Manual disassembly and inspection can disrupt normal equipment operation and make it difficult to detect early signs of minor wear. Furthermore, there is a lack of effective analytical methods for understanding the patterns of wear under different operating conditions, hindering the development of advance maintenance plans and making it difficult to ensure reliable equipment operation. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a hydraulic motor wear monitoring method and system, which achieves highly sensitive identification of early wear based on differential signal characteristics and multi-sensor fusion for online monitoring.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] Hydraulic motor wear monitoring methods include:

[0007] Real-time acquisition of multi-dimensional data signals from the hydraulic motor, followed by preprocessing;

[0008] The preprocessed multi-dimensional data of the hydraulic motor is processed to extract early wear characteristics of the hydraulic motor, including time domain features and frequency domain features. The processing includes decomposing the preprocessed multi-dimensional data of the hydraulic motor, constructing a decomposition tree, retaining nodes with preset frequencies in the decomposition tree, and reconstructing, calculating differences and suppressing window moving averages on the retained nodes.

[0009] Early wear characteristics of hydraulic motors are fused, and a comprehensive health index is generated through a dimensionality reduction algorithm;

[0010] Based on the comparison of comprehensive health indicators and dynamic thresholds, it is determined whether there is early minor wear in the hydraulic motor, and an alarm is triggered based on the determination result.

[0011] Specifically, the process of processing the pre-processed multi-dimensional data of the hydraulic motor to extract early wear characteristics of the hydraulic motor includes:

[0012] The pre-processed multi-dimensional data signal of the hydraulic motor is filtered according to the preset digital bandpass filter to eliminate phase distortion and retain the waveform characteristics of the multi-dimensional data signal of the hydraulic motor.

[0013] The multi-dimensional data signal of the hydraulic motor after bandpass filtering is decomposed, and the multi-dimensional data signal of the hydraulic motor at a preset frequency is retained;

[0014] Differential calculation is performed on the multi-dimensional data signal of the hydraulic motor with a preset frequency, and the differential calculation result is suppressed by window moving average to suppress numerical jitter.

[0015] The time-domain and frequency-domain features of the hydraulic motor multi-dimensional data signal after window moving average are extracted. The time-domain features include peak value, root mean square, skewness, and kurtosis. The frequency-domain features include: frequency band energy ratio, spectral entropy, and dominant frequency.

[0016] Specifically, the process of decomposing the bandpass-filtered multi-dimensional data signal of the hydraulic motor and retaining the multi-dimensional data signal of the hydraulic motor at a preset frequency includes:

[0017] Define the mother wavelet and determine the number of decomposition layers L based on the sampling rate and target bandwidth;

[0018] The multi-dimensional data signal of the hydraulic motor after bandpass filtering is decomposed and layered. At layer 0, the root node is initialized, and filtering and downsampling are performed to generate two child nodes. At layer 2, each node is further filtered and downsampled to generate four child nodes. This process is repeated up to layer L, forming two child nodes. L Build a decomposition tree from the last node;

[0019] The nodes with the preset frequency in the decomposition tree are retained and reconstructed to obtain the multi-dimensional data signal of the hydraulic motor with the preset frequency.

[0020] Specifically, the step of performing differential calculation on the multi-dimensional data signal of the hydraulic motor at a preset frequency, and then applying a window moving average to suppress numerical jitter in the differential calculation results includes:

[0021] Select the differential operator, perform differential calculation on each sampling point in the multi-dimensional data signal of the hydraulic motor at a preset frequency, process the first and last points, and use the same boundary extension or directly discard the first and last samples;

[0022] Set the window half-width to M, satisfying 2M+1∈[5,11]. Calculate the moving average of the difference calculation results based on the window half-width.

[0023] The results of the moving average calculation are normalized to obtain the multi-dimensional data signal of the hydraulic motor after window moving average.

[0024] Specifically, the process of fusing early wear characteristics of hydraulic motors and generating a comprehensive health index through a dimensionality reduction algorithm includes:

[0025] The time-domain and frequency-domain features of the multi-dimensional data signal of the hydraulic motor after window moving average are concatenated column by column to obtain the feature fusion matrix;

[0026] Calculate the covariance matrix of the feature fusion matrix, perform eigenvalue decomposition on the covariance matrix, and select the first k principal components in descending order of eigenvalue size to obtain the projection matrix;

[0027] Projecting the feature fusion matrix onto the first principal component yields the first comprehensive health index.

[0028] A bottleneck network is constructed, and the feature fusion matrix is ​​input into the bottleneck network. The training objective of the bottleneck network is set as minimizing the reconstruction error, and the reconstruction error output by the bottleneck network is used as the second comprehensive health indicator.

[0029] The minimum value between the first comprehensive health indicator and the second comprehensive health indicator is selected as the final comprehensive health indicator.

[0030] Specifically, the process of determining whether the hydraulic motor exhibits early minor wear based on a comparison of comprehensive health indicators and dynamic thresholds, and issuing an alarm based on the determination result, includes:

[0031] After the hydraulic motor is put into operation or maintained for the first time, the comprehensive health index sequence is collected under no-load conditions within a preset time. An initial comprehensive health index threshold is set, and the initial comprehensive health index threshold is updated in real time within the preset time to obtain the comprehensive health index threshold. The linear fitting slope k of the comprehensive health index within the preset time is calculated.

[0032] When the final comprehensive health index exceeds the comprehensive health index threshold, it is determined that the wear and tear zone has been entered, triggering a level one alarm.

[0033] Calculate the linear fit slope k of the final comprehensive health index at A time points. A If k A If the value is greater than k and continues for B sampling cycles, it is determined that the hydraulic motor has experienced early minor wear, triggering a level two alarm.

[0034] Specifically, the multi-dimensional data signals of the hydraulic motor include: vibration signals, pressure signals, temperature data, acoustic signals, and abrasive particle concentration; the preprocessing includes: noise reduction, normalization, compensation correction, and spatiotemporal synchronization.

[0035] The noise reduction is used to remove noise from vibration signals and pressure signals;

[0036] The normalization is used to normalize the abrasive particle concentration;

[0037] The compensation correction is used to perform compensation correction processing on the temperature data;

[0038] The spatiotemporal synchronization is used to process the timestamps corresponding to the multi-dimensional data signals of the hydraulic motor to a unified time grid, and to establish a spatial mapping relationship between the physical location of the sensor and the structure of the hydraulic motor.

[0039] A hydraulic motor wear monitoring system is used to implement a hydraulic motor wear monitoring method, including: a data processing module, an early feature extraction module, a score generation module, and a comparison and judgment module;

[0040] The data processing module is used to acquire multi-dimensional data signals from the hydraulic motor in real time and perform preprocessing.

[0041] The early feature extraction module is used to process the preprocessed multi-dimensional data of the hydraulic motor and extract the early wear features of the hydraulic motor, including time domain features and frequency domain features. The processing includes decomposing the preprocessed multi-dimensional data of the hydraulic motor, constructing a decomposition tree, retaining nodes with preset frequencies in the decomposition tree, and performing reconstruction, difference calculation and window moving average suppression.

[0042] The scoring generation module is used to fuse the early wear characteristics of the hydraulic motor and generate a comprehensive health index through a dimensionality reduction algorithm.

[0043] The comparison and determination module determines early minor wear of the hydraulic motor and issues an alarm based on a comparison of comprehensive health indicators and dynamic thresholds.

[0044] Specifically, the early feature extraction module includes: a decomposition unit, a difference suppression unit, and a feature extraction unit;

[0045] The decomposition unit decomposes the bandpass filtered hydraulic motor multi-dimensional data signal and retains the hydraulic motor multi-dimensional data signal of a preset frequency.

[0046] The differential suppression unit is used to perform differential calculation on the multi-dimensional data signal of the hydraulic motor with a preset frequency, and to perform window moving average suppression on the differential calculation result to suppress numerical jitter.

[0047] The feature extraction unit is used to extract the time-domain and frequency-domain features of the multi-dimensional data signal of the hydraulic motor after window moving average.

[0048] Specifically, the rating generation module includes: a feature fusion unit, a first rating generation unit, a second rating generation unit, and a final rating generation unit;

[0049] The feature fusion unit is used to concatenate the time-domain features and frequency-domain features of the multi-dimensional data signal of the hydraulic motor after window moving average by column to obtain a feature fusion matrix.

[0050] The first scoring generation unit is used to calculate the covariance matrix of the feature fusion matrix, perform eigenvalue decomposition on the covariance matrix, select the first k principal components in descending order of eigenvalue size to obtain the projection matrix, and project the feature fusion matrix onto the first principal component to obtain the first comprehensive health index.

[0051] The second rating generation unit is used to construct a bottleneck network, input the feature fusion matrix into the bottleneck network, set the bottleneck network training objective as minimizing the reconstruction error, and use the reconstruction error output by the bottleneck network as the second comprehensive health indicator.

[0052] The final score generation unit is used to select the minimum value of the first comprehensive health index and the second comprehensive health index as the final comprehensive health index.

[0053] Compared with the prior art, the beneficial effects of the present invention are:

[0054] This invention proposes a method and system for monitoring wear of hydraulic motors. It utilizes high-frequency bandpass filtering and wavelet packet decomposition to capture transient high-frequency signals generated by wear. Through differential methods, it extracts features such as peak value and kurtosis, significantly improving the ability to identify early wear and quantifying them into a comprehensive health index. Based on the comparison between the comprehensive health index and a dynamic threshold, and considering the fitting slope and usage period, it determines whether the wear has entered the wear stage or is in the early stage of minor wear. This achieves accurate identification and graded alarm for early minor wear of hydraulic motors, reducing the risk of equipment downtime due to sudden hydraulic motor failures and improving operational safety and production efficiency. Attached Figure Description

[0055] Figure 1 Flowchart of the hydraulic motor wear monitoring method provided by the present invention;

[0056] Figure 2 The flowchart for early feature extraction provided by this invention;

[0057] Figure 3 This is a schematic diagram of early wear determination provided by the present invention;

[0058] Figure 4 This is a diagram illustrating the architecture of a hydraulic motor wear monitoring system provided by the present invention. Detailed Implementation

[0059] The present application will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present application, but do not limit the present application in any way. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present application. These all fall within the protection scope of the present application.

[0060] 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.

[0061] It should be noted that, unless there is a conflict, the various features in the embodiments of this application can be combined with each other, all of which are within the protection scope of this application. Furthermore, although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than the module division in the device or the order in the flowchart. In addition, the terms "first," "second," and "third" used in this application do not limit the data or execution order, but only distinguish identical or similar items with essentially the same function and effect.

[0062] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items.

[0063] Example 1

[0064] Please see Figures 1-3 The present invention provides an embodiment of a hydraulic motor wear monitoring method, which includes the following specific steps:

[0065] Step S1: Install sensors for vibration, pressure, temperature, acoustics, and abrasive particle concentration at key parts of the hydraulic motor to collect multi-dimensional data signals of the hydraulic motor in real time and perform preprocessing.

[0066] In this embodiment, pressure sensors are installed at the inlet and outlet flanges to monitor minute pulsations and sudden changes in oil pressure in real time. Early wear causes slight changes in the geometry of the flow channel, resulting in characteristic pressure fluctuations at high frequencies. Vibration sensors are installed on the side plate of the housing near the main bearing. A metal bracket is welded first, and then the sensor is fixed with a magnetic base to ensure good coupling and easy maintenance and replacement. This sensor is used to detect high-frequency vibrations in the motor housing caused by the minute wear of internal components (gears, bearings). Temperature sensors are used to monitor changes in oil temperature. Early wear friction can cause local temperature rises. Acoustic sensors are used to capture high-frequency acoustic emission signals, such as ultrasonic waves generated by minute cracks, cavitation, or metal friction. Abrasive particle concentration sensors are used to monitor changes in the concentration of wear particles in the oil, and abnormal wear stages are determined by sudden changes in the number of abrasive particles.

[0067] The preprocessing includes: noise reduction, normalization, compensation correction, and spatiotemporal synchronization;

[0068] Denoising includes: performing three-level wavelet decomposition on the original vibration signal to obtain detail coefficients and approximation coefficients for different frequency bands; applying an adaptive threshold adjustment mechanism to each level of detail coefficients to retain effective signals in the high-frequency band, taking into account the non-stationary characteristics of the wear signal; reconstructing the denoised signal, calculating the kurtosis enhancement factor, and verifying the denoising effect; the specific denoising formula is as follows: Where w'(t) represents the denoised vibration signal, w(t) represents the original vibration signal, including the wear signal (effective signal) and the noise signal, λ represents the correction coefficient, σ represents the noise standard deviation, T represents the total duration of the denoised vibration signal, t represents the current time, i.e., the time point in the vibration signal sequence, used to dynamically adjust the threshold at the current time, and sgn() represents the sign function, used to determine the positive or negative sign of the vibration signal. This represents the time factor, which monotonically decreases from approaching infinity to as t increases. A nonlinear attenuation curve is formed, and the noise sensitivity and noise standard deviation are adjusted by the correction coefficient to adapt to different noise levels under different working conditions. The time factor matches the non-stationary characteristics of the wear signal, forming a dynamic adaptive threshold strategy. This solves the problem that traditional fixed thresholds cannot take into account both early weak signal detection and later noise suppression. It is especially suitable for extracting weak impact signals in the early stage of hydraulic motor wear.

[0069] Normalization is performed by collecting normal operating data of the motor under four typical operating conditions: no load, 30% load, 60% load, and full load (each operating condition lasts for 30 minutes with no obvious wear). The mean and standard deviation of abrasive particle concentration under each operating condition are calculated and processed using a normalization formula to eliminate the influence of oil flow fluctuations under different loads or oil temperatures on the abrasive particle concentration detection value.

[0070] The pressure signal is noise-removed using a filter, and the temperature signal is compensated and corrected.

[0071] Spatiotemporal synchronization: Add timestamps with microsecond accuracy to the data of each sensor, process the timestamps to a unified time grid, and establish a spatial mapping relationship between the physical location of the sensor and the structure of the hydraulic motor.

[0072] Step S2: Process the preprocessed multi-dimensional data of the hydraulic motor to extract the early wear characteristics of the hydraulic motor;

[0073] like Figure 2 As shown, the specific steps of step S2 are as follows:

[0074] Step S201: Filter the preprocessed hydraulic motor multi-dimensional data signal according to the preset digital bandpass filter to eliminate phase distortion and retain the waveform characteristics of the hydraulic motor multi-dimensional data signal;

[0075] It should be noted that finite impulse response digital bandpass filters are typically used to preserve the target frequency band and eliminate phase distortion. For hydraulic motors, the 5–50 kHz frequency band contains the high-frequency impact response caused by wear and is the main frequency domain region for wear identification.

[0076] Step S202: Decompose the multi-dimensional data signal of the hydraulic motor after bandpass filtering and retain the multi-dimensional data signal of the hydraulic motor at a preset frequency;

[0077] The specific steps of step S202 are as follows:

[0078] Step S2021: Set the mother wavelet and select the number of decomposition layers L according to the sampling rate and target bandwidth;

[0079] Specifically, such as Figure 2 As shown, Figure 2 The left side shows the signal decomposition hierarchy diagram, which shows that the filtered multi-dimensional data signal of the hydraulic motor is decomposed into multiple frequencies using the mother wavelet. The signal is divided into different frequency band sub-signals layer by layer. The mother wavelet is set to db4, the number of decomposition layers L=4, the 100kHz bandwidth is subdivided into 16, and the models between 5-50kHz are selectively retained to ensure the time domain positioning accuracy.

[0080] Step S2022: Decompose and layer the multi-dimensional data signal of the hydraulic motor after bandpass filtering. Initialize the root node in layer 0, and simultaneously filter and downsample the low and high frequencies to generate 2 child nodes. In layer 2, filter and downsample each node again to generate 4 child nodes. Repeat this process until layer L, forming 2 L Build a decomposition tree from the last node;

[0081] like Figure 2 As shown, Figure 2The middle left part is the decomposition tree structure diagram. The child nodes corresponding to the 5-50kHz frequency band are selected from the decomposition tree and retained to construct the decomposition tree.

[0082] Step S2023: Retain the nodes with the preset frequency in the decomposition tree and reconstruct them to obtain the multi-dimensional data signal of the hydraulic motor with the preset frequency.

[0083] Specifically, such as Figure 2 As shown, based on the sampling rate and number of layers, sub-nodes corresponding to the 5–50kHz frequency band are selected from the decomposition tree for retention, and reconstruction operations are performed on the selected nodes to accurately isolate the target frequency band, remove interference components unrelated to wear, retain the signal part with the most wear-characteristic response, discard irrelevant sub-bands, and remove background information related to the normal operating state of the hydraulic motor but unrelated to wear.

[0084] Step S203: Perform differential calculation on the multi-dimensional data signal of the hydraulic motor with a preset frequency, and perform window moving average suppression on the differential calculation result to suppress high-frequency numerical jitter;

[0085] The specific steps of step S203 are as follows:

[0086] Step S2031: Select the differential operator, perform differential calculation on each sampling point in the multi-dimensional data signal of the hydraulic motor at a preset frequency, process the first and last points, and use the same boundary extension or directly discard the first and last samples.

[0087] In this embodiment, the difference operator includes forward difference, central difference and higher-order difference. The purpose of difference calculation is to amplify the instantaneous rate of change, highlight the spikes or abrupt changes caused by wear, and facilitate the detection of edge events.

[0088] For each sampling point in the reconstructed target frequency band signal sequence, a differential operator is applied to obtain a new sequence. Figure 2 After signal reconstruction, a slight rise or abrupt change can be observed, and the slope of the abrupt change region can be increased by differential grading.

[0089] Step S2032: Set the window half-width to M, satisfying 2M+1∈[5,11]. Based on the window half-width, perform a moving average calculation on the difference calculation results.

[0090] In this embodiment, as Figure 2 As shown, the waveform after signal reconstruction shows a slight increase in drop after differential processing, indicating the presence of local spikes. During the moving average calculation and normalization stage, the signal is smoothed by window sliding, so that the abrupt change is still preserved, but its noise edge is suppressed, and the output signal shows a smoother trend.

[0091] Step S2033: Normalize the result after the moving average calculation to obtain the multi-dimensional data signal of the hydraulic motor after window moving average.

[0092] In this embodiment, Figure 2 The lower right corner shows the noise-reduced and normalized multi-dimensional data after window moving average. After differential, moving average and normalization processing, the signal retains the wear trend and eliminates the amplitude difference between multiple channels. This avoids the amplitude difference between different frequency bands or sensor channel signals interfering with feature extraction, and makes the data scale of each channel uniform.

[0093] The benefits of step S203 are: to amplify and extract transient changes caused by wear with low complexity and minimal delay, thereby improving the sensitivity and reliability of feature discrimination.

[0094] Step S204: Extract the time-domain and frequency-domain features of the multi-dimensional data signal of the hydraulic motor after window moving average. The time-domain features include peak value, root mean square, skewness, and kurtosis. The frequency-domain features include: frequency band energy ratio, spectral entropy, and dominant frequency.

[0095] In this embodiment, the peak value is calculated by the difference between the maximum and minimum values ​​of the signal within each window, amplifying the transient impact amplitude and sensitively capturing the maximum vibration or pressure fluctuations caused by wear spikes; the root mean square measures the overall energy level of the signal, reflecting the average intensity of continuous friction or abrasive impact; the skewness describes the waveform asymmetry, distinguishing between positive or negative abrupt changes, and identifying unilateral wear impacts earlier; the kurtosis captures occasional spikes, and high-amplitude pulses caused by tiny abrasive collisions can significantly improve the kurtosis value.

[0096] The frequency band energy ratio is used to calculate the energy of each sub-band; the spectral entropy measures the complexity of the spectral distribution, and the random high-frequency components caused by early microcracks or cavitation will increase; the dominant frequency is the frequency corresponding to the highest peak in the power spectrum;

[0097] The features extracted in step S204 constitute the input matrix of the subsequent comprehensive health assessment model. In practical applications, these statistical features can sensitively reflect small changes in wear and tear, and have a better diagnostic effect than manual inspection, especially in the early stages.

[0098] Step S3: Integrate the early wear characteristics of the hydraulic motor and generate a comprehensive health index through a dimensionality reduction algorithm.

[0099] The specific steps of step S3 are as follows:

[0100] Step S301: Concatenate the time-domain and frequency-domain features of the multi-dimensional data signal of the hydraulic motor after window moving average by column to obtain the feature fusion matrix;

[0101] Step S302: Calculate the covariance matrix of the feature fusion matrix, perform eigenvalue decomposition on the covariance matrix, select the first k principal components in descending order of eigenvalue size to obtain the projection matrix;

[0102] Specifically, the covariance matrix of the feature fusion matrix is ​​constructed, eigenvalue decomposition is performed on the covariance matrix, and the original feature fusion matrix is ​​linearly transformed through the principal component matrix to obtain the dimensionality-reduced feature representation. The projection matrix is ​​constructed by selecting the first k principal components.

[0103] Step S303: Project the feature fusion matrix onto the first principal component to obtain the first comprehensive health index;

[0104] In this embodiment, Principal Component Analysis (PCA) performs eigenvalue decomposition on the covariance matrix, selects the k directions with the largest variance, projects the high-dimensional data into a low-dimensional subspace, reduces the dimensionality of the feature fusion matrix, retains only the principal components with the largest variance, and removes low-variance signals. The first principal component carries the largest variance information among all the original features and serves as the first comprehensive health indicator, which can most sensitively reflect state changes.

[0105] Step S304: Construct a bottleneck network, input the feature fusion matrix into the bottleneck network, set the bottleneck network training objective to minimize the reconstruction error, and use the reconstruction error output by the bottleneck network as the second comprehensive health indicator.

[0106] In this embodiment, the bottleneck network is constructed as follows: a three-layer fully connected network is constructed, consisting of an input layer, a bottleneck layer, and an output layer; the activation function is ReLU, and the output layer is linearly activated; the input feature fusion matrix is ​​used, and the reconstruction error is calculated for any input window feature, and a reconstruction error sequence is obtained for all windows. Abnormal samples have poor reconstruction ability because they have not seen similar patterns, resulting in a significant increase in error. This can sensitively locate early micro-damage. Step S304 can capture nonlinear and coupled abnormal patterns in the original feature matrix that cannot be expressed by PCA, and amplify the small deviations caused by early wear into quantifiable reconstruction errors, which can then be used as a second comprehensive health indicator.

[0107] Step S305: Select the minimum value of the first comprehensive health index and the second comprehensive health index as the final comprehensive health index.

[0108] Specifically, the minimum value between the first and second comprehensive health indicators is selected to more promptly determine whether early wear has occurred, thus providing higher sensitivity and reliability for online fault early warning of hydraulic motors.

[0109] Step S4: Based on the comparison of comprehensive health indicators and dynamic thresholds, determine early minor wear and trigger an alarm.

[0110] like Figure 3 As shown, the specific steps of step S4 are as follows:

[0111] Step S401: After the hydraulic motor is put into operation for the first time or after maintenance, collect the comprehensive health index sequence within a preset time under no-load conditions, set the initial comprehensive health index threshold, update the initial comprehensive health index threshold in real time within the preset time, obtain the comprehensive health index threshold, and calculate the linear fitting slope k of the comprehensive health index within the preset time.

[0112] Specifically, there is no wear under no-load conditions after the initial commissioning or maintenance, so the value within this period is used as the threshold, including the linear fitting slope k.

[0113] Step S402: When the final comprehensive health index is greater than the comprehensive health index threshold, it is determined that the wear range has been entered, and a level one alarm is triggered;

[0114] Step S403: Calculate the linear fitting slope k of the final comprehensive health index at A time points. A If k A If the value is greater than k and continues for B sampling cycles, it is determined that the hydraulic motor has experienced early minor wear, triggering a level two alarm.

[0115] In this embodiment, a single threshold alarm can detect deviations in health indicators in a timely manner, but it is prone to false alarms due to short-term noise or transient fluctuations. Combining trend analysis can confirm the continuous upward trend of health indicators. The slope of the continuous rise can capture the trend inflection point before the health indicators exceed the threshold significantly, distinguishing between occasional pulses and true early minor wear and tear, and achieving earlier intervention. The dual strategy can achieve rapid response and high-confidence early warning, while suppressing false alarms.

[0116] exist Figure 3 In this context, A is set to 3. Those skilled in the art can adjust the value according to actual conditions. There exists a linear fitting slope k for the final comprehensive health index at the most recent A time points. A >k, but not a continuous B sampling period, so it does not meet the condition for early fine wear of the hydraulic motor. The points within the dashed box are the most recent A points. Calculate the slope of the data corresponding to these A time points. When k appears within a continuous B sampling period... A When the value is greater than k, early minor wear of the hydraulic motor is determined to have occurred; that is, the first point in B consecutive sampling cycles indicates early minor wear of the hydraulic motor. Because data fluctuations exist, and existing monitoring methods are inefficient and inaccurate, this application enables the detection of early minor wear signs within a very short period, providing maintenance personnel with ample time to develop reasonable maintenance plans and avoiding production interruptions due to sudden equipment failures. This reduces maintenance costs, improves equipment reliability and lifespan, ensures the continuity and safety of production processes in related industries, and enhances enterprise production efficiency.

[0117] Example 2

[0118] Please see Figure 4 Another embodiment of the present invention provides a hydraulic motor wear monitoring system, comprising: a data processing module, an early feature extraction module, a score generation module, and a comparison and judgment module;

[0119] The data processing module is used to acquire multi-dimensional data signals from the hydraulic motor in real time and perform preprocessing.

[0120] The early feature extraction module is used to process the preprocessed multi-dimensional data of the hydraulic motor and extract the early wear features of the hydraulic motor, including time domain features and frequency domain features. The processing includes decomposing the preprocessed multi-dimensional data of the hydraulic motor, constructing a decomposition tree, retaining nodes with preset frequencies in the decomposition tree, and performing reconstruction, difference calculation and window moving average suppression.

[0121] The scoring generation module is used to fuse the early wear characteristics of the hydraulic motor and generate a comprehensive health index through a dimensionality reduction algorithm.

[0122] The comparison and determination module determines early minor wear of the hydraulic motor and issues an alarm based on a comparison of comprehensive health indicators and dynamic thresholds.

[0123] The early feature extraction module includes: a decomposition unit, a difference suppression unit, and a feature extraction unit;

[0124] The decomposition unit uses wavelet packets to decompose the bandpass filtered multidimensional data signal of the hydraulic motor, and retains the multidimensional data signal of the hydraulic motor at a preset frequency.

[0125] The differential suppression unit is used to perform differential calculation on the multi-dimensional data signal of the hydraulic motor with a preset frequency, and to perform window moving average suppression on the differential calculation result to suppress high-frequency numerical jitter.

[0126] The feature extraction unit is used to extract the time-domain and frequency-domain features of the multi-dimensional data signal of the hydraulic motor after window moving average.

[0127] The rating generation module includes: a feature fusion unit, a first rating generation unit, a second rating generation unit, and a final rating generation unit;

[0128] The feature fusion unit is used to concatenate the time-domain features and frequency-domain features of the multi-dimensional data signal of the hydraulic motor after window moving average by column to obtain a feature fusion matrix.

[0129] The first scoring generation unit is used to calculate the covariance matrix of the feature fusion matrix, perform eigenvalue decomposition on the covariance matrix, select the first k principal components in descending order of eigenvalue size to obtain the projection matrix, and project the feature fusion matrix onto the first principal component to obtain the first comprehensive health index.

[0130] The second rating generation unit is used to construct a bottleneck network, input the feature fusion matrix into the bottleneck network, set the bottleneck network training objective as minimizing the reconstruction error, and use the reconstruction error output by the bottleneck network as the second comprehensive health indicator.

[0131] The final score generation unit is used to select the minimum value of the first comprehensive health index and the second comprehensive health index as the final comprehensive health index.

[0132] In addition, the parts of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of the corresponding technical solutions in the prior art have not been described in detail, so as to avoid excessive elaboration.

[0133] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements 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 method for monitoring wear of a hydraulic motor, characterized in that, include: Real-time acquisition of multi-dimensional data signals from the hydraulic motor, followed by preprocessing; The preprocessed multi-dimensional data of the hydraulic motor is processed to extract early wear characteristics of the hydraulic motor, including time domain features and frequency domain features. The processing includes decomposing the preprocessed multi-dimensional data of the hydraulic motor, constructing a decomposition tree, retaining nodes with preset frequencies in the decomposition tree, and reconstructing, calculating differences and suppressing window moving averages on the retained nodes. Early wear characteristics of hydraulic motors are fused, and a comprehensive health index is generated through a dimensionality reduction algorithm; Based on the comparison of comprehensive health indicators and dynamic thresholds, it is determined whether there is early minor wear in the hydraulic motor, and an alarm is triggered based on the determination result. The process of processing the pre-processed multi-dimensional data of the hydraulic motor to extract early wear characteristics of the hydraulic motor includes: The pre-processed multi-dimensional data signal of the hydraulic motor is filtered according to the preset digital bandpass filter to eliminate phase distortion and retain the waveform characteristics of the multi-dimensional data signal of the hydraulic motor. The multi-dimensional data signal of the hydraulic motor after bandpass filtering is decomposed, and the multi-dimensional data signal of the hydraulic motor at a preset frequency is retained; Differential calculation is performed on the multi-dimensional data signal of the hydraulic motor with a preset frequency, and the differential calculation result is suppressed by window moving average to suppress numerical jitter. The time-domain and frequency-domain features of the multi-dimensional data signal of the hydraulic motor after window moving average are extracted. The time-domain features include peak value, root mean square, skewness, and kurtosis. The frequency-domain features include: frequency band energy ratio, spectral entropy, and dominant frequency. The process of decomposing the bandpass-filtered multi-dimensional data signal of the hydraulic motor and retaining the multi-dimensional data signal of the hydraulic motor at a preset frequency includes: Define the mother wavelet and determine the number of decomposition layers L based on the sampling rate and target bandwidth; The multi-dimensional data signal of the hydraulic motor after band-pass filtering is decomposed and layered, a root node is initialized at the 0th layer, filtering and down-sampling are performed at the same time, 2 sub-nodes are generated, at the 2nd layer, filtering and down-sampling are performed on each node respectively, 4 sub-nodes are generated, and the process is repeated to the Lth layer, 2 L end nodes are formed, and a decomposition tree is constructed; The nodes with the preset frequency in the decomposition tree are retained and reconstructed to obtain the multi-dimensional data signal of the hydraulic motor with the preset frequency.

2. The hydraulic motor wear monitoring method as described in claim 1, characterized in that, The step of performing differential calculations on the multi-dimensional data signals of the hydraulic motor at a preset frequency, and then applying a window moving average to suppress numerical jitter in the differential calculation results includes: Select the differential operator to perform differential calculation on each sampling point in the multi-dimensional data signal of the hydraulic motor at a preset frequency, and process the first and last points by boundary extension or direct discard. Set the window half-width to M, satisfying 2M+1∈[5,11]. Calculate the moving average of the difference calculation results based on the window half-width. The results of the moving average calculation are normalized to obtain the multi-dimensional data signal of the hydraulic motor after window moving average.

3. The hydraulic motor wear monitoring method as described in claim 1, characterized in that, The process of fusing early wear characteristics of hydraulic motors and generating comprehensive health indicators through a dimensionality reduction algorithm includes: The time-domain and frequency-domain features of the multi-dimensional data signal of the hydraulic motor after window moving average are concatenated column by column to obtain the feature fusion matrix; Calculate the covariance matrix of the feature fusion matrix, perform eigenvalue decomposition on the covariance matrix, and select the first k principal components in descending order of eigenvalue size to obtain the projection matrix; Projecting the feature fusion matrix onto the first principal component yields the first comprehensive health index. A bottleneck network is constructed, and the feature fusion matrix is ​​input into the bottleneck network. The training objective of the bottleneck network is set as minimizing the reconstruction error, and the reconstruction error output by the bottleneck network is used as the second comprehensive health indicator. The minimum value between the first comprehensive health indicator and the second comprehensive health indicator is selected as the final comprehensive health indicator.

4. The hydraulic motor wear monitoring method as described in claim 3, characterized in that, The method of determining whether the hydraulic motor has early minor wear based on a comparison of comprehensive health indicators and dynamic thresholds, and issuing an alarm based on the determination result, includes: After the hydraulic motor is put into operation or maintained for the first time, the comprehensive health index sequence is collected under no-load conditions within a preset time. An initial comprehensive health index threshold is set, and the initial comprehensive health index threshold is updated in real time within the preset time to obtain the comprehensive health index threshold. The linear fitting slope k of the comprehensive health index within the preset time is calculated. When the final comprehensive health index exceeds the comprehensive health index threshold, it is determined that the wear and tear zone has been entered, triggering a level one alarm. Calculate the linear fit slope k of the final comprehensive health index at A time points. A If k A If the value is greater than k and continues for B sampling cycles, it is determined that the hydraulic motor has experienced early minor wear, triggering a level two alarm.

5. The hydraulic motor wear monitoring method as described in claim 1, characterized in that, The multi-dimensional data signals of the hydraulic motor include: vibration signal, pressure signal, temperature data, acoustic signal, and abrasive particle concentration; The preprocessing includes: noise reduction, normalization, compensation correction, and spatiotemporal synchronization; The noise reduction is used to remove noise from vibration signals and pressure signals; The normalization is used to normalize the abrasive particle concentration; The compensation correction is used to perform compensation correction processing on the temperature data; The spatiotemporal synchronization is used to process the timestamps corresponding to the multi-dimensional data signals of the hydraulic motor to a unified time grid, and to establish a spatial mapping relationship between the physical location of the sensor and the structure of the hydraulic motor.

6. A hydraulic motor wear monitoring system, used to implement the hydraulic motor wear monitoring method according to any one of claims 1-5, characterized in that, include: The module consists of a data processing module, an early feature extraction module, a score generation module, and a comparison and judgment module. The data processing module is used to acquire multi-dimensional data signals from the hydraulic motor in real time and perform preprocessing. The early feature extraction module is used to process the preprocessed multi-dimensional data of the hydraulic motor and extract the early wear features of the hydraulic motor. The scoring generation module is used to fuse early wear characteristics of hydraulic motors and generate a comprehensive health index through a dimensionality reduction algorithm. The comparison and judgment module determines whether the hydraulic motor has early minor wear based on the comparison of comprehensive health indicators and dynamic thresholds, and issues an alarm based on the judgment result.

7. The hydraulic motor wear monitoring system as described in claim 6, characterized in that, The early feature extraction module includes: a decomposition unit, a difference suppression unit, and a feature extraction unit; The decomposition unit decomposes the bandpass filtered hydraulic motor multi-dimensional data signal and retains the hydraulic motor multi-dimensional data signal of a preset frequency. The differential suppression unit is used to perform differential calculation on the multi-dimensional data signal of the hydraulic motor with a preset frequency, and to perform window moving average suppression on the differential calculation result to suppress numerical jitter. The feature extraction unit is used to extract the time-domain and frequency-domain features of the multi-dimensional data signal of the hydraulic motor after window moving average.

8. The hydraulic motor wear monitoring system as described in claim 7, characterized in that, The scoring generation module includes: a feature fusion unit, a first scoring generation unit, a second scoring generation unit, and a final scoring generation unit; The feature fusion unit is used to concatenate the time-domain features and frequency-domain features of the multi-dimensional data signal of the hydraulic motor after window moving average by column to obtain a feature fusion matrix. The first scoring generation unit is used to calculate the covariance matrix of the feature fusion matrix, perform eigenvalue decomposition on the covariance matrix, select the first k principal components in descending order of eigenvalue size to obtain the projection matrix, and project the feature fusion matrix onto the first principal component to obtain the first comprehensive health index. The second rating generation unit is used to construct a bottleneck network, input the feature fusion matrix into the bottleneck network, set the bottleneck network training objective as minimizing the reconstruction error, and use the reconstruction error output by the bottleneck network as the second comprehensive health indicator. The final score generation unit is used to select the minimum value of the first comprehensive health index and the second comprehensive health index as the final comprehensive health index.

Images