Intelligent online monitoring method and system for oil abrasion particles of hydraulic forcible entry equipment

Abstract

The invention discloses an intelligent on-line monitoring method and system for oil abrasion particles of hydraulic forcible entry equipment. The method comprises an on-line monitoring method and an on-line monitoring system. According to the online monitoring method, an electrostatic, inductive and optical multi-source sensor data acquisition module, a feature engineering processing module, a multi-modal feature fusion module, an abrasive particle quantitative and qualitative analysis module and an abrasion analysis module are included, and the online monitoring system comprises the online monitoring method, a monitoring sensor installation module and an abrasion intelligent decision-making module. According to the method, quantitative monitoring and abrasion mode analysis of metal abrasive particles, non-metal abrasive particles, ferromagnetic metal abrasive particles and non-ferromagnetic metal abrasive particles in oil of the hydraulic type forcible entry equipment can be synchronously completed through multi-mode fusion monitoring, online monitoring of the abrasion particles of the hydraulic type forcible entry equipment is achieved, and an intelligent maintenance decision is given.

Description

Technical Field

[0001] The invention relates to the field of wear monitoring of hydraulic systems of demolition equipment, specifically to an intelligent online monitoring method and system for oil wear particles in hydraulic demolition equipment. Background Art

[0002] The hydraulic system is the core power source of intelligent demolition equipment. Through the coordinated operation of hydraulic pumps, cylinders, valves, and intelligent control systems, it provides powerful demolition power and flexible motion control for specialized equipment. The hydraulic system delivers high pressure and torque, driving a variety of demolition tools for efficient operation. Intelligent technologies also enable precise operation, automated operation, and remote monitoring, adapting to complex operating environments. This significantly improves the efficiency, safety, and reliability of demolition equipment, providing crucial technical support for disaster relief and demolition projects. Therefore, the reliability and service life of the hydraulic transmission system are crucial to the proper operation of intelligent demolition equipment. According to domestic and international statistics, approximately 80% of hydraulic transmission system failures are caused by particle contamination in the hydraulic system. Among various contaminants, solid particles are the most widespread and harmful, representing the most significant cause of system failure, reduced reliability, and shortened component life. These solid particle contamination contributes to 60% to 70% of all hydraulic system failures.

[0003] In hydraulic systems, solid particle contamination of the oil includes metal chips, welding slag, and molding sand that remain unremoved during component processing and assembly; dust and mechanical impurities that infiltrate the system from the outside; wear debris and rust flakes generated during system operation; and sediment from oil oxidation and decomposition. The long-term interaction of solid particles with component surfaces causes various forms of wear, accelerating surface wear, increasing clearances, and increasing internal leakage, leading to performance degradation and eventual failure of subcomponents, shortening their service life. They also contribute to particle accumulation, blockage, and seizure, leading to sudden failures. Both metal and non-metallic components in hydraulic systems are particularly susceptible to this type of wear, especially those with high precision fittings, such as spool valves and plungers. Solid particle contamination is primarily determined by the wear characteristics of the contaminant and the relative relationship between particle size and the clearance dimensions of key moving parts. Particles smaller than the clearance between moving parts generally cause only slight friction. Particles larger than the clearance cannot enter the gap and therefore do not wear the surfaces of the moving parts. However, only particles equal to or slightly larger than the clearance can enter the gap under the action of fluid flow, causing wear on the surfaces of the moving parts. This particle size is the most sensitive to contamination wear on the moving parts of hydraulic components. It is not only a key parameter for assessing the contamination sensitivity and resistance of hydraulic components, but also a key indicator for determining the filtration requirements of hydraulic systems. Because the critical particle size corresponds to the clearance between the moving parts of hydraulic components, when severe wear causes the clearance between the moving parts to increase, larger particles will enter the gap and participate in the wear, thereby exacerbating the wear of the moving parts and causing a chain reaction of contamination wear. Therefore, it is urgent to take measures to monitor the size and wear pattern of wear solid particles in the hydraulic systems of special equipment.

[0004] As demolition equipment is increasingly developing towards intelligence, complexity and precision, and the research on the wear status assessment of hydraulic systems is gradually deepening, the online status monitoring technology of solid wear particles of hydraulic demolition equipment is becoming more and more important. It is very important to propose a relatively complete online monitoring device for oil wear particles of hydraulic demolition equipment suitable for actual industrial environments.

[0005] A Chinese patent document, published on September 16, 2015, with publication number CN 103541951 B, discloses a "hydraulic oil contamination monitoring system." This system uses an analytical processing device to monitor oil contamination in real time. However, compared to online monitoring of the specific number of wear particles in the oil, it is less precise, cannot monitor the material of the abrasive particles, or the type of wear on the friction pair, and the monitoring data is limited.

[0006] A Chinese patent, published on November 22, 2019, with publication number CN 107560984 B, discloses an "online monitoring device and method for oil contamination in a hydraulic system." The device assesses the degree of oil contamination by online monitoring of parameters such as the hydraulic system's temperature, pressure, and oil contamination particles. While this invention monitors the number of wear particles for fault warning, monitoring the number alone cannot distinguish the source and material of wear particles, making it impossible to locate wear or determine wear patterns, limiting detection capabilities.

[0007] Chinese patent publication number CN 108680579 A, ​​published on October 19, 2018, discloses a "machine vision-based online monitoring device and method for crane hydraulic oil contamination." This device uses an image acquisition unit to capture images of oil contamination at the filter membrane, enabling rapid and accurate measurement of the hydraulic oil contamination level when the crane's hydraulic cylinder is stopped. This invention uses a camera to capture images of oil contamination at the filter membrane, but its ability to detect quantity and material is limited. The problem it addresses is limited to static hydraulic oil collection, resulting in delayed detection and poor real-time performance. Summary of the Invention

[0008] In order to address the shortcomings of the existing technology, the present invention provides an intelligent online monitoring method and system for oil wear particles in hydraulic demolition equipment, aiming to perform online real-time monitoring of oil wear particles in the hydraulic system pipelines of demolition equipment by integrating electrostatic sensors, inductive sensors, and optical sensors.

[0009] Technical solution:

[0010] Preferably, in step 1, the oil flow rate is v, and the electrostatic signal ES is collected by an electrostatic sensor, which is installed on the outer wall of the oil pipeline; the inductive signal LS is collected by an inductive sensor, which is arranged behind the electrostatic sensor in the direction of oil flow, and the position of the electrostatic sensor is L1; the image data IMG is collected by an optical sensor, which is arranged in the visible window area of ​​the flow channel, behind the inductive sensor in the direction of oil flow, and the position of the inductive sensor is L2.

[0011] Preferably, in step 1, the charged particle signal ES collected by the electrostatic sensor is a continuous pulse voltage signal sequence e(t), and its sequence data is ES p , p represents the number of data collected by the electrostatic sensor, and its sampling frequency f ES ≥100kHz; the inductive signal LS collected by the inductive sensor is a continuous pulse voltage signal sequence l(t), and its sequence data is LS m , m represents the number of data collected by the inductive sensor, and its sampling rate f LS ≥50kHz; the optical sensor collects sequential images as IMG n, n represents the number of images captured by the image sensor, the image data IMG acquisition frequency ≥ 10 frames / second, and the resolution ≥ 640×480.

[0012] Preferably, in step 2, the time alignment processing of the electrostatic, inductance and image signals is as follows: the delay time between the electrostatic sensor and the inductance sensor is calculated as t1 = L1 / v, and the delay time between the inductance sensor and the image sensor is calculated as t2 = L2 / v; for the electrostatic sensor, the number of signal data points collected in the time period t1 + t2 is calculated as: (t1 + t2) * f ES , the electrostatic sequence data after time alignment is the sequence ES after t1+t2 time p At this time, the number of data collected by the aligned electrostatic sensor, p, is processed as: p-(t1+t2)*f ES ; For the inductive sensor, the number of signal data points collected during the t2 period is calculated as: t2*f LS , the electrostatic sequence data after time alignment is the sequence LS after time t2 m At this time, the number of data collected by the aligned electrostatic sensor m is processed as: m-t2*f LS ; For image sensors, the sequence images after alignment are still IMG n , the number of images collected by the sensor is still n.

[0013] As a preference, in step 2, the electrostatic signal is processed as follows: using the ICEEMDAN method to decompose the aligned electrostatic signal ES p The intrinsic mode function (IMF) components are obtained, where the noise standard deviation is set to 0.25, the number of noise additions is 100, and the maximum number of iterations is 500. The correlation coefficients of each IMF component and the original electrostatic signal are then calculated. IMF components less than the threshold are considered noise, and IMF components greater than the threshold are considered valid signals. The threshold is set to 0.5. The effective IMF components after screening are subjected to wavelet threshold denoising, where the wavelet basis function is sym8 and the number of layers is 6. The denoised signals are superimposed and reconstructed to form ES(i). Next, the mean of the reconstructed signal is calculated as the eigenvalue of the electrostatic signal:

[0014]

[0015] Where ES(i) represents the signal value of the i-th data point of the reconstructed electrostatic signal, N is the total number of samples in a 0.1 second time interval, and a mean is calculated every 0.1 second.

[0016] As a preference, in step 2, the processing of the inductance signal is as follows: first, the time-aligned inductance signal LS is processed. mTo sort the data, set the time segment length to 0.1 seconds and calculate the number of sampling points in each time segment as 0.1*f LS , calculate the total number of signal sampling points m cannot be 0.1*f LS The remainder Rem when divided, the number of rows the signal is rearranged into is (m-Rem) / (0.1*f LS ), the number of columns is 0.1*f LS The matrix is ​​then used to calculate the mean of each row of the matrix signal. The mean value of each row of the matrix signal is subtracted from the mean value of the segment (to remove the DC component). The skewness of the processed signal is then calculated as the eigenvalue of the inductance signal:

[0017]

[0018] Where, LS i represents the signal value of the i-th data point of the inductor signal after DC removal processing. N is the total number of samples in a 0.1 second time interval. A skewness eigenvalue is calculated every 0.1 second.

[0019] Preferably, in step 2, the image data is processed as follows: first, the RGB color image is converted into a LAB color space image, and the L channel is bilaterally filtered, wherein the intensity smoothness parameter is set to 40 and the spatial domain filtering parameter is set to 4, and then the A and B channels are fused and converted into an RGB space image to achieve image denoising; the processed RGB image is converted into a grayscale image, and an adaptive threshold segmentation method is used to automatically generate a foreground mask, wherein the local change response degree is set to 0.5, and the area where all abrasive particles exist is obtained, a binary mask for each abrasive particle is generated, and the RGB image data of the area is extracted; then the Canny operator is used to extract the edge of the abrasive particle and calculate the area, aspect ratio, texture entropy, and circularity characteristics of the abrasive particle pixels.

[0020] Preferably, in step 3, the electrostatic and inductance characteristics are sequentially rearranged into a vector matrix VE, wherein the first row of the matrix is ​​the eigenvalue of the mean value of the electrostatic signal every 0.1s, and the second row is the eigenvalue of the skewness of the inductance signal every 0.1s, finally forming a 2-row N-column matrix.

[0021] Preferably, in step 4, the method for identifying the particle size, quantity, and material type of the abrasive particles is as follows: a sliding window process is performed on the eigenvalue matrix of 2 rows and N columns, wherein the sliding window length is set to 100, and the minimum and maximum values ​​of the 100 mean values ​​and 100 skewness values ​​in the window are automatically and sequentially divided into 4 intervals, and the number of data points falling in each interval is counted, requiring that the data of the middle two intervals account for ≥90%, and the abnormal window where the abrasive particles exist is detected by analyzing the mean and skewness distribution through histogram, wherein the mean and skewness of the background noise are distributed in the middle interval, and the abrasive particles will cause the mean value of the electrostatic signal and the skewness of the inductance signal to change. If the data of the middle two intervals account for ≥90%, it is considered that there are no abrasive particles and it is background noise; otherwise, it is considered that there are abrasive particles; the particle size is determined by: The method for determining the number of abrasive particles is as follows: the peak value of the electrostatic signal and the inductance signal skewness in the minimum proportion interval and the maximum proportion interval is determined, and the larger the value, the larger the abrasive particle size, which can be determined by the pixel area of ​​the abrasive particle image; the method for determining the number of abrasive particles is as follows: the peak value of the electrostatic signal and the inductance signal skewness in the minimum proportion interval and the maximum proportion interval is determined, which can be verified by the number detected by the abrasive particle image; the method for determining the material is as follows: the abrasive particles in the minimum proportion mean interval are determined as negatively charged non-metallic abrasive particles, the abrasive particles in the maximum proportion mean interval are determined as positively charged metal abrasive particles, the abrasive particles in the minimum proportion skewness interval are determined as non-ferromagnetic metal abrasive particles, and the abrasive particles in the maximum proportion skewness interval are determined as ferromagnetic metal abrasive particles, which can also be verified by the morphological characteristics of the abrasive particle image.

[0022] Preferably, in step 4, the abrasive particle image area, aspect ratio, texture entropy, and circularity features are input into the ShuffleNet network to identify cutting abrasive particles, fatigue abrasive particles, sliding abrasive particles, spherical abrasive particles, and normal abrasive particles.

[0023] Preferably, in step 5, according to the analysis results of step 4, a trend sequence S(t) of the material, quantity, size and wear type of the abrasive particles changing with time is constructed respectively, wherein the method for constructing the material, quantity and size sequence of the abrasive particles is as follows: after performing a sliding window processing analysis with a length of 100 on the eigenvalue matrix of 2 rows and N columns, the sliding step of the sliding window is set to 1, and after each analysis of the abrasive particle material, quantity and size in the previous sliding window, the matrix data in the next window is automatically analyzed to form a time series of the material, quantity and size of the abrasive particles; similarly, for the processing of the abrasive wear type, the number of each type of abrasive wear type analyzed by the ShuffleNet network is counted every 0.1s, and finally a wear type time series is formed; the combination is the trend sequence S(t) of the material, quantity, size and wear type of the abrasive particles changing with time,

[0024] And combined with the exponential weighted moving average method to predict the degradation trend curve WD(t):

[0025] WD(t)=β·WD(t-1)+(1-β)·S(t)

[0026] In the formula, β∈(0,1) is the smoothing coefficient.

[0027] The present invention also discloses an intelligent online monitoring system for wear particles of oil in hydraulic demolition equipment, which is characterized by comprising:

[0028] The sensor array module includes electrostatic sensors, inductive sensors, and optical sensors arranged in sequence along the oil flow direction. The installation spacing satisfies that the distance between the electrostatic sensor and the inlet is ≤3 times the pipe diameter, and the distance between adjacent sensors is ≥5 times the pipe diameter. The sensors respectively collect the voltage signal ES of the charged abrasive particles in the oil, the voltage signal LS of the ferromagnetic metal abrasive particles and the non-ferromagnetic metal abrasive particles, and the abrasive particle image data IMG.

[0029] A data processing module for implementing the above-mentioned intelligent online monitoring method for wear particles in oil of hydraulic demolition equipment;

[0030] The intelligent decision-making module forms the pollution level G according to the oil contamination classification standard based on the number and size distribution of abrasive particles. t And the predicted degradation trend curve WD(t) is used to calculate the health index HI.

[0031] HI=α1·G t +α2·WD(t)

[0032] The wear level is divided according to the HI value and the corresponding maintenance strategy is triggered.

[0033] Compared with the existing technology, the present invention provides an intelligent online monitoring method and system for wear particles in oil of hydraulic rescue equipment, which has the following beneficial effects:

[0034] (1) The present invention can simultaneously monitor the morphology of metal and non-metal abrasive particles, ferromagnetic metal and non-ferromagnetic metal abrasive particles, and wear particles in the pipeline;

[0035] (2) The present invention does not require the hydraulic rescue equipment to be shut down to extract representative oil samples for offline monitoring. Through real-time online monitoring, it can realize real-time online monitoring of oil wear particles under the full flow circulation condition of the hydraulic system of the rescue equipment, avoiding the lack of real-time performance and the resulting hysteresis error during offline monitoring, and does not affect the normal operation of the rescue equipment;

[0036] (3) The present invention can be installed in the oil circulation circuit of any demolition equipment with a hydraulic system, with high flexibility and good practicality;

[0037] (4) The multi-principle sensor fusion monitoring adopted by the present invention can realize qualitative and quantitative online monitoring of wear particles, and the monitoring parameters are comprehensive;

[0038] (5) The intelligent decision-making provided by the present invention can provide intelligent maintenance suggestions in real time, which can avoid serious damage to the hydraulic system caused by excessive wear and tear, while not causing any impact on the entire equipment system. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A schematic diagram of a system flow chart of an embodiment of the present invention;

[0040] Figure 2 A schematic diagram of a sensor installation module according to an embodiment of the present invention;

[0041] Figure 3 This is a flow chart of feature extraction of electrostatic signals in the present invention;

[0042] Figure 4 This is a flow chart of feature extraction of inductance signal in the present invention;

[0043] Figure 5 This is a flow chart of the method for identifying the number of abrasive particles and the type of material in the present invention;

[0044] Figure 6 (a) The wear particle waveform detected by the electrostatic monitoring sensor after time alignment in the monitoring experiment of the intelligent online monitoring system for wear particles in hydraulic demolition equipment oil;

[0045] Figure 6 (b) The wear particle waveform detected by the inductive monitoring sensor after time alignment in the monitoring experiment of the intelligent online monitoring system for wear particles in hydraulic demolition equipment oil;

[0046] Figure 6 (c) is the wear particle image detected by the optical monitoring sensor after time alignment in the monitoring experiment of the intelligent online monitoring system for oil wear particles of hydraulic demolition equipment. DETAILED DESCRIPTION

[0047] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0048] It will be understood by those skilled in the art that, unless expressly stated otherwise, the singular forms "a", "an", "said" and "the" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the description of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, "connected" or "coupled" as used herein may include wireless connections or couplings. The term "and / or" used herein includes any unit and all combinations of one or more associated listed items.

[0049] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and, unless defined as such, will not be interpreted in an idealized or overly formal sense.

[0050] Currently, there is no systematic multi-parameter online monitoring method for wear particles in the oil of hydraulic rescue equipment in my country. Online monitoring technologies that focus solely on monitoring wear particles in hydraulic system oil each have their own advantages and disadvantages. To address these shortcomings, the present invention provides an intelligent online monitoring method and system for wear particles in the oil of hydraulic rescue equipment. This method can collect all the properties of wear particles in the hydraulic system of hydraulic rescue equipment in real time, thereby enabling real-time online monitoring of the wear of hydraulic system components, including hydraulic pumps, hydraulic cylinders, valves, seals, etc., and providing wear warnings.

[0051] The technical solution of the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0052] Example 1

[0053] like Figures 1 to 6 As shown, this embodiment discloses an intelligent online monitoring method for wear particles of oil in hydraulic rescue equipment, including the following steps:

[0054] Step 1: Collect multi-source sensor data: including the electrostatic signal ES of charged abrasive particles in the oil collected by the electrostatic sensor, the inductive signal LS of ferromagnetic metal abrasive particles and non-ferromagnetic metal abrasive particles collected by the inductive sensor, and the abrasive particle image data IMG collected by the optical sensor, and perform time alignment and preprocessing on the three types of signals.

[0055] Among them, the electrostatic sensor is installed on the outer wall of the oil pipeline, the inductive sensor is located behind the electrostatic sensor in the direction of oil flow, and the position of the electrostatic sensor is L1. The optical sensor is located in the visible window area of ​​the flow channel, behind the inductive sensor in the direction of oil flow, and the position of the inductive sensor is L2. The flow direction of oil in the pipeline is as follows: Figure 2 As shown, the current flows through the electrostatic sensor 2, the inductive sensor 3 and the optical sensor 4 respectively.

[0056] Step 2, feature processing: Time-align the electrostatic, inductance and image signals, perform ICEEMDAN decomposition and mean feature extraction on the aligned electrostatic signal ES. The flowchart is as follows: Figure 3 As shown; the data and sliding window skewness feature extraction of the aligned inductance signal LS is performed, and the flow chart is as follows Figure 4 As shown; filtering segmentation and morphological feature extraction are performed on the aligned image data IMG.

[0057] The time alignment processing of the electrostatic, inductive and image signals is as follows: the delay time between the electrostatic sensor and the inductive sensor is calculated as t1 = L1 / v, and the delay time between the inductive sensor and the image sensor is calculated as t2 = L2 / v; for the electrostatic sensor, the number of signal data points collected in the time period t1 + t2 is calculated as: (t1 + t2) * f ES , the electrostatic sequence data after time alignment is the sequence ES after t1+t2 time p At this time, the number of data collected by the aligned electrostatic sensor, p, is processed as: p-(t1+t2)*f ES ; For the inductive sensor, the number of signal data points collected during the t2 period is calculated as: t2*f LS , the electrostatic sequence data after time alignment is the sequence LS after time t2 m At this time, the number of data collected by the aligned electrostatic sensor m is processed as: m-t2*f LS ; For image sensors, the sequence images after alignment are still IMG n , the number of images collected by the sensor is still n.

[0058] The charged particle signal ES collected by the electrostatic sensor is a continuous pulse voltage signal sequence with a sampling frequency of no less than 100kHz. The output voltage signal sequence data of the electrostatic sensor is ES p, p represents the number of data collected by the electrostatic sensor. The improved complete adaptive noise ensemble empirical mode decomposition (ICEEMDAN) method is used to decompose the signal to obtain the intrinsic mode function (IMF) components. The ICEEMDAN method is used to decompose the signal to obtain the intrinsic mode function (IMF) components, where the noise standard deviation is set to 0.25, the number of noise additions is 100, and the maximum number of iterations is 500. Then, the correlation coefficient between each IMF component and the original electrostatic signal is calculated. IMF components less than the threshold are considered noise, and IMF components greater than the threshold are considered valid signals. The threshold is set to 0.5. Wavelet threshold denoising is performed on the filtered valid IMF components, where the wavelet basis function is sym8 and the number of layers is 6. The denoised signals are superimposed and reconstructed to form ES(i). Next, the mean of the reconstructed signal is calculated as the eigenvalue of the electrostatic signal:

[0059]

[0060] In the formula, ES(i) represents the signal value of the i-th data point of the reconstructed electrostatic signal, N is the total number of samples in a 0.1 second time interval, and a mean is calculated every 0.1 second.

[0061] The inductive signal LS collected by the inductive sensor is a continuous pulse voltage signal sequence, and its sampling rate is not less than 50kHz.

[0062] The output voltage signal sequence data of the inductive sensor is LS m , m represents the number of data collected by the inductance sensor. First, the inductance signal LS after time alignment is m To sort the data, set the time segment length to 0.1 seconds and calculate the number of sampling points in each time segment as 0.1*f LS , calculate the total number of signal sampling points m cannot be 0.1*f LS The remainder Rem when divided, the number of rows the signal is rearranged into is (m-Rem) / (0.1*f LS ), the number of columns is 0.1*f LS The matrix is ​​then used to calculate the mean of each row of the matrix signal. The mean value of each row of the matrix signal is subtracted from the mean value of the segment (to remove the DC component). The skewness of the processed signal is then calculated as the eigenvalue of the inductance signal:

[0063]

[0064] In the formula, LS irepresents the signal value of the i-th data point of the inductor signal after DC removal processing. N is the total number of samples in a 0.1 second time interval. A skewness eigenvalue is calculated every 0.1 second.

[0065] The image data IMG acquisition frequency is not less than 10 frames / second, and the information resolution is not less than 640×480. The optical sensor collects sequential images as IMG n , n represents the number of images captured by the image sensor. First, the RGB color image is converted to the LAB color space image, and the L channel is bilaterally filtered, where the intensity smoothness parameter is set to 40 and the spatial domain filtering parameter is set to 4. Then, the A and B channels are fused and converted to the RGB space image to achieve image denoising. The processed RGB image is converted into a grayscale image, and the foreground mask is automatically generated using the adaptive threshold segmentation method, where the local change response degree is set to 0.5. The area where all wear particles exist is obtained, and a binary mask is generated for each wear particle. The RGB image data of the area is extracted. Then, the Canny operator is used to extract the wear particle edge and calculate the area, aspect ratio, texture entropy, and circularity characteristics of the wear particle pixels.

[0066] The electrostatic, inductive and optical sensors automatically process data every 1 minute of data collection. The wear particle waveforms detected by the electrostatic monitoring sensor after time alignment in the monitoring experiment of the intelligent online monitoring system for wear particles in hydraulic demolition equipment are as follows: Figure 6 As shown in (a), the wear particle waveform detected by the inductive monitoring sensor of the intelligent online monitoring system for wear particles of hydraulic demolition equipment oil after time alignment in the monitoring experiment is as follows Figure 6 As shown in (b), the wear particle image detected by the optical monitoring sensor after time alignment in the monitoring experiment of the intelligent online monitoring system for wear particles of hydraulic demolition equipment oil is as follows Figure 6 (c) shown.

[0067] Step 3: Multimodal feature fusion: electrostatic and inductance feature fusion generates a unified state vector matrix VE = [f ES , f LS ], where f ES is the mean characteristic sequence of electrostatic sensor signal, f LS is the characteristic sequence of the inductance sensor signal skewness; the first row of the matrix is ​​the eigenvalue of the mean value of the electrostatic signal every 0.1s, and the second row is the eigenvalue of the skewness of the inductance signal every 0.1s, finally forming a 2-row N-column matrix.

[0068] Step 4, perform quantitative and qualitative analysis of abrasive particles: Identify the particle size, quantity, and material type of abrasive particles based on the state vector VE. The flow chart is as follows: Figure 5 As shown, the typical wear pattern of the abrasive particles is identified based on the abrasive particle image.

[0069] The method for identifying the particle size, quantity, and material type of abrasive particles is as follows: a sliding window process is performed on the eigenvalue matrix of 2 rows and N columns, where the sliding window length is set to 100, and the minimum and maximum values ​​of the 100 mean values ​​and 100 skewness values ​​in the window are automatically and evenly divided into 4 intervals respectively, and the number of data points falling in each interval is counted, requiring that the data of the middle two intervals account for ≥90%. The abnormal window where abrasive particles exist is detected by histogram analysis of the mean and skewness distribution, where the mean and skewness of the background noise are distributed in the middle interval. Abrasive particles will cause the mean of the electrostatic signal and the skewness of the inductance signal to change. If the data of the middle two intervals account for ≥90%, it is considered that there are no abrasive particles and it is background noise. Otherwise, it is considered that abrasive particles exist. The particle size is determined as follows: according to the most The number of abrasive particles is determined by the peak value of the electrostatic signal mean and the inductance signal skewness in the small proportion interval and the maximum proportion interval. The larger the value, the larger the abrasive particle size, which can be determined by the pixel area of ​​the abrasive particle image; the number is determined by the number of peak values ​​of the electrostatic signal mean and the inductance signal skewness in the minimum proportion interval and the maximum proportion interval, which can be verified by the number detected by the abrasive particle image; the material is determined by the abrasive particles in the minimum proportion mean interval being determined as negatively charged non-metallic abrasive particles, the abrasive particles in the maximum proportion mean interval being determined as positively charged metal abrasive particles, the abrasive particles in the minimum proportion skewness interval being determined as non-ferromagnetic metal abrasive particles, and the abrasive particles in the maximum proportion skewness interval being determined as ferromagnetic metal abrasive particles, which can also be verified by the morphological characteristics of the abrasive particle image.

[0070] The method for identifying typical wear patterns of abrasive particles is as follows: the image area, aspect ratio, texture entropy, and circularity features of the abrasive particles are input into the classic ShuffleNet network to identify cutting abrasive particles, fatigue abrasive particles, sliding abrasive particles, spherical abrasive particles, and normal abrasive particles.

[0071] Step 5, perform wear analysis: construct the wear state time series S(t), where the construction method of the abrasive material, quantity, and size series is as follows: after performing a sliding window processing analysis with a length of 100 on the eigenvalue matrix of 2 rows and N columns, the sliding step of the sliding window is set to 1. After each analysis of the abrasive material, quantity, and size in the previous sliding window, the matrix data in the next window is automatically analyzed to form the abrasive material, quantity, and size time series; similarly, for the processing of abrasive wear types, the number of various wear types of abrasives analyzed by the ShuffleNet network is counted every 0.1s, and finally a wear type time series is formed; the combination is the trend sequence S(t) of the abrasive material, quantity, size, and wear type changing over time.

[0072] The degradation trend curve WD(t) is predicted by exponentially weighted moving average:

[0073] WD(t)=β·WD(t-1)+(1-β)·S(t)

[0074] In the formula, β∈(0,1) is a smoothing coefficient, which is 0.95 in this embodiment.

[0075] Example 2

[0076] This embodiment discloses an intelligent online monitoring system for wear particles of oil in hydraulic demolition equipment. Figure 1 As shown, it includes a sensor module, a data processing module and a wear intelligent decision module.

[0077] The sensor module is multi-source sensor hardware, including an electrostatic sensor 2 located on the outer wall of the oil pipeline, an inductive sensor 3 located at a key node in the oil flow path, and an optical sensor 4 located in the visible window area of ​​the flow channel. The installation spacing ensures that the distance between the electrostatic sensor and the inlet is ≤3 pipe diameters, and the distance between adjacent sensors is ≥5 pipe diameters. The electrostatic sensor 2, inductive sensor 3, and optical sensor 4, arranged sequentially along the oil flow direction, respectively collect the voltage signal ES of charged abrasive particles in the oil, the voltage signals LS of ferromagnetic and non-ferromagnetic metal abrasive particles, and the abrasive particle image data IMG.

[0078] The data processing module is used to implement the monitoring method in Example 1, collect data from multiple source sensors and perform wear analysis, and output a predicted degradation trend curve WD(t).

[0079] Wear intelligent decision module based on pollution level G t The equipment health index HI is predicted in conjunction with the predicted degradation trend curve WD(t). The contamination level is calculated based on the number of wear particles detected by the optical sensor according to the ISO4406 standard, and the intelligent operation and maintenance decision of the upper system of the equipment is given. The equipment health index HI is expressed as

[0080] HI=α1·G t +α2·WD(t)

[0081] In the formula, α1 and α2 are empirical weight coefficients set according to different hydraulic equipment working conditions.

[0082] Finally, the equipment health index HI is normalized and divided into three wear levels to trigger corresponding maintenance strategies: Level A (HI ≥ 0.75): indicates that the equipment is in good operating condition, and the upper system recommends continuing operation and observing the wear trend in real time; Level B (0.3 ≤ HI < 0.75): indicates that the equipment wear is at moderate risk, and the upper system recommends shortening the operation and maintenance cycle appropriately; Level C (HI < 0.3): indicates that the equipment is in a state of severe wear or critical failure, and the upper system should immediately issue a shutdown command and enter the emergency maintenance process.

[0083] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. An intelligent online monitoring method for wear particles in hydraulic rescue equipment oil, characterized in that: include: Step 1: Collect multi-source sensor data: including the electrostatic signal ES of charged abrasive particles in the oil, the inductance signal LS of ferromagnetic metal abrasive particles and non-ferromagnetic metal abrasive particles, and the abrasive particle image data IMG, and perform time alignment and preprocessing on the three types of signals; Step 2: Signal processing and feature extraction: Time-align the electrostatic, inductance, and image signals; The aligned electrostatic signal ES is decomposed by ICEEMDAN and subjected to wavelet threshold denoising, and the mean of the reconstructed signal is extracted as the eigenvalue of the electrostatic signal; Perform data reshaping and DC component removal processing on the aligned inductance signal LS, and extract the skewness of the processed signal as the characteristic value of the inductance signal; Perform filtering segmentation and morphological feature extraction on the aligned image data IMG; Step 3: Perform multimodal feature fusion: fuse the electrostatic and inductance features to generate a unified state vector matrix VE = [f ES , f LS ], where f ES is the mean characteristic sequence of electrostatic sensor signal, f LS is the characteristic sequence of inductive sensor signal skewness; Step 4: Perform quantitative and qualitative analysis of the wear particles: identify the particle size, quantity, and material type of the wear particles based on the state vector VE, and identify the typical wear pattern of the wear particles based on the wear particle image; Step 5: Perform wear analysis: construct the wear state time series S(t) and predict the degradation trend curve WD(t) by exponentially weighted sliding average.

2. The method according to claim 1, characterized in that In step 1, the oil flow rate is v, and the electrostatic signal ES is collected by the electrostatic sensor. ES is a continuous pulse voltage signal sequence e(t), and its sequence data is ES p , p represents the number of data collected by the electrostatic sensor, the electrostatic sensor is installed on the outer wall of the oil pipeline; the inductance signal LS is collected by the inductance sensor, LS is a continuous pulse voltage signal sequence l(t), and its sequence data is LS m , m represents the number of data collected by the inductive sensor, the inductive sensor is arranged behind the electrostatic sensor in the oil flow direction, and the position of the electrostatic sensor is L1; The image data IMG is collected by an optical sensor, which is arranged in a visible window area of ​​the flow channel, behind the inductive sensor in the oil flow direction, and at a position L2 with the inductive sensor.

3. The method according to claim 1, characterized in that In step 2, the time alignment of the electrostatic, inductance and image signals is as follows: the delay time between the electrostatic sensor and the inductance sensor is calculated as t1 = L1 / v, and the delay time between the inductance sensor and the image sensor is calculated as t2 = L2 / v; for the electrostatic sensor, the number of signal data points collected in the time period t1 + t2 is calculated as: (t1 + t2) * f ES , the electrostatic sequence data after time alignment is the sequence ES after t1+t2 time p At this time, the number of data collected by the aligned electrostatic sensor, p, is processed as: p-(t1+t2)*f ES ; For the inductive sensor, the number of signal data points collected during the t2 period is calculated as: t2*f LS , the electrostatic sequence data after time alignment is the sequence LS after time t2 m At this time, the number of data collected by the aligned electrostatic sensor m is processed as: m-t2*f LS ; For image sensors, the sequence images after alignment are still IMG n , the number of images collected by the sensor is still n.

4. The method according to any one of claims 1 to 3, characterized in that In step 2, the electrostatic signal is processed as follows: the aligned electrostatic signal ES is decomposed using the ICEEMDAN method p The intrinsic mode function (IMF) components are obtained, and then the correlation coefficients of each IMF component and the original electrostatic signal are calculated. The IMF components smaller than the threshold are considered noise, and the IMF components larger than the threshold are considered valid signals. The effective IMF components after screening are subjected to wavelet threshold denoising, where the wavelet basis function is sym8. The denoised signals are superimposed and reconstructed to form ES(i). Next, the mean of the reconstructed signal is calculated as the eigenvalue of the electrostatic signal: Where ES(i) represents the signal value of the i-th data point of the reconstructed electrostatic signal, and N is the total number of samples in the time interval.

5. The method according to any one of claims 1 to 3, characterized in that In step 2, the inductance signal is processed as follows: first, the time-aligned inductance signal LS is processed. m To sort the data, set the time segment length to t seconds, and calculate the number of sampling points in each time segment as t*f LS , calculate the total number of signal sampling points m cannot be t*f LS The remainder Rem when divided, the number of rows the signal is rearranged into is (m-Rem) / (t*f LS ), the number of columns is t*f LS , then calculate the mean of each row of the matrix signal, subtract the mean value of each row of the signal from the segment, and then calculate the skewness of the processed signal as the eigenvalue of the inductance signal: Where, LS i represents the signal value of the i-th data point of the inductor signal after DC removal processing, and N is the total number of samples in the time interval.

6. The method according to any one of claims 1 to 3, characterized in that In step 2, the image data is processed as follows: first, the RGB color image is converted to the LAB color space image, the L channel is bilaterally filtered, and then the A and B channels are fused and converted to the RGB space image to achieve image denoising; The processed RGB image is converted into a grayscale image. The adaptive threshold segmentation method is used to automatically generate a foreground mask to obtain the area where all wear particles exist. A binary mask is generated for each wear particle, and the RGB image data of the area is extracted. The Canny operator is then used to extract the wear particle edge and calculate the area, aspect ratio, texture entropy, and circularity characteristics of the wear particle pixels.

7. The method according to claim 1, characterized in that In step 4, the method for identifying the particle size, quantity, and material type of the abrasive particles is as follows: a. Perform sliding window processing on the eigenvalue matrix, automatically and sequentially divide the minimum and maximum values ​​of several means and several skewness values ​​within the window into four intervals, and count the number of data points falling in each interval; b. Analyze the mean and skewness distribution through histogram analysis to detect abnormal windows where wear particles exist. The mean and skewness of the background noise are distributed in the middle range. If the data in the two middle ranges accounts for ≥ 90%, it is considered that there are no wear particles and the background noise is present. Otherwise, wear particles are considered to be present. c. Determine the particle size based on the mean value of the electrostatic signal and the peak value of the inductance signal skewness in the minimum and maximum proportion intervals. The larger the value, the larger the abrasive particle size. Determine the number of abrasive particles based on the mean value of the electrostatic signal and the peak value of the inductance signal skewness in the minimum and maximum proportion intervals. d. The material is determined as follows: the abrasive particles in the minimum proportion mean interval are determined as negatively charged non-metallic abrasive particles, the abrasive particles in the maximum proportion mean interval are determined as positively charged metal abrasive particles, the abrasive particles in the minimum proportion skewness interval are determined as non-ferromagnetic metal abrasive particles, and the abrasive particles in the maximum proportion skewness interval are determined as ferromagnetic metal abrasive particles.

8. The method according to claim 1 or 7, characterized in that In step 4, the abrasive particle image area, aspect ratio, texture entropy, and circularity features are input into the ShuffleNet network to identify cutting abrasive particles, fatigue abrasive particles, sliding abrasive particles, spherical abrasive particles, and normal abrasive particles.

9. The method according to claim 8, characterized in that In step 5, based on the analysis results of step 4, the trend sequence S(t) of the material, quantity, size and wear type of the abrasive particles over time is constructed respectively. The construction method of the material, quantity and size sequence of the abrasive particles is as follows: after performing sliding window processing analysis on the eigenvalue matrix, the sliding window sliding step is set. After each analysis of the material, quantity and size of the abrasive particles in the previous sliding window, the matrix data in the next window is automatically analyzed to form the material, quantity and size time series of the abrasive particles; similarly, for the processing of the abrasive wear type, the number of various wear types of the abrasive particles analyzed by the ShuffleNet network is counted once at a certain time interval, and finally the wear type time series is formed; the combination is the trend sequence S(t) of the material, quantity, size and wear type of the abrasive particles over time, and the degradation trend curve WD(t) is predicted by combining the exponentially weighted sliding average method: WD(t)=β·WD(t-1)+(1-β)·S(t) In the formula, β∈(0,1) is the smoothing coefficient.

10. An intelligent online monitoring system for wear particles of hydraulic rescue equipment oil, characterized by: include: The sensor array module includes electrostatic sensors, inductive sensors, and optical sensors arranged in sequence along the oil flow direction. The installation spacing satisfies that the distance between the electrostatic sensor and the inlet is ≤3 times the pipe diameter, and the distance between adjacent sensors is ≥5 times the pipe diameter. The sensors respectively collect the voltage signal ES of the charged abrasive particles in the oil, the voltage signal LS of the ferromagnetic metal abrasive particles and the non-ferromagnetic metal abrasive particles, and the abrasive particle image data IMG. A data processing module, configured to implement the method according to any one of claims 1 to 9; The intelligent decision-making module forms the pollution level G according to the oil contamination classification standard based on the number and size distribution of abrasive particles. t And the predicted degradation trend curve WD(t) is used to calculate the health index HI. HI=α1·G t +α2·WD(t) The wear level is divided according to the HI value and the corresponding maintenance strategy is triggered.

Images