A fault early warning method and system for an industrial equipment lubricating oil system

Abstract

The application discloses a kind of industrial equipment lubricating oil circulating system's fault early warning method: four different polarization states of polarized light are sequentially incident to lubricating oil, corresponding polarization holographic image is obtained by collection, and the phase difference corresponding to four different polarization states is obtained to obtain abrasive material quality;After three-dimensional reconstruction is carried out to polarization holographic image, feature extraction is carried out to obtain abrasive size and abrasive quantity;Abrasive material quality, abrasive size and abrasive quantity are carried out feature fusion and dimension reduction processing to obtain standardization feature vector;Standardization feature vector is classified and inferred, and the type of fault and corresponding probability distribution are output, and fault classification early warning is carried out;According to standardization feature vector, the current lubricating oil state is judged and abrasive classification early warning is carried out.The application also discloses a kind of industrial equipment lubricating oil circulating system's fault early warning system.The application realizes the direct output and optimization from image data to fault early warning by collecting the multi-polarization holographic image of lubricating oil, improves detection precision and response speed.

Description

Technical Field

[0001] This invention relates to the field of industrial equipment condition monitoring technology, specifically to a fault early warning method and system for industrial equipment lubrication oil systems. Background Technology

[0002] In modern industry, the lubrication system of critical equipment such as generator sets, heavy-duty CNC machine tools, and marine power systems is one of the core components ensuring stable equipment operation. During equipment operation, lubricating oil carries abrasive particles shed from friction pairs. The size, quantity, and material of these particles directly reflect the location and degree of wear. For example, steel abrasive particles correspond to metal wear in gears and bearings, while copper-based abrasive particles are associated with bearing failure. A surge in abrasive particle size often indicates that wear is entering a deterioration stage. Therefore, accurate and real-time monitoring of multi-dimensional parameters of lubricating oil abrasive particles is a key technology for achieving early warning of equipment failures and supporting predictive maintenance.

[0003] Traditional lubricating oil abrasive detection methods are based on standards such as GB / T 18854 and are mainly offline laboratory tests. They have the following inherent defects: (1) Offline detection lag: It is necessary to stop the machine to take samples and analyze them through equipment such as ferrometers and particle size analyzers. The detection cycle is as long as several hours to several days. It is impossible to capture the dynamic changes of oil abrasive particles in real time and it is difficult to give timely warning of sudden wear failures; (2) Missing parameter dimensions: Ferrography can only observe the abrasive particle morphology, laser particle size counting can only obtain size and quantity, and spectroscopy is not sensitive enough for micro abrasive particles and has low material differentiation. They cannot accurately identify the material properties of abrasive particles, resulting in the inability to correspond to specific wear parts and poor fault location accuracy; (3) Limitations of traditional holographic detection: Existing online holographic abrasive detection technology mostly uses single polarization state laser imaging. Since different materials of abrasive particles have specific differences in scattering and phase response to different polarized light, single polarized light can only restore the abrasive particle morphology and cannot invert its material through polarization characteristics. The abrasive particle parameter measurement dimensions are incomplete.

[0004] Current online monitoring technologies for lubricating oil in mechanical equipment mostly focus on the independent monitoring of single parameters. For example, Chinese patent CN117233040A discloses an online lubricating oil quality monitoring system that only integrates viscosity measurement and abrasive particle image acquisition functions, and can only obtain information on the size and quantity of abrasive particles, but cannot identify the abrasive particle material. Chinese patent CN116953038A, on the other hand, is based on the correlation between dielectric constant and oil deterioration, and does not involve multi-dimensional detection of abrasive particle parameters. At the same time, in existing technologies, the "abrasive particle detection-parameter analysis-fault early warning" are mostly decentralized modules, and the detection data needs to be manually transferred to the analysis system, resulting in low automation and the inability to achieve integrated operation of "real-time detection-instant analysis-automatic early warning".

[0005] Currently, there are no reports on an integrated system that utilizes multi-polarization holographic technology to simultaneously detect multiple parameters of lubricating oil abrasive particles, including size, quantity, and material, and integrates machine learning for fault mapping and early warning. Therefore, developing an online monitoring and fault warning system for multiple parameters of lubricating oil abrasive particles based on polarization holography has significant industrial application value and technological innovation. Summary of the Invention

[0006] The purpose of this invention is to provide a fault early warning method and system for industrial equipment lubricating oil circulation systems. By acquiring multi-polarization holographic images of lubricating oil, the invention enables direct output and optimization from image data to fault early warning, thereby improving detection accuracy and response speed.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A fault early warning method for an industrial equipment lubricating oil circulation system includes the following steps: S1. The laser emits a laser beam, which is then passed through a polarization modulator to obtain four different polarization states of polarized light. These polarized lights are then sequentially incident into the lubricating oil sample cell, and the corresponding polarization holographic images are acquired. The abrasive material is obtained based on the phase difference corresponding to the four different polarization states. S2. Perform three-dimensional reconstruction on the polarization holographic image and then extract features to obtain the abrasive grain size and number; S3. Perform feature fusion and dimensionality reduction on the abrasive material, abrasive size and abrasive number to obtain a standardized feature vector; S4. Perform classification reasoning on the standardized feature vectors, output the fault types and corresponding probability distributions of the industrial equipment lubricating oil circulation system, and perform fault classification and early warning. S5. Determine the current lubricating oil condition based on the standardized feature vector and perform abrasive particle classification early warning.

[0008] In S1, a material-polarization feature mapping table is first established based on the phase difference input corresponding to the standard material abrasive sample in four different polarization states and the abrasive material as the output. Then, the abrasive material is extracted based on the phase difference corresponding to the four different polarization states and the material-polarization feature mapping table.

[0009] In S2, the equivalent diameter of the abrasive grains is calculated by analyzing the connected components of the holographically reconstructed polarization holographic image; the number of abrasive grains is calculated by counting the number of connected components per unit volume of the holographically reconstructed polarization holographic image.

[0010] In S3, the method for fusing abrasive grain size, abrasive grain number, and abrasive grain material characteristics is as follows: Extract the mean, variance, and maximum value of abrasive parameters over a certain continuous time period; The classification results of abrasive materials are encoded as 4-dimensional unique heat vectors; Principal component analysis (PCA) is used to compress high-dimensional features into 6-dimensional feature vectors: ,in Average abrasive grain size For the number of abrasive grains, The principal component of the material encoding vector. For the maximum abrasive grain size, These represent the variances of size and quantity, respectively.

[0011] In S4, the fault types include gear wear and bearing failure, and the fault classification warning includes: Level 1 warning: When the probability of gear wear or bearing failure is ≥80%, an audible and visual alarm is triggered, and a "stop and check" command is sent. Level 2 warning: When the probability of gear wear or bearing failure is less than 80% (50% ≤ probability), a "strengthen monitoring" command is sent. Normal condition: When the probability of gear wear or bearing failure is less than 50%, continuous monitoring is performed.

[0012] In S5, the methods for determining whether the current lubricating oil condition is normal or abnormal include: By integrating abrasive material, abrasive size, and abrasive number, a predicted value for the overall condition of the lubricating oil is obtained; If the predicted value is greater than the set threshold, it is marked as abnormal; If the predicted value is less than or equal to the set threshold, it is marked as normal. After determining whether the current state of the lubricating oil is normal or abnormal, the next decision is generated: The current lubricating oil condition is normal; continue to monitor the lubricating oil condition. The current lubricating oil condition is abnormal, and an alarm is being issued.

[0013] In addition, a separate threshold is set for each parameter, and any parameter exceeding the threshold is also marked as abnormal; In S5, the abrasive particle classification early warning includes: The number of abrasive grains within a preset volume is counted, the abrasive grain concentration is calculated based on the number of abrasive grains, and the proportion of iron-based abrasive grains is calculated based on the abrasive grain material. If the abrasive concentration is ≥10 and the proportion of iron-based abrasive particles is >80%, a Level 1 alarm for "accelerated equipment wear" will be triggered. If the average size of the abrasive particles is greater than 50 μm and the morphology is cutting-like, a level 2 alarm for "component fatigue" will be triggered.

[0014] The present invention also provides a fault early warning system for an industrial equipment lubricating oil circulation system employing the above method, comprising: The polarization holographic module uses a laser to emit a laser beam, which is then passed through a polarization modulator to obtain four different polarization states of polarized light. These polarized lights are sequentially incident on a lubricating oil sample cell, and the corresponding polarization holographic images are acquired and reconstructed in three dimensions. Additionally, the abrasive material is extracted based on the phase difference and material-polarization feature mapping table corresponding to the four different polarization states. The image feature extraction module uses a convolutional neural network (CNN) to extract features from the reconstructed polarization holographic image to obtain the abrasive grain size and number. The multi-source feature fusion module performs feature fusion and dimensionality reduction on abrasive material, abrasive size and abrasive number to obtain a standardized feature vector; The fault classification and probability output module stores fault mapping and early warning models, which are used to classify and reason about standardized feature vectors and output the fault types and corresponding probability distributions of the lubricating oil circulation system of industrial equipment. The intelligent early warning module stores abrasive warning models, determines the current lubricating oil status based on standardized feature vectors, and performs graded early warnings for abrasive particles.

[0015] The fault mapping and early warning model uses a random forest (RF) network, and the abrasive particle early warning model uses an XGBoost network.

[0016] Using historical polarization holographic images as the training set, the abrasive particle size and number are extracted by the abrasive particle material extraction and image feature extraction modules. Then, the standardized feature vectors obtained by the multi-source feature fusion module through feature fusion and dimensionality reduction are used as input. The fault mapping and early warning models and the abrasive particle early warning model are trained with fault type and abrasive particle grade as labels, respectively.

[0017] During training, backpropagation is performed using the output values ​​to optimize the output results and parameter weights, continuously approximating the measured data for pre-training.

[0018] The polarization holographic module includes: a laser that emits a light source; a beam splitter that splits the beam into an object beam and a reference beam; a modulator that causes the object beam and the reference beam to be successively polarized over time into horizontally linearly polarized, vertically linearly polarized, left-handed circularly polarized, and right-handed circularly polarized light; and a high-speed camera that receives the object beam and the reference beam and sequentially obtains holographic images in four polarization directions.

[0019] Compared with the prior art, the beneficial effects of the present invention are: (1) This invention can utilize polarization holography to achieve simultaneous detection of multiple parameters, including the number, size, morphology, and material of lubricating oil abrasive particles; (2) The present invention also integrates a machine learning fault mapping and early warning system, which realizes real-time monitoring of lubricating oil status, automatic decision generation, and reduces equipment operation and maintenance costs; (3) This invention abandons the indirect process of "manual feature extraction - step-by-step modeling and judgment" in traditional abrasive particle detection and fault diagnosis. It directly establishes the end-to-end mapping relationship between multi-polarization holographic images and abrasive particle concentration, size, morphology, material parameters and lubricating oil fault status through machine learning models. It eliminates manual feature design and intermediate redundant links, realizes direct output and optimization from image data to fault warning, and improves detection accuracy and response speed. Attached Figure Description

[0020] Figure 1 A flowchart of a method for online monitoring and fault early warning of lubricating oil abrasive parameters provided by the present invention; Figure 2 A structural diagram of an online monitoring and fault early warning system for lubricating oil abrasive parameters provided by the present invention; Among them, 1. Fault classification and probability output module, 2. Polarization holographic acquisition module, 3. Holographic reconstruction module, 4. Polarization holographic module, 5. Threshold body, 6. Circulating oil tank, 7. Micron-level metal abrasive particles, 8. Submicron-level non-metallic abrasive particles, 9. Submicron-level metal abrasive particles, 10. Sealed protective shell, 11. Stabilized power supply unit. Figure 3 The monitoring results diagram provided by this invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0023] like Figure 1 As shown, the fault early warning method of the present invention, based on polarization holographic wear particle imaging and direct mapping optimization of machine learning, includes the following steps: S1. System Calibration Phase: Establish a "polarization state-abrasive material" response database using standard abrasive samples to provide a benchmark for subsequent material identification. S11. Select abrasive samples of standard materials (including four typical wear materials: steel, copper, rubber and aluminum) and prepare them into simulated lubricating oil with a known concentration (50 particles / mL). S12. Control the fast modulator to output four polarization states sequentially (horizontal linear polarization, vertical linear polarization, left-hand circular polarization, and right-hand circular polarization), and record the grayscale characteristics (mean and phase difference) of the interference image of the standard abrasive grain under each polarization state. S13. Establish a material-polarization feature mapping table: Using the phase difference corresponding to the four polarization states as input and the abrasive material as output, obtain the discriminant by least squares fitting: ; in, Let be the phase difference between the object light and the reference light in the i-th polarization state. is the weighting coefficient, and b is the bias term, determined by the standard sample calibration experiment.

[0024] S2, Polarization Holographic Image Acquisition Stage: S21. The laser output from the laser (a 532nm solid-state laser with a power of 10mW) is split into object light and reference light by a beam splitter, with a splitting ratio of 1:1. S22. A fast modulator (using a liquid crystal polarization modulator with a response time ≤50μs) sequentially applies four polarization states to the object light, with the switching frequency set to 100Hz. S23. After loading the polarized state, the object light is irradiated by the lubricating oil flow cell (containing abrasive-containing lubricating oil, with a flow rate of 0.5 m / s, and the cell body is made of quartz glass with a transmittance of ≥98%). After being scattered by the abrasive particles, the object light interferes with the reference light at the photosensitive surface of the high-speed camera (frame rate ≥400fps, resolution 1280×960). S24. The high-speed camera synchronously acquires interference images (polarization holographic images) corresponding to four polarization states, and acquires 100 sets of image data (each set contains 4 interference images of different polarization states) every 1 second, and transmits them to the data buffer unit.

[0025] S3. Polarization holographic image reconstruction stage: The acquired interference image is processed to reconstruct the multi-dimensional parameters of the abrasive particles. S31. Interference image preprocessing: Gaussian filtering is performed on each group of 4 images ( 1.5), eliminate background noise; S32. Holographic Reconstruction: Using the Fresnel holographic reconstruction algorithm, each interferogram is numerically reconstructed to obtain the amplitude and phase distribution of the abrasive particles. in, The grayscale values ​​of the interference image. Where is the laser wavelength and z is the reconstruction distance. Indicates Fourier transform, Wave number; S33, Parameter Extraction: Size parameters: The equivalent diameter of the abrasive grain is calculated by analyzing the connected components of the reconstructed image (taking the diameter of the circumcircle of the connected component). Quantitative parameter: The number of connected components of abrasive particles per unit volume; Material parameters: Phase difference under four polarization states Substitute the discriminant into S13 to output the abrasive material type.

[0026] S4. Abrasive Parameter Feature Extraction Stage: Feature fusion is performed on the reconstructed abrasive parameters to generate input features for fault mapping. S41. Statistical feature extraction: Extract the mean, variance, and maximum value of abrasive parameters (size, quantity, material) within a sliding window for 10 consecutive seconds. S42. Material feature encoding: Encode the abrasive material classification result into a 4-dimensional unique heat vector (e.g., steel corresponds to [1,0,0,0], copper corresponds to [0,1,0,0]). S43. Feature Dimensionality Reduction: Principal Component Analysis (PCA) is used to compress high-dimensional features into 6-dimensional feature vectors. ,in Average abrasive grain size For the number of abrasive grains, The principal component of the material encoding vector. For the maximum abrasive grain size, These represent the variances of size and quantity, respectively.

[0027] S5. Perform classification reasoning on the 6-dimensional feature vector obtained in S43, output the fault types and corresponding probability distributions of the industrial equipment lubricating oil circulation system, and perform fault classification and early warning: Fault types include gear wear and bearing failure; fault classification warnings include: Level 1 warning: When the probability of gear wear or bearing failure is ≥80%, an audible and visual alarm is triggered, and a "stop and check" command is sent. Level 2 warning: When the probability of gear wear or bearing failure is less than 80% (50% ≤ probability), a "strengthen monitoring" command is sent. Normal condition: When the probability of gear wear or bearing failure is less than 50%, continuous monitoring is performed.

[0028] S6. Determine the current lubricating oil condition based on the standardized feature vector and perform abrasive particle classification warning: S6-1. Integrate abrasive material, abrasive size, and abrasive number to obtain a predicted value of the overall condition of the lubricating oil; If the predicted value is greater than the set threshold, it is marked as abnormal; If the predicted value is less than or equal to the set threshold, it is marked as normal. After determining whether the current state of the lubricating oil is normal or abnormal, the next decision is generated: The current lubricating oil condition is normal; continue to monitor the lubricating oil condition. The current lubricating oil condition is abnormal, and an alarm is issued. S6-2, Abrasive Grading Early Warning includes: The number of abrasive grains within a preset volume is counted, the abrasive grain concentration is calculated based on the number of abrasive grains, and the proportion of iron-based abrasive grains is calculated based on the abrasive grain material. If the abrasive concentration is ≥10 and the proportion of iron-based abrasive particles is >80%, a Level 1 alarm for "accelerated equipment wear" will be triggered. If the average size of the abrasive particles is greater than 50 μm and the morphology is cutting-like, a level 2 alarm for "component fatigue" will be triggered.

[0029] Example 2 like Figure 2 As shown, the lubricating oil abrasive multi-parameter measurement and fault early warning system based on polarization holography provided in this embodiment includes: polarization holography module 4, image feature extraction module, multi-source feature fusion module and intelligent early warning module 1.

[0030] In this embodiment, the abrasive particles generated in the industrial equipment lubricating oil circulation system are classified into submicron-sized metallic abrasive particles 9, submicron-sized non-metallic abrasive particles 8, and micron-sized metallic abrasive particles 7 based on their size, morphology, and material. The oil is stored in the circulating oil 6. During system operation, the oil is sent to the polarization holographic module 4 via the threshold body 5 for real-time monitoring. To reduce external interference such as ambient light and vibration, and to ensure the accuracy of abrasive particle measurement and the reliability of early warning, the aforementioned core module and conveying device are integrated and installed in a sealed protective housing 10. The overall power supply for the system is provided by the voltage stabilizing power supply unit 11.

[0031] In this embodiment, the polarization holography module 4 includes a polarization holography acquisition module 2 and a holography reconstruction module 3. Through polarization state switching and digital holography reconstruction technology, it simultaneously monitors five core parameters of abrasive particles: size, morphology, material, concentration, and motion trajectory. In this embodiment, the polarization holography acquisition module 2 uses a high-speed liquid crystal polarizer integrated into an integrated optical valve block to reduce optical path offset and electromagnetic interference. It is also equipped with a gigabit Ethernet interface to transmit holographic image data to the holography reconstruction module 3 in real time via the UDP protocol.

[0032] In this embodiment, the laser in the polarization holographic acquisition module 2 emits a laser beam, which is then passed through a polarization modulator to obtain four different polarization states of polarized light. These beams are sequentially incident on the lubricating oil sample cell, and the corresponding polarization holographic images are acquired and reconstructed by the holographic reconstruction module 3. The abrasive material is extracted based on the phase difference and material-polarization feature mapping table corresponding to the four different polarization states.

[0033] In this embodiment, the image feature extraction module uses a convolutional neural network (CNN) to extract features from the reconstructed polarization holographic image to obtain the abrasive grain size and number. The CNN has a convolutional kernel size of 3×3, a pooling layer of 2×2, a learning rate of 0.001, and a batch size of 32.

[0034] In this embodiment, the multi-source feature fusion module performs feature fusion and dimensionality reduction on the abrasive material, abrasive size, and abrasive number to obtain a standardized feature vector; The intelligent early warning module 1 stores: a trained fault mapping and early warning model, which uses random forest (RF) for fault classification; and a trained abrasive fault classification prediction model (abrasive early warning model), which uses XGBoost, with a maximum depth of 6 layers, a learning rate of 0.08, and a column sampling ratio of 0.7.

[0035] The intelligent early warning module 1 also includes a real-time abrasive parameter dashboard, dynamically displaying abrasive size distribution, concentration change trends, and a heat map of fault risk levels, using abrasive parameters collected by the polarization holographic module 4 as input. Output the fault types and corresponding probability distributions of the lubricating oil circulation system of industrial equipment, and provide fault classification and early warning: Fault types include gear wear and bearing failure, and fault classification and early warning include: Level 1 warning: When the probability of gear wear or bearing failure is ≥80%, trigger an audible and visual alarm and push a "stop and check" command; Level 2 warning: When the probability of gear wear or bearing failure is 50% ≤ <80%, push a "strengthen monitoring" command; Normal state: When the probability of both gear wear and bearing failure is <50%, continuous monitoring is performed.

[0036] Output the equipment fault level corresponding to the abrasive particles; the graded early warning logic is as follows: if the abrasive particle concentration is ≥10 (corresponding to GB / T14039-2002 standard) and the proportion of iron-based abrasive particles is >80%, a first-level alarm of "accelerated equipment wear" is triggered; if the average size of the abrasive particles is >50μm and the morphology is cutting, a second-level alarm of "component fatigue" is triggered. The early warning threshold is set according to the ISO 4406-2021 lubricating oil cleanliness level standard, which meets the operation and maintenance needs of heavy industrial equipment; operation and maintenance decision push, generating targeted maintenance instructions (such as shutdown for maintenance, lubricating oil replacement, filter cleaning) according to the early warning level.

[0037] In this embodiment, taking mineral lubricating oil in a heavy machinery gearbox as an example, and taking parameters monitored at a certain moment during 72 hours of continuous operation as an example, the system uses a polarization holographic acquisition module and an image feature extraction module to collect parameters such as abrasive particle concentration (level 9), average size (48μm), material (Fe accounts for 85%), morphology (mainly cutting), and motion trajectory offset (+1.2μm) in the lubricating oil in real time. The system then generates standardized feature vectors through a multi-source feature fusion module and inputs them into the intelligent early warning module.

[0038] The intelligent early warning module includes two sub-models: a fault mapping and early warning model, and a wear particle early warning model. Taking the fault mapping and early warning model as an example, after receiving the standardized feature vector, it outputs the probability distribution of gear wear and bearing failure categories. Combined with the preset hierarchical logic, in this embodiment, the probability of gear wear is 72% and the probability of bearing failure is 21%, which corresponds to triggering a level-two early warning and pushing an "enhanced monitoring" command.

[0039] Taking the abrasive wear warning model as an example, XGBoost performs weighted calculations on multi-dimensional features, outputting a fault level of "Level 2 Alarm (Component Fatigue)" and predicting a remaining safe operating time of 168 hours. Simultaneous disassembly and inspection results showed that micro-fatigue cracks had appeared on the gearbox meshing surface, and the actual measured remaining safe operating time was 162 hours, with an error of only 6 hours and a relative error of 3.7%, meeting the accuracy requirements for industrial equipment operation and maintenance monitoring. Based on the prediction results, the system triggers the "Level 2 Alarm" flag and pushes an operation and maintenance instruction to "stop the machine within 48 hours to inspect the gear meshing surface."

[0040] The continuous 72-hour prediction results of the example are as follows: Figure 3 As shown, it should be noted that to make the difference between the predicted and actual abrasive particle concentration values ​​more intuitive, the y-axis uses a logarithmic scale, while the x-axis (monitoring duration) maintains a linear scale. Figure 3 It can be seen that with the increase of monitoring time, the abrasive particle concentration increases exponentially, the average size increases from the initial 12μm to 48μm, and the proportion of iron-based abrasive particles increases from 60% to 85%. This leads to the conclusion that the gearbox wear is continuously aggravated and fatigue failure is imminent. This is consistent with the actual physical law that during lubrication oil operation, the wear of the meshing surfaces of the equipment gradually worsens, and the amount and size of abrasive particles increase synchronously. Furthermore, it can be seen that the average relative error between the predicted values ​​of multiple abrasive parameters and the actual offline detection values ​​is ≤4%, meeting the inspection requirements of the ISO 18436-4 standard for oil analysis of mechanical equipment. This indicates that the system established in this invention can achieve real-time measurement of multiple abrasive parameters in lubricating oil and accurate early warning of equipment failure.

[0041] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A fault early warning method for an industrial equipment lubricating oil circulation system, characterized in that, Includes the following steps: S1. The laser emits a laser beam, which is then passed through a polarization modulator to obtain four different polarization states of polarized light. These polarized lights are then sequentially incident into the lubricating oil sample cell, and the corresponding polarization holographic images are acquired. The abrasive material is obtained based on the phase difference corresponding to the four different polarization states. S2. Perform three-dimensional reconstruction on the polarization holographic image and then extract features to obtain the abrasive grain size and number; S3. Perform feature fusion and dimensionality reduction on the abrasive material, abrasive size and abrasive number to obtain a standardized feature vector; S4. Perform classification reasoning on the standardized feature vectors, output the fault types and corresponding probability distributions of the industrial equipment lubricating oil circulation system, and perform fault classification and early warning. S5. Determine the current lubricating oil condition based on the standardized feature vector and perform abrasive particle classification early warning.

2. The method according to claim 1, characterized in that, In S1, a material-polarization feature mapping table is first established based on the phase difference input corresponding to the standard material abrasive sample in four different polarization states and the abrasive material as the output. Then, the abrasive material is extracted based on the phase difference corresponding to the four different polarization states and the material-polarization feature mapping table.

3. The method according to claim 1, characterized in that, In S2, the equivalent diameter of the abrasive grains is calculated by analyzing the connected components of the holographically reconstructed polarization holographic image; the number of abrasive grains is calculated by counting the number of connected components per unit volume of the holographically reconstructed polarization holographic image.

4. The method according to claim 1, characterized in that, In S3, the method for fusing abrasive grain size, abrasive grain number, and abrasive grain material characteristics is as follows: Extract the mean, variance, and maximum value of abrasive parameters over a certain continuous time period; The classification results of abrasive materials are encoded as 4-dimensional unique heat vectors; Principal component analysis (PCA) is used to compress high-dimensional features into 6-dimensional feature vectors: ,in Average abrasive grain size For the number of abrasive grains, The principal component of the material encoding vector. For the maximum abrasive grain size, These represent the variances of size and quantity, respectively.

5. The method according to claim 1, characterized in that, In S4, the fault types include gear wear and bearing failure. The fault classification warning includes: Level 1 warning: when the probability of gear wear or bearing failure is ≥80%, an audible and visual alarm is triggered, and a "stop and check" command is pushed. Level 2 warning: When the probability of gear wear or bearing failure is less than 80% (50% ≤ probability), a "strengthen monitoring" command is sent. Normal condition: When the probability of gear wear or bearing failure is less than 50%, continuous monitoring is performed.

6. The method according to claim 1, characterized in that, In S5, the methods for determining whether the current lubricating oil condition is normal or abnormal include: By integrating abrasive material, abrasive size, and abrasive number, a predicted value for the overall condition of the lubricating oil is obtained; If the predicted value is greater than the set threshold, it is marked as abnormal; If the predicted value is less than or equal to the set threshold, it is marked as normal. After determining whether the current state of the lubricating oil is normal or abnormal, the next decision is generated: The current lubricating oil condition is normal; continue to monitor the lubricating oil condition. The current lubricating oil condition is abnormal, and an alarm is being issued.

7. The method according to claim 1, characterized in that, In S5, the abrasive particle classification early warning includes: The number of abrasive grains within a preset volume is counted, the abrasive grain concentration is calculated based on the number of abrasive grains, and the proportion of iron-based abrasive grains is calculated based on the abrasive grain material. If the abrasive concentration is ≥10 and the proportion of iron-based abrasive particles is >80%, a Level 1 alarm for "accelerated equipment wear" will be triggered. If the average size of the abrasive grains is greater than 50 μm and the morphology is cutting, a level 2 alarm for "component fatigue" will be triggered.

8. A fault early warning system for an industrial equipment lubricating oil circulation system employing the method described in any one of claims 1-7, characterized in that, include: The polarization holographic module uses a laser to emit a laser beam, which is then passed through a polarization modulator to obtain four different polarization states of polarized light. These polarized lights are sequentially incident on a lubricating oil sample cell, and the corresponding polarization holographic images are acquired and reconstructed in three dimensions. Additionally, the abrasive material is extracted based on the phase difference and material-polarization feature mapping table corresponding to the four different polarization states. The image feature extraction module uses a convolutional neural network (CNN) to extract features from the reconstructed polarization holographic image to obtain the abrasive grain size and number. The multi-source feature fusion module performs feature fusion and dimensionality reduction on abrasive material, abrasive size and abrasive number to obtain a standardized feature vector; The fault classification and probability output module stores fault mapping and early warning models, which are used to classify and reason about standardized feature vectors and output the fault types and corresponding probability distributions of the lubricating oil circulation system of industrial equipment. The intelligent early warning module stores abrasive warning models, determines the current lubricating oil status based on standardized feature vectors, and performs graded early warnings for abrasive particles.

9. The system according to claim 8, characterized in that, The fault mapping and early warning model uses a random forest (RF) network, and the abrasive particle early warning model uses an XGBoost network.

10. The system according to claim 8, characterized in that, Using historical polarization holographic images as the training set, the abrasive particle size and number are extracted by the abrasive particle material extraction and image feature extraction modules. Then, the standardized feature vectors obtained by the multi-source feature fusion module through feature fusion and dimensionality reduction are used as input. The fault mapping and early warning models and the abrasive particle early warning model are trained with fault type and abrasive particle grade as labels, respectively.

Images