CN116228695A - Wearing state sensing system and method based on video image extraction technology - Google Patents

Abstract

The invention relates to the field of mechanical equipment maintenance research, and discloses a wear state sensing system and method based on a video image extraction technology, wherein the wear state sensing system comprises an image acquisition and preprocessing module, an abrasive particle image characteristic extraction algorithm module and a wear state analysis diagnosis and prediction module, a data output end of the image acquisition and preprocessing module is connected with a data input end of the abrasive particle image characteristic extraction algorithm module, a data output end of the abrasive particle image characteristic extraction algorithm module is connected with a data input end of the wear state analysis diagnosis and prediction module, abrasive particle video images are acquired in real time, acquired image identification is performed on the acquired images, equipment wear information contained in lubricating oil is acquired in real time, and the identification and evolution rule of equipment wear states in an operation period is output from a wear mechanism level, so that necessary evaluation basis and judgment criteria are provided for health state monitoring and life prediction of continuous operation equipment.

Description

Wearing state sensing system and method based on video image extraction technology

Technical Field

The invention relates to the field of maintenance and research of mechanical equipment, in particular to a wear state sensing system and method based on a video image extraction technology.

Background

Wear is the most common and predominant form of failure of mechanical parts. About 80% of the failures of mechanical devices are caused by wear, and the energy consumed by friction accounts for 1/3 to 1/2 of the total energy consumption. About 70% of the failures in diesel engines are due to oil pollution, while 50% of them are due to wear; about 40% of failures and damages in rolling bearings are due to improper lubrication; about 51% of failures in gears are associated with poor lubrication and abnormal wear; about 70% of the failures in the hydraulic system result from the contamination of the hydraulic medium, and the contamination level is too high.

Early people utilize electromagnetic race to carry out on-line monitoring to the abrasive grain, and it characterizes that the accumulated total amount of abrasive grain in the mechanical operation process, when the electromagnetic race sends out alarm signal, the table shows that the abrasive grain total amount of electromagnetic race absorption reaches the preset value, and the grinding of monitored part is already big enough. Attempts have been made to apply the iron pump technique to online monitoring of abrasive particles and to perform simple quantitative analysis of abrasive particles. In recent years, people continuously explore more online monitoring methods for monitoring abrasive particles, and research on online monitoring of the abrasive particles is promoted.

Many research institutions at home and abroad actively explore, and research on online monitoring of abrasive particles is performed by utilizing new technical means, such as laser technology, X-ray technology, ultrasonic technology and the like, so as to obtain many research results. However, in practical application, most of these researches are in the fumbling stage, and the technology is not very mature and needs to be further perfected.

Because of the complexity of the frictional wear state of the apparatus and its variation, and the unavoidable influence of various factors such as the operating state, the operating environment, etc., it is difficult to ensure the stability and comparability of the result. Most detection technologies and methods are still in laboratory research stage at present, many instruments can not fully meet the requirements of machine diagnosis, few instruments are used in the field, and commercial instruments are few, so that the detection technology and method are applied to the research field of rapid wear state identification modules of active equipment, and therefore, a wear state sensing system and method based on video image extraction technology are provided.

Disclosure of Invention

(one) solving the technical problems

Aiming at the defects of the prior art, the invention provides a wear state sensing system and a method based on a video image extraction technology, which solve the problems.

(II) technical scheme

In order to achieve the above purpose, the present invention provides the following technical solutions: the wear state sensing system and method based on the video image extraction technology consists of an online wear monitoring and analyzing instrument software platform, a kernel driver, an embedded operating system WinCE and an online wear monitoring and analyzing instrument software platform, wherein the online wear monitoring and analyzing instrument software platform is provided with a monitoring analyzer control function program and a monitoring analyzer interface control program;

the online wear monitoring and analyzer software platform comprises: the system comprises an image acquisition and preprocessing module, an abrasive particle image characteristic extraction algorithm module and a wear state analysis diagnosis and prediction module, wherein the data output end of the image acquisition and preprocessing module is connected with the data input end of the abrasive particle image characteristic extraction algorithm module, and the data output end of the abrasive particle image characteristic extraction algorithm module is connected with the data input end of the wear state analysis diagnosis and prediction module;

the using method of the sensing system comprises the following specific steps:

s1: designing a flow channel of the undisturbed micro-flow abrasive particle sensor;

s2: acquiring online video images of the abrasion particles through an image acquisition and preprocessing module, and acquiring abrasive particle video images in real time;

s3: after the image acquisition and preprocessing module acquires multi-surface information under the abrasive grain rolling state, rapidly reconstructing a three-dimensional image of the surface morphology based on luminosity stereoscopic vision;

s4: carrying out quantitative extraction on the multi-surface features such as color, particle size, surface morphology, concentration and the like of the abrasive particles flowing through the sensor through an abrasive particle image feature extraction algorithm module;

s5: the wear state analysis diagnosis and prediction module is combined with the equipment friction part characteristic information to perform consistency comparison, so that early warning, diagnosis and condition-based maintenance of faults are realized, and wear state monitoring is performed, so that dangerous conditions can be timely predicted.

Preferably, in the step S1, the flow field under two-phase and laminar flow states is simulated mainly by adopting Fluent software, and optimized values of flow channel structures and flow field parameters of stable and controllable flow of abrasive particles are obtained through a Reynolds equation and a particle stress equation to guide the design of the flow channel of the sensor.

Preferably, the image acquisition and preprocessing module in S2 acquires video image information by using a video stream-based abrasive grain multi-light-source microscopic imaging technology, and the construction of an abrasive grain multi-light-source imaging acquisition environment is completed starting from the analysis of an optical system under the influence of oil and optical glass multi-refractive indexes and the analysis of a large-depth optical structure under a micro-distance condition.

Preferably, the method for rapidly reconstructing the three-dimensional image of the surface morphology based on photometric stereo vision in S3 includes: the three-dimensional reconstruction of abrasive particles based on the stereoscopic vision technology is realized by adopting the steps of optical calibration, distortion correction, parameter extraction of a light source model, calibration and position determination of a light source direction, selection of an illumination model, namely a reflection model, normal vector solving, and the like, and adopting an acquisition surface field and structural components to initialize a calibration algorithm model.

Preferably, the algorithm module for extracting the image features of the abrasive particles in the S4 realizes functions by a method for quickly extracting morphological features of the abrasive particles and identifying a dynamic abrasion mechanism, namely a method for positioning the dynamic abrasive particles and quickly extracting features based on image frames, mainly adopts a chain code tracking method to identify all the abrasive particles and morphological features thereof in the images, and aims at feature matching of images of different frames to obtain the concentration, the morphology, the color and other features of the abrasive particles in the video stream. Developing a fuzzy matching relation between the color characteristics and the abrasion parts or the abrasion mechanisms based on the morphological characteristics and the abrasion mechanisms, and obtaining an abrasive particle characteristic characterization method of a typical abrasion mechanism by adopting a mathematical probability statistical model.

Preferably, the abrasion state analysis diagnosis and prediction module in the step S5 is provided with a multi-element feature vector dynamic matching algorithm, a two-dimensional abrasion state characterization system is established by adopting a real-time abrasion mechanism and an abrasion rate, a tribological state recognition mode of equipment is given, a digital characterization model of the abrasion state of the equipment is researched, a fuzzy matching model and an abnormal early warning algorithm are established according to the dynamic abrasive particle multi-element feature vector, finally an abrasion state intelligent recognition system based on abrasive particle features is established, and the abrasion state and fault early warning are recognized.

Preferably, the image acquisition and preprocessing module consists of four parts, namely a camera module, a runner module, a light field module and a flow control module;

performing Fluent simulation according to project target values, performing data analysis and modeling on physical entities according to extraction, forming a simulation process of multidisciplinary, multidisciplinary quantity, multidisciplinary time scale and multidisciplinary probability, establishing finite element model high-fidelity and high-confidence simulation through integrating data of multidisciplinary physical entities, and simultaneously respectively extracting a minimum quality unit and amplification factor range of abrasive particle images, a spatial three-dimensional size of a runner, a light source position of a light field, a light wave penetration mode and a sample flow volume range by combining discrete event simulation.

Preferably, the reconstructing stereoscopic vision in S3 includes:

calibrating an optical microscope: calibrating a microscope camera to obtain internal parameters of the optical microscope and realize the conversion of a coordinate system;

light source model and calibration: three-dimensional reconstruction is carried out based on luminosity stereo vision, one of the core problems is that information of an incident light source at each point of the surface of a grinding mark is required to be acquired, wherein the information comprises incident illuminance and incident light direction, so that a model of the light source is required to be selected and the light source information is required to be calibrated before calculation of luminosity stereo vision algorithm is carried out;

model selection of light source: firstly, a single light source is regarded as a parallel light source, the macroscopic distance between the light source and a shot object is regarded as parallel light at infinity in the observed micron-level visual field range, the position of the light source is calibrated, then another near-point light source with different light intensity attenuation degrees and different illumination angles along with different distances between the light source and the object is built, and a near-point light source model is adopted for building a photometric stereo vision algorithm model;

selecting a surface illumination model: in the application process, abrasive particles with various and more generation factors appear, the reflection condition of the surface is relatively complex, the project starts from a classical diffuse reflection surface, namely, a reflection model is defined as a lambertian model, normal solution is carried out on the abrasion surface, after the position of a light source, the direction of a main optical axis and main light intensity parameters are calibrated, incident light information at a first point of a surface sampling point is obtained, gray information of each pixel point of an image is obtained, after an illumination model is selected, surface method phase information is obtained by using a luminosity stereoscopic vision algorithm, and finally a normal quantity calculation model with the minimum error is obtained;

solving the depth information to obtain the surface shape: after converting the resulting normal vector into a gradient, the surface depth was calculated using the Frankot-chelappa algorithm to obtain the shape of the surface.

Preferably, the performing quantization extraction on the multifaceted characteristic of the abrasive particle in S4 further includes tracking the abrasive particle in a flowing state and counting physical characteristics:

a, extracting and matching characteristic points in the moving abrasive particle image after the correction of the sampling light scene parameters by using a SIFT algorithm, and then calculating space coordinates corresponding to each pair of matching points by solving a projection matrix of the image to reconstruct space point cloud;

b, calculating the relative height of the abrasive particle surface by using an SFS algorithm, and recovering the three-dimensional morphology of the abrasive particle surface;

c, carrying out gridding treatment on the point cloud by using a Power Curst algorithm, and then fusing the abrasive particle surface morphology reconstructed by the SFS algorithm with the gridded space point cloud by using a fusion algorithm to realize dense reconstruction;

and d, texture mapping, namely mapping the texture of the original image to the dense reconstruction result in the previous step by using an OpenGL texture mapping method, and obtaining the three-dimensional morphology of the multiple surfaces of the moving abrasive particles. .

(III) beneficial effects

Compared with the prior art, the invention provides a wearing state sensing system and a wearing state sensing method based on a video image extraction technology, which have the following beneficial effects:

1. according to the wear state sensing system and method based on the video image extraction technology, the wear type and the wear severity of the key friction pair of the mechanical equipment are identified through algorithms such as online video image acquisition, preprocessing, analysis and identification of wear particles, and important characteristics such as fault information and fault development trend of the equipment are acquired through calculation processing of the surface morphology of the abrasive particle video image, so that early warning, diagnosis and optionally maintenance of faults of the key friction pair of the mechanical equipment are realized. The wear state of the steel plates is monitored, dangerous situations can be timely predicted, and hidden danger is eliminated.

2. According to the wear state sensing system and method based on the video image extraction technology, the abrasive grain video image is obtained in real time, the obtained image is identified, the equipment wear information contained in the lubricating oil is obtained in real time, and the identification and evolution rule of the equipment wear state in the operation period is output from the wear mechanism level, so that necessary evaluation basis and judgment criteria are provided for the health state monitoring and life prediction of continuous operation equipment.

3. The abrasion state sensing system and the method based on the video image extraction technology firstly realize the acquisition of dynamic information flow based on operation equipment, and then realize the safe operation abrasion protection monitoring of the equipment through analysis, training and mining treatment after data accumulation; secondly, the self-diagnosis, prediction and maintenance guarantee capability of the equipment is further standardized and enhanced; thirdly, the equipment guarantee links are further reduced, the guarantee system and resources are optimized, the purposes of accuracy, maneuver, rapidness and economic guarantee are achieved, and favorable basic data are provided for intelligent transformation of enterprise equipment.

Drawings

FIG. 1 is a schematic diagram of a wear state sensing system based on video image extraction technology according to the present invention;

fig. 2 is a schematic diagram of a method for sensing a wear state based on a video image extraction technique according to the present invention.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

Referring to fig. 1-2, a wear state sensing system and method based on video image extraction technology is composed of an online wear monitoring and analyzer software platform, a kernel driver, an embedded operating system WinCE and an online wear monitoring and analyzer software platform, wherein the online wear monitoring and analyzer software platform is loaded with a monitoring analyzer control function program and a monitoring analyzer interface control program;

the online wear monitoring and analyzer software platform comprises: the device comprises an image acquisition and preprocessing module, an abrasive particle image characteristic extraction algorithm module and a wear state analysis diagnosis and prediction module, wherein the data output end of the image acquisition and preprocessing module is connected with the data input end of the abrasive particle image characteristic extraction algorithm module;

the using method of the sensing system comprises the following specific steps:

s1: designing a flow channel of the undisturbed micro-flow abrasive particle sensor;

s2: acquiring online video images of the abrasion particles through an image acquisition and preprocessing module, and acquiring abrasive particle video images in real time;

s3: after the image acquisition and preprocessing module acquires multi-surface information under the abrasive grain rolling state, rapidly reconstructing a three-dimensional image of the surface morphology based on luminosity stereoscopic vision;

s4: carrying out quantitative extraction on the multi-surface features such as color, particle size, surface morphology, concentration and the like of the abrasive particles flowing through the sensor through an abrasive particle image feature extraction algorithm module;

s5: the wear state analysis diagnosis and prediction module is combined with the equipment friction part characteristic information to perform consistency comparison, so that early warning, diagnosis and condition-based maintenance of faults are realized, and wear state monitoring is performed, so that dangerous conditions can be timely predicted.

The abrasion state sensing system and the method based on the video image extraction technology are used for acquiring abrasive grain video images in real time, identifying the acquired images, acquiring equipment abrasion information contained in lubricating oil in real time, outputting the identification and evolution rule of the equipment abrasion state in the operation period from the abrasion mechanism level, and providing necessary evaluation basis and judgment criterion for the health state monitoring and life prediction of continuous operation equipment.

S1, designing a flow channel of the undisturbed micro-flow abrasive particle sensor, wherein Fluent software is mainly adopted to simulate a flow field under two-phase and laminar flow states, and an optimized value of a flow channel structure and flow field parameters of abrasive particles with stable and controllable flow is obtained through a Reynolds equation and a particle stress equation, so that the design of the flow channel of the sensor is guided.

The image acquisition and preprocessing module in S2 acquires video image information, which is based on the video stream abrasive grain multi-light source microscopic imaging technology, and starts with the analysis of an optical system under the influence of oil and optical glass multi-refractive index and the analysis of a large depth of field optical structure under the micro-distance condition, so that the construction of an abrasive grain multi-light source imaging acquisition environment is completed.

The method for rapidly reconstructing the three-dimensional image of the surface morphology based on the photometric stereo vision in the S3 comprises the following steps: the three-dimensional reconstruction of abrasive particles based on the stereoscopic vision technology is realized by adopting the steps of optical calibration, distortion correction, parameter extraction of a light source model, calibration and position determination of a light source direction, selection of an illumination model, namely a reflection model, normal vector solving, and the like, and adopting an acquisition surface field and structural components to initialize a calibration algorithm model.

And S4, the abrasive particle image feature extraction algorithm module realizes functions through an abrasive particle morphology feature quick extraction and dynamic abrasion mechanism identification method, namely the dynamic abrasive particle positioning and feature quick extraction method based on image frames mainly adopts a chain code tracking method to identify all abrasive particles and morphological features thereof in images, and obtains the concentration, morphology, color and other features of the abrasive particles in video streams aiming at feature matching of images of different frames. Developing a fuzzy matching relation between the color characteristics and the abrasion parts or the abrasion mechanisms based on the morphological characteristics and the abrasion mechanisms, and obtaining an abrasive particle characteristic characterization method of a typical abrasion mechanism by adopting a mathematical probability statistical model. The abrasion of the movable equipment has different generation reasons and results, the project adopts a multi-light field and micro-disturbance technology to carry out video capture on the gesture in the process of lubricating oil flow, and carries out one-time extraction on a plurality of physical data such as the color, the particle size, the surface morphology, the concentration and the like of the abrasive particles of the flow sensor. And the information is subjected to induction comparison and output, such as standard comparison and output of the physical and chemical indexes of the existing lubricating oil, such as pollution degree, mechanical magazine content and the like.

And S5, carrying out dynamic matching algorithm of multiple feature vectors on the wear state analysis diagnosis and prediction module, establishing a two-dimensional wear state characterization system by adopting a real-time wear mechanism and wear rate, giving out a tribological state identification mode of equipment, researching a digital characterization model of the wear state of the equipment, establishing a fuzzy matching model and an abnormal early warning algorithm according to the multiple feature vectors of the dynamic abrasive particles, and finally establishing an intelligent wear state identification system based on the abrasive particle features, and identifying the wear state and the fault early warning. And tracking and capturing single abrasive particles in the flowing process of the lubricating oil, obtaining geometric parameters, parameters and texture parameters under different postures, carrying out three-dimensional reconstruction, and improving the accuracy of sample information acquisition. The multi-surface characteristics of the abrasive particles are quantitatively extracted, the matching degree comparison is carried out by combining the characteristic information of the friction part of the equipment, the tracing of the fault part is realized, more importantly, the development process of the abrasive particles carries a large amount of information of the gradual development of the abrasion process of the active equipment, and a large amount of data samples can be provided for the intelligent transformation of the equipment.

The image acquisition and preprocessing module consists of four parts, namely a camera module, a runner module, a light field module and a flow control module;

performing Fluent simulation according to a project target value, performing data analysis and modeling on a physical entity according to extraction to form a simulation process of multidisciplinary, multidisciplinary quantity, multidisciplinary time scale and multidisciplinary probability, establishing finite element model high-fidelity and high-confidence simulation through integrating data of the multidisciplinary physical entity, and simultaneously respectively extracting a minimum quality unit and amplification factor range of an abrasive particle image, a spatial three-dimensional size of a runner, a light source position of a light field, a light wave penetration mode and a sample flow volume range by combining discrete event simulation;

the control logic is implemented in two modes, the acquisition state and the flush state. And the oil is supplied by a micro-conveying unit to obtain action instructions, then the sample oil is conveyed to a collecting area, and meanwhile, the unit volume calculation is carried out according to the working time so as to calculate the concentration content. And after the sample is collected, the oil sample is discharged out of the system.

The stereoscopic reconstruction in S3 includes:

calibrating an optical microscope: calibrating a microscope camera to obtain internal parameters of the optical microscope and realize the conversion of a coordinate system;

light source model and calibration: three-dimensional reconstruction is carried out based on luminosity stereo vision, one of the core problems is that information of an incident light source at each point of the surface of a grinding mark is required to be acquired, wherein the information comprises incident illuminance and incident light direction, so that a model of the light source is required to be selected and the light source information is required to be calibrated before calculation of luminosity stereo vision algorithm is carried out;

model selection of light source: firstly, a single light source is regarded as a parallel light source, the macroscopic distance between the light source and a shot object is regarded as parallel light at infinity in the observed micron-level visual field range, the position of the light source is calibrated, then another near-point light source with different light intensity attenuation degrees and different illumination angles along with different distances between the light source and the object is built, and a near-point light source model is adopted for building a photometric stereo vision algorithm model;

selecting a surface illumination model: in the application process, abrasive particles with various and more generation factors appear, the reflection condition of the surface is relatively complex, the project starts from a classical diffuse reflection surface, namely, a reflection model is defined as a lambertian model, normal solution is carried out on the abrasion surface, after the position of a light source, the direction of a main optical axis and main light intensity parameters are calibrated, incident light information at a first point of a surface sampling point is obtained, gray information of each pixel point of an image is obtained, after an illumination model is selected, surface method phase information is obtained by using a luminosity stereoscopic vision algorithm, and finally a normal quantity calculation model with the minimum error is obtained;

solving the depth information to obtain the surface shape: after converting the obtained normal vector into a gradient, the surface depth is calculated by using the Frankot-Chellappa algorithm to obtain the shape of the surface, and the construction factors in the process are shown in the following table:

in S4, carrying out quantitative extraction on the multi-surface characteristics of the abrasive particles further comprises the following steps of abrasive particle tracking in a flowing state and physical characteristic statistics:

a, extracting and matching characteristic points in the moving abrasive particle image after the correction of the sampling light scene parameters by using a SIFT algorithm, and then calculating space coordinates corresponding to each pair of matching points by solving a projection matrix of the image to reconstruct space point cloud;

b, calculating the relative height of the abrasive particle surface by using an SFS algorithm, and recovering the three-dimensional morphology of the abrasive particle surface;

c, carrying out gridding treatment on the point cloud by using a Power Curst algorithm, and then fusing the abrasive particle surface morphology reconstructed by the SFS algorithm with the gridded space point cloud by using a fusion algorithm to realize dense reconstruction;

and d, texture mapping, namely mapping the texture of the original image to the dense reconstruction result in the previous step by using an OpenGL texture mapping method, and obtaining the three-dimensional morphology of the multiple surfaces of the moving abrasive particles.

According to the wear state sensing system and method based on the video image extraction technology, the wear type and the wear severity of the key friction pair of the mechanical equipment are identified through algorithms such as online video image acquisition, preprocessing, analysis and identification of wear particles, and important characteristics such as fault information and fault development trend of the equipment are acquired through calculation processing of the surface morphology of the abrasive particle video image, so that early warning, diagnosis and optionally maintenance of faults of the key friction pair of the mechanical equipment are realized. The wear state of the steel plates is monitored, dangerous situations can be timely predicted, and hidden danger is eliminated;

firstly, realizing dynamic information flow acquisition based on operation equipment by adopting technologies such as video flow extraction and identification, data modeling and the like of lubricating oil wear particles, and then realizing equipment safe operation wear guarantee monitoring through analysis, training and mining treatment after data accumulation; secondly, the self-diagnosis, prediction and maintenance guarantee capability of the equipment is further standardized and enhanced; thirdly, the equipment guarantee links are further reduced, the guarantee system and resources are optimized, and the aims of accuracy, maneuver, rapidness and economic guarantee are achieved. Therefore, if the project can be successfully implemented, the project can provide favorable basic data for intelligent transformation of the enterprise equipment while ensuring normal operation of the active equipment.

Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.

Claims (9)

1. A wear state sensing system and method based on video image extraction technology is characterized in that: the system consists of an online wear monitoring and analyzing instrument software platform, a kernel driver, an embedded operating system WinCE and an online wear monitoring and analyzing instrument software platform, wherein the online wear monitoring and analyzing instrument software platform is provided with a monitoring analyzer control function program and a monitoring analyzer interface control program;

the online wear monitoring and analyzer software platform comprises: the system comprises an image acquisition and preprocessing module, an abrasive particle image characteristic extraction algorithm module and a wear state analysis diagnosis and prediction module, wherein the data output end of the image acquisition and preprocessing module is connected with the data input end of the abrasive particle image characteristic extraction algorithm module, and the data output end of the abrasive particle image characteristic extraction algorithm module is connected with the data input end of the wear state analysis diagnosis and prediction module;

the using method of the sensing system comprises the following specific steps:

s1: designing a flow channel of the undisturbed micro-flow abrasive particle sensor;

s2: the image acquisition and preprocessing module is used for carrying out video capture on the gesture in the process of lubricating oil flowing by adopting a multi-light field and micro-disturbance technology, so as to acquire an abrasive particle video image in real time;

s3: then, the image acquisition and preprocessing module tracks and captures single abrasive particles in the process of lubricating oil flowing, acquires geometric parameters, parameters and texture parameters under different postures, and performs three-dimensional reconstruction;

s4: carrying out quantitative extraction on the multi-surface features such as color, particle size, surface morphology, concentration and the like of the abrasive particles flowing through the sensor through an abrasive particle image feature extraction algorithm module;

s5: the wear state analysis diagnosis and prediction module is combined with the equipment friction part characteristic information to perform consistency comparison, so that early warning, diagnosis and condition-based maintenance of faults are realized, and wear state monitoring is performed, so that dangerous conditions can be timely predicted.

2. The system and method for detecting the abrasion state based on the video image extraction technology according to claim 1, wherein: the design of the flow channel of the undisturbed micro-flow abrasive particle sensor in the step S1 mainly adopts Fluent software to simulate a flow field under two-phase and laminar flow states, and obtains the optimized values of the flow channel structure and flow field parameters of stable and controllable flow of abrasive particles through a Reynolds equation and a particle stress equation to guide the design of the flow channel of the sensor.

3. The system and method for detecting the abrasion state based on the video image extraction technology according to claim 1, wherein: the image acquisition and preprocessing module in S2 acquires video image information based on the video stream abrasive grain multi-light source microscopic imaging technology, and the construction of an abrasive grain multi-light source imaging acquisition environment is completed by starting with the analysis of an optical system under the influence of oil and optical glass multi-refractive indexes and the analysis of a large depth-of-field optical structure under the micro-distance condition.

4. The system and method for detecting the abrasion state based on the video image extraction technology according to claim 1, wherein: the method for rapidly reconstructing the three-dimensional image of the surface morphology based on the photometric stereo vision in the S3 comprises the following steps: the three-dimensional reconstruction of abrasive particles based on the stereoscopic vision technology is realized by adopting the steps of optical calibration, distortion correction, parameter extraction of a light source model, calibration and position determination of a light source direction, selection of an illumination model, namely a reflection model, normal vector solving, and the like, and adopting an acquisition surface field and structural components to initialize a calibration algorithm model.

5. The system and method for detecting the abrasion state based on the video image extraction technology according to claim 1, wherein: the abrasive particle image feature extraction algorithm module in the S4 realizes functions through an abrasive particle morphology feature quick extraction and dynamic abrasion mechanism identification method, namely the dynamic abrasive particle positioning and feature quick extraction method based on image frames mainly adopts a chain code tracking method to identify all abrasive particles and morphological features thereof in images, obtains the concentration, morphology, color and other features of the abrasive particles in video streams aiming at feature matching of different frames of images, develops a fuzzy matching relation between the morphological features and the abrasion mechanism, and obtains an abrasive particle feature characterization method of a typical abrasion mechanism by adopting a mathematical probability statistical model.

6. The system and method for detecting the abrasion state based on the video image extraction technology according to claim 1, wherein: and S5, carrying out multi-element feature vector dynamic matching algorithm on the abrasion state analysis diagnosis and prediction module, establishing a two-dimensional abrasion state characterization system by adopting a real-time abrasion mechanism and abrasion rate, giving out a tribological state identification mode of equipment, researching a digital characterization model of the abrasion state of the equipment, establishing a fuzzy matching model and an abnormal early warning algorithm according to the dynamic abrasive particle multi-element feature vector, finally establishing an abrasion state intelligent identification system based on abrasive particle features, and identifying the abrasion state and fault early warning.

7. The system and method for detecting the abrasion state based on the video image extraction technology according to claim 1, wherein: the image acquisition and preprocessing module consists of four parts, namely a camera module, a runner module, a light field module and a flow control module;

performing Fluent simulation according to project target values, performing data analysis and modeling on physical entities according to extraction, forming a simulation process of multidisciplinary, multidisciplinary quantity, multidisciplinary time scale and multidisciplinary probability, establishing finite element model high-fidelity and high-confidence simulation through integrating data of multidisciplinary physical entities, and simultaneously respectively extracting a minimum quality unit and amplification factor range of abrasive particle images, a spatial three-dimensional size of a runner, a light source position of a light field, a light wave penetration mode and a sample flow volume range by combining discrete event simulation.

8. The system and method for detecting the abrasion state based on the video image extraction technology according to claim 1, wherein: the stereoscopic reconstruction in S3 includes:

calibrating an optical microscope: calibrating a microscope camera to obtain internal parameters of the optical microscope and realize the conversion of a coordinate system;

light source model and calibration: before the calculation of the luminosity stereoscopic vision algorithm is carried out, a model of the light source is selected, and the light source information is calibrated;

model selection of light source: firstly, a single light source is regarded as a parallel light source, the macroscopic distance between the light source and a shot object is regarded as parallel light at infinity in the observed micrometer-level visual field range, the position of the light source is calibrated, then another near-point light source with different light intensity attenuation degrees and different illumination angles along with the different distances between the light source and the object is built, and a near-point light source model is adopted for building a photometric stereo vision algorithm model;

selecting a surface illumination model: starting from a classical diffuse reflection surface, namely, a reflection model is defined as a lambertian model, normal solving is carried out on a worn surface, after the position of a light source, the direction of a main optical axis and main light intensity parameters are calibrated, incident light information at a first point of a surface sampling point is obtained, gray information of each pixel point of an image is obtained, after an illumination model is selected, surface method phase information is obtained by using a photometric stereo vision algorithm, and finally a normal calculation model with minimum error is obtained;

solving the depth information to obtain the surface shape: after converting the resulting normal vector into a gradient, the surface depth was calculated using the Frankot-chelappa algorithm to obtain the shape of the surface.

9. The system and method for detecting the abrasion state based on the video image extraction technology according to claim 1, wherein: the step S4 of quantitatively extracting the multi-face characteristics of the abrasive particles further comprises the following steps of flowing state abrasive particle tracking and physical characteristic statistics:

a, extracting and matching characteristic points in the moving abrasive particle image after the correction of the sampling light scene parameters by using a SIFT algorithm, and then calculating space coordinates corresponding to each pair of matching points by solving a projection matrix of the image to reconstruct space point cloud;

b, calculating the relative height of the abrasive particle surface by using an SFS algorithm, and recovering the three-dimensional morphology of the abrasive particle surface;

c, carrying out gridding treatment on the point cloud by using a Power Curst algorithm, and then fusing the abrasive particle surface morphology reconstructed by the SFS algorithm with the gridded space point cloud by using a fusion algorithm to realize dense reconstruction;

and d, texture mapping, namely mapping the texture of the original image to the dense reconstruction result in the previous step by using an OpenGL texture mapping method, and obtaining the three-dimensional morphology of the multiple surfaces of the moving abrasive particles.

Images