A method and system for dynamically detecting sediment particles

Abstract

The application relates to a sediment particle dynamic detection method and a detection system, and relates to the technical field of river sediment detection technology. The method comprises the following steps: S1, obtaining a high-pressure sediment sample, performing metering, and then performing pressure reduction treatment to form a to-be-detected sample; S2, screening the to-be-detected sample according to particle size to form a first sample and a second sample; S3, performing turbidity detection and dilution on the sample by using a transmission-scattering ratio method; S4, irradiating the first sample with multi-wavelength combined shaping light, and obtaining particle morphology and component information based on parameter analysis; S5, performing imaging detection on the second sample, obtaining a polarization image and a spectral image, and fusing and analyzing the polarization image and the spectral image to obtain particle morphology and component information; and S6, integrating detection data, identifying and classifying through a machine learning model, and outputting a detection result. Through the steps of high-pressure sampling, pressure reduction, particle size screening, turbidity detection and dilution, double-channel detection and data integration, the labor intensity is reduced, and the detection efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of river sediment detection technology, and in particular to a method and system for dynamic detection of sediment particles. Background Technology

[0002] River sediment detection is a crucial task involving the determination and analysis of sediment content, particle size distribution, and sediment types in rivers. It provides critical data support for water conservancy project planning, river management, and hydropower station operation and maintenance. Particularly in the hydropower sector, accurate detection of sediment passing through turbines is of paramount engineering significance for assessing turbine wear, predicting equipment lifespan, and developing operation and maintenance strategies. In recent years, with the large-scale development of hydropower resources in areas with complex geological conditions such as the Yarlung Tsangpo River, the high sediment content, wide particle size distribution, and complex composition of rivers have placed higher demands on sediment detection technology, necessitating a high-precision, real-time online sediment detection solution capable of adapting to a wide particle size range.

[0003] Currently, river sediment detection primarily employs a combination of manual sampling and laboratory analysis. Specifically, sediment samples are collected at different cross-sections and locations using sampling equipment. These samples are then brought back to the laboratory, dried to remove moisture, and their weight is measured. Finally, sieving is used to classify the sediment particles and determine their particle size distribution. In recent years, with the development of optical detection technology, several optically based sediment detection devices have emerged, such as laser scattering and image analysis. Laser scattering involves irradiating sediment particles with a laser and calculating particle size parameters based on the angle and intensity distribution of the scattered light. Image analysis involves acquiring images of sediment particles and using image processing techniques to obtain particle morphology information. Additionally, there are techniques using the transmission-to-scattering ratio method for turbidity detection, which assesses the sediment content in water bodies by monitoring changes in the intensity of transmitted and scattered light.

[0004] However, existing technologies still have many shortcomings in practical applications. First, manual sampling and drying methods are labor-intensive and inefficient, unable to achieve real-time online detection, and their accuracy is greatly affected by the sampling location and operator, making it difficult to meet the real-time and accurate monitoring requirements of modern hydropower stations. Second, existing optical detection technologies have limitations when dealing with sediments of a wide particle size range: for small-particle sediments, although laser scattering methods have high resolution, single-wavelength illumination is easily interfered with by the irregular shape of particles, and it is difficult to achieve single-particle classification based on composition; for large-particle sediments, image analysis methods are limited by imaging resolution and particle overlap issues, and both detection accuracy and species identification capabilities need improvement. In addition, the detection accuracy of existing detection equipment decreases significantly in high-concentration sediment environments, and its ability to identify the composition of sediment particles is limited, making it difficult to distinguish the differences in the degree of wear on turbine units caused by different types of sediment. Existing technologies generally lack effective means to treat the wastewater generated during the detection process, making it difficult to meet increasingly stringent environmental protection requirements. Especially under complex hydrological conditions such as the Yarlung Tsangpo River, existing technologies are unable to simultaneously meet the comprehensive requirements of wide particle size range, high precision, real-time online operation, multi-parameter identification, and environmentally friendly operation. Summary of the Invention

[0005] The purpose of this application is to provide a method and system for dynamic detection of sediment particles to solve the problems existing in the prior art.

[0006] To achieve the above objectives, the technical solution adopted in this application is as follows: This application provides a method for dynamic detection of sediment particles, including the following steps: S1. Obtain high-pressure sediment samples, measure them, and depressurize them to form the sample to be tested; S2. The sample to be tested is sieved according to the preset particle size to form a first sample with a first particle size range and a second sample with a second particle size range, wherein the first particle size range is smaller than the second particle size range. S3. Turbidity of the first and second samples was detected by the transmission scattering ratio method, and the samples were diluted according to the detection results. S4. The first sample is processed by sheath fluid focusing to form a single-row particle stream, and then irradiated with multi-wavelength beam shaping light. The scattering signal of the particles is collected by a multi-channel polarization detector, and the particle morphology and composition information of the first sample is obtained based on Stokes full vector parameter analysis. S5. The second sample is subjected to imaging detection processing. The polarization image and spectral image of the particles are obtained by using a backlight source. The polarization image and spectral image are fused and analyzed to obtain the particle morphology and composition information of the second sample. S6. Integrate the detection data of the first and second samples, identify and classify them through a machine learning model, and output the detection results of sediment particles.

[0007] Furthermore, in S1, the pressure of the high-pressure sediment sample is monitored. Based on the pressure feedback, the high-pressure sediment sample is depressurized to the normal pressure range by adjusting the flow rate and providing transport power. The turbidity, pressure, and flow rate parameters of the sediment sample after depressurization are collected in real time.

[0008] This application also provides a detection system for implementing the dynamic detection method for sediment particles according to any one of the above, comprising: Sampling unit, comprising: A pressurized water sampling probe for use in constant motion sampling; The pressure-reducing sampling device connected in series downstream of the pressure-bearing water intake probe includes a pressure-resistant main body, a pressure regulating component, and a parameter detection component connected in sequence. The pressure-resistant main body is used to introduce high-pressure water flow, the pressure regulating component is used to reduce the pressure of the high-pressure water flow in stages, and the parameter detection component is used to collect the turbidity, pressure, and flow parameters of the water flow after pressure reduction in real time. A preprocessing unit, comprising: A diffusion suspension section and a static mixer are sequentially connected downstream of the parameter detection component to eliminate bubbles and homogenize. A particle size classification module connected downstream of a static mixer includes an inclined plate settling tank, a hydrocyclone, and an inertial microfluidic chip for separating a sample into a first sample and a second sample, which are output from a first outlet and a second outlet, respectively. The first branch detection unit has its input end connected to the first outlet. The first branch detection unit includes a first turbidity meter, a first flow meter, a first dilution device and a small particle detection device connected in series. The small particle detection device includes a sheath fluid focusing module, a multi-band laser source and a lateral polarization detection module, and a first data processing module based on Stokes vector analysis. The second branch detection unit has its input end connected to the second outlet. The second branch detection unit includes a second turbidity meter, a second flow meter, a second dilution device and a large particle detection device connected in series. The large particle detection device includes a polarization imaging module, a spectral imaging module and a second data processing module for image fusion analysis. The central data processing unit is connected to the first data processing module and the second data processing module respectively. The central data processing unit includes a machine learning model for particle identification and classification, which integrates the two detection data and outputs the sediment particle identification and classification results. The waste liquid treatment unit has its input terminals connected to the pretreatment unit, the first branch detection unit, and the second branch detection unit, respectively, and is used to collect and treat the generated waste liquid in a unified manner. The small particle detection device and / or large particle detection device are equipped with an easy-to-maintain optical module.

[0009] Furthermore, the front end of the pressure water intake probe is equipped with a replaceable ceramic anti-wear guide cover, and its water intake hole diameter is 2-3mm.

[0010] Furthermore, a high-pressure shut-off valve and a microporous damping assembly are connected in series between the pressurized water intake probe and the bypass pipeline to configure a dual-valve isolation structure.

[0011] Furthermore, the diffusion suspension section is a vertically installed expansion pipe with a honeycomb rectifier grid inside and an automatic exhaust valve at the top; the static mixer is an SK type or SMV type online mixer.

[0012] Furthermore, the inclined plate settling tank contains a set of parallel inclined plates with an inclination angle of 55°-60°, and the hydrocyclone is a 25mm diameter polyurethane hydrocyclone tube with a median particle size cutoff. The inertial microfluidic chip has a diameter of 50 μm and is a glass-silicon composite chip with etched arc-shaped curved channels or micropillar arrays.

[0013] Furthermore, the multi-band laser source of the small particle detection device is a three-band semiconductor laser with wavelengths of 405nm, 532nm, and 650nm, and the lateral polarization detection module is a linear polarization analyzer set in the 90° direction.

[0014] Furthermore, the polarization imaging module of the large particle detection device includes a switchable linear polarizer and a quarter-wave plate, and the spectral imaging module is an RGB camera, a liquid crystal tunable filter, or a pushbroom hyperspectral sensor, with a working wavelength range of 400-1000nm.

[0015] The beneficial effects of the technical solution provided in this application include at least the following: (1) This application significantly reduces labor intensity and improves detection efficiency by moving from high-pressure sampling, pressure reduction treatment, particle size screening, turbidity detection and dilution to dual-channel detection and data integration. It can reflect the dynamic changes of sediment in the water flow of the hydropower station in real time and provide timely data support for the operation and maintenance of the turbine unit.

[0016] (2) By splitting the sample into small particles and large particles, the small particles are analyzed by sheath scattering and the large particles are analyzed by imaging, which can simultaneously detect a wide range of particle sizes from micron-sized fine particles to millimeter-sized coarse particles, significantly improving the accuracy of particle size, shape and type identification.

[0017] (3) The classification model based on machine learning in this application can automatically learn the characteristic patterns of different sediment particles, and achieve relatively fast and accurate identification of species and particle size classification, reducing the problems of time-consuming, labor-intensive and subjective traditional manual microscopic examination or sieving methods. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the present application and form part of the specification. They are used together with the embodiments of the present application to explain the application and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating the detection method in one embodiment of the present invention; Figure 2 This is a schematic diagram of a deep learning algorithm in one embodiment of the present invention; Figure 3 This is a schematic diagram of the output result of the detection method in one embodiment of the present invention; Figure 4 This is a schematic diagram of the structural framework of the sampling unit of the system in one embodiment of the present invention; Figure 5 This is a schematic diagram of the structural framework of the preprocessing unit, the first branch detection unit, and the second branch detection unit of the system in one embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of a small particle detection device in one embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of a large particle detection device in one embodiment of the present invention.

[0019] Explanation of key figure labels: 10. Sampling unit; 11. Pressure-resistant main body; 12. High-precision valve; 13. Regulating pump; 14. Parameter detection component; 20. Pretreatment unit; 21. Diffusion suspension section; 22. Static mixer; 23. Particle size classification module; 231. First outlet; 232. Second outlet; 30. First branch detection unit; 31. First turbidity meter; 32. First flow meter; 33. First dilution device; 34. Small particle detection device; 40. Second branch detection unit; 41. Second turbidity meter; 42. Second flow meter; 43. Second dilution device; 44. Large particle detection device. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] In this specification, identical components are represented by the same reference numerals. It should be noted that the terms "front," "rear," "left," "right," "upper," and "lower" used in the following description refer to directions in the accompanying drawings, while the terms "bottom surface," "top surface," "inner," and "outer" refer to directions towards or away from a specific component, respectively. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this specification, "multiple" means two or more. Example 1

[0022] Please refer to Figures 1-3 A method for dynamic detection of sediment particles includes the following steps: S1. Obtain high-pressure sediment samples, measure them, and depressurize them to form the sample to be tested; S2. The sample to be tested is sieved according to the preset particle size to form a first sample with a first particle size range and a second sample with a second particle size range, wherein the first particle size range is smaller than the second particle size range. S3. Turbidity of the first and second samples was detected by the transmission scattering ratio method, and the samples were diluted according to the detection results. S4. The first sample is processed by sheath fluid focusing to form a single-row particle stream, and then irradiated with multi-wavelength beam shaping light. The scattering signal of the particles is collected by a multi-channel polarization detector, and the particle morphology and composition information of the first sample is obtained based on Stokes full vector parameter analysis. S5. The second sample is subjected to imaging detection processing. The polarization image and spectral image of the particles are obtained by using a backlight source. The polarization image and spectral image are fused and analyzed to obtain the particle morphology and composition information of the second sample. S6. Integrate the detection data of the first and second samples, identify and classify them through a machine learning model, and output the detection results of sediment particles.

[0023] In this embodiment, as Figure 1 As shown, in step 1, a sampling interface is set at the end of the water pressure pipeline. High-pressure sediment samples are obtained through a depressurization sampling device. During the sampling process, an electromagnetic flowmeter accurately measures the sample flow rate. Under real-time feedback from the pressure monitoring instrument, the depressurization sampling device precisely adjusts the flow rate through a high-precision valve, and works with a delivery pump to provide stable power, gradually reducing the high-pressure water flow of 3-5 MPa to the normal pressure range of 0.1-0.3 MPa. Simultaneously, the detection equipment collects parameters such as turbidity, pressure, and flow rate of the depressurized water flow in real time and transmits the electrical signals to the control system to ensure that the sediment sample entering the subsequent detection stage is stable and controllable.

[0024] In step S2, the sample to be tested is split and sieved according to a preset particle size threshold, which can be set according to actual detection needs, for example, 50 μm. Particles smaller than 50 μm are defined as the first particle size range, and sediment particles smaller than 50 μm are used as the first sample and enter the small particle detection channel; particles larger than 50 μm are defined as the second particle size range, and sediment particles larger than 50 μm are used as the second sample and enter the large particle detection channel. Automatic splitting of samples from different particle size ranges can be achieved through physical sieving.

[0025] In step S3, the first and second samples after splitting are subjected to turbidity detection using the transmission-scattering ratio method. The transmission method monitors the change in transmitted light intensity based on Lambert-Beer's law, while the scattering method monitors the intensity of scattered light in the 90° direction. The computer collects and analyzes turbidity data in real time. When the turbidity of the sample is detected to be too high, the peristaltic pump is automatically controlled to pump in a certain amount of clean water for dilution. At the same time, the volume of clean water pumped in is accurately recorded to obtain a sample with moderate turbidity suitable for subsequent detection.

[0026] In step S4, the first sample enters the small particle detection channel. First, sheath fluid focusing is used to arrange the particles in the sample into a single row, allowing them to pass through the detection area sequentially, thus avoiding particle overlap and ensuring that each particle is detected independently. In the detection area, multi-wavelength beam-shaping light is used to illuminate the particle stream: different wavelengths of laser light are combined and shaped into a uniform beam, such as a one-dimensional flat-top beam, to improve the uniformity of illumination and energy utilization. After being illuminated, the particles generate scattering signals, which are collected in parallel by a multi-channel polarization detector. The multi-channel polarization detector can simultaneously acquire the intensity information of scattered light in different polarization directions. Subsequently, the acquired signals are processed based on Stokes full vector parameter analysis: the polarization state of the scattered light is comprehensively described by calculating the Stokes vectors (S0, S1, S2, S3), and feature information related to particle morphology and composition, such as particle size, shape, and refractive index, is extracted from them, ultimately obtaining the morphology and composition information of the particles in the first sample.

[0027] In step S5, the second sample enters the large particle imaging detection channel. The sample is uniformly distributed on a glass slide or other transparent carrier, and the particles are illuminated from behind using a backlight source, allowing for clear imaging of the particle's outline and internal structure. A polarization camera is used to acquire polarization images of the particles, recording the light intensity distribution under different polarization states. Simultaneously, a spectral imaging device (such as a multispectral camera or hyperspectral reconstruction system) is used to acquire spectral images of the particles, recording the particle's reflection or transmission characteristics to different wavelengths of light. Then, the polarization and spectral images are fused and analyzed: for example, after registering the two images, an image fusion algorithm, such as based on color space transformation or feature-level fusion, is used to integrate polarization and spectral information, thereby obtaining the morphology and composition information of the particles in the second sample.

[0028] In step S6, the first sample detection data obtained in step S4 and the second sample detection data obtained in step S5 are aggregated to the central processing unit. This unit has a built-in machine learning model for identifying sediment particles. The trained model classifies and identifies the integrated data and outputs the final detection results. The results may include parameters such as sediment particle size distribution, particle shape statistics, type identification, and concentration, and are presented in the form of visual reports or data streams for reference by hydropower station operation and maintenance personnel.

[0029] The aforementioned method automates the entire process, from high-pressure sampling, depressurization, particle size sieving, turbidity detection and dilution to dual-channel detection and data integration, eliminating the need for manual intervention. This significantly reduces labor intensity and improves detection efficiency. It can reflect the dynamic changes of sediment in the hydropower station's water flow in real time, providing timely data support for the operation and maintenance of the turbine units. Simultaneously, it achieves stable collection and precise pretreatment of sediment samples under high water pressure, broadening the applicability of the detection device to meet the detection needs of sediments with different concentrations and particle sizes.

[0030] Furthermore, by setting a preset particle size threshold, samples are separated into small and large particles, and different detection techniques are employed for each: sheath flow scattering analysis is used for small particles, while imaging analysis is used for large particles. This allows for the simultaneous detection of a wide range of particle sizes, from micrometer-scale fine particles to millimeter-scale coarse particles, making it particularly suitable for water bodies such as the Yarlung Tsangpo River, where sediment particle size distribution is extensive and composition is complex. Specifically, small particle detection utilizes multi-wavelength beam-shaping light and multi-channel polarization detection, combined with Stokes full vector analysis, to extract rich polarization features from the scattering signal, effectively overcoming the interference of irregular particle shapes on the measurement. Large particle detection integrates polarization and spectral images, taking into account both morphological and compositional information, thereby significantly improving the accuracy of particle size, shape, and type identification.

[0031] Machine learning-based classification models can automatically learn the characteristic patterns of different sediment particles, enabling faster and more accurate identification of species and particle size classification, reducing the time-consuming, labor-intensive, and subjective problems of traditional manual microscopic examination or sieving methods.

[0032] Furthermore, in S1, the pressure of the high-pressure sediment sample is monitored. Based on the pressure feedback, the high-pressure sediment sample is depressurized to the normal pressure range by adjusting the flow rate and providing transport power. The turbidity, pressure, and flow rate parameters of the sediment sample after depressurization are collected in real time.

[0033] More specifically, in S2, the preset particle size is 50μm, with sediment smaller than 50μm being the first sample and sediment larger than 50μm being the second sample.

[0034] In this embodiment, sample splitting can be performed through physical sieving. Specifically, after the depressurization process in step S1, the sample to be tested is introduced into the splitting device and physically sieved according to a preset particle size threshold. In this embodiment, the preset particle size threshold is 50 μm, thereby dividing the sediment sample into two sub-samples with different particle size ranges, which are used for different subsequent detection and analysis. A woven filter screen with a pore size slightly smaller than 50 μm is installed at the outlet of the dilution tank. This filter screen is installed at the first outlet of the dilution tank, so that the sample fluid must pass through the filter screen when flowing through this outlet.

[0035] Mud and sand particles with a diameter of less than 50 μm and water molecules in the diluent can pass smoothly through the mesh of the woven filter and flow out from the first outlet of the dilution pool with the fluid. This part of the sample is defined as the first sample, whose particle size range is less than 50 μm, mainly including fine mud and sand such as clay and silt. The first sample is guided to the subsequent small particle detection channel.

[0036] Silt particles with a diameter greater than or equal to 50 μm cannot pass through the mesh of the woven filter and are trapped on one side of the filter. These large silt particles are propelled by the fluid and flow out from the second outlet of the dilution tank along another passage of the dilution tank. This part of the sample is defined as the second sample, whose particle size range is greater than 50 μm, mainly including fine sand, medium sand and larger coarse silt particles. The second sample is guided to the subsequent large particle detection channel.

[0037] To ensure screening effectiveness, the flow rates through the first and second outlets can be monitored and adjusted to maintain stable output of both samples. An auxiliary rinsing or drainage device can be installed at the second outlet to prevent large particles from accumulating and clogging the filter screen, thus ensuring the continuity and stability of the screening process.

[0038] Specifically, in S3, the transmission-to-scattering ratio method includes both transmission and scattering methods; In the transmission method, the Lambert-Beer law is satisfied: ,in The intensity of transmitted light. Where α is the incident light intensity, β is the absorption coefficient, and L is the scattering coefficient; In the scattering method, the intensity of scattered light at a 90° direction Satisfying the relation: Where K is a coefficient related to the instrument's geometry, N is the number of particles per unit volume, and V is the volume of the particles. The incident light wavelength; under certain conditions, when the particle size and the incident light wavelength are relatively fixed, If it is a constant, then It is directly proportional to N, that is, directly proportional to turbidity; The turbidity value was calculated based on the data obtained from the transmission and scattering methods, and then diluted by pumping in clean water, with the volume of clean water pumped in recorded.

[0039] In this embodiment, the transmission-scattering ratio method is used for turbidity detection. The transmission-scattering ratio method is as follows: Where T is the turbidity value, and I S The intensity of the scattered light. Let K be the intensity of transmitted light, and K be a proportionality constant related to the liquid.

[0040] In the transmission method, a sensor can be used to monitor the transmitted light. When the intensity of the light emitted by the light source is... When light passes through a cuvette containing a water sample, the intensity of the transmitted light decreases due to the absorption and scattering of incident light by suspended solids and impurities in the liquid. Let the intensity of the absorbed and scattered light be I. A and I S ,in, Therefore, these three factors satisfy the Lambert-Beer Law: α is the absorption coefficient, reflecting the absorption of light by the sediment sample; β is the scattering coefficient, reflecting the scattering of light by sediment particles; and L is the optical path length. The intensity of transmitted light is negatively correlated with sediment concentration: the higher the sediment concentration, the weaker the transmitted light.

[0041] In the scattering method, the scattered light can be monitored by a sensor. When the light beam passes through the water sample being tested, the intensity of the scattered light in the 90° direction is measured. It can be expressed by the following formula: ,in, The incident light intensity This refers to the number of particles per unit volume. Let V be the volume of the particle. Where is the incident light wavelength, and K is a coefficient. During the detection process, under certain conditions, it is visible... and If it is a constant, then The coefficient K is directly proportional to the total number of particles per unit volume, i.e., directly proportional to the turbidity. It is proportional to N, that is, the intensity of scattered light is proportional to turbidity: the higher the turbidity, the stronger the scattered light in the 90° direction.

[0042] The system collects transmitted light intensity in real time. and the intensity of scattered light at 90° direction The turbidity value of the current sample is calculated based on the aforementioned optical relationship. Since the sensitivity of the transmission method decreases at high turbidity and the scattering method is more sensitive at low turbidity, the combination of the two can achieve accurate measurement of turbidity over a wide range.

[0043] When the calculated turbidity value exceeds the preset suitable detection range—for example, excessively high turbidity may cause particle overlap or scattering signal saturation in subsequent detection—the control system automatically initiates a dilution procedure: a certain amount of clean water is pumped into the sediment sample using a peristaltic pump or other precision pumping device for dilution, while accurately recording the volume of clean water pumped in. During the dilution process, the system continuously monitors turbidity changes until the sample turbidity is adjusted to a suitable target range for subsequent detection.

[0044] Recording the volume of pumped clean water serves two purposes: firstly, it enables closed-loop control to ensure dilution accuracy; secondly, this volume data, combined with the initial sample volume, can be used to infer the original turbidity of the original sample, providing complete concentration information for subsequent data analysis.

[0045] By using the transmission scattering ratio method and automatic dilution control, a sample with moderate turbidity and stable state was obtained, providing reliable input conditions for subsequent particle size sieving and detection analysis.

[0046] The principle of the transmission scattering ratio method is to measure turbidity using a turbidity meter, connect it to a pressure reducing device and a controllable dilution device, and use a computer to collect the measurement data from the turbidity meter. The computer analyzes the data obtained by the turbidity meter, and an automated device drives a peristaltic pump to pump in a certain amount of clean water, recording the volume of the pumped clean water.

[0047] In addition, in the specific steps of small particle detection, S4 includes the following steps: S41. The first sample forms a single-row particle stream and is irradiated with multi-wavelength beam-shaping light; S42. Scattered signals are acquired in parallel through the four polarization channels of the Stokes full vector sensor and the eight channels of the photodetector. S43. The acquired signal is amplified and converted from analog to digital. Based on FPGA, 8-channel parallel acquisition and 4-polarization channel Stokes full vector acquisition are implemented. The analog signal is then processed by Fourier transform to realize real-time spectrum analysis of the scattered signal. S44. Calculate the scattering characteristics of sediment particles using Mie scattering theory, extract the particle size and refractive index optical parameters, and analyze the scattering characteristics of particles of arbitrary shape using the T-matrix method to determine the particle shape and size.

[0048] In S42, the four polarization channels of the Stokes full vector sensor are polarization filters set at angles of 0°, 45°, 90° and 135° on different pixels. Every four pixels form a computing unit. A microlens array is set above each pixel, and a quarter-wave plate is used to obtain the circular polarization component.

[0049] In S43, the data captured by the polarization camera sensor is represented by a Stokes vector, and the Stokes vector S satisfies the following relationship: in, The total intensity of light, 0° ) and 90° Intensity difference under linear polarization state, 45° ) and 135° Intensity difference under linear polarization; Right-handed circular polarization ( ) and left-handed circular polarization ( The intensity difference between states; The linear polarization degree DoLP and polarization angle AoLP are calculated using the following formulas: A fusion method is used to fuse features from S0, DoLP, and AoLP: Identify and preserve overlapping features among S0, DoLP, and AoLP: ; Using subtraction to highlight the unique characteristics of each parameter: Synthesize and fuse images using a weighted average method: ,in + + =1.

[0050] In S43, parallel Fourier transforms of the 8-channel photoelectric signal and the 4-channel polarization signal are performed using the following formula: in, f(t) is a frequency domain signal, and f(t) is a time domain analog signal. ω is the angular frequency.

[0051] In S44, the Mie scattering theory satisfies the following relationship: in, Scattering angle The intensity of scattered light at that location The incident light intensity where λ is the incident light wavelength and r is the observation distance. , Let be the Mie scattering amplitude function.

[0052] In S41, the multi-wavelength beam combining and shaping light adopts TTL modulation trigger mode, and the trigger signal is sent at a frequency of 50MHz through an arbitrary waveform signal generator; In S42, a photodiode and two avalanche photodiodes are used to capture pulse signals, and precise control of the light source is achieved by setting a trigger threshold.

[0053] In this embodiment, regarding the detection of small-particle sediment: An optimized method based on the first term Stokes full vector polarization information for fine detection is proposed, which can achieve real-time high-precision classification of different particle components.

[0054] The laser measurement process can be described by the laser equation: in, and Let r represent the parallel and perpendicular components of the scattered signal received by the polarization lidar at a distance r. The laser's emission power, and These are the system constants for the parallel component detection channel and the horizontal component detection channel, respectively. and These are the parallel and perpendicular components of the scattering coefficient, respectively. and These are the parallel and perpendicular components of the extinction coefficient, respectively.

[0055] Based on the different depolarization ratios of various aerosol particles, the composition of particle groups can be identified and classified by analyzing the depolarization ratio. The depolarization ratio is defined as: In the configuration of the polarization image sensor embedded in the polarization camera, four polarization filters are set on pixels at four different angles (0°, 45°, 90°, and 135°). Every four pixels form a computing unit, and the degree of polarization and polarization direction can be calculated by the relationship between the polarization filters in different orientations. In addition, to solve the crosstalk problem caused by faulty pixels and thus avoid polarization angle detection errors, a microlens array is used to improve detection accuracy. Each pixel has a microlens array, which helps to effectively guide light to the sensor and can be combined with a quarter-wave plate to obtain the circular polarization component.

[0056] Subsequently, Stokes vector characterization can be performed using data captured by a polarization camera sensor. The polarization state of light can be represented by a four-component Stokes vector, denoted as S, with components... , , and Specifically, Indicates the total intensity of light; 0° ) and 90° Intensity difference under linear polarization; 45° ) and 135° Intensity difference under linear polarization; Indicates right-handed circular polarization ( ) and left-handed circular polarization ( The intensity difference between states. Its mathematical expression is as follows: Even more conveniently, the linear polarization degree DoLP and polarization angle AoLP can be directly calculated from the Stokes components: DoLP and AoLP are important tools for understanding the interaction of light with different particles, providing crucial insights into the optical properties of materials. In particular, DoLP quantifies the degree of influence of particles on linearly polarized light, providing a basis for distinguishing the behavior of different materials under different light polarization states. AoLP, on the other hand, reveals the change in polarization direction after light interacts with particles. This parameter is crucial for understanding how particles alter the direction of polarized light, contributing to a deeper understanding of particle shape, orientation, and surface properties. By using these two parameters in combination, a more comprehensive understanding of the physical and optical properties of particles can be achieved, leading to more accurate identification and classification of materials.

[0057] Among the various representations of the polarization state of the target particles, some features show significant overlap. Therefore, to improve the accuracy and robustness of the classification process, a fusion method is adopted to combine complementary polarization information to optimize the representation of particle features, thereby further enhancing image contrast, improving classification accuracy, and providing a more reliable basis for particle identification.

[0058] First, identify and retain Overlapping features and parameter information between DoLP and AoLP: Next, to highlight the unique characteristics of each parameter, a subtraction process was used. This step emphasizes the unique properties of each parameter, thus more clearly showcasing the characteristics of the particles: Further image processing is performed to highlight the role of each parameter. The extracted information ensures... The unique details provided by DoLP and AoLP are presented more significantly and clearly in the final image: Finally, the processed values ​​were analyzed using a weighted average method. The DoLP and AoLP parameter information are combined to generate a fused image. Among them, the weighting coefficient , and The sum equals 1, that is + + =1 to ensure that the contribution of each parameter to the final output remains balanced. This fusion method combines particle brightness, polarization information, and structural details, providing a more comprehensive characterization. Therefore, by capturing a wider range of particle characteristics, it improves the accuracy of particle identification and classification, thereby achieving more reliable and robust analysis.

[0059] In addition, the FPGA data processing module performs parallel Fourier transforms on the 8-channel photoelectric signals and the 4-channel polarization signals to achieve real-time spectrum analysis of the scattered signals. in, f(t) is a frequency domain signal, and f(t) is a time domain analog signal. ω is the angular frequency.

[0060] Then, by combining Mie scattering theory to calculate the scattering characteristics of sediment particles, optical parameters such as particle size and refractive index are extracted. The scattering characteristics of particles of arbitrary shape are analyzed using the T-matrix method to determine the particle shape and size. in, The intensity of the scattered light at the scattering angle. The incident light intensity where λ is the incident light wavelength and r is the observation distance. , Let be the Mie scattering amplitude function.

[0061] In addition, the sheath fluid flow cell is equipped with a sample inlet, a sheath flow inlet, and an outlet. The internal flow channel of the sheath flow has a size of 2 mm × 0.2 mm, forming a collimated laminar flow within the imaging window. This design effectively maintains a consistent flow profile during the measurement process. The central sample flow is focused by hydrodynamics and enveloped by the sheath flow, ensuring that particles remain within the depth of focus of the imaging system while preventing them from adhering to the channel walls and reducing contamination of the optical components.

[0062] The optical design of the particulate matter online detection device based on multi-wavelength laser illumination and modulation trigger acquisition mode utilizes TO-packaged semiconductor laser diodes for continuous particulate matter monitoring. Simultaneously, a laser module with TTL modulation capability performs trigger modulation. To accurately capture the scattered signal, a photodiode (PD) is used to collect the intensity of the light pulses scattered by the particles. Data synchronization is achieved through an arbitrary waveform signal generator card, sending the trigger signal at a frequency of 50MHz to a second laser controlled by TTL. By setting a trigger threshold, uncertainties caused by particle size differences can be eliminated, enabling precise control of the second laser. Finally, two APD detectors convert the light signals scattered by the particles into electrical signals. A high-speed data acquisition card collects and outputs the scattered light signals received by the PD and APD modules to determine the polarization ratio of each particle, significantly improving the accuracy and reliability of particulate matter detection.

[0063] In addition, an FPGA-based architecture was adopted, which leverages the advantages of high-speed parallel processing to achieve synchronous acquisition and processing of multiple data channels, further improving the overall efficiency and accuracy of the data acquisition system. It is also equipped with an advanced analog-to-digital converter (AD converter) to convert the acquired analog signals into digital signals, thereby ensuring the accuracy and stability of the data.

[0064] In traditional particle size analysis, instruments rely on a single-wavelength Gaussian beam for illumination. However, this method is susceptible to interference from irregular particle shapes, affecting the accuracy of particle size measurement and failing to achieve composition-based single-particle classification. To overcome these limitations, an improved scheme employing multi-wavelength structured light illumination and a multi-classification intelligent algorithm is proposed to enhance detection accuracy and energy utilization efficiency.

[0065] Specifically, multi-wavelength laser beams are combined using dichroic mirror technology, ensuring the combined light uniformly illuminates the particles. During this process, some light rays scatter to both sides of the channel, exhibiting a certain divergence angle. Next, the focused beam is equally split into two beams: one maintains its original path, while the other is deflected by 90°. The beam following the original path passes through a filter and polarizer before being captured by a photodetector. The deflected beam undergoes special processing at the filter to remove different wavelengths of light (e.g., green and red light) before being collected by the photodetector as well. This differential detection method effectively eliminates noise and environmental interference, further improving detection accuracy.

[0066] In addition, an aspherical lens was specially designed to shape the Gaussian beam emitted by the semiconductor laser diode and convert it into a one-dimensional flat-top beam. This beam shaping technology greatly improves the uniformity of the beam, with a uniformity level of over 95%, thereby significantly improving the accuracy of particle detection and energy utilization efficiency.

[0067] To ultimately achieve high-precision real-time particle classification, deep learning algorithms are further introduced to fuse and identify the temporal features of dual-channel scattering signals. A one-dimensional convolutional neural network (1D-CNN) is used for the classification task. Through sliding window convolution operations, it can efficiently extract key local features from the temporal signal and achieve hierarchical abstraction and dimensionality reduction of features using pooling layers. This approach maintains the ability to model the overall pattern of the temporal signal while enhancing the capture of subtle local features, thus supporting the improvement of classification performance for one-dimensional temporal electrical pulse signals. Figure 2 As shown, Figure 2 a describes the 1D-CNN architecture. Figure 2 b illustrates the classification results of multi-wavelength illumination particles using a 1D-CNN architecture.

[0068] In addition, in the specific steps of large particle detection, in S5, the spectral image of the particle is obtained by acquiring the RGB image of the particle through an RGB camera, and reconstructing the RGB image into a hyperspectral image using a pre-built feature dictionary and sparse coding algorithm.

[0069] Specifically, in S5, the fusion analysis of polarization image and spectral image is as follows: the polarization image and the reconstructed hyperspectral image are mapped to the HSI color space respectively and then fused; and the optimized U-Net semantic segmentation model MSA-SE-Unet is used to extract features and segment the fused image.

[0070] In this embodiment, regarding the detection of large-particle sediment: A particle detection channel based on the fusion of polarization and spectral features was designed. A beam emitted from a backlight source is split by a beam splitter onto a glass slide covered with particles. The particles on the slide reflect the beam through the beam splitter, a lens, and then to a polarization camera and an RGB camera, achieving polarization and spectral imaging. By analyzing the polarization images acquired by the polarization camera, a comprehensive Stokes vector feature description of the particles can be performed. Simultaneously, the polarization information and morphological statistical information of the particles are extracted for identification, mitigating the impact of background noise, particle shape, and particle size on particle size and species detection.

[0071] After acquiring RGB images using an RGB camera, relevant artificial intelligence algorithms can be used to reconstruct hyperspectral images from these images. Hyperspectral features describe the spectral differences of particles, enabling high-precision particle classification and significantly reducing the cost of purchasing hyperspectral cameras. Fusing polarization and hyperspectral images using relevant algorithms can improve information dimensionality and recognition accuracy, eliminating biases from single-dimensional detection results, thereby achieving high-precision classification.

[0072] In this embodiment, a sparsity dictionary suitable for different scenarios was constructed. Hyperspectral images were reconstructed through RGB sampling, and principal component analysis was performed on the hyperspectral images to fuse them into a single image. This image was then fused with a polarization image mapped to the HSI color space, further improving the information dimensionality and recognition accuracy of the particles.

[0073] To further improve classification accuracy, the U-Net semantic segmentation model was introduced and optimized to propose the MSA-SE-Unet model. The optimizations include the addition of a multi-level receptive field module (MRM), a channel self-attention mechanism, and SqueezeandExcitationNetworks (SE-Net).

[0074] Hyperspectral reconstruction and sparse coding first determine the vectors constituting the hyperspectral feature dictionary and the RGB feature dictionary, then derive the weights for mapping the RGB feature dictionary to a given RGB image. Finally, the hyperspectral image is estimated using the corresponding feature dictionary. After reconstructing the hyperspectral image from the RGB channels, rich spectral information of particles can be obtained, enabling particle classification and recognition. Both the polarization image and the hyperspectral image are mapped to the HSI color space, thus facilitating fusion. By integrating spectral information into the polarization image, interference from background noise in the polarization image can be reduced, thereby improving the performance of particle fingerprint recognition.

[0075] By adding graph pooling and depooling layers, spectral spatial features and multi-resolution features are extracted, and a dual residual convolutional network model is used for training, which significantly improves the performance of particle classification.

[0076] More specifically, in S6, the machine learning model is a classification model trained based on labeled particle samples; preferably, the classification model is a shallow machine learning model or a convolutional neural network model; and it is trained based on the sample labeled data.

[0077] In S6, particle overlap is handled by combining a segmentation algorithm with edge detection technology, and particle size error is corrected by a calibration model. For example... Figure 3 As shown, Figure 3 From left to right, column a represents the input value, the actual value, and the predicted value of PS, Si, and PMMA particles in the classification model. Figure 3 b represents the detection accuracy, recall, and F1 score of the classification model.

[0078] In this embodiment, when processing the collected data, a shallow machine learning model or a convolutional neural network model is used to train and classify the particulate matter samples.

[0079] Linear kernel support vector machines (L-SVMs) demonstrate good classification performance in ternary classification tests of PSL, silica (SiO2), and carbon black particles due to their simplicity and low computational requirements. Similarly, deep learning methods also show good classification performance under dual-angle, dual-polarization conditions. However, when faced with more diverse particle signal classification tasks, deep learning-based methods exhibit higher stability and accuracy.

[0080] During the imaging detection process (especially in step S5, large particle imaging), particles may overlap, affecting the feature extraction and recognition of individual particles. Therefore, a segmentation algorithm combined with edge detection technology is used to solve the particle overlap problem.

[0081] The segmentation algorithm employs a cascaded segmentation architecture, progressively refining the detection results through multiple stages: Phase 1: Rapidly generate candidate particle regions and initially locate the possible locations of particles; The second stage involves refining the candidate regions, filtering out a large number of non-particle regions, and initially distinguishing between single-particle and overlapping particle regions. The third stage involves finely segmenting the overlapping regions and outputting the bounding box and key point information for each particle.

[0082] Based on the segmentation algorithm, edge detection technology is used for fine-tuning. Edge detection algorithms such as the Canny operator or Sobel operator are used to extract edge information of the particle region. The particle boundaries are further confirmed through edge continuity analysis. For adhering particles that cannot be completely segmented, secondary segmentation is performed using the edge information.

[0083] The combination of segmentation algorithm and edge detection achieves the following processing flow: For the input large-particle image, the overlapping particles are first quickly located and preliminarily segmented by the segmentation algorithm; then the segmentation results are verified by edge detection; for the detected discontinuous edges or suspected under-segmented regions, local optimization processing is performed, and finally the complete image region of each independent particle is output.

[0084] In addition, the detection method also includes step S7: collecting the raw water, diluent and sheath fluid waste generated in steps S1 to S5 into a sealed container for waste treatment.

[0085] In this embodiment, after the sediment sample is analyzed in steps S1 to S5, all types of waste liquid generated in each step are collected and treated uniformly to achieve environmentally friendly operation of the testing process. Specifically, independent waste liquid outlets are set up at each waste liquid generation point, and all outlets are connected to a central waste liquid collection pipeline through pipelines. The end of the central waste liquid pipeline is connected to a sealed waste liquid collection container. This sealed container is made of corrosion-resistant material (such as polyethylene), has good sealing performance, prevents waste liquid leakage and odor volatilization, and can be equipped with a liquid level monitoring device to monitor the waste liquid level in real time. When the liquid level reaches the warning line, an automatic prompt is issued to facilitate timely removal and treatment.

[0086] The raw water, diluent, sheath fluid, and other waste liquids generated during the testing process are collected in sealed containers for treatment, which avoids the direct discharge of wastewater containing silt particles into the natural environment. This meets the increasingly stringent environmental protection regulations of the country and is particularly suitable for hydropower projects in ecologically sensitive areas, such as hydropower development projects in the Yarlung Tsangpo River Basin. Example 2

[0087] Please refer to Figures 4-7 A detection system for implementing the dynamic detection method for sediment particles as described above, comprising: Sampling unit, comprising: A pressurized water sampling probe for use in constant motion sampling; The pressure-reducing sampling device connected in series downstream of the pressure-bearing water intake probe includes a pressure-resistant main body, a pressure regulating component, and a parameter detection component connected in sequence. The pressure-resistant main body is used to introduce high-pressure water flow, the pressure regulating component is used to reduce the pressure of the high-pressure water flow in stages, and the parameter detection component is used to collect the turbidity, pressure, and flow parameters of the water flow after pressure reduction in real time. A preprocessing unit, comprising: A diffusion suspension section and a static mixer are sequentially connected downstream of the parameter detection component to eliminate bubbles and homogenize. A particle size classification module connected downstream of a static mixer includes an inclined plate settling tank, a hydrocyclone, and an inertial microfluidic chip for separating a sample into a first sample and a second sample, which are output from a first outlet and a second outlet, respectively. The first branch detection unit has its input end connected to the first outlet. The first branch detection unit includes a first turbidity meter, a first flow meter, a first dilution device and a small particle detection device connected in series. The small particle detection device includes a sheath fluid focusing module, a multi-band laser source and a lateral polarization detection module, and a first data processing module based on Stokes vector analysis. The second branch detection unit has its input end connected to the second outlet. The second branch detection unit includes a second turbidity meter, a second flow meter, a second dilution device and a large particle detection device connected in series. The large particle detection device includes a polarization imaging module, a spectral imaging module and a second data processing module for image fusion analysis. The central data processing unit is connected to the first data processing module and the second data processing module respectively. The central data processing unit includes a machine learning model for particle identification and classification, which integrates the two detection data and outputs the sediment particle identification and classification results. The waste liquid treatment unit has its input terminals connected to the pretreatment unit, the first branch detection unit, and the second branch detection unit, respectively, and is used to collect and treat the generated waste liquid in a unified manner. The small particle detection device and / or large particle detection device are equipped with an easy-to-maintain optical module.

[0088] Specifically, the front end of the pressure water intake probe is equipped with a replaceable ceramic anti-wear guide cover, and its water intake hole has a diameter of 2-3mm.

[0089] In addition, a high-pressure shut-off valve and a microporous damping assembly are connected in series between the pressurized water intake probe and the bypass pipeline to configure a dual-valve isolation structure.

[0090] Furthermore, the diffusion suspension section is a vertically installed expansion pipe with a honeycomb rectifier grid inside and an automatic exhaust valve at the top; the static mixer is an SK type or SMV type online mixer.

[0091] Furthermore, the inclined plate settling tank has a set of parallel inclined plates with an inclination angle of 55°-60°, the hydrocyclone is a polyurethane hydrocyclone tube with a diameter of 25mm and a median cut particle size of 50μm, and the inertial microfluidic chip is a glass-silicon composite chip with an arc-shaped curved channel or a micro-pillar array.

[0092] Furthermore, the multi-band laser source of the small particle detection device is a three-band semiconductor laser with wavelengths of 405nm, 532nm, and 650nm, and the lateral polarization detection module is a linear polarization analyzer set in the 90° direction.

[0093] Furthermore, the polarization imaging module of the large particle detection device includes a switchable linear polarizer and a quarter-wave plate, and the spectral imaging module is an RGB camera, a liquid crystal tunable filter, or a pushbroom hyperspectral sensor, with a working wavelength range of 400-1000nm.

[0094] In this embodiment, the detection system of the dynamic detection method for sediment particles draws water from the high-pressure zone in front of the water intake pipe or the movable guide vane. Through the bypass loop, the pressure of the sample liquid is adapted, homogenized and controlled, and the optical module is maintenance-free, ensuring that the monitoring data accurately represents the sediment conditions of the power station and meets the reliability requirements for long-term industrial operation.

[0095] For water intake conditions before the movable guide vane with a design head of 1000m (corresponding to a pressure of 9.8MPa), a pressure-bearing water intake probe is used at the sampling port. The body of this probe is made of 316L stainless steel or titanium alloy, and the front end is equipped with a replaceable ceramic anti-wear guide cover. The diameter of the water intake hole is controlled at 2-3mm, and the sample liquid is drawn out by gravity flow using pressure difference. A high-pressure shut-off valve and a microporous damping component are connected in series between the pressure-bearing water intake probe and the bypass pipeline. Through multi-stage throttling, the inlet pressure is reduced from 9.8MPa to 0.3-0.6MPa in stages. This not only meets the normal pressure operating range of the downstream hydrocyclone and microfluidic chip, but also avoids cavitation bubbles induced by sudden changes in flow velocity during the decompression process. The entire high-pressure interface has been verified by hydrostatic testing and is equipped with a dual-valve isolation structure, which allows for safe replacement of the probe filter element without shutting down the system.

[0096] To ensure that the sand content and particle size distribution of the bypass sample are consistent with those of the original pipeline, and to eliminate air bubbles and uneven mixing, the water intake circuit is designed as a whole according to the following principles: The pressurized water sampling probe is positioned in the stable flow velocity zone of the pipe cross-section, with the sampling direction directly facing the incoming flow. The sampling speed is adjusted by regulating the back pressure valve to approximately match the mainstream flow velocity in the pipe, achieving isodynamic sampling and avoiding representativeness bias caused by the inertial deflection of large particles. After entering the loop, the sample liquid first flows through an upward-flowing diffusion suspension section. This diffusion suspension section is a vertically installed expanding pipe (with an inner diameter gradually increasing from 10mm to 50mm) with multiple layers of honeycomb rectifier grids, reducing the flow velocity to below 0.1m / s. Coarse particles with a density greater than water naturally sink and converge to the bottom sand collection hopper in this section, while the upward-flowing mainstream carries fine particles and remains suspended, achieving the dual functions of separating and storing coarse particles and uniformly transporting fine particles. This design also utilizes the bubble buoyancy characteristics in the low-velocity zone, causing bubbles with a diameter >50μm to float to the air vent and be automatically discharged, completely eliminating the interference of bubbles on optical detection.

[0097] Following the diffusion suspension section, the sample solution is subjected to forced turbulent mixing via an online static mixer (SK or SMV type). The mixing element continuously cuts and reverses the flow, ensuring a uniform radial concentration distribution of sediment particles with a relative standard deviation of <5%. A bypass circulation pump is installed downstream of the mixer (only activated when the loop resistance is insufficient), and frequency conversion is used to maintain a constant flow velocity (0.5-1.0 m / s) within the loop, preventing particle deposition in long, straight horizontal pipe sections.

[0098] In addition, an isokinetic sampling interface is reserved in the loop, which can be periodically connected to the laboratory comparison sampler. The online monitoring values ​​can be traced and calibrated by offline weighing and laser particle size analyzer data to ensure measurement accuracy during long-term operation.

[0099] Among them, such as Figure 4 As shown, the high-pressure, sediment-laden water first enters the pressure-resistant main body, which is made of high-strength stainless steel that is pressure-resistant and corrosion-resistant. A large-diameter interface at the top connects to the tail end of the hydropower station's water intake pressure pipeline, introducing the high-pressure water flow. A pressure monitoring instrument integrated in the middle of the pressure-resistant main body displays the internal water pressure in real time, providing data support for subsequent pressure reduction operations. Subsequently, the water flow is transported through pipelines to the pressure regulating component on the right. This component includes a high-precision valve and a regulating pump. The high-precision valve can precisely adjust the water flow rate according to the control system's instructions, gradually reducing the high-pressure water flow (typically 3-5 MPa) to a pressure range suitable for subsequent testing (0.1-0.3 MPa) in conjunction with the regulating pump. The regulating pump provides power for the water flow, ensuring a stable flow through the sampling device. The parameter detection component can collect key parameters such as turbidity, pressure, and flow rate in real time and transmit electrical signals to the control system, providing basic data for subsequent sediment testing and ensuring the stability and controllability of the sediment samples entering the subsequent testing stages.

[0100] The particle size classification module consists of a cascaded inclined plate settling tank, a hydrocyclone, and an inertial microfluidic chip. None of the three components have moving parts, and the precise particle size division is achieved by relying on the geometric constraints of the flow channel and fluid dynamics. The inclined plate settling tank is a coarse particle buffer and interception stage. Sample liquid enters the tank at a low flow rate. The interior is densely packed with parallel inclined plates at an angle of 55°-60°, reducing the settling height from the traditional meter level to the centimeter level. Based on the principle of shallow settling, coarse sand, rust, and fibers with a density greater than water (>200μm) settle rapidly into the collection hopper under gravity and are periodically discharged, preventing blockage of downstream microchannels. This stage, as a zero-energy protection unit, does not pursue precise classification; it only removes extremely large interfering materials.

[0101] The hydrocyclone is a 50μm main cutting stage. The coarsely filtered sample solution enters tangentially into a 25 mm diameter polyurethane hydrocyclone tube under a constant pressure (0.2-0.4 MPa). Through theoretical calculations and experimental pre-calibration, the overflow pipe insertion depth, underflow orifice diameter, and inlet flow velocity are coordinated and adjusted to the peak of separation sharpness, so that the volumetric median cutting particle size of the hydrocyclone is precisely locked at 50μm. The underflow continuously discharges the viscous phase rich in particles >50μm to the large particle channel; the overflow carries particles ≤50μm and a small amount of residual large particles into the fine-tuning stage.

[0102] The inertial microfluidic chip is a boundary sharpening stage. Overflow sample liquid is introduced into a glass-silicon composite chip etched with symmetrical arc-shaped curved channels or micropillar array barrier fields. In the inertial flow region with Reynolds number Re~100, particles are coupled with the inertial lift induced by the wall and the secondary flow of the Dean vortex: particles >50μm, due to their longer relaxation time, migrate inertially to the vicinity of the inner wall of the channel and are introduced into the large particle channel through a dedicated sidewall bifurcation; particles ≤50μm remain focused in the low-shear zone at the center of the channel and enter the fine particle channel with the mainstream. This stage reduces the classification misclassification rate at the 50μm boundary to below 5%, achieving near-ideal narrow-band sieving.

[0103] After the above three-stage processing, the original broadly distributed sediment sample is transformed in real time into two representative samples with uniform particle size, providing a clear particle size range for subsequent optical detection.

[0104] In addition, for the optical windows and light sources in optical sensors (laser scattering cell of small particle detection device and polarization hyperspectral imaging channel of large particle detection device) that are easily contaminated and damaged, a modular quick-change plug-in structure and an active protective air curtain are used for dual protection: Both the scattering cell and the imaging flow cell have independently sealed lens assemblies for their light transmission windows, which are sealed to the flow channel body by O-rings on the end faces. A clamp-type quick-release mechanism is installed around the lens assembly; operators can remove the entire window module by simply rotating it 90° without special tools, replace the lens, and then reinsert and lock it in place—the entire process takes less than 2 minutes. The windows are made of synthetic sapphire or high-hardness coated quartz, which is wear-resistant and scratch-resistant, with a lifespan more than 5 times longer than ordinary glass.

[0105] In each optical window, on the side facing the sample liquid, an integrated micro-jet self-cleaning nozzle periodically rinses the mirror surface at an angle with high-pressure clean water (taken from the circuit itself and finely filtered through 0.45μm) to remove adhering sticky particles. Simultaneously, a laminar flow air curtain generator is installed on the outside of the window (air side) to continuously blow dry, clean air, preventing water vapor condensation or external dust contamination of the mirror's outer surface. The air source is taken from the power station's instrument ventilation system and supplied after precise pressure regulation and filtration.

[0106] In addition, both the laser diode and the hyperspectral illumination source adopt a redundant hot-swap design, automatically and seamlessly switching to the backup source when the main source fails. The light source module is packaged as a standard plug-in card, supporting hot-swapping; after replacement, no manual optical calibration is required, relying on the built-in reference optical path and standard scattering body to automatically complete the self-calibration of light intensity and wavelength, with a monitoring recovery time of less than 10 seconds.

[0107] Through the aforementioned high-pressure water intake adaptation, loop representativeness assurance, and easy-to-maintain optical module design, the laboratory-level multi-stage particle size sorting and polarization-spectral fusion monitoring technology has been successfully applied to the real-world operating conditions of a 1000m head turbine, facilitating the transition from technical feasibility to engineering reliability. In exemplary operation, the system maintains good detection stability and ease of maintenance, providing support for online monitoring of turbine sediment wear.

[0108] To achieve continuous and dynamic sorting of sediment particles in water with a critical size of 50 μm, a multi-stage coupled inertial centrifugal classification system can be constructed. This system directs particles larger than 50 μm into the heavy phase channel, while particles smaller than 50 μm enter the light phase channel. An inclined plate settling tank is used to pre-intercept most of the coarse sand and fibers >200 μm using shallow settling principles with zero energy consumption, preventing downstream channel blockage. The coarsely filtered sample then enters a hydrocyclone. By adjusting the inlet pressure and underflow diameter, the hydrocyclone separates and cuts the particle size (…). The hydrocyclone is precisely set at 50 μm. Particles >50 μm are enriched by the underflow and continuously discharged into the large particle collection channel, while the overflow carries finer particles <50 μm into the next stage. To further sharpen the classification boundaries, the overflow liquid is introduced into an inertial microfluidic chip. Utilizing the inertial lift and Dean vortex secondary flow generated by the fluid in the curved microchannel or obstacle array, the remaining small number of >50 μm particles are deflected by inertia and move along the wall, entering the large particle channel through a dedicated sidewall outlet. Meanwhile, <50 μm particles remain focused at the center of the main channel and enter the fine particle collection channel. The entire device has no moving parts and can operate continuously online, achieving high-throughput, narrow-band sorting based on a 50 μm particle size threshold, providing representative samples with a single particle size for subsequent laser scattering or holographic imaging monitoring.

[0109] like Figure 4 As shown, the small particle detection device employs multispectral / polarization scattering analysis. For suspended particles with a diameter ≤50 μm, laser scattering is used, based on a joint model of Rayleigh scattering (particle size d << λ) and Mie scattering (d ≈ λ to d ≤50λ). By selecting 405 nm, 532 nm, and 650 nm semiconductor lasers as the detection source, monochromatic light is sequentially and perpendicularly incident into the flow cell to ensure: For submicron-sized wear particles (d ~ 0.1-1 μm), Rayleigh scattering is dominant, and the intensity of the scattered light is related to d.6 It is directly proportional to the particle size distribution and can be highly sensitively inverted to determine the number concentration of fine particles. For particles of 1-50 μm, Mie scattering is dominant, and the ratio of forward and side scattered light intensity is a monotonic function of particle size. Combined with a multi-wavelength scattering matrix, the particle size distribution and volume concentration can be calculated.

[0110] Multispectral scattering analysis: By utilizing the differences in the imaginary part of the refractive index of particles at different wavelengths, the dispersion curves of particle materials (such as quartz, metal wear debris, and clay minerals) can be derived to achieve coarse classification of components. Polarization scattering analysis: A linear polarization analyzer is set up in the 90° lateral detection channel to measure the depolarization ratio. Non-spherical particles (such as sheet-like mica and elongated grinding debris) exhibit significant depolarization due to anisotropic scattering, which can effectively distinguish them from spherical sediment particles.

[0111] Following the applicability of Mie scattering theory in the particle size ≤ 50 times the wavelength range (taking 650 nm wavelength as an example, 50 times the wavelength = 32.5 μm, close to the 50 μm boundary; if the near-infrared 1064 nm wavelength is used, 50 times the wavelength can reach 53.2 μm, completely covering the ≤ 50 μm range), the theory is complete and has been verified by simulation.

[0112] like Figure 5 As shown, the large particle detection device employs polarization-hyperspectral fusion imaging. For coarse particles >50 μm, a high-speed area array camera combined with a telecentric microscopic optical system is used for flow imaging. Large particles exhibit stable posture and clear edges under fluid shearing, and traditional methods can already achieve the statistical analysis of shape factors (aspect ratio, convexity) and particle count.

[0113] Specifically, switchable linear polarizers and quarter-wave plates are placed in the illumination path to acquire particle reflection / transmission images under 0°, 45°, 90°, and circularly polarized light, respectively. The polarization anisotropy parameters of the particle surface are extracted by calculating the Mueller matrix elements. Metal abrasive shavings (iron-manganese oxides) exhibit strong polarization-preserving properties, while non-metallic minerals such as quartz sand and feldspar show significant depolarization, thus enabling rapid identification of material properties.

[0114] Using liquid crystal tunable filters (LCTF) or pushbroom hyperspectral sensors, continuous spectral cubes of particles are acquired in the 400-1000 nm visible-near infrared band with a step size of 5-10 nm. The spectral curve of each particle implicitly contains its molecular vibrational and electronic transition fingerprints: for example, chlorite has a characteristic iron absorption band near 900 nm, and chalcopyrite exhibits a double absorption valley at 550 nm and 800 nm. Through spectral angle mapping (SAM) and support vector machine classification, the mineral phase and wear source of the particles can be identified simultaneously.

[0115] Polarization-hyperspectral fusion imaging elevates traditional two-dimensional morphology measurement to a four-dimensional detection dimension of morphology and material, effectively filling the technological gap in existing turbine sediment monitoring, which can only count concentrations and cannot trace the source of wear.

[0116] In the above structure, the wide-distribution particles are divided into two optimal optical methods with 50 μm as the boundary. The scattering method is good at statistically analyzing a large number of subvisible fine particles, while the imaging method is good at resolving the morphology of millimeter-scale coarse particles. The two complement each other, and there are no blind spots in the entire particle size range.

[0117] By introducing polarization / multispectral features into the scattering module and polarization / hyperspectral features into the imaging module, an intelligent monitoring terminal capable of answering questions about the material of particles and the origin of wear can be effectively constructed.

[0118] The grading unit relies entirely on the channel geometry and fluid inertia, with no rotating or vibrating parts, and can be embedded in the turbine bypass sampling circuit for a long time without maintenance.

[0119] This solution demonstrated good sorting and identification performance in exemplary laboratory prototype testing, providing a new technical approach for hydraulic machinery condition monitoring.

[0120] The aforementioned structure aims to address the common technical challenges in real-time monitoring of wear particles in water turbine guide vanes, namely, the large particle size range, the difficulty of optical methods in covering the entire range, and the inability of traditional imaging to identify components beyond counting. It employs an integrated architecture of graded preprocessing and dual-mode optical detection, using 50 μm as the critical size. Through a series of three-stage non-powered sorting units—sedimentation, swirling, and inertial shearing—the original water sample is continuously separated into two independent streams: one rich in coarse particles >50 μm and the other rich in fine particles ≤50 μm. These streams are then adapted to imaging and scattering optical detection modules, respectively. Furthermore, a polarization / hyperspectral dimension is introduced to achieve integrated identification of particle morphology and material composition.

[0121] In the embodiments disclosed in this application, the terms "installation," "connection," "linking," and "fixing" should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; "linking" can be a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in the embodiments disclosed in this application according to the specific circumstances.

[0122] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for dynamic detection of sediment particles, characterized in that, Includes the following steps: S1. Obtain high-pressure sediment samples, measure them, and depressurize them to form the sample to be tested; S2. The sample to be tested is sieved according to the preset particle size to form a first sample with a first particle size range and a second sample with a second particle size range, wherein the first particle size range is smaller than the second particle size range. S3. Turbidity of the first sample and the second sample is detected by the transmission scattering ratio method, and they are diluted according to the detection results. S4. The first sample is processed by sheath fluid focusing to form a single-row particle stream, and then irradiated with multi-wavelength beam shaping light. The scattering signal of the particles is collected by a multi-channel polarization detector, and the particle morphology and composition information of the first sample is obtained based on Stokes full vector parameter analysis. S5. The second sample is subjected to imaging detection processing. The polarization image and spectral image of the particles are obtained by using a backlight source. The polarization image and spectral image are fused and analyzed to obtain the particle morphology and composition information of the second sample. S6. Integrate the detection data of the first sample and the second sample, identify and classify them through a machine learning model, and output the detection results of sediment particles.

2. The method for dynamic detection of sediment particles according to claim 1, characterized in that, In S2, the preset particle size is 50 μm. The silt smaller than 50 μm is the first sample, and the silt larger than 50 μm is the second sample.

3. The method for dynamic detection of sediment particles according to claim 1, characterized in that, In S3, the transmission-to-scattering ratio method includes both transmission and scattering methods; In the aforementioned transmission method, the Lambert-Beer law is satisfied: ,in The intensity of transmitted light. Where α is the incident light intensity, β is the absorption coefficient, and L is the scattering coefficient; In the scattering method described above, the intensity of scattered light at a 90° angle is... Satisfying the relation: Where K is a coefficient related to the instrument's geometry, N is the number of particles per unit volume, and V is the volume of the particles. The incident light wavelength; under certain conditions, when the particle size and the incident light wavelength are relatively fixed, If it is a constant, then It is directly proportional to N, that is, directly proportional to turbidity; The turbidity value is calculated based on the data obtained by the transmission method and the scattering method, and then diluted by pumping in clean water, with the volume of clean water pumped in recorded.

4. The method for dynamic detection of sediment particles according to claim 1, characterized in that, S4 includes the following steps: S41. The first sample forms a single-row particle stream and is irradiated with multi-wavelength beam-shaping light. The multi-wavelength beam-shaping light adopts TTL modulation trigger mode and sends a trigger signal at a frequency of 50MHz through an arbitrary waveform signal generator. S42. Scattered signals are acquired in parallel through the four polarization channels of the Stokes full vector sensor and the eight channels of the photodetector. The four polarization channels of the Stokes full vector sensor are polarization filters set at angles of 0°, 45°, 90° and 135° on different pixels. Every four pixels form a calculation unit. A microlens array is set above each pixel and combined with a quarter-wave plate to obtain the circular polarization component. S43. The acquired signal is amplified and converted from analog to digital. Based on FPGA, 8-channel parallel acquisition and 4-polarization channel Stokes full vector acquisition are implemented. The analog signal is then processed by Fourier transform to realize real-time spectrum analysis of the scattered signal. S44. Calculate the scattering characteristics of sediment particles using Mie scattering theory, extract the particle size and refractive index optical parameters, and analyze the scattering characteristics of particles of arbitrary shape using the T-matrix method to determine the particle shape and size. The Mie scattering theory satisfies the following relationship: in, Scattering angle The intensity of scattered light at that location The incident light intensity where λ is the incident light wavelength and r is the observation distance. , Let be the Mie scattering amplitude function.

5. The method for dynamic detection of sediment particles according to claim 4, characterized in that, In step S43, the data captured by the polarization camera sensor is represented by a Stokes vector, and the Stokes vector S satisfies the following relationship: in, The total intensity of light, 0° ) and 90° Intensity difference under linear polarization state, 45° ) and 135° Intensity difference under linear polarization; Right-handed circular polarization ( ) and left-handed circular polarization ( The intensity difference between states; The linear polarization degree DoLP and polarization angle AoLP are calculated using the following formulas: A fusion method is used to fuse features from S0, DoLP, and AoLP: Identify and preserve overlapping features among S0, DoLP, and AoLP: ; Using subtraction to highlight the unique characteristics of each parameter: Synthesize and fuse images using a weighted average method: ,in + + =1.

6. The method for dynamic detection of sediment particles according to claim 5, characterized in that, In step S43, a parallel Fourier transform is performed on the 8-channel photoelectric signal and the 4-channel polarization signal using the following formula: in, f(t) is a frequency domain signal, and f(t) is a time domain analog signal. ω is the angular frequency.

7. The method for dynamic detection of sediment particles according to claim 1, characterized in that, In step S5, the spectral image of the particle is obtained by acquiring the RGB image of the particle through an RGB camera, and reconstructing the RGB image into a hyperspectral image using a pre-built feature dictionary and sparse coding algorithm. The fusion analysis of the polarization image and the spectral image is performed by mapping the polarization image and the reconstructed hyperspectral image to the HSI color space respectively and then fusing them. The MSA-SE-Unet semantic segmentation model is then used to extract features and segment the fused image.

8. The method for dynamic detection of sediment particles according to claim 1, characterized in that, In step S6, the machine learning model is a classification model trained based on labeled particle samples; Preferably, the classification model is a shallow machine learning model or a convolutional neural network model; And training is performed based on labeled sample data; Particle overlap is handled by combining segmentation algorithms with edge detection technology, and particle size error is corrected by calibration models.

9. The method for dynamic detection of sediment particles according to claim 1, characterized in that, It also includes S7: collecting the raw water, diluent and sheath fluid waste liquid generated in steps S1 to S5 into a sealed container for waste liquid treatment.

10. A detection system for implementing the dynamic detection method for sediment particles according to any one of claims 1-9, characterized in that, include: Sampling unit, comprising: A pressurized water sampling probe for use in constant motion sampling; The pressure-reducing sampling device connected in series downstream of the pressure-bearing water intake probe includes a pressure-resistant main body, a pressure regulating component, and a parameter detection component connected in sequence. The pressure-resistant main body is used to introduce high-pressure water flow, the pressure regulating component is used to reduce the pressure of the high-pressure water flow in stages, and the parameter detection component is used to collect the turbidity, pressure, and flow parameters of the water flow after pressure reduction in real time. A preprocessing unit, comprising: A diffusion suspension section and a static mixer are sequentially connected downstream of the parameter detection component to eliminate bubbles and homogenize. A particle size classification module connected downstream of the static mixer, the particle size classification module including an inclined plate settling tank, a hydrocyclone and an inertial microfluidic chip, is used to separate the sample into a first sample and a second sample, and output them from a first outlet and a second outlet, respectively; The first branch detection unit has its input end connected to the first outlet. The first branch detection unit includes a first turbidity meter, a first flow meter, a first dilution device and a small particle detection device connected in series. The small particle detection device includes a sheath fluid focusing module, a multi-band laser source and a lateral polarization detection module, and a first data processing module based on Stokes vector analysis. The second branch detection unit has its input end connected to the second outlet. The second branch detection unit includes a second turbidity meter, a second flow meter, a second dilution device and a large particle detection device connected in series. The large particle detection device includes a polarization imaging module, a spectral imaging module and a second data processing module for image fusion analysis. A central data processing unit is connected to the first data processing module and the second data processing module respectively. The central data processing unit includes a machine learning model for particle identification and classification, which integrates the two detection data and outputs the sediment particle identification and classification results. The waste liquid treatment unit has its input terminals connected to the pretreatment unit, the first branch detection unit, and the second branch detection unit, respectively, for the unified collection and treatment of the generated waste liquid; The small particle detection device and / or the large particle detection device are equipped with an easy-to-maintain optical module.

Images