Image sensor for online detection of oil quality

Abstract

The embodiment of the invention provides an image sensor for online detection of oil quality. The image sensor comprises a sensor body, a multi-view imaging unit, a sound wave separation unit, a dynamic focusing unit and an analysis unit. The sensor body includes a multi-view observation chamber in the shape of a polygonal pyramid. The multi-view-angle imaging unit obtains a top view image and a side view angle image of the abrasive particles through the main imaging assembly and the auxiliary imaging assembly. The sound wave separation unit comprises an ultrasonic transmitter array, a phase control circuit and a sound wave focusing lens, and the gathered abrasive particles are physically separated by generating focused sound waves. The dynamic focusing unit comprises a liquid lens assembly and a focal length control circuit, and clear imaging of different depth levels is achieved by adjusting the optical focal length of a liquid lens. The analysis unit comprises an aggregation detection network, a separation planning network and a feature extraction network and is used for recognizing abrasive particle aggregation, generating a separation scheme and extracting abrasive particle features. According to the invention, the problem of inaccurate abrasive particle counting is solved.

Description

Technical Field

[0001] The present application relates to the technical field of oil quality detection, and in particular to an image sensor for online detection of oil quality. Background Art

[0002] In the field of condition monitoring and fault diagnosis of modern industrial equipment, online oil quality detection technology plays an important role. By analyzing the characteristic parameters of abrasive particles in the oil, such as the number, size, shape and material, it can effectively evaluate the wear state of the equipment and provide key information for equipment maintenance and fault prediction. At present, the existing online oil quality detection image sensor adopts a technical route that combines magnetic adsorption and image acquisition, such as Figure 1 As shown, the ferromagnetic abrasive particles in the oil are adsorbed onto the surface of the transparent protective plate by a magnetic adsorption component, and then a two-dimensional image of the abrasive particles is obtained by an image acquisition component, and then the abrasive particles are identified and feature analyzed by an image processing unit.

[0003] However, in actual industrial applications, abrasive particles in oil are often not individually dispersed, but often aggregate, overlap, and obscure each other. When multiple abrasive particles are simultaneously attracted to the surface of a transparent protective plate by a magnetic field, they tend to form clusters due to the uneven distribution of the magnetic field and the interaction between the abrasive particles. This aggregation phenomenon is more pronounced with small abrasive particles, as they have a larger surface area to volume ratio and higher surface energy, making them more susceptible to mutual adhesion.

[0004] When abrasive particles aggregate, existing image segmentation and recognition methods face significant challenges. Traditional edge detection and contour extraction algorithms struggle to accurately identify the individual boundaries of abrasive particles within a cluster, leading to inaccurate particle counts. Furthermore, because the true morphological features of abrasive particles in an aggregated state are partially obscured or deformed, key characteristic parameters such as the shape, size, and surface texture of individual particles cannot be accurately extracted, hindering accurate assessment of the wear status of equipment.

[0005] Furthermore, when abrasive particles overlap in the thickness direction, particles in the lower layer may be completely or partially obscured by particles in the upper layer. Traditional 2D image acquisition methods can only capture information about visible particles on the surface, failing to capture the complete topography of obscured particles, further exacerbating inaccurate detection results. This information loss is particularly pronounced in oil samples with high abrasive concentrations, severely impacting the reliability and practicality of the detection system.

[0006] Therefore, it is necessary to propose a new technical solution to solve the above technical problems. Summary of the Invention

[0007] The purpose of the embodiments of the present application is to provide an image sensor for online detection of oil quality, aiming to solve the technical problem of inaccurate wear particle counting in existing image sensors for online detection of oil quality.

[0008] An embodiment of the present application provides an image sensor for online detection of oil quality, comprising: a sensor body, the sensor body comprising a multi-view observation chamber, the multi-view observation chamber being in the shape of a multi-faceted pyramid, the multi-view observation chamber being provided with a multi-layer optical window assembly, the multi-layer optical window assembly comprising a main observation window located at the center of the top of the chamber and a plurality of auxiliary observation windows radially distributed around the main observation window, the plurality of auxiliary observation windows being inclined toward the inside of the chamber; a multi-view imaging unit comprising a main imaging assembly and an auxiliary imaging assembly, the main imaging assembly being used to obtain a top-view image of the abrasive particles through the main observation window, the auxiliary imaging assembly being used to obtain a side-view angle image of the abrasive particles through the auxiliary observation window; an acoustic wave separation unit comprising an ultrasonic transmitter array, a phase control circuit and an acoustic wave focusing lens, the ultrasonic transmitter array being located at the bottom of the detection chamber, the phase The control circuit is used to calculate the phase delay value of each transducer according to the target focusing position and generate a multi-channel driving signal with a specific phase relationship. The acoustic wave focusing lens is used to perform a phase modulation operation on the incident acoustic wave; the dynamic focusing unit includes a liquid lens assembly and a focal length control circuit. The liquid lens assembly contains a double-layer liquid medium of conductive liquid and insulating oil, and a transparent electrode surrounds the lens cavity to form an electric field control area. The focal length control circuit is used to apply a variable voltage to the transparent electrode to generate a change in electric field intensity to adjust the optical focal length of the liquid lens; and the analysis unit includes an aggregation detection network, a separation planning network and a feature extraction network. The aggregation detection network is used to perform a feature mapping extraction operation on the input image to identify the wear particle aggregation phenomenon. The separation planning network is used to generate an acoustic wave separation execution plan based on the aggregation detection results. The feature extraction network is used to perform deep feature analysis on the separated independent wear particles.

[0009] In the above image sensor, the circumferential angle interval between every two adjacent auxiliary windows is 45 degrees, and the angle between the window plane of each auxiliary observation window and the top surface of the chamber is 30 degrees.

[0010] In the above-mentioned image sensor, the multi-view imaging unit also includes a field of view calculation unit and a stereo vision processing module. The field of view calculation unit is used to calculate the field of view boundary data of each window, and the stereo vision processing module is used to allocate the image data to the corresponding processing channel according to the view identification.

[0011] In the above-mentioned image sensor, the acoustic wave separation unit also includes an amplitude modulator and an acoustic field distribution controller. The amplitude modulator is used to calculate the required acoustic radiation force based on the abrasive aggregation information and convert it into a transducer driving power requirement. The acoustic field distribution controller is used to control the acoustic wave focus to move within the detection area along a predetermined path.

[0012] In the above-mentioned image sensor, the dynamic focusing unit also includes a focus position detector and a multi-layer image acquisition module. The focus position detector is used to emit a reference laser beam to the detection area and calculate the focal distance based on the reflected laser. The multi-layer image acquisition module is used to continuously collect image data at different focal length positions and construct a three-dimensional image data stack.

[0013] The above-mentioned image sensor also includes a spectral analysis unit, which includes a multi-wavelength LED array and a spectral filter group. The multi-wavelength LED array is used to activate light sources of different wavelengths in sequence according to a timing configuration table, and the spectral filter group is used to receive a wavelength synchronization signal and rotate to a predetermined angle to achieve wavelength switching.

[0014] In the above-mentioned image sensor, the aggregation detection network includes a spatial density analysis module, a shape abnormality evaluation module and a boundary continuity detection module. The spatial density analysis module is used to count the number of abrasive feature points in the local area to calculate the density value. The shape abnormality evaluation module is used to calculate the geometric characteristic parameters of the abrasive contour and compare them with the standard template. The boundary continuity detection module is used to analyze the breakpoints and curvature changes of the contour to determine the degree of aggregation tightness.

[0015] In the above-mentioned image sensor, the separation planning network includes a frequency selection module, a power allocation calculation module and an action time planning module. The frequency selection module is used to calculate the optimal acoustic wave frequency according to the abrasive size distribution, the power allocation calculation module is used to determine the required acoustic wave power level according to the degree of aggregation, and the action time planning module is used to calculate the duration of the acoustic wave action and divide it into multiple short pulse periods.

[0016] In the above-mentioned image sensor, the feature extraction network includes a boundary detection convolution layer, a contour refinement module and a multi-scale feature fusion module. The boundary detection convolution layer is used to detect the irregular shape contour of the wear particle using a deformable convolution kernel, the contour refinement module is used to perform sub-pixel level processing on the boundary contour, and the multi-scale feature fusion module is used to process image data of different resolutions and extract detail features and shape features.

[0017] The above-mentioned image sensor also includes an adaptive learning unit, which includes an aggregation pattern recognition module, a parameter optimization module and a knowledge base update module. The aggregation pattern recognition module is used to establish a characteristic data index table of the abrasive aggregation pattern, the parameter optimization module is used to dynamically adjust the acoustic wave parameters according to the separation effect, and the knowledge base update module is used to identify high-quality processing cases and update the database.

[0018] The image sensor for online detection of oil quality provided in this application effectively solves the technical problem of inaccurate wear particle counting in the prior art through the coordinated cooperation of a multi-perspective observation chamber, a multi-perspective imaging unit, an acoustic wave separation unit, a dynamic focusing unit and an analysis unit.

[0019] The image sensor of the present application adopts a multi-angle observation chamber in the shape of a multi-cone with a multi-layer optical window assembly. The main observation window obtains a complete top-view image of the abrasive particles, and multiple auxiliary observation windows observe the abrasive particles from different side angles, which can obtain three-dimensional morphological information of the abrasive particles from multiple dimensions. This multi-angle imaging method overcomes the limitations of traditional single-angle observation. When the abrasive particles occlude each other, the areas blocked by certain angles are still visible in other angles. Through the information fusion of multi-angle images, the complete outline of the blocked abrasive particles can be reconstructed, thereby obtaining the comprehensive morphological characteristics of the abrasive particles. The stereoscopic vision processing module of the multi-angle imaging unit can determine the position coordinates of the abrasive particles in three-dimensional space by calculating the disparity values ​​of feature points in images with different angles, realize the spatial positioning and hierarchical recognition of overlapping abrasive particles, and solve the problem that traditional two-dimensional images cannot obtain depth information.

[0020] The acoustic wave separation unit generates a controllable acoustic wave phase distribution through an ultrasonic transmitter array. A phase control circuit calculates the phase delay of each transducer based on the location of abrasive particle accumulation, enabling constructive interference of the acoustic wave energy at the target location and focusing the acoustic wave. An acoustic focusing lens performs phase modulation on the incident acoustic wave, converting the plane acoustic wave into a focused acoustic wave with a specific wavefront shape. This generates a concentrated acoustic radiation force in the area where the abrasive particles are concentrated. When the ultrasonic wave encounters the abrasive particles, acoustic scattering occurs. The scattered sound waves and the incident sound wave form an uneven sound pressure distribution around the abrasive particles, generating a net radiation force on the abrasive particles. This radiation force overcomes the adhesion between the abrasive particles and achieves physical separation of the aggregated particles. An amplitude modulator adjusts the driving power of the transducer based on the strength of the inter-particle bonding to ensure sufficient separation force. An acoustic field distribution controller controls the trajectory of the acoustic wave focus within the detection area, enabling selective processing of different accumulation areas, effectively resolving the problem of inaccurate counting caused by abrasive particle accumulation.

[0021] The dynamic focusing unit uses a liquid lens assembly to achieve clear imaging of wear particles at different depths. The liquid lens consists of a dual-layer dielectric medium consisting of a conductive liquid and an insulating oil. A focus control circuit applies a variable voltage to a transparent electrode to change the electric field strength. This change in electric field intensity modulates the wetting properties of the conductive liquid surface, thereby changing the geometry of the liquid interface and the optical focal length of the lens. By continuously adjusting the focal length, the multi-layer image acquisition module can capture clear images of wear particles at different depths, constructing a three-dimensional image data stack containing information from multiple depths. The focus position detector dynamically adjusts the focus position by measuring the reflection of the laser beam, ensuring optimal focus at each depth. This dynamic focusing capability ensures that each individual wear particle within overlapping wear particles is clearly imaged at the corresponding depth. The overlapping wear particle recognition module analyzes the pixel intensity distribution at each depth level to identify individual wear particles at different depths and reconstruct the complete outline of obscured wear particles, effectively solving the problem of information loss in overlapping wear particle detection.

[0022] The analysis unit's cluster detection network accurately identifies wear particle clustering through deep learning processing. The spatial density analysis module calculates the density distribution of wear particle feature points within a local area. The shape anomaly assessment module calculates the degree to which the wear particle contour deviates from the standard shape. The boundary continuity detection module analyzes contour breaks and curvature changes. These three modules work together to comprehensively assess the type and compactness of wear particle clustering. The separation planning network generates a targeted separation plan based on the cluster detection results. The frequency selection module determines the optimal acoustic frequency based on the wear particle size distribution. The power allocation calculation module calculates the required acoustic power based on the degree of clustering. The action time planning module divides the continuous action time into multiple pulse periods to avoid over-action, ensuring that the separation parameters are highly aligned with the actual situation. The feature extraction network performs deep feature analysis on the separated individual wear particles. The boundary detection convolutional layer uses a deformable convolution kernel to adapt to the irregular shape of the wear particles. The contour refinement module improves the accuracy of boundary recognition through sub-pixel processing. The multi-scale feature fusion module extracts both detailed and overall features. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a schematic diagram of a traditional image sensor used for online detection of oil quality.

[0024] Figure 2 Schematic diagram of an image sensor for online detection of oil quality provided in an embodiment of the present application.

[0025] Figure 3 This is a block diagram of an image sensor for online detection of oil quality provided in an embodiment of the present application.

[0026] Figure 4 yes Figure 3The block diagram of the multi-view imaging unit of the image sensor for online detection of oil quality is shown.

[0027] Figure 5 yes Figure 3 The block diagram of the spectral analysis unit of the image sensor used for online detection of oil quality is shown.

[0028] Figure 6 yes Figure 3 The block diagram of the dynamic focusing unit of the image sensor used for online detection of oil quality is shown.

[0029] Figure 7 yes Figure 3 The block diagram of the acoustic wave separation unit of the image sensor for online detection of oil quality is shown.

[0030] Figure 8 yes Figure 3 The block diagram of the analysis unit of the image sensor used for online detection of oil quality is shown.

[0031] Figure 9 yes Figure 3 The block diagram of the real-time monitoring feedback unit of the image sensor for online detection of oil quality is shown.

[0032] Figure 10 yes Figure 3 The block diagram of the adaptive learning unit of the image sensor for online detection of oil quality is shown. DETAILED DESCRIPTION

[0033] The specific implementation methods of this application are described in detail below with reference to the accompanying drawings.

[0034] The terms "first", "second" and similar words do not indicate any order, quantity or importance, but are only used to distinguish different technical features. The term "plurality" and similar words mean two or more, unless otherwise expressly limited.

[0035] The embodiments of the present application may be combined with each other.

[0036] like Figure 1As shown, existing image sensors for online oil quality monitoring use a magnetic adsorption component to attract ferromagnetic abrasive particles to the lower surface of a transparent protective plate. The image acquisition component then captures images of the particles, and the image processing unit and feature analysis unit in the processing circuit identify and analyze the particles. However, in actual industrial applications, abrasive particles in oil often do not exist in isolation, but often aggregate, overlap, and occlude each other. When multiple abrasive particles are simultaneously attracted to the surface of a transparent protective plate by a magnetic field, they tend to form clusters due to the uneven distribution of the magnetic field and the interaction forces between the particles. This aggregation phenomenon is more pronounced with small abrasive particles, as they have a larger surface area-to-volume ratio, higher surface energy, and are more likely to adhere to each other. When abrasive particles aggregate, traditional image segmentation methods struggle to accurately identify the boundaries of individual abrasive particles within the cluster, resulting in inaccurate particle counting. Furthermore, because the morphological features of the aggregated abrasive particles are obscured or distorted, the true shape, size, and surface texture of individual abrasive particles cannot be accurately extracted, hindering the assessment of the wear status of the equipment. In addition, when the abrasive particles overlap in the thickness direction, the lower layer of abrasive particles are completely or partially blocked by the upper layer of abrasive particles. The traditional two-dimensional image acquisition method cannot obtain complete information of the blocked abrasive particles, further exacerbating the inaccuracy of the detection results.

[0037] In order to solve the above technical problems, this application constructs an image sensor for online detection of oil quality based on the original image sensor, such as Figure 3 As shown, the image sensor realizes effective separation and accurate detection of aggregated abrasive particles by integrating a multi-view imaging unit, a spectral analysis unit, a dynamic focusing unit, an acoustic wave separation unit and an analysis unit.

[0038] like Figure 2 As shown, the sensor body of this image sensor is an improvement on the existing detection chamber, featuring a multi-view observation chamber. This chamber adopts a three-dimensional polyhedron configuration, expanding its overall shape from a simple cylinder to a multi-faceted pyramid. The chamber incorporates a multi-layered optical window assembly, comprising a primary observation window located at the center of the chamber's top and multiple (e.g., eight) auxiliary observation windows radially distributed around it. The primary observation window maintains a vertical downward viewing angle, with the window plane parallel to the bottom of the detection chamber, allowing for a complete top-down image of the abrasive particles. The window's effective observation area covers the entire magnetic adsorption area.

[0039] Eight auxiliary observation windows are positioned at equal angles around the main observation window, with a 45-degree angular interval between each pair of adjacent auxiliary windows. Each auxiliary observation window is tilted toward the chamber interior, with the window plane at a 30-degree angle to the chamber ceiling, and the window normal pointing toward the center of the inspection area. This tilted configuration allows each auxiliary observation window to observe abrasive particles adsorbed on the transparent protective plate from a specific side-view angle, capturing their three-dimensional morphology.

[0040] The sensor body includes a multi-view observation chamber, which is a multi-faceted pyramid in shape. The multi-view observation chamber is provided with a multi-layer optical window assembly. The multi-layer optical window assembly includes a main observation window located in the center of the top of the chamber and multiple auxiliary observation windows distributed radially around the main observation window. The multiple auxiliary observation windows are all inclined toward the inside of the chamber.

[0041] like Figure 4 As shown, in order to ensure that there is no visual interference between the various observation windows, the multi-view imaging unit also includes a field of view calculation unit and a stereo vision processing module. The field of view calculation unit calculates the field of view boundary data of each window, and the stereo vision processing module allocates the image data to the corresponding processing channel according to the view identifier. The field of view calculation unit of the multi-view imaging unit calculates the field of view boundary data of each window respectively, and performs a spatial intersection operation on the coordinates of the field of view cone area of ​​each window at the center of the detection area. The field of view cone areas of each window converge at the center of the detection area, but remain independent of each other near the window to avoid physical interference between imaging components. The window spacing baffle of the multi-view observation chamber is used to block the cross-irradiation of the lighting sources of different windows, and absorbs the excess light through the light-absorbing material to convert the light energy into heat energy for dissipation, so as to prevent the light sources from generating light intensity superposition interference in the adjacent window areas, thereby improving the clarity and contrast of image acquisition.

[0042] like Figure 2 and Figure 4 As shown, the multi-view imaging unit includes a main imaging component and an auxiliary imaging component. The main imaging component obtains a top view image of the abrasive particles through a main observation window, and the auxiliary imaging component obtains a side view angle image of the abrasive particles through an auxiliary observation window.

[0043] The multi-view observation chamber's trapezoidal sidewalls optimize the geometry of the optical path. The chamber gradually expands from bottom to top to ensure that the multiple observation windows and the optical axes of the imaging components do not obstruct each other. The inner walls of the multi-view observation chamber are coated with a low-reflectivity black coating to reduce secondary reflections of stray light and absorb incident light, reducing the intensity of reflected light to a negligible level. This ensures that imaging quality is not affected by ambient light, improves image clarity and contrast, and enables high-quality capture of wear particle images.

[0044] The main imaging component of the multi-view imaging unit performs distributed image acquisition and data transmission operations, converting optical signals into charge signals. These charge signals are then converted into digital pixel values ​​via an analog-to-digital conversion circuit. The digital image data is then packaged into data packets and transmitted to the internal image processing module. After receiving the data packets, the multi-view imaging unit parses the image's pixel matrix and performs a convolution operation on the pixel matrix through edge detection to identify wear particle boundaries. A gradient operator is used to calculate the grayscale difference between adjacent pixels. Pixels with grayscale differences exceeding a threshold are marked as edge points, and adjacent edge points are connected to form a continuous contour line. The multi-view imaging unit further performs chain code encoding on the contour line data, converting the coordinates of each point on the contour into a directional code sequence. The coordinates of the contour's geometric center point are calculated as the wear particle position. The unit then outputs the encoded data for the wear particle position coordinates and boundary contour, enabling the location and shape description of individual wear particles, effectively addressing the shortcomings of traditional methods in wear particle boundary identification.

[0045] The auxiliary imaging components of the multi-view imaging unit synchronously receive trigger signals for image acquisition. A unified clock pulse is sent to the eight auxiliary imaging components via a synchronous trigger circuit, causing them to simultaneously initiate exposure and data reading on the rising edge of the clock pulse. The multi-view imaging unit adds a viewpoint identifier to the oblique view image data and transmits it to the stereo vision processing module. Based on the viewpoint identifier, the image data is assigned to the corresponding processing channel. The multi-view imaging unit performs corner detection on each channel image, calculating the gradient matrix eigenvalue of each pixel. Pixels whose eigenvalues ​​meet the corner point criteria are selected as candidate feature points. Local extreme value screening is performed on these candidate points to identify stable feature points. The multi-view imaging unit calculates a description vector for each feature point, extracts the gradient direction histogram of the neighborhood surrounding the feature point, and normalizes the histogram data into a fixed-length description vector. The multi-view imaging unit calculates the similarity of the description vectors of feature points in images from different viewpoints, using the Euclidean distance metric to measure the differences between the description vectors. The feature point pair with the smallest distance is considered a match. The multi-view imaging unit calculates the pixel coordinate difference based on the matching point pairs, subtracts the pixel coordinates of the same feature point at different viewpoints to obtain the disparity value, and converts the disparity value into depth distance in combination with the camera intrinsic parameter matrix. The depth distance is merged with the plane coordinates of the main imaging component to construct the wear particle position vector containing three-dimensional coordinates, and outputs a list of spatial position data of the wear particle to achieve three-dimensional spatial positioning of the wear particle, effectively solving the problem that traditional two-dimensional image detection technical solutions cannot obtain depth information.

[0046] The 3D reconstruction module of the multi-view imaging unit receives the spatial position data of the abrasive particles, calculates the spatial distance between the abrasive particles and constructs a distance matrix. It uses a clustering discriminant function to classify abrasive particles with similar distances into clusters and assigns a unique identification code to each cluster. The 3D reconstruction module performs depth sorting on the abrasive particles within the cluster, establishes the front-to-back order of the abrasive particles based on the Z coordinate value, identifies the pixel areas where the rear abrasive particles are blocked by the front abrasive particles, and calculates the proportion of the blocked area to the total area of ​​the abrasive particles. The 3D reconstruction module calculates the acoustic wave action parameters based on the cluster analysis results, determines the acoustic wave frequency and power configuration for each cluster, encodes the parameter data into a control instruction format, and transmits the instruction data to the acoustic wave separation control module to achieve targeted separation operations on the aggregated abrasive particles, solving the problems of inaccurate counting and difficulty in extracting morphological features caused by abrasive particle aggregation.

[0047] The spectrum analysis unit stores a timing configuration table for multi-wavelength scanning, generates a time reference signal according to the configuration table, and distributes the reference signal to each LED drive circuit as an activation trigger. Figure 5 As shown, the spectral analysis unit includes a multi-wavelength LED array and a spectral filter group. The multi-wavelength LED array sequentially activates light sources of different wavelengths according to a timing configuration table. The spectral filter group receives a wavelength synchronization signal and rotates to a predetermined angle to achieve wavelength switching. After receiving the trigger signal, the multi-wavelength LED array of the spectral analysis unit drives the corresponding LED with current, adjusting the current to control the LED's light intensity output. During wavelength switching, it performs gradual control to avoid sudden changes in light intensity. The spectral filter group of the spectral analysis unit receives the wavelength synchronization signal, calculates the target angular position of the filter, drives the stepper motor to rotate the filter to the predetermined angle, and confirms the filter is in place through a position encoder, which feedbacks the ready status. This enables continuous wavelength switching and spectral scanning, providing a spectral basis for accurate identification of abrasive particles of different materials.

[0048] The spectral data acquisition module of the spectral analysis unit reads pixel data from the image sensor during each wavelength illumination period, classifies and stores the pixel data according to wavelength identifiers, and constructs a multidimensional spectral data matrix containing image information for each wavelength. The spectral data acquisition module extracts pixel values ​​from the abrasive region from the spectral data matrix, performs a ratio calculation on the abrasive pixel values ​​with the pixel values ​​in the reference region, and generates a spectral feature vector containing multi-wavelength reflectivity for each abrasive. The spectral data acquisition module correlates the spectral feature vector with pre-stored material spectral templates, calculates the correlation coefficient between the feature vector and each template, selects the template with the highest correlation coefficient as the material identification result, and outputs the material type identifier of the abrasive particle, enabling accurate classification of abrasive particles of different materials, resolving the problem that traditional methods cannot distinguish between abrasive particles of different materials.

[0049] The spectral depth analysis module of the spectral analysis unit establishes a depth relationship model based on the attenuation characteristics of light of different wavelengths in oil, calculates the effective penetration distance of light of different wavelengths in oil, and establishes a corresponding relationship between depth and light intensity attenuation for each wavelength. The spectral depth analysis module analyzes the variation in reflected light intensity of abrasive particles at different wavelengths, compares the degree of light intensity attenuation of each wavelength, and infers the depth layer position of the abrasive particles based on the attenuation difference. It assigns depth identifiers of the front and back layers to overlapping abrasive particles and outputs the depth layer sorting information of the abrasive particles. This enables depth layer identification and three-dimensional spatial distribution analysis of overlapping abrasive particles, effectively solving the problem of information loss caused by obstruction of the lower layer of abrasive particles during overlapping abrasive particle detection.

[0050] like Figure 6 As shown, the dynamic focusing unit includes a liquid lens assembly, a focal length control circuit, a focus position detector, and a multi-layer image acquisition module. The liquid lens assembly of the dynamic focusing unit contains a double-layer liquid medium of conductive liquid and insulating oil. Transparent electrodes surround the lens cavity to form an electric field control region. A variable voltage is applied to the transparent electrodes through the focal length control circuit to generate changes in the electric field intensity to adjust the optical focal length of the liquid lens. The change in electric field intensity changes the wetting properties of the conductive liquid surface. The change in wetting properties directly affects the geometric shape of the interface between the conductive liquid and the insulating oil. The change in the curvature of the interface shape adjusts the optical focal length of the liquid lens. The dynamic focusing unit stores multiple focal length preset value data tables. The data tables contain voltage configurations corresponding to different depth levels within the detection area. The corresponding control voltage signals are output in sequence according to a preset scanning sequence to achieve accurate focusing of wear particles at different depth levels, solving the problem that traditional fixed focal length methods cannot simultaneously observe multiple layers of overlapping wear particles.

[0051] The dynamic focusing unit's focus position detector emits a reference laser beam into the detection area and calculates the focal distance based on the reflected laser light. The multi-layer image acquisition module continuously collects image data at different focal lengths and constructs a three-dimensional image data stack. The focus position detector emits a reference laser beam into the detection area. After reflecting off the abrasive surface, the laser beam is received by a position-sensitive detector, which converts the position offset of the reflected laser light into an electrical signal. By calculating the geometric relationship between the laser's incident and reflection angles, the actual distance from the focal point to the surface of the transparent protective plate is calculated using a triangulation formula. The focus position detector digitally converts the distance measurement data and feeds it back to the focal length control circuit. The detector compares the deviation between the measured distance and the target distance and adjusts the voltage output based on the deviation to achieve dynamic correction of the focus position, ensuring optimal focusing at each depth level.

[0052] The dynamic focusing unit's multi-layer image acquisition module sends acquisition instructions to the image sensor after each focal length stabilizes, continuously capturing sharply focused image data at different focal lengths. The image data for each focal length layer is categorized and stored according to depth identifiers, constructing a three-dimensional image data stack containing information from multiple depth levels. The multi-layer image acquisition module performs gradient amplitude calculations on each layer of the three-dimensional image stack, counting the gradient intensity of each pixel position at different depth levels and selecting the pixel with the largest gradient intensity as the optimally focused pixel at that location. The multi-layer image acquisition module extracts the optimally focused pixels from each depth level to form a panoramic depth image, records the depth source information for each pixel, and outputs a fused, clear image and pixel depth mapping table, enabling clear imaging and depth information extraction of overlapping wear particles.

[0053] The overlapping wear particle recognition module of the dynamic focusing unit reads the pixel intensity distribution data of the three-dimensional image stack, calculates the intensity change curve of each pixel at different depth levels, and identifies the focal depth of the wear particle corresponding to the peak position in the intensity curve. The overlapping wear particle recognition module classifies pixels with similar focal depths as the same wear particle, analyzes the wear particle distribution pattern at different depth levels, and identifies the spatial relationship between foreground and background wear particles in the overlapping area. The overlapping wear particle recognition module reconstructs the complete outline of the obscured wear particle based on the clear pixels at each depth level, completes the boundary information of the background wear particle obscured by the foreground wear particle, and outputs the depth coordinates and complete outline data of each wear particle, effectively solving the problems of incomplete morphological feature extraction and counting errors caused by overlapping wear particles.

[0054] like Figure 7As shown, the acoustic wave separation unit includes an ultrasonic transmitter array, a phase control circuit, an acoustic wave focusing lens, an amplitude modulator, and an acoustic field distribution controller. The ultrasonic transmitter array is located at the bottom of the detection chamber. The phase control circuit calculates the phase delay value of each transducer based on the target focus position and generates multiple drive signals with specific phase relationships. The acoustic wave focusing lens performs phase modulation on the incident acoustic wave. The ultrasonic transmitter array of the acoustic wave separation unit converts electrical signals into mechanical vibrations. The matrix arrangement of the transducers produces a controllable acoustic wave phase distribution, forming an acoustic wave emission surface at the bottom of the detection chamber. The amplitude modulator calculates the required acoustic radiation force based on wear particle accumulation information and converts it into a required transducer drive power. The acoustic field distribution controller controls the movement of the acoustic wave focus along a predetermined path within the detection area. The acoustic wave separation unit calculates the required phase delay value for each transducer based on the target focus position. Using wavefront reconstruction, it determines the phase configuration that creates constructive interference at the target point and generates independent phase control parameters for each transducer. After receiving the phase control parameters, the acoustic wave separation unit generates a corresponding drive signal, performs phase shifting on the reference signal according to the calculated phase delay, and outputs a multi-channel drive signal with a specific phase relationship for use by the transducer array, thereby achieving precise focusing of the acoustic wave energy at the abrasive accumulation location.

[0055] The acoustic wave focusing lens of the acoustic wave separation unit performs phase modulation on the incident acoustic wave. The geometric dimensions and arrangement of the multi-layer metamaterial units determine the phase delay of the acoustic wave, converting the plane acoustic wave into a focused acoustic wave with a specific wavefront shape. The acoustic wave focusing lens calculates the propagation path of the acoustic wave based on the metamaterial parameters, predicts the focusing effect of the acoustic wave in the oil medium, and determines the spatial position and sound pressure distribution of the acoustic wave focal point. The acoustic wave focusing lens adjusts the operating state of the metamaterial unit, changes the position of the acoustic wave focal point within the detection area, and concentrates the acoustic wave energy at the target wear particle location, thereby improving the targeted and efficient acoustic wave separation.

[0056] The amplitude modulator of the acoustic wave separation unit receives the abrasive particle aggregation information output by the image analysis module, calculates the required acoustic radiation force based on the bond strength and aggregation density between the abrasive particles, and converts the required acoustic radiation force into a required transducer drive power. The amplitude modulator distributes the drive power based on the position and effectiveness of each transducer in the array, adjusts the voltage amplitude of each transducer to achieve the required power output, and outputs a power control scheme that matches the target sound field. The amplitude modulator amplifies the power control signal to the voltage level required for transducer operation, maintains the phase relationship between the various signals, and provides stable drive power to the transducer array, ensuring the controllability and effectiveness of the acoustic wave separation process.

[0057] The acoustic field distribution controller of the acoustic wave separation unit synchronously manages the operating states of the phase control circuit and amplitude modulator, generates a path plan for acoustic wave scanning based on the spatial distribution of the abrasive particle separation task, and controls the movement of the acoustic wave focus along a predetermined path within the detection area. The acoustic field distribution controller calculates the movement trajectory and dwell time of the acoustic wave focus, optimizes the scanning sequence to improve separation efficiency, and ensures that abrasive particles accumulated in different locations receive appropriate acoustic wave action. The acoustic field distribution controller manages the time allocation of acoustic wave action, coordinates the processing order of each accumulation area, and implements selective and sequential acoustic wave processing of abrasive accumulation areas, effectively avoiding unnecessary waste of acoustic wave energy.

[0058] The working process of the acoustic wave separation unit is based on the physical effect of acoustic radiation force on tiny particles. When ultrasonic waves propagate through an oil medium and encounter abrasive particles, acoustic scattering occurs. The scattered sound waves and the incident sound waves interfere with each other in the space surrounding the abrasive particles. The uneven sound pressure distribution caused by this interference produces a net radiation force on the abrasive particles. The acoustic wave separation unit establishes a scattering model based on the geometric dimensions and material properties of the abrasive particles, calculates the interaction strength between the scattered wave and the incident wave, and determines the vector direction and magnitude of the acoustic radiation force. The acoustic wave separation unit decomposes the acoustic radiation force into components along different spatial directions, analyzes the contribution of each directional component to the motion of the abrasive particles, and predicts the motion trajectory of the abrasive particles under the influence of the acoustic field, thereby achieving precise control of the abrasive particle separation process.

[0059] The multi-frequency acoustic controller in the acoustic wave separation unit simultaneously generates multiple ultrasonic signals of varying frequencies, modulating the amplitude and phase of each frequency signal to produce a composite frequency acoustic field distribution in the abrasive particle accumulation area. The multi-frequency acoustic controller selects the appropriate acoustic wave frequency combination based on the abrasive particle size distribution, leveraging the selective effect of different frequency acoustic waves on abrasive particles of varying sizes to preferentially separate abrasive particles of a specific size. The multi-frequency acoustic controller analyzes the spatial superposition of the multi-frequency acoustic waves, calculates the sound pressure amplitude and phase distribution at each location in the composite acoustic field, and determines the optimal frequency configuration and power allocation scheme to improve separation adaptability to abrasive particle accumulations of varying sizes.

[0060] The acoustic wave separation unit calculates the minimum separation force threshold required to overcome inter-abrasive adhesion, analyzes the effects of van der Waals forces and surface tension between the abrasive particles, and determines the acoustic wave parameter configuration to generate sufficient separation force. The acoustic wave separation unit adjusts the combination of acoustic wave frequency, phase, and power to produce the optimal acoustic radiation force distribution at the target location, and controls the duration and intermittent pattern of the acoustic wave action. The acoustic wave separation unit monitors the real-time position changes of the abrasive particles, adjusts the acoustic wave action parameters based on the separation progress, and achieves a gradual and controllable physical separation of the aggregated abrasive particles, effectively addressing the problem of reduced detection accuracy caused by abrasive particle aggregation.

[0061] The acoustic separation unit's separation effect monitoring module collects real-time images of abrasive particles during the acoustic wave process, compares changes in the spatial distribution of abrasive particles before and after separation, and counts the number of successfully separated abrasive particles and the degree of residual aggregation. The separation effect monitoring module analyzes the position change trajectory of each individual abrasive particle, calculates the abrasive particle movement distance and separation speed, and evaluates the efficiency and integrity of the separation process. The separation effect monitoring module compares and analyzes the separation results with the expected goals, generates quantitative evaluation indicators for the separation effect, and provides optimization feedback to the acoustic wave controller. Based on this feedback, the acoustic wave controller dynamically adjusts the acoustic wave parameter configuration to improve the separation effect, forming a closed-loop optimized separation process control.

[0062] like Figure 8 As shown, the analysis unit consists of three components: an aggregation detection network, a separation planning network, and a feature extraction network, specifically designed to perform deep learning processing on the wear particle aggregation problem. The aggregation detection network performs feature map extraction on the input image to identify wear particle aggregation. The separation planning network generates an acoustic wave separation execution plan based on the aggregation detection results. The feature extraction network performs deep feature analysis on the separated individual wear particles. The aggregation detection network of the analysis unit receives image data from the multi-view imaging unit, performs feature map extraction on the input image, and performs gradient filtering on the image pixel matrix using the first-layer convolution kernel. The gradient filtering operation is calculated between the grayscale differences between adjacent pixels to generate an edge response map, highlighting the boundary between the wear particles and the background. The aggregation detection network performs texture pattern recognition on the feature map output by the first layer using the second-layer convolution kernel. A Gabor filter bank is used to extract texture features of different directions and frequencies, generating a texture feature map that can be used to distinguish the roughness and smoothness characteristics of the wear particle surface. The aggregation detection network performs spatial correlation analysis on the feature map of the previous layer through the third layer of convolution kernel, calculates the spatial distance and directional relationship between feature points, and generates a spatial correlation feature map to show the aggregation pattern and distribution law between wear particles, thereby accurately identifying the initial position and range of wear particle aggregation, effectively solving the problem that traditional methods are difficult to identify complex aggregation patterns.

[0063] The aggregation detection network includes a spatial density analysis module, a shape anomaly assessment module and a boundary continuity detection module. The spatial density analysis module counts the number of abrasive feature points in the local area to calculate the density value. The shape anomaly assessment module calculates the geometric characteristic parameters of the abrasive contour and compares them with the standard template. The boundary continuity detection module analyzes the breakpoints and curvature changes of the contour to determine the degree of aggregation.

[0064] The spatial density analysis module of the cluster detection network reads the feature map data output by the convolutional layer, sets a circular window of a fixed radius at each pixel position, and counts the number of wear particle feature points detected within the window to calculate the local density value. The spatial density analysis module compares the local density value with a preset threshold, marks pixel positions with density values ​​exceeding the threshold as high-density candidate points, and connects adjacent high-density candidate points to form the boundary outline of the cluster area. The spatial density analysis module calculates the coordinates of the geometric center point from the cluster area outline and outputs the location information of the high-density area for subsequent processing, thereby quickly locating key areas with dense wear particle distribution and improving the efficiency and accuracy of cluster detection.

[0065] The shape anomaly assessment module of the aggregation detection network reads the pixel coordinate sequence of the abrasive particle contour, calculates the contour's geometric characteristic parameters, and finds the contour's minimum circumscribed rectangle by dividing the rectangle's long side by the short side to obtain the aspect ratio. The shape anomaly assessment module measures the contour's perimeter and area, calculates its similarity to a standard circle using a roundness formula, and constructs a convex hull polygon for the contour. The convex hull area ratio is calculated by comparing the contour area with the convex hull area. The shape anomaly assessment module calculates the difference between the calculated geometric parameters and those of a standard abrasive particle template, identifies abrasive particle contours whose geometric parameters significantly deviate from the normal range, and outputs an identifier for suspected aggregation areas with abnormal shapes. This allows for the detection of abrasive particle aggregation from the perspective of shape change, improving the accuracy of aggregation identification.

[0066] The boundary continuity detection module of the aggregation detection network performs boundary tracking processing on the abrasive particle contour, moving pixel by pixel along the boundary direction starting from the contour starting point, recording the coordinates and movement direction of each boundary point, and detecting the position where the direction change exceeds the threshold and marking it as a turning point. The boundary continuity detection module analyzes the distance between adjacent boundary points, identifies the position where the distance exceeds the continuity threshold and marks it as a breakpoint, and counts the number and distribution of breakpoints in the contour. The boundary continuity detection module calculates the local curvature value for each point on the contour, calculates the curvature size using the three-point arc fitting method, and counts the standard deviation of the curvature change as a smoothness indicator. The boundary continuity detection module judges the tightness of the abrasive particle aggregation based on the number of breakpoints and the degree of curvature change, and outputs a quantitative score of the aggregation tightness, thereby evaluating the bonding strength of the abrasive particle aggregation and providing an important reference for the selection of separation parameters.

[0067] The aggregation type classifier of the aggregation detection network receives the output data of the above three modules, combines the density value, shape parameters and boundary features into a multi-dimensional feature vector, and performs branch judgment according to the numerical range of the feature vector. The aggregation type classifier first determines whether there is aggregation based on the density value, then determines the degree of aggregation based on the shape abnormality, and finally determines the type of aggregation based on the boundary continuity. The aggregation type classifier divides the aggregation state into four levels: slight contact, partial overlap, complete aggregation and multi-layer stacking, assigns a digital coding identifier to each level, and outputs the aggregation level identifier and the bounding box coordinate data of the corresponding area, providing accurate target positioning information for subsequent separation operations, and realizing accurate classification and positioning of aggregated abrasive particles.

[0068] The analysis unit's separation planning network receives the clustering area information output by the clustering detection network and the material data provided by the spectral analysis unit. Based on this input data, it generates an acoustic separation execution plan, correlating and matching the clustering information with the material information. The separation planning network establishes a correspondence table between clustering areas and material types, assigning each clustering area a corresponding material attribute identifier. Based on this correspondence table, the network queries a preset separation parameter template, selects the parameter configuration that best matches the current clustering type and material combination, and outputs a complete separation plan encompassing frequency, power, and timing. This ensures that the separation parameters closely match the actual situation, improving the pertinence and success rate of the separation operation.

[0069] The separation planning network includes a frequency selection module, a power allocation calculation module and an action time planning module. The frequency selection module calculates the optimal acoustic wave frequency based on the abrasive size distribution, the power allocation calculation module determines the required acoustic wave power level based on the degree of aggregation, and the action time planning module calculates the duration of the acoustic wave action and divides it into multiple short pulse periods.

[0070] The frequency selection module of the separation planning network calculates the optimal acoustic wave frequency based on the abrasive particle size distribution and material density, extracts the equivalent diameter of each abrasive particle from the image data, compiles a statistical histogram of the abrasive particle size distribution, and calculates the mean and standard deviation of the size distribution. The frequency selection module uses the acoustic resonance formula to calculate the resonant frequency of each abrasive particle based on the abrasive particle size and material density, establishes a functional relationship between the abrasive particle size and the resonant frequency, and generates a size-frequency correspondence data table. The frequency selection module queries the data table based on the size range of the abrasive particles in the aggregation area, determines the optimal frequency range covering this size range, and outputs a frequency configuration scheme with multiple frequency points to achieve selective separation of abrasive particles of different sizes, improving the targetedness and efficiency of the separation process.

[0071] The power allocation calculation module of the separation planning network determines the required acoustic power level based on the degree of aggregation and the strength of the bonding force between the abrasive particles. It calculates the adhesion force between the abrasive particles based on the abrasive material and surface characteristics, calculates the bonding force value using the van der Waals force model and surface energy theory, and establishes a mapping relationship between the aggregation type and the bonding force strength. The power allocation calculation module calculates the acoustic radiation force required to overcome the adhesion based on the bonding force. It uses the acoustic radiation force formula to convert the force requirement into an acoustic power requirement and assigns a corresponding power coefficient to each transducer. The power allocation calculation module checks whether the power allocation of each transducer exceeds the safety limit, adjusts the power allocation ratio while ensuring the separation effect, and outputs a power allocation plan that meets safety requirements. This ensures the separation effect while avoiding equipment overload and ensuring safe and stable operation of the equipment.

[0072] The action time planning module of the separation planning network calculates the duration of the acoustic wave action based on the stability of the abrasive aggregation and the strength of the adhesion force, evaluates the response characteristics of the aggregated abrasive under the action of the acoustic wave, calculates the energy and time required for the abrasive to break away from the aggregated state, and establishes a relationship model between aggregation stability and separation time. The action time planning module divides the continuous action time into multiple short pulse periods, inserts pauses between pulses for effect observation and parameter adjustment, and calculates the optimal ratio of pulse width and pause time. The action time planning module encodes the pulse timing information into digital control instructions, generates a timing data packet containing the start time, duration, and interval time, and outputs timing instructions that can be directly used for hardware control, realizing time control of the separation process and avoiding abrasive damage caused by excessive action.

[0073] The action sequence optimization module of the separation planning network analyzes the spatial distribution and mutual influence of clustered areas, calculates the relative position and distance of each clustered area, identifies the possibility of acoustic interference between adjacent areas, and establishes a relationship diagram of the mutual influence between areas. The action sequence optimization module determines the processing order based on the independence of the clustered areas and the difficulty of separation, prioritizing areas with strong independence and low separation difficulty, and avoiding simultaneous processing of adjacent areas to prevent acoustic interference. The action sequence optimization module generates a processing queue for the clustered areas based on priority, assigns a processing time window to each area, and outputs an execution schedule containing the processing order and time schedule, maximizing separation efficiency and avoiding mutual interference, thereby improving the effectiveness and efficiency of the overall separation operation.

[0074] The safety assessment module of the separation planning network examines the safety impact of the generated separation plan on the abrasive particles. It calculates the acoustic pressure stress on the abrasive particle surface based on the acoustic wave parameters, predicts the stress distribution and magnitude using an acoustic theory model, and compares the predicted stress value with the mechanical strength of the abrasive material. The safety assessment module determines whether the acoustic wave stress exceeds the yield strength or fracture strength of the material, identifies risk parameters that may cause the abrasive particles to break or deform, and triggers parameter adjustments when safety risks are detected. The safety assessment module automatically reduces the power parameters that exceed the limit or modifies the mode of action, recalculates the adjusted stress level to ensure safety, and outputs a revised safety separation plan. This protects the integrity of the abrasive particles while achieving the separation goal, ensuring the accuracy and reliability of the test results.

[0075] The analysis unit's feature extraction network performs deep feature analysis on the separated individual wear particles, enabling pixel-level wear particle boundary identification. It then performs foreground and background classification for each pixel in the image, extracting contextual features from the pixels using a deep convolutional network, and assigning pixels to corresponding wear particle instances based on feature similarity. The feature extraction network assigns a unique instance identifier to each wear particle, ensuring that pixels belonging to the same wear particle share the same identifier. It also generates an instance segmentation mask for subsequent boundary extraction, enabling accurate separation and identification of individual wear particles and resolving the challenge of identifying individual wear particles after they have been separated.

[0076] The feature extraction network includes a boundary detection convolution layer, a contour refinement module, and a multi-scale feature fusion module. The boundary detection convolution layer uses a deformable convolution kernel to detect the irregular shape contour of the wear particle. The contour refinement module performs sub-pixel level processing on the boundary contour. The multi-scale feature fusion module processes image data of different resolutions and extracts detail features and shape features.

[0077] The boundary detection convolutional layer of the feature extraction network uses a deformable convolution kernel to detect the irregular contours of wear particles. It dynamically adjusts the sampling point positions based on local gradient information, calculates the gradient direction and intensity to determine the optimal sampling position, and concentrates the convolution kernel's sampling points near the boundary. The boundary detection convolutional layer uses these adjusted sampling points to extract boundary features, capturing detailed feature information of sharp edges and small protrusions. It then generates a high-resolution boundary feature map, accurately detecting boundary details of complex wear particles and improving the accuracy and integrity of boundary recognition.

[0078] The edge enhancement processing module of the feature extraction network performs morphological operations on the boundary detection results, performing morphological dilation on the boundary image, expanding the width of the boundary line using circular elements, and connecting broken boundary segments with similar distances. The edge enhancement processing module performs morphological erosion on the dilation results, shrinking the boundary line width using elements of the same size, restoring the original thickness of the boundary line and maintaining connectivity. The edge enhancement processing module identifies and removes pseudo-boundaries caused by noise, selects valid boundaries based on the length and continuity of boundary segments, and outputs a clean wear particle boundary outline, obtaining complete and accurate wear particle boundary information, providing reliable basic data for subsequent feature extraction.

[0079] The feature extraction network's contour refinement module performs sub-pixel processing on the rough boundary contour, extracting the contour's pixel coordinates from the boundary image, sorting the coordinates according to the contour direction, and generating an ordered sequence of contour points. The module then performs spline curve fitting on the sequence of contour points, reconstructing a smooth, continuous contour curve using B-spline functions, and interpolating to generate sub-pixel-precise contour point coordinates. The module then locally optimizes the fitting results to eliminate oscillations and overfitting during the fitting process. It then outputs a highly accurate sequence of wear particle contour coordinates, providing sub-pixel contour measurement accuracy and significantly improving the accuracy of wear particle size and shape measurements.

[0080] The multi-scale feature fusion module of the feature extraction network processes image data of varying resolutions, decomposing the input image into multiple image layers of varying resolutions. It then uses the Gaussian pyramid method to generate a multi-scale representation of the image while maintaining the spatial correspondence between images at each scale. The multi-scale feature fusion module extracts the texture features of the abrasive surface from the detail image, identifies surface scratches, particles, and reflective properties, and generates a detail feature vector. Simultaneously, it extracts the shape features of the abrasive from the overall image, calculates the geometric parameters and shape descriptors of the contour, and generates a shape feature vector. The multi-scale feature fusion module performs a weighted combination of detail and shape features, assigning weight coefficients based on feature importance. It then generates a complete feature vector containing multi-level information, capturing both the local details and overall characteristics of the abrasive, providing a comprehensive and accurate description of the abrasive's characteristics.

[0081] The local detail analysis module of the feature extraction network specifically processes the microscopic features of the abrasive surface, identifying scratches, pits, and oxidation spots on the abrasive surface. It uses edge detection and region segmentation methods to locate surface defects and calculates the geometric size and distribution density of the defects. The local detail analysis module measures the height variation of the abrasive surface, estimates the surface undulation using grayscale gradient variations, calculates the root mean square value of the surface roughness, and calculates the main direction of the surface texture. It uses gradient statistics to determine the dominant direction of the texture and outputs a directional index of the texture. The local detail analysis module converts surface features into numerical evaluation results, generates a comprehensive score for surface quality, and outputs a standardized description of the abrasive surface state. It quantitatively evaluates the surface quality and degree of wear of the abrasive, providing an important basis for equipment status diagnosis.

[0082] The overall shape analysis module of the feature extraction network calculates the geometric shape parameters of the abrasive particle and calculates the equivalent circle diameter based on the abrasive particle area. This converts the abrasive particle area into the diameter of a circle of the same area, providing a standardized representation of the abrasive particle size. The overall shape analysis module calculates the maximum length and width of the abrasive particle, fits the minimum circumscribed ellipse of the abrasive particle, extracts the lengths of the major and minor axes of the ellipse, and assesses the degree to which the abrasive particle shape deviates from a regular geometric shape. The module compares the abrasive particle profile with the standard geometric shape and outputs a shape irregularity coefficient. The overall shape analysis module identifies the basic shape types of abrasive particles based on the geometric parameters, classifying them as spherical, lamellar, acicular, or irregular, and calculating similarity indices to each type. The overall shape analysis module converts all geometric parameters into a unified descriptive data format, generating a digital archive of the abrasive particle shape and outputting standardized data that can be stored in a database and used for comparative analysis. This module establishes a comprehensive database of abrasive particle shape characteristics for equipment condition monitoring and trend analysis, providing a scientific basis for equipment maintenance decisions.

[0083] The workflow of the image sensor used for online oil quality monitoring utilizes an iterative optimization approach. The image sensor first sends a start command to the multi-view imaging unit, which then begins image data acquisition and establishes a data communication link with the cluster detection network. The image sensor encapsulates the initial image data into a data packet containing the image pixel matrix, timestamp, and viewpoint identification information, and sends the packet to the input buffer of the cluster detection network. The cluster detection network in the analysis unit reads the image data packet from the buffer, extracts the pixel matrix, and reconstructs the complete image data, initiating the cluster recognition process to ensure image data integrity and timing accuracy.

[0084] The analysis unit's cluster detection network generates processing results, including a list of pixel coordinates of clustered areas and a numerical code of the clustering degree. These results are converted into a standardized data format, generating a data packet containing spatial coordinates and a level identifier. Based on the packet type identifier, the image sensor routes the cluster detection results to the spectral analysis control module, establishes a priority queue for data transmission, and ensures the timely transmission of critical data, thereby reducing data transmission delays, improving processing efficiency, and achieving efficient coordination among functional modules.

[0085] The image sensor's spectral analysis control module analyzes the clustered area information, extracts the spatial coordinates and range data of the clustered areas, and generates a list of target areas for spectral scanning. Based on the target area list, the spectral analysis control module generates a spectral analysis execution sequence, assigns a scanning time window to each area, and sends execution instructions to the spectral analysis unit. Following the instruction sequence, the spectral analysis unit sequentially activates light sources of different wavelengths, records the reflected light intensity data of the target area at each wavelength, and constructs a multi-wavelength spectral data matrix. The spectral analysis unit performs material identification operations on the spectral data matrix, calculates the spectral feature vector at each pixel position, and matches this feature vector with the material database to obtain material identification results, thereby determining the material composition information of the abrasive particles within the clustered area.

[0086] The image sensor converts the material identification results and spectral data into a data format recognizable by the separation planning network. It then compresses the spectral data and extracts key characteristic parameters, generating a data packet containing the material identification and characteristic parameters. The image sensor receives the aggregation information and material data, establishes a mapping table linking aggregation areas and material types, and provides a complete input data set for the separation planning network. The analysis unit's separation planning network optimizes the acoustic wave parameters based on the input data set, using a multi-objective optimization method to determine the optimal parameter combination. It then generates parameter configuration instructions, including frequency, power, and timing, to ensure an optimal match between the acoustic wave parameters and the actual aggregation conditions, significantly improving the success rate of the separation operation.

[0087] The image sensor encodes parameter configuration instructions into a hardware control instruction format, converts floating-point parameters into hardware-recognizable digital control codes, and adds a checksum to ensure the accuracy of instruction transmission. The image sensor sends the control instructions to the dynamic focusing control module and the acoustic wave separation control module, establishing a bidirectional communication channel to ensure reliable instruction transmission and monitoring the instruction execution status to provide feedback. The dynamic focusing unit's control module parses the received control instructions, extracts the focus depth and scanning range parameters, and sends a focusing control signal to the dynamic focusing unit. The dynamic focusing unit collects image data at different depth levels according to the control signal, records the focus quality evaluation indicators at each depth level, and constructs a three-dimensional image data stack. The dynamic focusing unit performs spatial coordinate calculations on the image stack, determines the position coordinates of each abrasive particle in three-dimensional space, and generates a digital model of the abrasive particle's spatial distribution.

[0088] The image sensor converts the three-dimensional position data into an acoustic wave control coordinate system, performs coordinate system rotation and translation operations, and converts the image coordinates into the physical coordinates of the acoustic wave transmitter array. The image sensor merges the three-dimensional position data with the separation parameter data to generate a comprehensive control instruction containing the target position and action parameters, and transmits the complete separation task data to the acoustic wave separation control module. The control module of the acoustic wave separation unit analyzes the comprehensive control instruction, extracts the target coordinates and acoustic wave parameter information, and generates the drive signal configuration for the transducer array. Based on the drive signal configuration, the acoustic wave separation unit generates multiple phase- and amplitude-controllable electrical signals and outputs the electrical signals to the transducer array. The transducer array converts the electrical signals into acoustic wave energy and generates acoustic radiation force in the target area, achieving a directional separation effect on the aggregated abrasive particles, effectively solving the detection accuracy problem caused by abrasive particle aggregation.

[0089] After the separation operation is complete, the image sensor detects the completion signal from the acoustic separation unit, triggering a new round of image acquisition and restarting the multi-view imaging unit to obtain the separated image data. The image sensor receives the image data before and after separation and performs a pixel-level difference operation on the two sets of image data, identifying the areas and pixel locations that have changed in the image. The image sensor segments and labels the difference results, calculates the area and location of each changed area, and determines whether the changed area corresponds to a successfully separated wear particle. This quantitatively evaluates the effectiveness of the separation operation and provides accurate feedback data for iterative optimization.

[0090] The image sensor's performance evaluation module performs numerical statistics on the separation results, counting the number of successfully separated abrasive particles and the number of remaining aggregated areas, and calculates the separation success rate and the percentage of improvement in aggregated particles. The performance evaluation module scans the image data after separation, identifies the coordinates of areas where aggregates still exist, and analyzes the characteristics and distribution patterns of the remaining aggregated areas. The performance evaluation module encodes the statistical data and residual aggregate information into an evaluation report, generating a data package containing a success rate indicator and the coordinates of failed areas. This evaluation report is then transmitted to the parameter adjustment module, providing a quantitative basis for subsequent parameter optimization, enabling closed-loop control and continuous improvement of the separation process.

[0091] The image sensor's parameter adjustment module receives the evaluation report data, compares the expected separation performance with the actual performance, and identifies the key factors that led to separation failure. Based on the failure analysis results, the parameter adjustment module recalculates the acoustic wave parameters, increases the power value or modifies the frequency combination to improve separation performance, and optimizes the parameter configuration using feedback control methods. The parameter adjustment module converts the adjusted parameters into new control instructions, generates a specialized separation plan for the remaining clustered areas, and transmits the secondary separation instructions to the acoustic wave separation control module. This addresses the areas where the initial separation was incomplete, thereby improving the overall separation completion rate.

[0092] The control module of the acoustic wave separation unit receives the secondary separation instruction, replaces the original control configuration with the new parameters, and reconfigures the driving parameters of the transducer array. The acoustic wave separation unit restarts the separation operation under the adjusted parameter configuration, specifically targets the remaining accumulation area with acoustic waves, and monitors the real-time progress of the secondary separation. The iterative control module of the image sensor records the number of times the separation operation is executed, monitors whether the number of cycles has reached the preset upper limit, and monitors whether the separation success rate has reached the target threshold. The iterative control module decides whether to terminate the iterative process based on the number of cycles and success rate data. When the termination condition is met, it sends a stop signal to the process control module, triggering the final feature extraction processing flow to ensure that the separation task is completed within a reasonable time and avoid over-processing.

[0093] After the iterative process terminates, the image sensor activates the feature extraction network and transmits the separated wear particle image data to its input port, establishing a feature extraction processing queue. The feature extraction network in the analysis unit performs feature analysis on each individual wear particle, extracting its shape, texture, and surface characteristics, and generating a feature vector containing multidimensional feature parameters. The image sensor stores each wear particle's feature vector in a feature database, establishing an index relationship between the wear particle identifier and the feature vector. This output is a complete wear particle feature dataset for comprehensive oil quality assessment and a comprehensive wear particle feature profile for equipment status diagnosis, providing a scientific basis for equipment maintenance decisions.

[0094] The image sensor used for online detection of oil quality is integrated with a real-time monitoring feedback unit to ensure the safety and effectiveness of the separation process. Figure 9 As shown in the figure, the real-time monitoring and feedback unit includes a high-speed image acquisition module, a motion state analysis module, a force sensing monitoring module, and a safety control module. The high-speed image acquisition module receives continuous image frames from a high-frame-rate camera in real time, arranges and stores the image frames in a time sequence, and establishes a circular buffer queue to ensure data continuity. The high-speed image acquisition module transmits the continuous image frame data to the motion state analysis module, maintaining the temporal relationship of the image frames and adding timestamps to ensure the time accuracy of motion analysis, thus enabling real-time tracking of the wear particle motion state.

[0095] The motion state analysis module of the real-time monitoring feedback unit performs pixel-level differential operations on consecutive image frames, calculates the pixel grayscale difference between adjacent frames, and generates a binary mask image of the moving area. The motion state analysis module identifies and segments the moving abrasive target from the mask image, calculates the center of mass coordinates and boundary contours of the moving target, and establishes the identification and tracking record of the abrasive target. The motion state analysis module calculates the motion trajectory based on the position changes of the abrasive in consecutive frames, uses the position differential method to calculate the velocity vector, and uses the velocity differential method to calculate the acceleration vector. The motion state analysis module monitors whether the motion parameters of the abrasive exceed the normal range, identifies abnormal high-speed motion or abnormal rotation behavior, determines whether the abrasive deviates from the expected separation trajectory, and promptly detects abnormal conditions during the separation process to ensure the safety and controllability of the separation operation.

[0096] The motion analysis module generates a warning signal upon detecting abnormal motion, encodes the warning signal into a digital message format, and transmits the message to the signal receiving port of the safety control module. The aggregation warning module within the motion analysis module calculates the distance between separated abrasive particles, monitors the changing trend of the distance between the particles, and identifies pairs of abrasive particles whose distance is rapidly decreasing. The aggregation warning module activates a warning upon detecting the risk of secondary aggregation, generates a secondary aggregation risk signal, and transmits the risk signal to the safety control module to prevent the reaggregation of the abrasive particles after separation and ensure the continuity of the separation effect.

[0097] The force sensing monitoring module of the real-time feedback unit receives the analog signal output by the pressure sensor, amplifies and filters the signal, and converts it into digital data. The force sensing monitoring module performs spectral analysis on the digital data, extracting the frequency components and amplitude information of the vibration signal, and monitoring the changes in mechanical stress within the test chamber. The force sensing monitoring module compares the measured data with preset safety thresholds, identifying stress levels exceeding the safety limit. If overstress is detected, an overload warning signal is generated, protecting the test equipment from excessive stress damage and ensuring long-term stable operation.

[0098] The safety control module of the real-time monitoring feedback unit simultaneously receives status information from each monitoring module, prioritizes and sorts the signals, and establishes a priority queue for signal processing. The safety control module selects the appropriate safety response measure based on the type of warning signal. It generates a power reduction command when abnormal motion is detected, a pause operation command when a secondary accumulation risk is detected, and an emergency shutdown command when overload stress is detected. The safety control module sends the response command to the corresponding control module to ensure the priority execution of the safety command, monitor the effectiveness of the safety measures, and quickly implement protective measures to ensure the safety of the equipment and samples when abnormal conditions occur, thus achieving comprehensive safety protection for the separation process.

[0099] Image sensors for online oil quality monitoring are used to automatically optimize processing parameters based on the separation effect of different types of abrasive particles. Figure 10 As shown, the image sensor's adaptive learning unit includes an aggregation pattern recognition module, a parameter optimization module, an effect evaluation module, and a knowledge base update module. The aggregation pattern recognition module establishes a feature data index table for wear particle aggregation patterns. The parameter optimization module dynamically adjusts acoustic wave parameters based on separation results. The knowledge base update module identifies high-quality processing cases and updates the database. The adaptive learning unit's aggregation pattern recognition module establishes a feature data index table for wear particle aggregation patterns, associates and stores image feature vectors of different aggregation types with corresponding optimal separation parameter combinations, and uses a hash table to establish a fast retrieval relationship between feature vectors and parameter configurations. The aggregation pattern recognition module performs multi-layer feature extraction operations on newly encountered aggregation situations, calculates the spatial distribution characteristics, shape characteristics, and density characteristics of the aggregation area, and organizes the extracted feature data into a standardized feature vector format. The aggregation pattern recognition module compares the new feature vectors with known patterns in the database one by one, performs vector inner product and modulus length calculations to obtain similarity values, and quickly identifies the degree of similarity between the current aggregation pattern and historical patterns.

[0100] The similarity calculation module of the adaptive learning unit calculates the Euclidean distance between the new feature vector and the database vector, taking the square root of the sum of the squared feature differences in each dimension to obtain the distance value. It also calculates the cosine similarity between the vectors as an angular similarity indicator. The similarity calculation module combines the distance and angle metrics to calculate the final matching score, combining the values ​​of the two metrics using a weighted summation method to output the highest matching score and the corresponding historical clustering pattern identifier. The similarity calculation module compares the matching score with a preset threshold. If the matching score falls below the threshold, the new pattern learning process is activated, triggering the expansion and update of the feature database, automatically identifying unknown clustering pattern types, and ensuring the effective processing of newly emerging clustering patterns.

[0101] The parameter optimization module of the adaptive learning unit extracts the corresponding separation parameter configuration from the database based on the matching results, uses the historical optimal parameters as the starting parameters for the current separation operation, and establishes a mapping relationship between the parameter configuration and the separation task. The parameter optimization module monitors the real-time changes in the separation effect during the separation process, dynamically adjusts the acoustic wave frequency, power, and action time parameters based on the effect feedback data, and records the numerical changes and effect responses of each parameter adjustment. The parameter optimization module analyzes the mathematical relationship between parameter changes and effect improvements, calculates the partial derivatives of the effect function with respect to each parameter, and determines the search direction and step size for parameter optimization. The parameter optimization module saves the complete historical trajectory of parameter adjustments, establishes a time correspondence between the parameter sequence and the effect sequence, generates learning sample data for parameter optimization, continuously improves the accuracy of parameter selection and the separation effect, and achieves continuous optimization of the separation process.

[0102] The performance evaluation module of the adaptive learning unit performs numerical calculations on multiple performance indicators of the separation operation, counting the number of successfully separated abrasive particles and the total processing time, and calculating the percentage of separation success rate and processing efficiency. The performance evaluation module records the power consumption of each equipment module during the separation process, cumulatively calculates the total energy consumption, and analyzes the relationship between energy consumption and separation performance. The performance evaluation module verifies the shape and surface integrity of the separated abrasive particles, compares the changes in the geometric parameters of the abrasive particles before and after separation, and calculates a quantitative indicator of the degree of abrasive damage. The performance evaluation module statistically compares the current evaluation results with historical data, identifies the changing trends and improvement range of performance indicators, and generates a performance evaluation report for optimization reference, comprehensively evaluating the overall performance level of the separation operation.

[0103] The knowledge base update module of the adaptive learning unit identifies processing cases with high separation success rates and excellent efficiency, filters high-quality cases based on preset performance thresholds, and extracts the feature vectors and parameter configuration data of high-quality cases. The knowledge base update module integrates new high-quality cases with similar cases in the database, calculates the mean feature vector of similar cases, and updates the feature vectors and parameter configurations of corresponding entries in the database. The knowledge base update module identifies historical records in the database that have poor results or are rarely used, deletes redundant data based on performance scores and access frequency, and maintains the storage efficiency and query speed of the database. The knowledge base update module regularly checks the practicality and accuracy of each entry in the database, verifies the applicability of historical data in the current environment, updates the database index and retrieval efficiency, and ensures the continuous optimization and practical value of the knowledge base.

[0104] The image sensor's new pattern learning mechanism activates the learning process when encountering an unknown aggregation type, analyzes the feature vector of the current aggregation pattern, and searches the database for the closest historical pattern as a basis for learning. The new pattern learning mechanism establishes an initial processing plan based on the most similar historical case, uses the parameter configuration of the similar case as a starting point, and assigns a temporary identification code and parameter configuration to the new pattern. The new pattern learning mechanism performs small parameter changes during the separation process, making exploratory adjustments to frequency, power, and timing parameters, and monitoring the impact of each adjustment on the separation effect. The new pattern learning mechanism establishes a new feature pattern record based on the results of the trial adjustments, associates and stores the optimal parameter combination with the feature vector, and adds the new pattern to the knowledge base for subsequent use. It continuously expands processing capabilities and adapts to new types of aggregation patterns, improving the image sensor's adaptability to complex aggregation situations.

[0105] The comprehensive application of multi-dimensional separation detection technology in image sensors for online oil quality monitoring effectively solves the problems of abrasive particle aggregation, overlap, and occlusion. The image sensor's aggregation solution significantly improves the accuracy of abrasive particle counting through the synergistic effect of spatial analysis and acoustic wave separation. The aggregation detection network identifies the location of the aggregation area, and the acoustic wave separation unit performs a directional separation operation on the aggregation area. After separation, the individual abrasive particles can be accurately counted and analyzed. The image sensor eliminates the counting errors in traditional methods where multiple abrasive particles are mistaken for a single large particle. When processing densely aggregated areas, multiple rounds of iterative separation maintain stable recognition performance, significantly improving the quantitative accuracy and reliability of abrasive particle detection, and solving the core problem of insufficient counting accuracy of traditional methods.

[0106] The image sensor's ability to resolve overlapping issues effectively enables depth separation through dynamic focusing and 3D reconstruction technology. The dynamic focusing unit captures clear images at different depth levels, while the 3D reconstruction module of the multi-layer image acquisition module identifies the spatial relationship between overlapping wear particles. The image sensor reconstructs the outline of obscured wear particles based on information fusion from multiple depth images. The visible portion of the wear particle is extracted from each depth level, and these visible portions are stitched together to form a complete wear particle boundary. The shape characteristics of the wear particle are extracted from the reconstructed outline to obtain complete topographic information of the obscured wear particle, effectively resolving the information loss problem in overlapping wear particle detection.

[0107] Image sensor occlusion issues are addressed through multi-view imaging and stereo vision reconstruction, enabling omnidirectional observation. The multi-view imaging unit captures image data of the same abrasive particle from different angles, while the stereo vision processing module calculates the spatial correspondence between images from each perspective. The image sensor constructs a complete spatial model of the abrasive particle based on the multi-view image data, using 3D point cloud data to represent the particle's surface geometry. This generates a 3D mesh model of the abrasive particle for feature analysis, overcoming the limitations of single-view observation and providing comprehensive topographic information.

[0108] The reliability of image sensor topographic feature extraction is enhanced through multi-dimensional information fusion and high-precision image processing. The feature extraction network extracts detailed texture features of the wear particle surface from high-resolution images, using a multi-scale filter to preserve the original texture details. The high-precision geometric parameter measurement of the image sensor is derived from sub-pixel boundary detection and contour fitting technology. Sub-pixel interpolation is used to improve boundary positioning accuracy, and curve fitting is used to reduce measurement errors, ensuring high fidelity and measurement accuracy of wear particle feature information.

[0109] Comparisons between image sensors and traditional two-dimensional image analysis methods demonstrate that they provide more comprehensive wear particle information. Traditional methods only provide projection profiles of wear particles on a single plane, while three-dimensional reconstruction technology captures the complete spatial shape of wear particles. Image sensors identify material composition through multi-wavelength spectral analysis. The spectral analysis unit extracts the reflective properties of wear particles at different wavelengths and matches these spectral signatures with a material database for identification. The image sensor's comprehensive information captures multi-dimensional data on shape, surface texture, material composition, and spatial position, providing more complete and accurate wear particle feature information for oil quality analysis, significantly improving the comprehensiveness and accuracy of detection and analysis.

[0110] To validate the technical effectiveness of the image sensor of this application, a standardized test protocol was designed to compare and evaluate the conventional two-dimensional image detection solution with the multi-dimensional detection solution of this application. The test environment utilized a standard oil quality testing laboratory platform equipped with a constant temperature control system to maintain the oil temperature at 25±1°C. A standard abrasive sample library containing ferromagnetic, copper, and aluminum abrasives of varying materials, ranging in size from 5μm to 50μm, was used. During sample preparation, standard abrasives of known quantity and characteristics were dispersed in a transparent carrier oil. A magnetic field was then used to control the formation of varying degrees of aggregation, including four typical aggregation patterns: slight contact, partial overlap, complete aggregation, and multi-layer stacking. Comparative tests utilized a conventional single-view CCD image sensor with conventional image processing software. Example 1 employed the complete technical solution of this application but with a fixed parameter configuration, while Example 2 employed the technical solution of this application with the parameter optimization function of the adaptive learning unit activated. During the tests, each method processed the same 100 sets of standard samples, each containing 20 to 80 abrasives of varying sizes and materials. Manual microscopy was used to establish a database of standard answers for accuracy evaluation. The performance indicators are calculated as follows: the abrasive particle identification accuracy is equal to the number of correctly identified abrasive particles divided by the total number of standard abrasive particles; the success rate of aggregated abrasive particle separation is equal to the number of successfully separated aggregated abrasive particles divided by the total number of initial aggregated abrasive particles; the integrity of overlapping abrasive particle detection is equal to the number of detected complete outlines of overlapping abrasive particles divided by the total number of actual overlapping abrasive particles; and the accuracy of morphological feature extraction is calculated by the relative error between the measurement results and the standard value. The test results are as follows: Table 1 Performance comparison test results of the technical solution of this application and the traditional technical solution As shown in Table 1, test data demonstrates that the multi-dimensional detection technology solution of this application achieves significant improvements across all key performance indicators compared to traditional two-dimensional image detection technology solutions. Example 1 of this application utilizes a standard configuration of a multi-view imaging unit, spectral analysis unit, dynamic focusing unit, and acoustic wave separation unit, improving wear particle identification accuracy from 72.3% of traditional methods to 94.6%, and significantly increasing the success rate of aggregated wear particle separation from 31.5% to 87.2%. Example 2, building on Example 1, utilizes an optimized configuration of the adaptive learning unit, further improving wear particle identification accuracy to 96.8% and the success rate of aggregated wear particle separation to 91.4%. This application effectively addresses the issue of missing information in overlapping wear particle detection through multi-view stereo imaging, improving the integrity of overlapping wear particle detection from 45.8% of traditional methods to 92.7% in Example 2. The spectral analysis unit enables this application to possess material recognition capabilities, achieving material recognition accuracy rates of 88.7% to 91.3%, a feature not available with traditional methods. The dynamic focusing unit enables this application to support up to 12 layers of depth-level recognition, and the three-dimensional spatial positioning accuracy reaches ±1.8μm, significantly improving the stereoscopic level and accuracy of detection.

[0111] Image sensors have significantly enhanced their technical support for oil quality assessment and equipment status diagnosis. The multi-dimensional wear particle data they output is used to establish quantitative analysis models for the wear process. The wear state of equipment components can be inferred based on the size distribution, shape characteristics, and material composition of the wear particles. Image sensors determine the type and severity of wear based on wear particle characteristics, distinguishing between normal, abnormal, and severe wear. Historical wear particle data is used to train a predictive model for equipment health status, establishing a relationship between wear particle characteristics and equipment's remaining lifespan. This enables wear particle analysis-based equipment status prediction and maintenance decision support.

[0112] The above is a detailed introduction to the embodiments of the present application. The contents of this specification should not be understood as limiting the scope of protection of the present application.

Claims

1. An image sensor for online detection of oil quality, characterized in that: include: A sensor body, the sensor body including a multi-view observation chamber, the multi-view observation chamber being in the shape of a polygonal frustum, the multi-view observation chamber being provided with a multi-layer optical window assembly, the multi-layer optical window assembly including a main observation window located at the center of the chamber top and a plurality of auxiliary observation windows radially distributed around the main observation window, the plurality of auxiliary observation windows being arranged inclined toward the inside of the chamber; A multi-view imaging unit, comprising a main imaging component and an auxiliary imaging component, wherein the main imaging component is used to obtain a top-view image of the abrasive particles through the main observation window, and the auxiliary imaging component is used to obtain a side-view angle image of the abrasive particles through the auxiliary observation window; An acoustic wave separation unit, comprising an ultrasonic transmitter array, a phase control circuit, and an acoustic wave focusing lens. The ultrasonic transmitter array is located at the bottom of the detection chamber. The phase control circuit is used to calculate the phase delay value of each transducer based on the target focus position and generate a multi-channel drive signal with a specific phase relationship. The acoustic wave focusing lens is used to perform a phase modulation operation on the incident acoustic wave. A dynamic focusing unit includes a liquid lens assembly and a focal length control circuit. The liquid lens assembly includes a double-layer liquid medium of conductive liquid and insulating oil. Transparent electrodes surround the lens cavity to form an electric field control area. The focal length control circuit is used to apply a variable voltage to the transparent electrodes to generate a change in electric field intensity to adjust the optical focal length of the liquid lens. as well as The analysis unit includes an aggregation detection network, a separation planning network and a feature extraction network. The aggregation detection network is used to perform a feature mapping extraction operation on the input image to identify the wear particle aggregation phenomenon. The separation planning network is used to generate an execution plan for acoustic wave separation based on the aggregation detection results. The feature extraction network is used to perform deep feature analysis on the separated independent wear particles.

2. The image sensor according to claim 1, wherein The circumferential angle interval between every two adjacent auxiliary windows is 45 degrees, and the angle between the window plane of each auxiliary observation window and the top surface of the chamber is 30 degrees.

3. The image sensor according to claim 1, wherein The multi-view imaging unit also includes a field of view calculation unit and a stereo vision processing module. The field of view calculation unit is used to calculate the field of view boundary data of each window, and the stereo vision processing module is used to allocate image data to corresponding processing channels according to the view identification.

4. The image sensor according to claim 1, wherein The acoustic wave separation unit also includes an amplitude modulator and an acoustic field distribution controller. The amplitude modulator is used to calculate the required acoustic radiation force based on the abrasive aggregation information and convert it into a transducer driving power requirement. The acoustic field distribution controller is used to control the acoustic wave focus to move within the detection area along a predetermined path.

5. The image sensor according to claim 1, wherein The dynamic focusing unit also includes a focus position detector and a multi-layer image acquisition module. The focus position detector is used to emit a reference laser beam to the detection area and calculate the focus distance based on the reflected laser. The multi-layer image acquisition module is used to continuously acquire image data at different focal length positions and construct a three-dimensional image data stack.

6. The image sensor according to claim 1, wherein It also includes a spectrum analysis unit, which includes a multi-wavelength LED array and a spectrum filter group. The multi-wavelength LED array is used to activate light sources of different wavelengths in sequence according to a timing configuration table, and the spectrum filter group is used to receive a wavelength synchronization signal and rotate to a predetermined angle to achieve wavelength switching.

7. The image sensor according to claim 1, wherein The aggregation detection network includes a spatial density analysis module, a shape anomaly assessment module and a boundary continuity detection module. The spatial density analysis module is used to count the number of abrasive feature points in a local area and calculate the density value. The shape anomaly assessment module is used to calculate the geometric characteristic parameters of the abrasive profile and compare them with the standard template. The boundary continuity detection module is used to analyze the breakpoints and curvature changes of the profile to determine the degree of aggregation tightness.

8. The image sensor according to claim 1, wherein The separation planning network includes a frequency selection module, a power allocation calculation module and an action time planning module. The frequency selection module is used to calculate the optimal acoustic wave frequency according to the abrasive size distribution, the power allocation calculation module is used to determine the required acoustic wave power level according to the degree of aggregation, and the action time planning module is used to calculate the duration of the acoustic wave action and divide it into multiple short pulse periods.

9. The image sensor according to claim 1, wherein: The feature extraction network includes a boundary detection convolution layer, a contour refinement module and a multi-scale feature fusion module. The boundary detection convolution layer is used to detect the irregular shape contour of the wear particle using a deformable convolution kernel. The contour refinement module is used to perform sub-pixel level processing on the boundary contour. The multi-scale feature fusion module is used to process image data of different resolutions and extract detail features and shape features.

10. The image sensor according to claim 1, wherein It also includes an adaptive learning unit, which includes an aggregation pattern recognition module, a parameter optimization module and a knowledge base update module. The aggregation pattern recognition module is used to establish a characteristic data index table of the abrasive aggregation pattern, the parameter optimization module is used to dynamically adjust the acoustic wave parameters according to the separation effect, and the knowledge base update module is used to identify high-quality processing cases and update the database.

Images