A physical field perception-based stirring mixing uniformity evaluation method and system

Abstract

The present application belongs to the technical field of monitoring, evaluation and regulation of stirring mixing process, and discloses a stirring mixing uniformity evaluation method and system based on physical field sensing, which comprises the following steps: acquiring a plurality of image sequences of the flow process in the stirring tank; constructing an image-physical fusion mixing uniformity evaluation model and training the same; inputting the plurality of image sequences into the trained mixing uniformity evaluation model to realize real-time monitoring and intelligent regulation of the stirring process. The present application realizes adaptive optimization of stirring power and mixing effect through an energy efficiency-uniformity coupling control algorithm, and can display images, flow field vectors and uniformity index changes in real time on the host computer interface. The method has high precision and strong universality, and is suitable for flow field structure analysis and intelligent monitoring and energy consumption optimization of the stirring process.

Description

Technical Field

[0001] This invention belongs to the field of monitoring, evaluation and control technology of mixing process, and specifically relates to a method and system for evaluating the uniformity of mixing based on physical field perception. Background Technology

[0002] The mixing process is widely used in industrial production such as hydrometallurgy, chemical industry, food processing, and pharmaceuticals. It is a crucial unit operation for achieving mass transfer, energy exchange, and stable operation of the reaction process. The uniformity of mixing directly affects the reaction rate, product quality, and energy consumption. However, due to factors such as differences in fluid viscosity, diverse impeller structures, and variations in operating speed, the flow state within the stirred tank is usually quite complex, easily forming areas of insufficient mixing such as flow dead zones, low-shear zones, or stagnant zones. This leads to inadequate local mixing, reduced energy utilization efficiency, and decreased process stability.

[0003] Currently, research and evaluation of mixing processes mainly rely on numerical simulation and experimental measurement methods. While numerical simulation methods can describe the flow field, they typically rely on many idealized assumptions, have high computational costs, and struggle to meet the real-time and flexibility requirements of engineering applications. Experimental measurement methods, such as sampling analysis or tracer dye methods, have complex procedures, slow response times, and limited applicability under opaque or high-viscosity fluid conditions. On the other hand, commonly used online monitoring signals, such as temperature, pressure, or motor torque, are mostly parameters of the overall order of magnitude, which are difficult to reflect the spatial distribution characteristics and transient evolution of the flow field, making it difficult to achieve a unified and comparable evaluation of mixing uniformity under different impeller types, rotational speeds, and fluid systems.

[0004] In recent years, diagnostic techniques based on flow visualization and physical field reconstruction have gradually attracted attention. High-speed imaging and particle image velocimetry can acquire velocity vector field information, providing new means to reveal the structure and mixing mechanism of stirred flows. However, existing methods often focus on single physical quantities or image information, lacking the synergistic fusion of image grayscale features with shear rate, eddy current, Q-criterion, and energy-related physical quantities. This makes it difficult to achieve accurate identification of different flow regions and unified quantitative evaluation of the mixing state. Furthermore, the lack of an effective coupling mechanism between mixing evaluation results and energy consumption feedback makes it difficult to achieve energy efficiency regulation and operational optimization of the stirring process while ensuring mixing effects.

[0005] Therefore, there is an urgent need for a physical field-sensing-based method and system for evaluating the uniformity of mixing, which can systematically characterize the flow field characteristics and mixing state during the mixing process under different impeller types, rotation speeds, and fluid conditions, establish a unified and comparable evaluation index system for mixing uniformity, realize quantitative analysis of the entire process from image acquisition, physical feature calculation, flow region identification to uniformity score output, and provide a reliable basis for the operation control and energy efficiency optimization of the mixing process. Summary of the Invention

[0006] To address the problems of insufficient real-time performance, weak feature correlation, lack of energy consumption coupling control, and difficulty in adapting to different mixing conditions in existing methods for evaluating mixing uniformity, this invention provides a mixing uniformity evaluation method and system based on physical field perception. This method integrates image features and fluid physics features to achieve accurate identification and quantitative evaluation of the flow state under different impeller types, rotation speeds, and fluid systems. Furthermore, it achieves adaptive optimization control of the mixing process through an energy efficiency-uniformity coupling algorithm.

[0007] To achieve the above objectives, the present invention provides the following solution: A method for evaluating the uniformity of stirring and mixing based on physical field sensing, the method comprising: Acquire a multi-frame image sequence of the flow process inside the stirred tank; Construct and train a hybrid uniformity evaluation model for image-physical fusion; The multi-frame image sequence is input into the trained mixing uniformity evaluation model to achieve real-time monitoring and intelligent control of the stirring process.

[0008] Preferably, the method for real-time monitoring and intelligent control of the stirring process by inputting the multi-frame image sequence into the trained mixing uniformity evaluation model includes: The target area of ​​the study is obtained by cropping a multi-frame image sequence of the flow process in the stirred tank; the target area image is then processed by flat field correction, background subtraction and impeller masking; the flow field velocity vector is calculated using the particle image velocimetry method to obtain the velocity components and velocity amplitude distribution, thereby realizing the spatiotemporal reconstruction of the flow field. Based on the velocity field, shear rate, vorticity, Q criterion, and energy consumption proxy are calculated, and physical features are fused with image grayscale features to form a multi-channel input tensor for identification. The tensor identifies and segments the "dead flow zone", "stagnant zone" and "low shear zone" in the stirring tank and outputs the corresponding segmentation mask. Calculate the dead area ratio (DR), low shear ratio (LSR), gray entropy (H), and average energy consumption based on the segmentation mask. Construct a unified mixed homogeneity score U This enables quantitative evaluation and visual analysis of the stirring process under different impeller types, rotation speeds, and fluid systems.

[0009] Preferably, the method for performing flat-field correction, background subtraction, and blade masking on the image of the target area includes: Let the original image be , Represents the horizontal spatial coordinates in the image; Represents the spatial coordinates in the vertical direction of the image; The time dimension indicates the acquisition time of the corresponding image. First, the target region is cropped from the original image to obtain the image to be analyzed. Then, flat-field correction is performed, and the corrected image is... Represented as: ; in, This is a dark field image. This is a flat-field reference image acquired under particle-free, uniform illumination conditions. To prevent the use of tiny constants with a denominator of zero, after flat-field correction, background subtraction is employed to suppress reflections from the container walls and fix background interference. The image after background subtraction... Represented as: ; in, The background image is constructed from images of a flow-free state or averaged images from multiple frames; based on this, a binary mask is constructed for the blade region and its highly reflective regions. The image is then processed by blade masking to obtain a preprocessed image. : ; in, This indicates the blades and their reflective area. This indicates the effective flow field region.

[0010] Preferably, the method for calculating the flow field velocity vector using particle image velocimetry to obtain the velocity components and velocity amplitude distribution, and realizing the spatiotemporal reconstruction of the flow field includes: Preprocessed images at adjacent time points and The calculation is divided into multiple PIV calculation windows, and the average displacement of the tracer particles within each window is obtained by cross-correlation matching. The velocity components are obtained: ; in, This indicates the time interval, which is the time difference between the acquisition of two adjacent image frames, in seconds. The velocity amplitude is: .

[0011] Preferably, the method for calculating shear rate, vorticity, Q criterion, and energy consumption proxy based on velocity field, and fusing physical features with image grayscale features to form a multi-channel input tensor for identification, identifying and segmenting "dead flow zones," "stagnant zones," and "low shear zones" within the stirred tank, and outputting corresponding segmentation masks includes: ; ; in, For pixels Belongs to the The probability of a class region; Number of categories; Represents pixels Belongs to the Predicted scores for class regions.

[0012] Preferably, the dead area ratio (DR), low shear ratio (LSR), gray entropy (H), and average energy consumption are calculated based on the segmentation mask. Construct a unified mixed homogeneity score U The methods include: Among them, weight parameters , , , The settings are determined based on different mixing objectives, process stages, and energy consumption constraints. To normalize the grayscale entropy, This represents the normalized average energy consumption.

[0013] The present invention also provides a physical field-sensing-based system for evaluating the uniformity of mixing, the system being used to implement the aforementioned method, the system comprising: The stirring unit includes a stepper motor, a stirring paddle, and a motor driver; the stepper motor is controlled by a control module via a motor driver to set its speed and direction, thereby achieving precise adjustment of the stirring process; the stirring paddle is disposed in a stirring tank, which is preferably a transparent structure, for driving the fluid containing tracer particles to flow. The optical illumination unit includes a laser source and a light sheet imaging device. The laser is formed into a thin light sheet through a cylindrical lens to illuminate a designated plane in the stirring tank, making the movement trajectory of the tracer particles visible. The image acquisition unit includes a high-speed camera with a sampling frame rate of 500 to 2000 frames per second, used to acquire images of particle motion in the fluid in real time from the illuminated area; the high-speed camera is synchronized with the stepper motor control terminal through a hardware trigger signal to realize the time correspondence between image acquisition and stirring action; The data processing unit includes a host computer and an image processing software module, which is used to receive the raw image sequence transmitted by the high-speed camera, perform flat field correction, background subtraction and blade masking on the image, and use the particle image velocimetry algorithm to calculate the velocity vector field and extract shear rate, vorticity, Q criterion and energy consumption proxy. The hybrid uniformity analysis unit includes an image-physical fusion feature recognition module and a uniformity calculation module. It fuses image grayscale features with physical field features, identifies "flow dead zones," "stagnant zones," and "low-shear zones" through a physically guided fusion network, and calculates the dead zone area ratio, low-shear zone proportion, grayscale entropy, and average energy consumption to construct a unified hybrid uniformity score. U It also outputs a uniformity score; The control and visualization unit includes a host computer display interface and a control module. The host computer displays the original image, velocity vector field, segmentation results, and mixing uniformity score change curve in real time; when U When the speed is below a set threshold, the control module automatically increases the motor speed. U When the target is reached but energy consumption is high, the control module reduces the input power of the motor by sending speed reduction, frequency reduction, or current limiting commands to the motor driver, thereby achieving closed-loop optimization control of stirring energy efficiency.

[0014] Preferably, the stirring tank is a transparent structure, which facilitates optical observation and particle image velocity measurement; the stirring paddle is an arc-shaped paddle, a straight-blade paddle, or a spiral paddle structure, and the paddle surface is treated with matte sandblasting and black matte coating.

[0015] Preferably, the host computer display interface includes an original image display area, a velocity field vector overlay area, a region segmentation display area, and a mixing uniformity score curve display area. The user sets the target mixing uniformity score in the control module. U Energy consumption upper limit and control strategy parameters are set to achieve automatic control under different mixing conditions.

[0016] Preferably, the control module is connected to the stepper motor driver via Ethernet or Modbus industrial communication protocol, and achieves real-time data interaction with the host computer; the control module has a built-in energy efficiency-uniformity control algorithm, which determines the mixing uniformity score. U When the speed is below a set threshold, the control module outputs an acceleration command to increase the motor speed; when... U When the target is reached but energy consumption is too high or exceeds the upper limit, the control module reduces the input power of the motor by sending speed reduction, frequency reduction or current limiting commands to the motor driver, thereby achieving closed-loop optimization control of stirring energy efficiency.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention integrates the image information of the stirring process with physical features such as velocity field, shear rate, vorticity, Q criterion and energy consumption related characterization quantities, which can identify and spatially segment different flow regions such as flow dead zone, stagnation zone and low shear zone in the stirring tank, thereby more comprehensively characterizing the spatial distribution characteristics of the flow field.

[0018] (2) The present invention constructs a unified mixing uniformity scoreU By integrating evaluation factors such as dead zone area ratio, low shear zone ratio, gray entropy, and average energy consumption, the evaluation of the mixing state under different propeller types, rotational speeds, and fluid system conditions has better consistency and comparability.

[0019] (3) The system of the present invention has online calculation and visualization capabilities, and can display the original image, velocity field information, segmentation results and... U The scoring curve is used to correlate the evaluation results with operating parameters such as stirring power and speed, providing a basis for adjusting the operation of the stirring process, so as to achieve energy efficiency management under energy consumption constraints while meeting mixing requirements.

[0020] (4) This invention is applicable to a variety of stirring conditions and media conditions, and has good versatility and scalability, making it easy to promote in laboratory scale and industrial application scenarios. Attached Figure Description

[0021] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the system structure according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the image preprocessing results of stirring and mixing in an embodiment of the present invention, wherein (a) is the original image acquired by a high-speed camera, (b) is the image after flat field correction and background subtraction, and (c) is the image after blade masking processing based on (b). In the diagram: 1. Stirring unit; 1-1. Stepper motor; 1-2. Stirring paddle; 1-3. Motor driver; 1-4. Stirring tank; 2. Optical illumination unit; 2-1. Laser light source; 2-2. Light sheet imaging device; 3. Image acquisition unit; 4. Data processing unit; 5. Mixing uniformity analysis unit; 6. Control and visualization unit; 6-1. Host computer display interface; 6-2. Control module. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0025] Example 1 This invention provides a method for evaluating the uniformity of mixing based on physical field perception. This method combines image information and fluid dynamics characteristics to analyze the flow state during mixing. It quantitatively evaluates the mixing uniformity using shear rate, vorticity, grayscale characteristics, and energy consumption-related parameters, providing a basis for adjusting the operation of the mixing process. This invention is applicable to industrial and laboratory applications involving fluid or slurry mixing processes, such as hydrometallurgy, chemical engineering, food processing, and pharmaceuticals. Specifically, the method is suitable for laboratory-scale simulation and measurement of actual industrial mixing processes. It is used to quantitatively characterize the mixing state under controlled experimental conditions for different mixing speeds, different mixing media, different impeller structures, different fluid rheological properties, device structural parameters, and operating conditions. By changing the mixing speed, medium properties, fluid rheological properties, impeller structure, and device structural parameters in a laboratory-scale mixing device, image information and corresponding physical field characteristics or physical field proxy characteristics of the mixing process are obtained, enabling comparative evaluation of mixing uniformity under different operating conditions. The evaluation results are used to analyze the impact of different combinations of industrial mixing parameters on mixing effects and energy consumption characteristics, providing experimental basis for parameter selection, process scale-up, and intelligent optimization of industrial mixing processes.

[0026] The method includes: Acquire a multi-frame image sequence of the flow process within stirring tanks 1-4; Construct and train a hybrid uniformity evaluation model for image-physical fusion; The multi-frame image sequence is input into the trained mixing uniformity evaluation model to achieve real-time monitoring and intelligent control of the stirring process.

[0027] In this embodiment, the method for real-time monitoring and intelligent control of the stirring process by inputting the multi-frame image sequence into the trained mixing uniformity evaluation model includes: The target area of ​​the study was obtained by cropping a multi-frame image sequence of the flow process in the stirred tank 1-4; the target area image was processed by flat field correction, background subtraction and blade masking; the flow field velocity vector was calculated by the particle image velocimetry method to obtain the velocity components and velocity amplitude distribution, and the spatiotemporal reconstruction of the flow field was realized. Based on the velocity field, shear rate, vorticity, Q criterion, and energy consumption proxy are calculated, and physical features are fused with image grayscale features to form a multi-channel input tensor for identification. The tensor identifies and segments the "flow dead zone", "stagnant zone" and "low shear zone" in the stirring tank 1-4, and outputs the corresponding segmentation mask. Calculate the dead area ratio (DR), low shear ratio (LSR), gray entropy (H), and average energy consumption based on the segmentation mask. Construct a unified mixed homogeneity score U This enables quantitative evaluation and visual analysis of the stirring process under different impeller types, rotation speeds, and fluid systems.

[0028] In this embodiment, energy consumption agent The calculation formula is: ; Where μ(γ̇) is the equivalent viscosity function, γ̇ is the shear rate, and μ(γ̇) is expressed as: ; in, is the consistency coefficient, and n is the rheological index.

[0029] In this embodiment, the dead area ratio (DR), low shear ratio (LSR), gray entropy (H), and average energy consumption are calculated based on the segmentation mask. Construct a unified mixed homogeneity score U The methods include: Among them, weight parameters , , , The settings are determined based on different mixing objectives, process stages, and energy consumption constraints. To normalize the grayscale entropy, This represents the normalized average energy consumption.

[0030] Example 2 This embodiment provides a method for evaluating the uniformity of stirring and mixing based on physical field perception, applicable to visualized stirring processes using glycerol-water solutions as the medium and incorporating tracer particles. For example... Figure 1 As shown, it includes the following steps: S1: Construct an image-physical fusion model for evaluating the uniformity of mixing, and train the model using labeled sample data to match the output mask with the real mixing region, thereby enabling the model to identify the mixing state based on image-physical information.

[0031] S2: A multi-frame image sequence of the flow process in the stirred tank 1-4 is acquired by a high-speed camera. First, the target area of ​​the study is cropped out, and then the images are processed by flat field correction, background subtraction and blade masking. The flow field velocity vector is calculated by the particle image velocimetry (PIV) method to obtain the velocity components and velocity amplitude distribution, so as to realize the spatiotemporal reconstruction of the flow field.

[0032] S3: Based on the aforementioned mixing uniformity evaluation model, calculate physical parameters such as shear rate, vorticity, Q criterion, and energy consumption proxy, and fuse the physical features with image grayscale features to form a multi-channel input tensor for identification. The feature fusion network input to the model identifies and segments the "flow dead zone," "stagnant zone," and "low shear zone" within the mixing tanks 1-4, outputting the corresponding segmentation mask; specifically: Based on the preprocessed multi-frame image sequence, the velocity field at each moment within the measurement sections 1-4 of the stirred tank is calculated using the Particle Image Velocimetry (PIV) method, yielding the velocity components. , and velocity amplitude ,in: in, Represents the horizontal spatial coordinates in the image; Represents the spatial coordinates in the vertical direction of the image; It represents the time dimension, indicating the time point when the corresponding image was acquired.

[0033] Based on the aforementioned velocity field, the velocity gradient tensor is further calculated: The shear rate is calculated based on the velocity gradient tensor. vorticity Q criterion characterization quantity and energy consumption agency The velocity amplitude, shear rate, vorticity, Q-criterion characterization, and energy consumption proxy are used to construct a physical feature tensor according to their corresponding spatial locations in the image: Meanwhile, the preprocessed image is denoted as The image feature encoding branch is then used to extract image grayscale features that characterize the distribution of tracer particles, grayscale changes, texture boundaries, and local blending states. ,in: in, Represents the image feature coding branch. Layer encoding mapping; The physical feature tensor Input the physical feature encoding branch to extract physical features. ,in: in, The physical feature coding branch is represented by the first Layer encoding mapping; The feature fusion network, located after the image feature coding branch and the physical feature coding branch, includes a physically constrained attention fusion unit and a multi-scale decoder. It receives the output features from the two branches and performs cross-branch fusion. During the fusion process, physically constrained attention weights are generated from the physical features. And perform weighted enhancement on the image grayscale features: in, This is the activation function.

[0034] Then enhance the grayscale features of the image With physical characteristics Channel concatenation and convolution fusion are performed to obtain fused features: The fused features from each scale are then input into a multi-scale decoder to restore the spatial resolution and output prediction score maps for each category: in, This indicates a multi-scale decoder. Represents pixels Belongs to the The predicted score of each region; after softmax normalization, the probability of each pixel belonging to a different region is obtained: in, The number of categories is defined, and the categories include at least background regions, dead zones, stagnant regions, and low-shear regions; a segmentation mask is generated based on the maximum probability principle. This enables the identification and spatial segmentation of flow dead zones, stagnant zones, and low-shear zones within the mixing tanks 1-4, and outputs the corresponding segmentation masks.

[0035] Among them, the flow dead zone is used to characterize regions with low local velocity amplitude and shear rate, and weak fluid renewal ability; the stagnant zone is used to characterize regions with low local velocity and small positional changes in continuous time-series images; the low-shear zone is used to characterize regions with local shear rate below a preset threshold, but where some slow flow may still exist. In one embodiment, segmentation can be based on the region probability output by the network; in another embodiment, the segmentation result can also be constrained and corrected by combining physical parameter thresholds, wherein: The corresponding pixel is then prioritized as a dead zone. The corresponding pixel is then identified as a stagnant region. The corresponding pixel is then prioritized as a low-shear region.

[0036] in, , , , and These are threshold parameters determined based on the fluid system, impeller structure, operating speed, or calibration sample.

[0037] S4: Calculate the dead area ratio (DR), low shear ratio (LSR), gray entropy (H), and average energy consumption based on the segmentation mask. Construct a unified mixed homogeneity score U It outputs uniformity scores and visualization results, and associates the evaluation indicators with parameters such as stirring power and speed to achieve real-time monitoring and intelligent control of the stirring process.

[0038] In a preferred embodiment, the image-physical fusion hybrid uniformity evaluation model may include an image preprocessing module, a physical feature calculation module, a feature fusion and segmentation module, and a uniformity calculation module; the above module division is used to illustrate a preferred processing flow for implementing hybrid uniformity evaluation according to the present invention, and does not constitute a limitation on the internal structure division of the model: Image preprocessing module: Performs target region cropping, flattening correction, background subtraction, and impeller masking on the acquired images of the mixing process to eliminate interference from uneven lighting, container reflections, fixed backgrounds, and impeller occlusion on subsequent velocity field reconstruction and region segmentation; whereby the original image is... First, the target region is cropped from the original image to obtain the image to be analyzed. Then, flat-field correction is performed, and the corrected image is... It can be represented as: in, This is a dark field image. This is a flat-field reference image acquired under particle-free, uniform illumination conditions. To prevent the use of tiny constants with a denominator of zero, after flat-field correction, background subtraction is employed to suppress reflections from the container walls and fix background interference. The image after background subtraction... It can be represented as: in, The background image is constructed from images of a flow-free state or averaged images from multiple frames; based on this, a binary mask is constructed for the blade region and its highly reflective regions. The image is then processed by blade masking to obtain a preprocessed image. : in, This indicates the blades and their reflective area. This represents the effective flow field region. After the above processing, the preprocessed image... Used for subsequent particle image velocimetry calculations and image feature extraction.

[0039] Unlike existing methods that directly perform particle image velocimetry or grayscale analysis on the original stirring images, this invention introduces flat field correction, background subtraction, and impeller masking processing sequentially before velocity field reconstruction. This suppresses the influence of uneven illumination, wall reflection, and periodic impeller occlusion on velocity field calculation and subsequent region identification, thereby improving the stability of image-physics co-analysis under high viscosity systems and complex stirring conditions.

[0040] Physical Feature Calculation Module: Used to perform PIV calculation on the image sequence processed by the image preprocessing module to obtain the velocity field, and extract the shear rate based on the velocity field. vorticity Q criterion characterization quantity and energy consumption agency Among them, preprocessed images at adjacent time points are... and The calculation is divided into multiple PIV calculation windows, and the average displacement of the tracer particles within each window is obtained by cross-correlation matching. The velocity components are obtained: in, This indicates the time interval, which is the time difference between the acquisition of two adjacent image frames, in seconds.

[0041] The velocity amplitude is: Based on the aforementioned velocity field, the velocity gradient tensor is further calculated. The shear rate calculation unit is used to characterize the local fluid deformation intensity, and its calculation formula is: The vorticity calculation unit is used to characterize the local rotation intensity, and its calculation formula is as follows: The Q criterion calculation unit is used to distinguish between rotation-dominated and strain-dominated regions, and its characterization quantity can be written as: in, For strain rate tensor, The rotation tensor is used to characterize the local energy dissipation intensity per unit volume of fluid during stirring; its calculation formula is: For fluids exhibiting power-law rheological properties, the equivalent viscosity function satisfies: Therefore: in, This is the consistency coefficient. The rheological index is used. Furthermore, the velocity amplitude, shear rate, vorticity, Q-criterion characterization, and energy consumption proxy are spatially aligned in the image coordinate system to form a physical feature tensor: Used as input for subsequent physical feature encoding branches.

[0042] In a preferred embodiment, the feature fusion and segmentation module can be implemented using a dual-branch feature extraction and fusion structure; wherein, an image feature coding branch for extracting image appearance information and a physical feature coding branch for extracting flow physical features can be provided. Preferably, the feature fusion and segmentation module may further include a physical constraint fusion unit and a multi-scale decoding structure to achieve feature fusion and region segmentation.

[0043] In one embodiment, the image feature encoding branch can be used to extract grayscale distribution, grain texture, edge changes, and local mixing interface information from the preprocessed stirring and mixing image, thereby forming multi-scale image features, which can be represented as: in, The image after preprocessing. Represents the image feature coding branch. Layer encoding mapping, Indicates the first Layer image features.

[0044] In one implementation, the physical feature encoding branch can be used to encode the physical feature tensor composed of velocity amplitude, shear rate, vorticity, Q-criterion characterization, and energy consumption surrogate, to extract multi-scale physical features characterizing flow intensity, shear variation, rotational structure, and energy distribution differences, which can be expressed as: in, For physical characteristic tensors, The physical feature coding branch is represented by the first Layer encoding mapping, Indicates the first Layer physical characteristics.

[0045] Preferably, the physical constraint fusion unit can be positioned after the image feature coding branch and the physical feature coding branch, and is used to receive the features output by the two branches and perform cross-branch fusion. During the fusion process, attention weights can be generated from the physical features. : And perform weighted enhancement on image features: Then enhance the image features With physical characteristics Channel concatenation and convolution fusion are performed to obtain fused features: in, For activation function, This represents element-wise multiplication. This indicates channel splicing.

[0046] In one implementation, the multi-scale decoding structure can be used to perform stepwise upsampling and feature recovery on the fused features at each scale, and output a prediction score map of each pixel belonging to different flow regions: After softmax normalization, the pixels are obtained. Belongs to the Probability of class region: in, It includes at least the background region, dead flow region, stagnant region, and low shear region; a segmentation mask is generated based on the maximum probability principle: This enables the identification and spatial segmentation of different flow zones within the mixing tanks 1-4.

[0047] Uniformity Calculation Module: The uniformity calculation module is used to calculate the dead zone area ratio (DR), low shear zone ratio (LSR), gray-level entropy (H), and average energy consumption. A uniformity score for mixing is generated based on the following unified evaluation formula. : in, dead zone area ratio, For low shear zone ratio, To normalize the grayscale entropy, This represents the normalized average energy consumption. The weighting parameters... , , , The settings are configured according to different mixing objectives, process stages and energy consumption constraints, so as to constrain energy consumption indicators while meeting the requirements of mixing uniformity, or to prioritize the mixing effect under energy-limited conditions, thereby achieving a coordinated trade-off between mixing effect and energy consumption level.

[0048] In this embodiment, the weighting parameter The weighting parameters are set or adjusted according to the stirring objectives, process stages, and energy consumption constraints. Preferably, the weighting parameters satisfy: And the dead zone area is greater than Low shear zone ratio Gray entropy and average energy consumption Normalization is performed before substituting into the unified mixing uniformity scoring formula.

[0049] When in the initial mixing stage or when high uniformity is required, priority should be given to reducing... and And enhance the response to changes in mixed states, taking a larger value. and And appropriately increase the size based on the sensitivity of grayscale texture to changes in blending state. At the same time reduce Preferably, in one embodiment, the following can be taken: When the mixed state meets the requirements and it is necessary to reduce energy consumption or meet the energy consumption upper limit constraint, increase and reduce accordingly. Preferably, in another embodiment, the following can be taken: Furthermore, the weighting parameters can also be based on the current mixing uniformity score. Target score The weights are adaptively updated based on energy consumption constraints. The above weight settings and adjustments are examples and do not constitute a limitation on the range of weight values ​​or the update method.

[0050] A further implementation method is that the image-physical fusion hybrid uniformity evaluation model is trained using a semi-supervised learning strategy, including pre-training with weakly labeled samples and performing consistency distillation optimization through high-confidence pseudo-labels and a small number of manually refined samples; the training process uses the Sharpness-Aware Minimization (SAM) algorithm to improve the generalization performance of the model under different lighting conditions and fluid systems.

[0051] A further implementation involves a mixing uniformity score. With stirring power and speed An energy efficiency-uniformity coupling relationship is formed; wherein, under fixed rotational speed or fixed power conditions, the coupling relationship can be represented as a two-dimensional projection curve, while simultaneously considering... , and When the three factors are related, it can be represented as an energy efficiency-uniformity coupled surface.

[0052] The formation process of the coupling relationship includes: at different stirring speeds and corresponding power Under these conditions, image acquisition and mixing uniformity evaluation were performed on the stirring process, resulting in multiple sets of operational samples. ,in The power and rotational speed are normalized. in, The minimum power value represents the minimum power value observed during measurement or experimentation. The maximum power value represents the maximum power value observed during the measurement or experiment. This represents the minimum rotational speed, indicating the minimum rotational speed observed during the measurement or experiment. The maximum value of the rotational speed represents the maximum rotational speed observed during the measurement or experiment.

[0053] Based on the aforementioned samples, a coupling relationship between the uniformity score and the operating parameters is constructed: in, It can be determined through interpolation fitting, piecewise fitting, regression model, or table lookup.

[0054] To characterize the hybrid effect per unit of energy consumption, energy efficiency indicators are further defined. for: in, To prevent the use of tiny constants with a denominator of zero, an operating point that meets the target uniformity requirements and has relatively low energy consumption can be determined based on the aforementioned energy efficiency index and uniformity score. ,For example: in, To set the target uniformity score, This represents the maximum allowable power.

[0055] The control system performs closed-loop optimization and control of the stirring process based on the energy efficiency-uniformity coupling relationship; when the real-time uniformity score is... Below the set target When, control module 6-2 outputs an acceleration command to increase the rotational speed; when And real-time power When the energy efficiency index is lower than the set threshold, the control module 6-2 outputs a power reduction or speed reduction command to achieve energy-saving optimization.

[0056] In one implementation, the control module 6-2 can update the speed command according to the following adjustment rule: in, and For adjustment coefficients, This is the speed command for the next control moment.

[0057] The velocity gradient calculation unit is used to calculate the velocity components. and The velocity gradient tensor at each location within the measurement cross section is calculated, and its expression is as follows: The velocity gradient tensor serves as the basis for subsequent calculations of shear rate, vorticity, and the Q criterion.

[0058] The shear rate calculation unit is used to characterize the local fluid deformation intensity, and its calculation formula is: in, This represents the shear rate.

[0059] The vorticity calculation unit is used to characterize the local rotation intensity, and its calculation formula is as follows: in, It represents vorticity.

[0060] The Q criterion calculation unit is used to distinguish between rotation-dominated and strain-dominated regions, and its characterization quantity can be expressed as: in, For strain rate tensor, For rotation tensors; when This indicates that the local area is dominated by rotational effects. This indicates that the local area is dominated by strain effects.

[0061] The energy consumption proxy calculation unit is used to construct an energy consumption proxy model based on the relationship between local fluid viscosity and shear rate, so as to characterize the local energy dissipation intensity per unit volume of fluid during the stirring process; the local energy consumption proxy The formula for calculation is: in, Let be the equivalent viscosity function. For fluids exhibiting power-law rheological properties, the equivalent viscosity function satisfies: Therefore: in, This is the consistency coefficient. The rheological index; the parameter and It can be determined through experimental calibration, empirical models, or function fitting based on the fluid solid content, temperature, and operating conditions.

[0062] The average energy consumption Local energy consumption agent It is obtained through time averaging, spatial averaging, or statistical processing. In one embodiment, the average energy consumption can be obtained by analyzing the target area. Internal sampling period The local energy consumption agent within the region is used to perform spatiotemporal averaging to obtain: Or it can be expressed in continuous form: in, This indicates the area of ​​the target region or the size of the effective sampling area.

[0063] Furthermore, the velocity gradient, shear rate, vorticity, Q criterion characterization quantity, and local energy consumption proxy can be spatially aligned in the image coordinate system and constitute a physical feature tensor for input to the subsequent physical feature encoding branch.

[0064] Example 3 This invention also provides a physical field-sensing-based stirring and mixing uniformity evaluation system for evaluating and controlling the stirring and mixing uniformity of hydrometallurgical slurry systems. This system is particularly suitable for identifying and quantitatively characterizing flow weakening, stagnation, and low-shear risk regions near the wall of stirring tanks 1-4. Figure 2 As shown, it includes the following modules: Stirring unit 1 includes a stepper motor 1-1, a stirring paddle 1-2, and a motor driver 1-3; the stepper motor 1-1 is controlled by the motor driver 1-3 and the control module 6-2 to set the speed and direction, so as to achieve precise adjustment of the stirring process; the stirring paddle 1-2 is disposed in the stirring tank 1-4, which is preferably a transparent structure, for driving the fluid containing tracer particles to flow.

[0065] Optical illumination unit 2: includes laser light source 2-1 and light sheet imaging device 2-2. The laser is formed into a thin light sheet through a cylindrical lens to illuminate a designated plane in the stirring tank 1-4, so that the movement trajectory of the tracer particles is clearly visible.

[0066] Image acquisition unit 3 includes a high-speed camera with a sampling frame rate of 500 to 2000 frames per second, used to acquire images of particle motion in the fluid in real time from the illuminated area; the high-speed camera is synchronized with the stepper motor 1-1 control terminal through a hardware trigger signal to realize the time correspondence between image acquisition and stirring action.

[0067] Data processing unit 4: Receives the raw image sequence transmitted from the high-speed camera, performs flat-field correction, background subtraction, and blade masking on the images, and calculates the velocity vector field using the Particle Image Velocimetry (PIV) algorithm to extract physical characteristic parameters such as shear rate, vorticity, Q-criterion, and energy consumption proxy. Figure 3 As shown, the image preprocessing includes flat field correction, background subtraction, and blade masking, wherein... Figure 3 (a) is the original image captured by the high-speed camera. Figure 3 (b) is the image after flat field correction and background subtraction. Figure 3 (c) is in Figure 3 (b) The image after blade masking processing. A further implementation method is that flat field correction is used to eliminate uneven illumination, background subtraction is used to suppress container reflection and fix background interference, and blade masking processing is used to shield the blades and their reflective areas to avoid interfering with PIV velocity field calculation and subsequent region segmentation.

[0068] Hybrid Uniformity Analysis Unit 5: This unit fuses image grayscale features with physical field features. Through a physically guided fusion network, it identifies "flow dead zones," "stagnant zones," and "low-shear zones," and calculates the dead zone area ratio, low-shear zone proportion, grayscale entropy, and average energy consumption to construct a unified hybrid uniformity score.U It also outputs a uniformity score.

[0069] Control and visualization unit 6: includes a host computer display interface 6-1 and a control module 6-2. The host computer display interface 6-1 displays the original image, velocity vector field, segmentation results, and mixing uniformity score change curve in real time; when U When the speed is below the set threshold, control module 6-2 automatically increases the motor speed. U When the target is achieved but energy consumption is high, the control module 6-2 sends speed reduction and frequency reduction commands to the motor driver 1-3 to reduce the input power of the motor, thereby achieving closed-loop optimization control of stirring energy efficiency.

[0070] A further embodiment is that the host computer display interface 6-1 includes an original image display area, a velocity field vector overlay area, a region segmentation display area, and a mixing uniformity score curve display area. The user can set the target uniformity score in the control module 6-2. U Energy consumption upper limit and control strategy parameters are set to achieve automatic control under different mixing conditions.

[0071] A further embodiment is that the stirring tank 1-4 is a transparent structure, which facilitates optical observation and particle image velocity measurement; the stirring paddle 1-2 is an arc-shaped paddle, a straight-blade paddle, or a spiral paddle structure, and the paddle surface is treated with matte sandblasting and black matte coating to reduce light reflection interference and improve the clarity of flow field imaging.

[0072] A further implementation involves the control module 6-2 being connected to the motor driver 1-3 via Ethernet or Modbus industrial communication protocol, and interacting with the host computer display interface 6-1 in real time; the control module 6-2 incorporates an energy efficiency-uniformity control algorithm, which, when the mixing uniformity score... U When the speed is below the set threshold, control module 6-2 controls the motor speed to increase. U When the target is achieved but energy consumption is high, the control module 6-2 sends speed reduction, frequency reduction, or current limiting commands to the motor driver 1-3 to reduce the motor input power, thereby achieving closed-loop optimization control of stirring energy efficiency.

[0073] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for evaluating the uniformity of mixing based on physical field perception, characterized in that, The method includes: Acquire a multi-frame image sequence of the flow process inside the stirred tank; Construct and train a hybrid uniformity evaluation model for image-physical fusion; The multi-frame image sequence is input into the trained mixing uniformity evaluation model to achieve real-time monitoring and intelligent control of the stirring process.

2. The method according to claim 1, characterized in that, The method for real-time monitoring and intelligent control of the mixing process by inputting the multi-frame image sequence into the trained mixing uniformity evaluation model includes: The target area of ​​the study is obtained by cropping a multi-frame image sequence of the flow process in the stirred tank; the target area image is then processed by flat field correction, background subtraction and impeller masking; the flow field velocity vector is calculated using the particle image velocimetry method to obtain the velocity components and velocity amplitude distribution, thereby realizing the spatiotemporal reconstruction of the flow field. Based on the velocity field, shear rate, vorticity, Q criterion, and energy consumption proxy are calculated. The physical features are fused with image grayscale features to form a multi-channel input tensor for identification. The tensor identifies and segments the "dead flow zone", "stagnant zone" and "low shear zone" in the stirring tank and outputs the corresponding segmentation mask. Calculate the dead area ratio (DR), low shear ratio (LSR), gray entropy (H), and average energy consumption based on the segmentation mask. Construct a unified mixed homogeneity score U This enables quantitative evaluation and visual analysis of the stirring process under different impeller types, rotation speeds, and fluid systems.

3. The method according to claim 2, characterized in that, Methods for flat-field correction, background subtraction, and blade masking of images of the target region include: Let the original image be , Represents the horizontal spatial coordinates in the image; Represents the spatial coordinates in the vertical direction of the image; The time dimension indicates the acquisition time of the corresponding image. First, the target region is cropped from the original image to obtain the image to be analyzed. Then, flat-field correction is performed, and the corrected image is... Represented as: ； in, This is a dark field image. This is a flat-field reference image acquired under particle-free, uniform illumination conditions. To prevent the use of tiny constants with a denominator of zero, after flat-field correction, background subtraction is employed to suppress reflections from the container walls and fix background interference. The image after background subtraction... Represented as: ； in, The background image is constructed from images of a flow-free state or averaged images from multiple frames; based on this, a binary mask is constructed for the blade region and its highly reflective regions. The image is then processed by blade masking to obtain a preprocessed image. : ； in, This indicates the blades and their reflective area. This indicates the effective flow field region.

4. The method according to claim 2, characterized in that, Methods for calculating the velocity vector of a flow field using particle image velocimetry, obtaining velocity components and velocity amplitude distribution, and realizing spatiotemporal reconstruction of the flow field include: Preprocessed images at adjacent time points and The calculation is divided into multiple PIV calculation windows, and the average displacement of the tracer particles within each window is obtained by cross-correlation matching. The velocity components are obtained: ； in, This indicates the time interval, which is the time difference between the acquisition of two adjacent image frames, in seconds. The velocity amplitude is: 。 5. The method according to claim 2, characterized in that, The method involves calculating shear rate, vorticity, Q criterion, and energy consumption proxy based on velocity field, fusing physical features with image grayscale features to form a multi-channel input tensor for identification, and identifying and segmenting "dead flow zones," "stagnant zones," and "low shear zones" within a stirred tank, outputting corresponding segmentation masks. ； ； in, For pixels Belongs to the The probability of a class region; Number of categories; Represents pixels Belongs to the Predicted scores for class regions.

6. The method according to claim 2, characterized in that, Calculate the dead area ratio (DR), low shear ratio (LSR), gray entropy (H), and average energy consumption based on the segmentation mask. Construct a unified mixed homogeneity score U The methods include: Among them, weight parameters , , , The settings are determined based on different mixing objectives, process stages, and energy consumption constraints. To normalize the grayscale entropy, This represents the normalized average energy consumption.

7. A physical field sensing-based system for evaluating the uniformity of mixing, the system being used to implement the method described in any one of claims 1-6, characterized in that, The system includes: The stirring unit (1) includes a stepper motor (1-1), a stirring paddle (1-2), and a motor driver (1-3); the stepper motor (1-1) is controlled by the motor driver (1-3) and the control module (6-2) to set the speed and direction of the stirring process, thereby achieving precise adjustment of the stirring process; the stirring paddle (1-2) is disposed in the stirring tank (1-4), which is preferably a transparent structure, for driving the fluid containing tracer particles to flow; The optical illumination unit (2) includes a laser source (2-1) and a light sheet imaging device (2-2). The laser is formed into a thin light sheet through a cylindrical lens to illuminate a designated plane in the stirring tank (1-4) so ​​that the movement trajectory of the tracer particles can be seen. The image acquisition unit (3) includes a high-speed camera with a sampling frame rate of 500 to 2000 frames per second, which is used to acquire images of particle motion in the fluid in real time from the illuminated area; the high-speed camera is synchronized with the control terminal of the stepper motor (1-1) through a hardware trigger signal to realize the time correspondence between image acquisition and stirring action; The data processing unit (4) includes a host computer and an image processing software module, which is used to receive the original image sequence transmitted by the high-speed camera, perform flat field correction, background subtraction and blade masking on the image, and use the particle image velocimetry algorithm to calculate the velocity vector field and extract the shear rate, vorticity, Q criterion and energy consumption proxy. The hybrid uniformity analysis unit (5) includes an image-physical fusion feature recognition module and a uniformity calculation module, which are used to fuse image grayscale features with physical field features, identify "flow dead zones", "stagnant zones" and "low shear zones" through a physical-guided fusion network, and calculate the dead zone area ratio, low shear zone ratio, grayscale entropy and average energy consumption to construct a unified hybrid uniformity score. U It also outputs a uniformity score; The control and visualization unit (6) includes a host computer display interface (6-1) and a control module (6-2). The host computer display interface (6-1) displays the original image, velocity vector field, segmentation results, and mixing uniformity score change curve in real time. U When the speed is below the set threshold, the control module (6-2) automatically increases the motor speed. U When the target is achieved but energy consumption is high, the control module (6-2) sends speed reduction, frequency reduction or current limiting commands to the motor driver (1-3) to reduce the input power of the motor, thereby achieving closed-loop optimization control of stirring energy efficiency.

8. The system according to claim 7, characterized in that, The stirring tank (1-4) is a transparent structure, which facilitates optical observation and particle image velocity measurement; the stirring paddle (1-2) is an arc-shaped paddle, a straight-bladed paddle, or a spiral paddle structure, and the paddle surface is treated with matte sandblasting and black matte coating.

9. The system according to claim 7, characterized in that, The host computer display interface (6-1) includes an original image display area, a velocity field vector overlay area, a region segmentation display area, and a mixing uniformity score curve display area. The user sets the target mixing uniformity score in the control module (6-2). U Energy consumption upper limit and control strategy parameters are set to achieve automatic control under different mixing conditions.

10. The system according to claim 7, characterized in that, The control module (6-2) is connected to the motor driver (1-3) via Ethernet or Modbus industrial communication protocol, and interacts with the host computer display interface (6-1) in real time; the control module (6-2) has a built-in energy efficiency-uniformity control algorithm, which is used when the mixing uniformity score is... U When the speed is below the set threshold, the control module (6-2) controls the motor speed to increase. U When the target is reached and the energy consumption exceeds the upper limit, the control module (6-2) sends speed reduction and frequency reduction commands to the motor driver (1-3) to reduce the input power of the motor and achieve a dynamic balance between energy consumption and hybrid effect.

Images