Micro-flow field measurement method of mechanical biological net port flowmeter

Abstract

The invention provides a micro-flow field measurement method of a mechanical biological net port flowmeter, and belongs to the technical field of micro-flow field measurement, and the micro-flow field measurement method comprises the following steps: firstly, constructing a net port cavity digital model, and realizing micro-flow field high-resolution characterization by using fluorescent particle tracing liquid and a laser confocal system; and recording a fluid movement track and processing original data by applying a track optimization function under the condition that the microfluidic driving system generates a constant pressure gradient. And solving fluid pressure distribution by adopting an orthogonal grid system and a Navier-Stokes equation, and processing speed field and pressure distribution data by utilizing a pre-trained micro-flow field analysis neural network model. Finally, the flow and pressure relation curve is obtained by calculating the flow flux of the cross section of the network port, a micro-flow field measurement correction coefficient table is generated, and the technical problems that in the prior art, the accurate measurement difficulty of the biological network port micro-flow field is large, and multi-scale flow characteristic representation is insufficient are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of microfluidic field measurement. Specifically, it relates to a microfluidic field measurement method for a mechanical bio-orifice flowmeter. Background Art

[0002] Bio-orifice flowmeters are important measurement tools for biomedical fluid mechanics research and medical device evaluation, and are widely used in biomedical engineering fields such as artificial heart valves and vascular stents. The measurement techniques of traditional bio-orifice flowmeters mainly rely on the determination of the relationship between macroscopic flow rate and pressure, and methods such as hot film type, electromagnetic type or ultrasonic Doppler are used to obtain overall flow rate parameters. These methods have achieved certain results in clinical applications and product evaluations, and can realize the quantitative measurement and recording of flow rate.

[0003] However, traditional measurement techniques are difficult to accurately characterize the local flow field characteristics in the orifice cavity at the microscale. Existing technologies usually can only obtain the average flow rate or pressure values at the inlet and outlet of the orifice, and lack effective means to characterize the micro-flow characteristics such as the complex three-dimensional flow structure, fluid shear stress distribution and laminar-turbulent transition region inside the orifice. Especially at the micron scale, traditional flowmeters cannot accurately capture the details of boundary layer flow and transient flow field changes, resulting in a significant difference between the measurement data and the actual biomechanical environment.

[0004] Due to the complex structure and small size of the bio-orifice, existing technologies are difficult to simultaneously meet the measurement requirements of high spatial resolution and high temporal resolution, and cannot accurately characterize the complex flow characteristics inside the orifice at the microscale, thus affecting the accuracy of bio-orifice design optimization and performance evaluation. This technical problem urgently needs to be solved by a new type of microfluidic field precise measurement method. That is to say, there are technical problems in the prior art such as the difficulty in precisely measuring the microfluidic field of the bio-orifice and the insufficient characterization of multi-scale flow characteristics. Summary of the Invention

[0005] In view of this, the present invention provides a microfluidic field measurement method for a mechanical bio-orifice flowmeter, which can solve the technical problems in the prior art such as the difficulty in precisely measuring the microfluidic field of the bio-orifice and the insufficient characterization of multi-scale flow characteristics.

[0006] The present invention is implemented as follows: The present invention provides a method for measuring the microfluidic field of a mechanical biological network orifice flowmeter, including: constructing a three-dimensional digital model of the network orifice cavity and determining the main measurement area of the microfluidic field; injecting a tracer fluid containing fluorescent particles into the network orifice cavity; adjusting the laser confocal system to the main measurement area of the microfluidic field; starting the microfluidic driving system to generate a constant pressure gradient, and synchronously recording the fluid movement trajectory in the network orifice cavity; analyzing the fluid movement trajectory data, processing the original data using a trajectory optimization function, and calculating the fluid velocity field and acceleration field; constructing an orthogonal grid system, mapping the fluid velocity field data to the orthogonal grid system, and applying the Navier-Stokes equation to solve the fluid pressure distribution; using a pre-trained microfluidic field analysis neural network model to process the fluid velocity field and fluid pressure distribution data, automatically identifying the laminar-turbulent transition region, and calculating the fluid shear stress; based on the fluid velocity field and fluid pressure distribution, calculating the flow flux of the network orifice cross-section, and obtaining the flow-pressure relationship curve through integration; comparing the flow-pressure relationship curve with the standard biological network orifice theoretical model to generate a microfluidic field measurement correction coefficient table.

[0007] Among them, the step of constructing the three-dimensional digital model of the network orifice cavity is specifically to perform a transverse scan of the biological network orifice flowmeter using a three-dimensional microscopy imaging system, and set the Gaussian beam scan step to 10 to 20 micrometers. The Gaussian beam scan step refers to the distance between two adjacent scans when the laser scanning system performs a transverse scan of the sample, and the Gaussian beam scan step determines the transverse resolution of the three-dimensional imaging.

[0008] Among them, the main measurement area of the microfluidic field refers to the spatial range in the network orifice cavity where the fluid flow is most significant and has the greatest impact on the measurement results, including three key parts: the inlet area, the contraction area, and the expansion area.

[0009] Among them, in the step of injecting the tracer fluid containing fluorescent particles into the network orifice cavity, the diameter of the fluorescent particles in the tracer fluid is 0.5 to 2 micrometers, the density of the fluorescent particles is 10,000 to 30,000 per milliliter, and the fluorescence wavelength range is 520 to 580 nanometers.

[0010] Among them, in the step of adjusting the laser confocal system to the main measurement area of the microfluidic field, the laser power is set to 15 to 25 milliwatts, the scanning frequency is 200 to 500 Hertz, and the thickness of the collected optical section is 5 to 10 micrometers. The optical section thickness refers to the thickness of the sample collected once by the confocal microscopy system in the depth direction, and the optical section thickness determines the longitudinal resolution of the three-dimensional imaging.

[0011] Among them, in the step of starting the microfluidic driving system to generate a constant pressure gradient, the pressure gradient range is 0.05 to 0.2 Pascal per micrometer, and the duration is 60 to 120 seconds.

[0012] Among them, in the step of processing the original data using the trajectory optimization function, the trajectory optimization function is used to perform noise reduction and smoothing processing on the originally collected fluid motion trajectory data, improve the accuracy and continuity of the fluid motion trajectory data. The inputs include the three-dimensional coordinate sequence of fluorescent particles at consecutive time points, the time interval between adjacent frames, the evaluation value of the flow field background noise level, the signal-to-noise ratio threshold for fluorescent particle recognition, and the velocity constraint range. The output is a high-quality particle trajectory data set after noise filtering and trajectory connection processing.

[0013] Among them, the microfluidic field analysis neural network model adopts a multi-level residual network structure. The network depth is determined by the ratio of the optical section thickness to the Gaussian beam scanning step size. The convolution kernel size is determined by the ratio of the fluorescent particle diameter to the grid spacing. The width of the attention mechanism is determined by the size of the main measurement area of the microfluidic field.

[0014] Among them, the specific structure of the microfluidic field analysis neural network model is a hybrid architecture that combines a multi-scale residual convolution network and a graph attention network. The network input layer receives three-dimensional fluid velocity field and fluid pressure distribution data. The encoder part uses eight three-dimensional convolutional layers to extract multi-scale flow field features. Residual connections are set every two layers to avoid the problem of gradient disappearance. The middle layer uses an adaptive graph attention mechanism to establish long-distance flow field correlations, and the weight distribution is automatically adjusted based on fluid physical constraint conditions. The decoder part consists of six deconvolutional layers and is used to reconstruct the fluid shear stress distribution. The final output layer is a fully connected layer that generates the fluid shear stress.

[0015] Among them, the steps for establishing the training data set during the pre-training process of the microfluidic field analysis neural network model include collecting fluid numerical simulation data under various typical bio-net orifice geometric structures, covering two states of laminar flow conditions and turbulent flow conditions. For each geometric structure, thirty groups of flow field data are generated at different Reynolds numbers, including complete information on the fluid velocity field, fluid pressure distribution, and fluid shear stress. All data are normalized to eliminate the influence of dimensions and randomly divided into a training set, a validation set, and a test set. The training set accounts for 70% of the total data volume, the validation set accounts for 20%, and the test set accounts for 10%.

[0016] Compared with the prior art, for the microfluidic field measurement method of a mechanical bio-net orifice flowmeter provided by the present invention, the proposed microfluidic field measurement method of the mechanical bio-net orifice flowmeter combines high-precision three-dimensional microscopy, laser confocal technology, and fluorescent particle tracing technology to establish a complete microfluidic field characterization system. By precisely controlling the Gaussian beam scanning step size and the optical section thickness, this method realizes the high-resolution three-dimensional reconstruction of the microfluidic field in the bio-net orifice cavity and captures micro-scale flow details that cannot be obtained by traditional techniques.

[0017] The pre-trained microfluidic field analysis neural network model adopted in the present invention effectively solves the limitations of traditional technologies in microfluidic data processing. Through a multi-level residual network structure and an adaptive graph attention mechanism, this model can accurately identify the laminar-turbulent transition region, precisely calculate the distribution of boundary fluid shear stress, and achieve a comprehensive characterization of complex microfluidic field characteristics. Especially when processing flow field data in a high-noise environment, the trajectory optimization function of the present invention significantly improves the measurement accuracy.

[0018] By constructing a microfluidic measurement correction coefficient table, the present invention effectively solves the technical problem of accurate measurement of the microfluidic field of the biological network port. It can not only accurately characterize the multi-scale flow characteristics inside the network port but also provide hydrodynamic parameters at the microscale, providing reliable technical support for the design optimization and performance evaluation of the biological network port and significantly improving the accuracy and reliability of microfluidic measurement in biomedical engineering. Brief Description of the Drawings

[0019] Figure 1 It is a flowchart of the method of the present invention.

[0020] Figure 2 It is a schematic diagram of the composition of the mechanical biological network port flowmeter in Embodiment 2.

[0021] Figure 3 It is a schematic diagram of the composition of the mechanical biological network port flowmeter in Embodiment 3. Detailed Description of the Embodiments

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0023] As Figure 1 shown, it is a flowchart of a microfluidic measurement method of a mechanical biological network port flowmeter provided by the present invention. This method includes the following steps: S01. Horizontally scan the biological network port flowmeter using a high-precision three-dimensional microscopic imaging system, set the Gaussian beam scanning step size to 10 to 20 micrometers, construct a three-dimensional digital model of the network port cavity, and determine the main microfluidic measurement region based on the three-dimensional digital model of the network port cavity; S02. Inject a tracer solution containing fluorescent particles into the network port cavity. The diameter of the fluorescent particles in the tracer solution is 0.5 to 2 micrometers, the density of the fluorescent particles is 10,000 to 30,000 per milliliter, and the fluorescence wavelength range is 520 to 580 nanometers; S03. Adjust the laser confocal system to the main microfluidic measurement region, set the laser power to 15 to 25 milliwatts, the scanning frequency to 200 to 500 hertz, and the optical slice thickness for acquisition to 5 to 10 micrometers;

[0024] S04. Activate the microfluidic driving system to generate a constant pressure gradient, where the pressure gradient ranges from 0.05 to 0.2 Pascal per micrometer, and the duration is 60 to 120 seconds. Synchronously record the fluid motion trajectory in the mesh oral cavity body; S05. Analyze the fluid motion trajectory data, process the original data using a trajectory optimization function. The input parameters include the particle coordinate sequence, time interval, flow field background noise level, particle recognition threshold, and velocity constraint range. Output the optimized particle trajectory dataset, calculate the fluid velocity field through the first derivative of the displacement function, and calculate the fluid acceleration field through the second derivative; S06. Construct an orthogonal grid system with a grid spacing of 2 to 5 micrometers, map the fluid velocity field data to the orthogonal grid system, and apply the Navier - Stokes equation to solve for the fluid pressure distribution; S07. Use a pre - trained microfluidic field analysis neural network model to process the fluid velocity field and the fluid pressure distribution data. The microfluidic field analysis neural network model adopts a multi - level residual network structure, where the network depth is determined by the ratio of the optical section thickness to the Gaussian beam scanning step size, the convolution kernel size is determined by the ratio of the fluorescent particle diameter to the grid spacing, the width of the attention mechanism is determined by the size of the main measurement area of the microfluidic field, automatically identify the laminar - turbulent transition region, and calculate the fluid shear stress along the boundary of the mesh oral cavity body; S08. Based on the fluid velocity field and the fluid pressure distribution, calculate the flow flux of the mesh cross - section, and obtain the flow - pressure relationship curve of the bio - mesh orifice flowmeter under different pressure conditions through integration; S09. Compare the flow - pressure relationship curve with the standard bio - mesh orifice theoretical model, calculate the measurement deviation, and generate a microfluidic field measurement correction coefficient table.

[0025] Among them, the Gaussian beam scanning step size specifically refers to the distance between two adjacent scans when the laser scanning system performs a transverse scan on the sample. The Gaussian beam scanning step size determines the transverse resolution of three - dimensional imaging.

[0026] Among them, the main measurement area of the microfluidic field specifically refers to the spatial range in the mesh oral cavity body where the fluid flow is most significant and has the greatest impact on the measurement results, usually including three key parts: the inlet area, the contraction area, and the expansion area.

[0027] Among them, the optical section thickness specifically refers to the thickness of the sample collected once by the confocal microscopy system in the depth direction. The optical section thickness determines the longitudinal resolution of three - dimensional imaging.

[0028] Among them, the Navier-Stokes equations are specifically partial differential equations that describe the motion of fluids and are used to calculate the relationship between the fluid pressure distribution and the fluid velocity field distribution of an incompressible fluid under given boundary conditions.

[0029] Among them, particle image velocimetry is specifically a non-contact measurement technique that calculates the fluid velocity field distribution by analyzing the position changes of the fluorescent particles in consecutive images and is applicable to micron-scale flow measurements.

[0030] Among them, fluid shear stress is specifically a mathematical expression that describes the stress distribution state in all directions inside the fluid, is calculated through the gradient of the fluid velocity field, and is used to evaluate the magnitude of the shear force of the fluid near the surface of the mesh orifice body.

[0031] Among them, the flow flux is specifically the volume of fluid passing through the cross-section of the mesh orifice per unit time, is calculated through the integral of the fluid velocity field over the cross-section of the mesh orifice, and is the core parameter for measuring the performance of the biological mesh orifice flowmeter.

[0032] Among them, the microfluidic field measurement correction coefficient table is specifically a data table that lists the ratio between the measured flow value and the theoretical flow value according to different working conditions and is used to correct the measurement results to improve the measurement accuracy.

[0033] Among them, the trajectory optimization function is used to perform noise reduction and smoothing processing on the originally collected fluid motion trajectory data, improve the accuracy and continuity of the fluid motion trajectory data. The input includes the three-dimensional coordinate sequence of the fluorescent particles at consecutive time points, the time interval between adjacent frames, the evaluation value of the flow field background noise level, the signal-to-noise ratio threshold for fluorescent particle recognition, and the velocity change range under fluid physical constraints. The output is a high-quality dataset of the particle trajectories after noise filtering and trajectory connection processing.

[0034] Among them, the specific structure of the microfluidic field analysis neural network model is a hybrid architecture that combines a multi-scale residual convolutional network and a graph attention network. The network input layer receives three-dimensional fluid velocity field and fluid pressure distribution data. The encoder part uses an eight-layer three-dimensional convolutional layer to extract multi-scale flow field features, and residual connections are set every two layers to avoid the problem of gradient disappearance. The middle layer uses an adaptive graph attention mechanism to establish long-distance flow field correlations, and the weight assignment is automatically adjusted based on fluid physical constraint conditions. The decoder part consists of six layers of transposed convolutional layers and is used to reconstruct the fluid shear stress distribution. The final output layer is a fully connected layer that generates the fluid shear stress.

[0035] Among them, the steps for establishing the training dataset in the pre-training process of the microfluidic field analysis neural network model specifically include collecting fluid numerical simulation data under various typical biological orifice geometries, covering two states of laminar flow conditions and turbulent flow conditions. For each geometry, thirty sets of flow field data are generated at different Reynolds numbers, including the complete information of the fluid velocity field, the fluid pressure distribution, and the fluid shear stress. All the data are normalized to eliminate the influence of dimensions, and then randomly divided into a training set, a validation set, and a test set. The training set accounts for 70% of the total data volume, the validation set accounts for 20%, and the test set accounts for 10%. And the data augmentation technology is used to expand the dataset scale to three times the original data volume by adding different degrees of random noise and geometric deformations to the original data.

[0036] Among them, the steps for pre-training the microfluidic field analysis neural network model specifically include first initializing the network parameters with a Gaussian distribution having a mean of zero and a standard deviation of 0.01, training the network using the mini-batch stochastic gradient descent algorithm with a batch size of 16, setting the initial learning rate to 0.001, and decaying the learning rate to 90% of the original every 50 rounds of training. The weighted loss function is used to simultaneously consider the physical conservation constraints of the flow field and the prediction accuracy of the fluid shear stress. The weight of the physical constraint part is 0.3, and the weight of the prediction accuracy part is 0.7. During the training process, monitor the loss function value of the validation set. When the loss function value of the validation set does not decrease for ten consecutive rounds of training, start the early stopping mechanism. Finally, select the model parameters with the best performance on the validation set as the pre-trained model, and fine-tune it on the actual flow field data of the biological orifice flowmeter to adapt to the actual application scenario.

[0037] Among them, the structure of the mechanical biological orifice flowmeter specifically includes five parts: a microfluidic chip body, an orifice cavity structure, an inlet and outlet fluid channel, a pressure sensor, and a fluid driving system. The microfluidic chip body is prepared from polydimethylsiloxane material, with a thickness of 2 to 5 mm and length and width dimensions of 40 to 60 mm. The orifice cavity structure is located in the center of the microfluidic chip body, with a diameter of 0.5 to 1 mm and a height of 100 to 300 microns, and the inner wall surface roughness is less than 0.1 micron. The inlet and outlet fluid channels are connected to the orifice cavity structure, with a diameter of 50 to 100 microns and a length of 5 to 10 mm. The pressure sensor is installed at both ends of the inlet and outlet fluid channels, with a measurement range of 0 to 100 Pascal and an accuracy of 0.01 Pascal. The fluid driving system includes a micro peristaltic pump and a precision syringe pump, with a flow control range of 0.1 to 10 microliters per minute and a pressure control range of 0 to 200 Pascal, constituting the complete biological orifice flowmeter system and providing a hardware basis for the microfluidic field measurement method.

[0038] The following describes the specific implementation manners of the above steps in detail.

[0039] The specific implementation of step S01 is to perform a lateral scan on the biological orifice flowmeter using a high-precision three-dimensional microscopic imaging system. First, set the scanning area, fix the biological orifice flowmeter on a precision displacement stage, and adjust the displacement stage so that the orifice cavity is located at the center of the microscopic system's field of view. Then, optimize the Gaussian beam parameters, set the beam diameter to 5 to 8 microns, the focal length to 200 to 300 microns, the numerical aperture to 0.6 to 0.8, and the wavelength to 488 to 532 nanometers. Next, perform a lateral scan, set the Gaussian beam scan step to 15 microns, which ensures a 30% to 40% overlapping area between adjacent scan positions to ensure imaging continuity. Subsequently, perform a depth scan, collect one layer of images every 5 microns from the top to the bottom of the cavity, and collect 40 to 60 optical slices in total. After the collection is completed, use a voxel reconstruction algorithm to convert the two-dimensional slice sequence into a three-dimensional digital model. This algorithm is based on the Markov random field theory and reconstructs the cavity structure through maximum a posteriori probability estimation, with a reconstruction accuracy better than 1 micron. Finally, identify the main measurement regions of the microfluidic field based on the three-dimensional digital model, including three key parts: the orifice inlet region, the contraction region, and the expansion region. These regions have a decisive impact on the flow measurement results. The purpose of this step is to construct a high-precision three-dimensional digital model of the orifice cavity to provide an accurate geometric basis for subsequent microfluidic field measurements.

[0040] The specific implementation of step S02 is to inject a tracer solution containing fluorescent particles into the orifice cavity. First, prepare the fluorescent particle tracer solution. Select polystyrene fluorescent particles with a diameter of 1 micron, a carboxylated surface to enhance hydrophilicity, a density of 1.05 g / cm³, which is close to the density of the aqueous solution, and a fluorescence wavelength of 550 nanometers. Then, mix the fluorescent particles with deionized water, adjust the particle density to 20,000 per milliliter, add 0.01% non-ionic surfactant to prevent particle aggregation, and add 0.02% anti-bleaching agent to extend the fluorescence lifetime. Next, perform ultrasonic treatment on the tracer solution at a frequency of 40 kHz for 15 minutes to fully disperse the fluorescent particles. Subsequently, use a micro-injection pump to inject the tracer solution into the orifice cavity at a flow rate of 0.5 μL / min to ensure no bubbles are generated. Monitor the fluorescence signal intensity in the cavity during the injection process and stop the injection when the signal intensity reaches more than 10 times the background signal. The purpose of this step is to introduce fluorescent particles that can trace the fluid motion to achieve visual measurement of the microfluidic field.

[0041] The specific implementation of step S03 is to adjust the laser confocal system to the main measurement area of the microfluidic field. First, calibrate the confocal system. Use standard fluorescent microspheres for three-dimensional spatial calibration, and control the spatial resolution error within 0.2 microns. Optimize the laser excitation wavelength to 488 nm for a fluorescence wavelength of 550 nm. Then set the scanning parameters. Set the laser power to 20 mW. This power value can generate a strong enough fluorescence signal without causing sample photobleaching or local heating. Next, adjust the scanning frequency to 350 Hz. This frequency can capture the continuous movement of particles at the maximum expected flow rate in the microfluidic field. Subsequently, optimize the optical slice thickness and set it to 7 microns. This value is less than the maximum expected displacement of the fluorescent particles in the flow direction, ensuring that the particles do not "jump" beyond the acquisition range between adjacent time frames. Finally, set the gain coefficient to 1.2 to 1.5, and increase the signal-to-noise ratio by at least 30 dB through a wavelength-selective filter to ensure that the fluorescent particles are clearly distinguishable. The purpose of this step is to optimize the parameters of the laser confocal system to ensure high-quality acquisition of the movement trajectories of fluorescent particles in the microfluidic field.

[0042] The specific implementation of step S04 is to start the microfluidic driving system to generate a constant pressure gradient. First, calibrate the microfluidic driving system. Use a precision pressure sensor to measure the system output pressure, and control the error within ±0.5%. Set the pressure feedback cycle frequency to 100 Hz to ensure that the pressure fluctuation is less than 1% of the set value. Then set the pressure gradient. Calculate the required pressure difference according to the length of the mesh orifice body, and the set value is 0.1 Pascal per micron. This value can generate a stable and measurable flow in the laminar flow region. Next, start the constant flow mode. Set the flow rate to 1 μL / min, and the system automatically adjusts the inlet and outlet pressures to maintain a constant flow rate. At the same time, the pressure difference between the inlet and outlet is monitored in real time through a pressure sensor. Subsequently, start synchronous data recording. Trigger the synchronous signal of pressure recording and image acquisition, with a time synchronization accuracy better than 1 ms, and continuously acquire data for 90 s. During the recording, pressure data is acquired every 5 s. The purpose of this step is to generate stable and controllable flow conditions in the microfluidic field and synchronously record the fluid movement trajectories to provide basic data for subsequent flow field analysis.

[0043] The specific implementation of step S05 is to analyze the fluid motion trajectory data. First, preprocess the original image sequence, use the Gaussian filtering algorithm to remove background noise, and then extract fluorescent particles through the adaptive threshold segmentation algorithm. The threshold is set to 1.5 times the local background mean. During the segmentation process, the cross-entropy loss function is used to optimize the boundary recognition accuracy. Then, perform particle center localization, apply the centroid method to calculate the centroid coordinates of the particles, with a sub-pixel accuracy better than 0.2 pixels. For overlapping particles, the ellipse fitting algorithm is used for separation, and the separation threshold is 0.7 times the particle diameter. Subsequently, use the trajectory optimization function to process the original data. The function is based on the Kalman filter principle, and the input parameters include the three-dimensional coordinate sequence of the particles, the time interval between adjacent frames of 3 milliseconds, the background noise level of 5%, the particle recognition signal-to-noise ratio threshold of 20 decibels, and the velocity constraint range of 0 to 100 micrometers per second. The fourth-order Runge-Kutta numerical integration method is used in the trajectory optimization process, and at the same time, the least squares fitting algorithm is combined for trajectory smoothing. The Hungarian algorithm is used for trajectory connection, and the connection threshold is set to 2 particle diameters. Finally, calculate the fluid velocity field and acceleration field. The velocity field is calculated through the first derivative of the displacement function, and the smoothing factor is set to 0.8. The acceleration field is calculated through the second derivative of the displacement function, using the central difference format, and the time step is 3 milliseconds. The purpose of this step is to extract fluid velocity and acceleration information from the fluorescent particle motion trajectory and provide accurate data for subsequent flow field analysis.

[0044] The specific implementation of step S06 is to construct an orthogonal grid system. First, determine the computational domain boundary. Based on the three-dimensional digital model constructed in step S01, expand the computational domain boundary to 3 times the diameter upstream of the flow inlet and 5 times the diameter downstream of the flow outlet to ensure capturing the complete flow development process. Then, generate an orthogonal grid. The grid spacing is set to 3 micrometers, which is less than 1 / 5 of the expected minimum flow feature scale to ensure computational accuracy. The elliptical grid generation algorithm is used for grid generation, and the grids at the boundary are encrypted, with the minimum grid size being 1 micrometer. Then, perform flow field data mapping, map the unstructured velocity field data calculated in step S05 to the orthogonal grid system through the radial basis function interpolation method, and the interpolation weight coefficient is determined based on the cubic exponential function of the distance from the particle to the grid point. Subsequently, solve the fluid pressure distribution, apply the discretized Navier-Stokes equations, use the finite volume method for the discretization method, and use the SIMPLE algorithm for the pressure-velocity coupling. The convergence criterion is set to the residual being less than 10⁻ 4 . The purpose of this step is to reconstruct the unstructured experimental measurement data into flow field data on a regular grid and calculate the pressure distribution based on the fluid mechanics equations.

[0045] The specific implementation of step S07 is to process the fluid velocity field and fluid pressure distribution data using a pre-trained microfluidic field analysis neural network model. First, the microfluidic field analysis neural network model is loaded. This model adopts a multi-level residual network structure, and the network depth is determined by the ratio of the optical slice thickness to the Gaussian beam scanning step size. The calculation result is 7 / 15 = 0.467. After rounding, the network depth is set to 8 layers. The size of the convolutional kernel is determined by the ratio of the fluorescent particle diameter to the grid spacing. The calculation result is 1 / 3 = 0.333. Taking the lower integer 1 and considering the receptive field requirements, the size of the convolutional kernel is finally set to 3×3×3. The width of the attention mechanism is determined by the size of the main measurement area of the microfluidic field and is set to 1 / 4 of the area diameter, approximately 125 microns. Then, data preprocessing is performed. The velocity field and pressure field data are normalized. The normalization coefficient of the velocity field is the reciprocal of the maximum velocity value, and the normalization coefficient of the pressure field is the reciprocal of the maximum pressure difference. Next, model inference is performed. The preprocessed data is input into the network, and the network automatically identifies the laminar-turbulent transition region. The judgment basis is the local Reynolds number threshold, and the critical value is set to 100. Finally, the fluid shear stress is calculated. The velocity gradient is extracted along the boundary of the mesh orifice body and multiplied by the dynamic viscosity coefficient of the fluid (1.002×10⁻³ Pascal·second for water at 20°C) to obtain the shear stress distribution. The purpose of this step is to use deep learning technology to extract deeper fluid dynamics characteristics from experimental flow field data, especially the boundary shear stress distribution information.

[0046] The specific implementation of step S08 is to calculate the flow flux of the mesh orifice cross-section based on the fluid velocity field and fluid pressure distribution. First, the position of the mesh orifice cross-section is identified, and the minimum cross-section of the mesh orifice is determined in the three-dimensional digital model. This cross-section is usually located at the end of the contraction zone. Then, the cross-section velocity distribution is calculated. The velocity values of each point on this cross-section are extracted from the fluid velocity field, and trilinear interpolation is used to ensure data continuity. Next, the flow flux is calculated. Area integration is performed on the cross-section velocity distribution, and the Gaussian integration method is used. The number of integration points is 4 times the number of cross-section grid points, and the integration error is controlled within 1%. Subsequently, the flow rate under different pressure conditions is measured. The output pressure of the microfluidic drive system is adjusted, and 5 to 7 measurement points are set. The pressure range covers 0.05 to 0.15 Pascal per micron, and each pressure point is measured 3 times to evaluate the repeatability error. Finally, the relationship curve between the flow rate and the pressure is plotted, and the least squares method is used for fitting. The form of the fitting function is a second-order polynomial, and the goodness of fit R² value is required to be greater than 0.98. The purpose of this step is to obtain the flow rate characteristic curve of the biological mesh orifice flowmeter under different working conditions and evaluate its metering performance.

[0047] The specific implementation of step S09 is to compare the flow rate-pressure relationship curve with the standard biological network port theoretical model. First, select the theoretical model. According to the characteristics of the network port cavity structure, the modified Hagen-Poiseuille formula or Richardson equation is selected as the theoretical model, which takes into account the influence of flow channel geometry, entrance effect, and exit effect on flow rate calculation. Then calculate the theoretical flow rate. Based on the measured pressure conditions, the theoretical model is applied to calculate the theoretical flow rate values at each pressure point. Next, calculate the measurement deviation. The measured flow rate is compared with the theoretical flow rate, and the relative error is calculated. The deviation calculation formula is (measured flow rate - theoretical flow rate) / theoretical flow rate × 100%. Subsequently, analyze the deviation law, study the variation trends of the deviation with pressure, Reynolds number, and other parameters, and identify the systematic error and random error components. Finally, generate a correction coefficient table. For the deviation under different pressure conditions, calculate the correction coefficient, which is defined as the ratio of the theoretical flow rate to the measured flow rate. The correction coefficient table is arranged from small to large according to the pressure gradient, with an interval of 0.01 Pascal per micron, covering the actual application range. The purpose of this step is to obtain the systematic error and generate the correction coefficient by comparing the experimental results with the theoretical prediction, so as to improve the measurement accuracy of the flowmeter.

[0048] The specific implementation of the microfluidic field analysis neural network model adopts a hybrid architecture that combines a multi-scale residual convolutional network and a graph attention network. This architecture can simultaneously process local flow field details and global flow field structures. First, the network input layer is designed to receive three-dimensional fluid velocity field and fluid pressure distribution data. The input tensor size is the number of grid points × the number of features (4 dimensions, namely the velocity components in three directions and pressure). Data preprocessing includes standardization and non-linear mapping. The standardization formula is x(standardized) = (x - mean) / standard deviation, and the non-linear mapping uses the hyperbolic tangent function to compress the data range to [-1, 1]. Then, the encoder part is constructed, and eight layers of three-dimensional convolutional layers are used to extract multi-scale flow field features. The number of convolutional kernels in the first layer is 32, doubling with each subsequent layer, and reaching 256 convolutional kernels in the deepest layer. Residual connections are set every two layers to avoid the problem of gradient disappearance. The activation function uses a rectified linear unit with a leakage parameter, and the leakage coefficient is 0.2. Next, the adaptive graph attention mechanism is implemented. The flow field data is constructed into a graph structure, with nodes being grid points and edges being spatial adjacency relationships. The attention weights are calculated based on the similarity of node features. The attention weight calculation uses the scaled dot product formula, and the scaling factor is the square root of the feature dimension. The weight assignment is automatically adjusted based on the fluid physical constraints, and the physical constraints include mass conservation and momentum conservation constraints. The constraint conditions are transformed into prior biases in the attention mechanism. Subsequently, the decoder part is designed, which consists of six layers of three-dimensional transposed convolutional layers and is used to reconstruct the fluid shear stress distribution. The number of channels in the transposed convolutional layers decreases by half layer by layer starting from 256, and the final output layer has 32 channels. After each transposed convolution, a batch normalization layer and a rectified linear unit activation function are connected. The momentum parameter of the batch normalization layer is set to 0.9. Finally, the output layer is constructed using a fully connected layer. The input is the output feature map of the decoder, and the output is a three-dimensional tensor of fluid shear stress. The output activation function uses a linear function to ensure that the output range is not restricted. The advantage of this implementation is that it can accurately capture multi-scale flow features in complex microfluidic fields and infer the fluid shear stress distribution that is difficult to directly measure.

[0049] The specific implementation of establishing the pre-training dataset for the microfluidic field analysis neural network model includes first collecting data on various typical bio-network orifice geometric structures. Five basic shapes and three variant shapes are designed, resulting in a total of eight different geometric structures. The geometric parameters include the inlet diameter, contraction section length, orifice diameter, and expansion angle, etc., and the parameter range covers the geometric variation range of actual applications. Then, fluid numerical simulations are carried out. Using commercial computational fluid dynamics software, the Reynolds number range is set from 0.1 to 50, covering the laminar flow and the early turbulent transition region. The mesh division uses hexahedral structured meshes, with the near-wall region meshes being refined, and the minimum mesh size being 0.5 microns. For each geometric structure, 30 sets of flow field data are generated under different Reynolds number conditions, totaling 240 sets of original data. Each set of data contains complete information on the fluid velocity field, fluid pressure distribution, and fluid shear stress. Next, data normalization is performed. The maximum-minimum normalization method is used to eliminate the dimensional influence between different physical quantities. The normalization benchmark for the velocity field is the characteristic velocity, the normalization benchmark for the pressure field is the characteristic pressure difference, and the normalization benchmark for the shear stress is the characteristic shear stress. Subsequently, the dataset is divided. The stratified random sampling method is used to ensure that the data under each geometric structure and Reynolds number condition are evenly distributed in each subset. The training set accounts for 70% (168 sets) of the total data volume, the validation set accounts for 20% (48 sets), and the test set accounts for 10% (24 sets). Finally, data augmentation is carried out. The dataset is expanded by adding different degrees of random noise and geometric deformations to the original data. The noise amplitude range is from 0 to 10% of the standard deviation of the original data. The geometric deformations include scaling, rotation, and non-rigid deformation, and the deformation amplitude is controlled within ±5% of the original geometric size. After expansion, the dataset size reaches three times the original data volume, totaling 720 sets of training data. The purpose of this implementation is to construct a training dataset with a wide coverage and representativeness, providing sufficient learning samples for the neural network model.

[0050] The following describes in detail the mathematical models or calculation processes involved in the present invention.

[0051] In step S05, the fluid velocity field is calculated through the first derivative of the displacement function, and the acceleration field is calculated through the second derivative. The specific expressions are as follows: ; ; Among them, represents the three-dimensional position vector of the fluorescent particle at time ; is the fluid velocity vector at position and time ; is the fluid acceleration vector. The position vector data is obtained by tracking the fluorescent particles through a confocal microscopy system at a frequency of 350 Hz, and the time interval between adjacent frames is 3 milliseconds.

[0052] For calculating the velocity field using the displacement function, the central difference method is adopted: ; where is the time interval between adjacent frames, which is set to 3 milliseconds; the smoothing factor is set to 0.8 to reduce the noise in velocity calculation.

[0053] For the calculation of the acceleration field, the second-order central difference formula is used: ; The mathematical expression of the trajectory optimization function for processing the original fluid motion trajectory data is: ; where is the optimized trajectory data; is the original trajectory data; is the time interval between adjacent frames (3 milliseconds); is the level of background noise in the flow field (5%); is the signal-to-noise ratio threshold for fluorescence particle recognition (20 decibels); and are the minimum and maximum values of the velocity constraint (0 and 100 microns per second), respectively. The function is based on the Kalman filtering principle and uses the fourth-order Runge-Kutta numerical integration method and the least squares fitting algorithm for trajectory smoothing.

[0054] In step S06, the Navier-Stokes equations are used to solve for the fluid pressure distribution. For incompressible fluid flow, these equations can be expressed as: ; ; where is the fluid density (for water at 20°C, kg / m³); is the velocity vector field; is the time; is the gradient operator; is the pressure field; is the dynamic viscosity (for water at 20°C, Pa·s); is the Laplace operator; represents the external force per unit volume, which is usually negligible in this microfluidic environment.

[0055] For the discrete form for numerical solution using the finite volume method, the momentum equation becomes:

[0056] ;

[0057] where the superscript represents the time step. The SIMPLE algorithm (Semi-Implicit Method for Pressure Linked Equations) is used for pressure-velocity coupling, and the convergence criterion is set to a residual less than .

[0058] In step S07, the shear stress at the boundary is calculated from the velocity gradient: ; where is the wall shear stress; is the dynamic viscosity; is the tangential velocity component; is the direction perpendicular to the wall. The velocity gradient is extracted from the velocity field data of points near the wall.

[0059] For the determination of the laminar-turbulent transition region, the local Reynolds number is calculated as: ; where is the local Reynolds number; is the fluid density; is the local velocity magnitude; is the local characteristic length; is the dynamic viscosity. The critical threshold is set to .

[0060] In step S08, the flow flux through the mesh opening cross-section is calculated by integrating the velocity field over the area: ; where is the volume flow rate; is the cross-sectional area of the mesh opening; is the velocity vector field; is the unit normal vector of the cross-section. Using the Gaussian integration method, this becomes: ; where is the number of integration points (set to 4 times the number of cross-section grid points); is the integration point 's weight factor; is the integration point 's velocity vector at; is related to the integration point 's area element. The integration error is controlled within 1%.

[0061] For the relationship between the flow rate and the pressure gradient, a second-order polynomial fitting is used: ; wherein is the flow rate; is the pressure gradient; 、 and are fitting coefficients determined using the least squares method. The goodness of fit (R² value) is required to be greater than 0.98.

[0062] In step S09, the measurement deviation is calculated as: ; wherein is the percentage relative error; is the measured flow rate; is the theoretical flow rate calculated according to the modified Hagen - Poiseuille formula or Richardson equation, depending on the characteristics of the mesh orifice structure.

[0063] The correction coefficient is defined as: ; wherein is the correction coefficient used to adjust the measured flow rate to improve accuracy.

[0064] The theoretical flow rate is calculated using the modified Hagen - Poiseuille formula: ; wherein is the radius at the narrowest part of the mesh orifice; is the pressure gradient; is the dynamic viscosity; is the entrance effect coefficient; is the exit effect coefficient. These coefficients take into account the pressure losses due to flow contraction at the entrance and expansion at the exit of the mesh orifice.

[0065] The theoretical flow rate is calculated using the Richardson equation, which is more applicable to meshes with complex geometries: ; wherein is the effective length of the mesh orifice; is a function of the Reynolds number, taking into account the inertial effect, and can be expressed as: ; wherein and are empirical constants determined by calibration with a standard flow rate; is the Reynolds number, defined as: ; Data standardization in neural network pre - processing uses the following standard formula: ; wherein is the original data value; is the average value of the data set; is the standard deviation of the data set.

[0066] The non - linear mapping uses the hyperbolic tangent function: ; This compresses the data range to [-1, 1], which is beneficial to neural network training.

[0067] For the calculation of the attention weights in the graph attention mechanism, the scaled dot - product formula is used: ; where represents the query matrix; represents the key matrix; represents the value matrix; is the dimension of the key, which is used as a scaling factor to prevent overly large values after the dot - product operation.

[0068] Specifically, the principle of the present invention is as follows: The technical principle of the present invention is based on the deep integration of multi - scale hydrodynamics measurement and intelligent data analysis. First, a biological mesh opening is scanned and modeled by a high - precision three - dimensional microscopic imaging system to determine the main measurement area of the micro - flow field, laying a foundation for subsequent accurate measurement. A Gaussian beam scanning step of 10 - 20 microns is adopted, ensuring that the lateral resolution is sufficient to capture microscopic flow details; while the optical section thickness setting of 5 - 10 microns ensures that the longitudinal resolution can restore the three - dimensional structure of the fluid.

[0069] During the micro - flow field measurement process, the present invention injects a tracer fluid containing fluorescent particles with a diameter of 0.5 - 2 microns into the mesh cavity, and combines with the high - frequency scanning (200 - 500 Hz) of the laser confocal system to achieve high spatio - temporal resolution tracking of fluid motion. This design follows the principle in hydrodynamics that the tracer particles must be small enough to follow the fluid motion and large enough to be captured by the optical system, overcoming the resolution limitation in the micro - scale flow characterization of traditional technologies.

[0070] The core innovation of the present invention lies in using deep - learning methods to process complex micro - flow field data. Through a trajectory optimization function, the original data is pre - processed, effectively reducing the influence of measurement noise; while the pre - trained micro - flow field analysis neural network model is constructed based on hydrodynamics principles, and its network depth, convolution kernel size, and attention mechanism width are all associated with physical parameters, ensuring that the model has physical interpretability. In particular, the multi - level residual network structure solves the problem of gradient disappearance in the training of deep networks, and the graph attention mechanism effectively captures the long - distance flow field correlation. These two technologies enable the model to accurately identify complex flow patterns and calculate the shear stress distribution.

[0071] Finally, by combining the measured fluid velocity field and pressure distribution data with the Navier-Stokes equations, the flow flux of the net opening section was calculated, a pressure-flow relationship curve was established, and a correction coefficient table was generated by comparing with the theoretical model, achieving an accurate conversion from microscopic flow characteristics to macroscopic flow parameters, providing a scientific basis for the accurate measurement and calibration of the biological net opening flowmeter.

[0072] A specific Embodiment 1 of the present invention is provided below. The specific implementation manners of each step in this Embodiment 1 are described in detail as follows.

[0073] The specific implementation manners of steps S01 - S04 are the same as those described above and will not be elaborated here.

[0074] The specific implementation manner of step S05 is to analyze the fluid motion trajectory data. First, the original image sequence is preprocessed. The Gaussian filtering algorithm is used to remove background noise, and then the fluorescent particles are extracted by the adaptive threshold segmentation algorithm. The threshold is set to 1.5 times the local background mean value. The cross-entropy loss function is used to optimize the boundary recognition accuracy during the segmentation process. Then, the particle center positioning is carried out. The centroid method is applied to calculate the centroid coordinates of the particles, and the sub-pixel accuracy is better than 0.2 pixels. For overlapping particles, the ellipse fitting algorithm is used for separation, and the separation threshold is 0.7 times the particle diameter. Subsequently, the original data is processed using the trajectory optimization function. The mathematical expression of this function is: ; where is the optimized trajectory data; is the original trajectory data; is the time interval between adjacent frames, set to 3 milliseconds; is the flow field background noise level, set to 5%; is the signal-to-noise ratio threshold for fluorescent particle recognition, set to 20 decibels; and are respectively the minimum and maximum values of the velocity constraint, set to 0 and 100 microns per second respectively. The function is based on the Kalman filtering principle, adopts the fourth-order Runge-Kutta numerical integration method, and at the same time combines the least squares fitting algorithm for trajectory smoothing. The Hungarian algorithm is used for trajectory connection, and the connection threshold is set to 2 particle diameters. When calculating the fluid velocity field, it is calculated through the first derivative of the displacement function: ; where is the fluid velocity vector; is the position vector of the fluorescent particle at time ; is the time step, set to 3 milliseconds. The smoothing factor is set to 0.8 to reduce the noise in velocity calculation. When calculating the acceleration field, it is calculated through the second derivative of the displacement function: ; where is the fluid acceleration vector. Calculated using the central difference scheme to ensure numerical stability. The purpose of this step is to extract fluid velocity and acceleration information from the motion trajectories of fluorescent particles, providing accurate data for subsequent flow field analysis.

[0075] The specific implementation of step S06 is to construct an orthogonal grid system. First, determine the computational domain boundary. Based on the three-dimensional digital model constructed in step S01, extend the computational domain boundary 3 times the diameter upstream of the flow inlet and 5 times the diameter downstream of the flow outlet to ensure capturing the complete flow development process. Then generate an orthogonal grid with a grid spacing of 3 microns, which is less than 1 / 5 of the expected minimum flow feature scale to ensure calculation accuracy. The grid generation uses the elliptical grid generation algorithm, and the grid is refined at the boundaries with a minimum grid size of 1 micron. Next, perform flow field data mapping, mapping the unstructured velocity field data calculated in step S05 to the orthogonal grid system through the radial basis function interpolation method, and determining the interpolation weight coefficients based on the cubic exponential function of the distance from the particle to the grid point. Subsequently, solve for the fluid pressure distribution, applying the discretized Navier - Stokes equations: ; ; where is the fluid density, which is 998.2 kg / m³ for water at 20°C; is the velocity vector field; is the time; is the gradient operator; is the pressure field; is the dynamic viscosity, which is Pascal - seconds for water at 20°C; is the Laplace operator; represents the external force per unit volume, which is usually negligible in the microfluidic environment. The discrete form is: ; where the superscript represents the time step. The discrete method uses the finite volume method, and the pressure - velocity coupling uses the SIMPLE algorithm, with the convergence criterion set to a residual less than . The purpose of this step is to reconstruct the unstructured experimental measurement data into flow field data on a regular grid and calculate the pressure distribution based on the fluid mechanics equations.

[0076] The specific implementation of step S07 is to process the fluid velocity field and fluid pressure distribution data using a pre-trained microfluidic field analysis neural network model. First, the microfluidic field analysis neural network model is loaded. This model adopts a multi-level residual network structure, and the network depth is determined by the ratio of the optical slice thickness to the Gaussian beam scanning step size. The calculation result is 7 / 15 = 0.467, and after rounding, the network depth is set to 8 layers. The size of the convolutional kernel is determined by the ratio of the diameter of the fluorescent particle to the grid spacing. The calculation result is 1 / 3 = 0.333, and the lower integer 1 is taken. Considering the requirement of the receptive field, the size of the convolutional kernel is finally set to 3×3×3. The width of the attention mechanism is determined by the size of the main measurement area of the microfluidic field and is set to 1 / 4 of the area diameter, about 125 microns. Then, data preprocessing is carried out, and data standardization is performed: ; where is the original data value; is the average value of the data set; is the standard deviation of the data set. The non-linear mapping uses the hyperbolic tangent function: ; Then, model inference is executed. The network automatically identifies the laminar-turbulent transition region, and the judgment basis is the local Reynolds number: ; where is the local Reynolds number; is the fluid density; is the local velocity magnitude; is the local characteristic length; is the dynamic viscosity. The critical value is set to 100. Calculate the fluid shear stress: ; where is the wall shear stress; is the dynamic viscosity; is the tangential velocity component; is the direction perpendicular to the wall. The purpose of this step is to use deep learning technology to extract deeper hydrodynamic characteristics from the experimental flow field data, especially the boundary shear stress distribution information.

[0077] The specific implementation of step S08 is to calculate the flow flux of the net outlet cross-section based on the fluid velocity field and fluid pressure distribution. First, identify the position of the net outlet cross-section, and determine the minimum cross-section of the net outlet in the three-dimensional digital model. This cross-section is usually located at the end of the contraction zone. Then, calculate the cross-section velocity distribution, extract the velocity values of each point on this cross-section from the fluid velocity field, and use trilinear interpolation to ensure data continuity. Next, calculate the flow flux by performing area integration on the cross-section velocity distribution: ; where is the volume flow rate; is the cross-sectional area of the mesh opening; is the velocity vector field; is the unit normal vector of the cross-section. The Gaussian integral method is used for numerical calculation: ; where is the number of integration points, set to 4 times the number of mesh points of the cross-section; is the integration point 's weight factor; is the integration point 's velocity vector at; is related to the integration point 's area element. The integration error is controlled within 1%. Subsequently, the flow rate under different pressure conditions is measured, the output pressure of the microfluidic driving system is adjusted, 5 to 7 measurement points are set, the pressure range covers 0.05 to 0.15 pascals per micrometer, and each pressure point is measured 3 times to evaluate the repeatability error. Finally, the relationship curve between the flow rate and the pressure is plotted and fitted with a second-order polynomial: ; where is the flow rate; is the pressure gradient; , and are the fitting coefficients determined using the least squares method. The goodness-of-fit R² value is required to be greater than 0.98. The purpose of this step is to obtain the flow rate characteristic curve of the biological mesh opening flowmeter under different working conditions and evaluate its metering performance.

[0078] The specific implementation of step S09 is to compare the relationship curve between the flow rate and the pressure with the standard biological mesh opening theoretical model. First, the theoretical model is selected. According to the structural characteristics of the mesh opening cavity, the modified Hagen-Poiseuille formula or the Richardson equation is selected as the theoretical model. The modified Hagen-Poiseuille formula is: ; where is the radius of the narrowest part of the mesh opening; is the pressure gradient; is the dynamic viscosity; is the entrance effect coefficient; is the exit effect coefficient. The Richardson equation is: ; where is the effective length of the mesh opening; is a function of the Reynolds number, expressed as: ; ; where and are empirical constants; is the Reynolds number; is the fluid density; is the flow rate; is the dynamic viscosity. Then calculate the theoretical flow rate. Based on the measured pressure conditions, apply the theoretical model to calculate the theoretical flow rate values at each pressure point. Then calculate the measurement deviation: ; where is the relative error; is the measured flow rate; is the theoretical flow rate. Subsequently, analyze the deviation law, study the variation trend of the deviation with pressure, Reynolds number and other parameters, and identify the systematic error and random error components. Finally, generate a correction coefficient table. The correction coefficient is defined as: ; The correction coefficient table is arranged in ascending order of pressure gradient, with an interval of 0.01 Pascal per micrometer, covering the actual application range. The purpose of this step is to obtain the systematic error and generate the correction coefficient by comparing the experimental results with the theoretical prediction, so as to improve the measurement accuracy of the flowmeter.

[0079] The process of establishing the pre-training dataset for the microfluidic field analysis neural network model first collects data on various typical bio-net orifice geometric structures. Design 5 basic shapes and 3 variant shapes, a total of 8 different geometric structures. The geometric parameters include the inlet diameter, contraction section length, orifice diameter, and expansion angle, etc. The parameter range covers the geometric variation range of actual applications. Then perform fluid numerical simulations. Using commercial computational fluid dynamics software, set the Reynolds number range from 0.1 to 50, covering the laminar flow and early turbulent transition region. The mesh is divided into hexahedral structured meshes, and the near-wall region is mesh-refined, with the minimum mesh size of 0.5 micrometers. For each geometric structure, 30 sets of flow field data are generated under different Reynolds number conditions, totaling 240 sets of original data. Then perform data normalization, using the maximum-minimum normalization method to eliminate the dimensional influence between different physical quantities. Subsequently, divide the dataset. Using the stratified random sampling method to ensure that the data under each geometric structure and Reynolds number condition is evenly distributed in each subset. The training set accounts for 70% of the total data volume, the validation set accounts for 20%, and the test set accounts for 10%. Finally, perform data augmentation. Expand the dataset by adding different degrees of random noise and geometric deformations to the original data. The noise amplitude range is 0 to 10% of the standard deviation of the original data. The geometric deformations include scaling, rotation, and non-rigid deformation, and the deformation amplitude is controlled within ±5% of the original geometric size. After expansion, the dataset size reaches 3 times the original data volume.

[0080] Through the specific implementation of the above steps, a complete microfluidic field measurement method for a mechanical biological network port flowmeter is formed. This method starts from high-precision three-dimensional microscopic imaging, and through steps such as injecting fluorescent particle tracer liquid, setting up a laser confocal system, microfluidic driving, analyzing the fluid motion trajectory, constructing an orthogonal grid, processing with a neural network model, calculating the flow flux, and finally generating a correction coefficient table, a systematic microfluidic field measurement system is constructed. This method combines multi-disciplinary technologies such as experimental fluid mechanics, optical measurement, numerical simulation, and deep learning, achieving high-precision measurement of micron-scale flow, and providing a powerful tool for the optimized design and performance evaluation of microfluidic devices.

[0081] In this Embodiment 1, the composition structure of the mechanical biological network port flowmeter mainly includes five core parts, and the following is a detailed description of each part: The main body of the microfluidic chip is the basic carrier of the entire biological network port flowmeter, which is prepared from polydimethylsiloxane material. This material has good biocompatibility, transparency, and processability. The thickness of the chip body is 2 to 5 millimeters, providing sufficient mechanical strength to withstand the pressure during the measurement process. The length and width dimensions of the chip are 40 to 60 millimeters, facilitating operation and connection with the measurement system. The surface of the microfluidic chip body has been precisely processed to ensure compatibility with the optical measurement system.

[0082] The network orifice cavity structure is located at the central position of the microfluidic chip body and is the core measurement area of the entire system. The diameter of the network orifice cavity is 0.5 to 1 millimeter, and the height is 100 to 300 micrometers, forming a microfluidic chamber. The surface roughness of the inner wall of the cavity is controlled below 0.1 micrometer to ensure that the fluid does not generate additional turbulence and interference due to surface roughness when passing through. The precise geometric shape of the network orifice cavity directly affects the measurement characteristics and calibration parameters of the flowmeter.

[0083] The inlet and outlet fluid channels are connected to the network orifice cavity structure and are responsible for guiding the fluid into and out of the network orifice cavity. The diameter of these channels is 50 to 100 micrometers, and the length is 5 to 10 millimeters. The size design ensures stable fluid input and output conditions. The inner wall of the channel also maintains low roughness to reduce the influence of the boundary layer effect on fluid flow. The geometric design of the channel takes into account fluid mechanics characteristics to reduce the influence of the inlet effect and outlet effect on the measurement accuracy.

[0084] Pressure sensors are installed at both ends of the inlet and outlet fluid channels to monitor the pressure difference before and after the network orifice cavity in real time. The measurement range of these sensors is 0 to 100 Pascals, and the measurement accuracy reaches 0.01 Pascal, which can accurately capture minute pressure changes. The pressure sensors are manufactured using microelectromechanical system (MEMS) technology, with small volume, fast response, and high sensitivity, capable of providing pressure data with high time resolution and providing key parameters for flow calculation.

[0085] The fluid drive system consists of a micro peristaltic pump and a precision syringe pump, which is responsible for controlling the fluid flow rate and pressure flowing through the net orifice body. The system flow control range is from 0.1 to 10 microliters per minute, and the pressure control range is from 0 to 200 Pascals, which can meet the requirements of different measurement conditions. The fluid drive system adopts a closed-loop control mechanism, and adjusts the working parameters of the pump in real time according to the feedback information of the pressure sensor to ensure that the fluid maintains a stable flow state during the measurement process. The drive system is also equipped with an anti-pulsation design to minimize the fluctuations of fluid flow and improve the measurement accuracy.

[0086] To better understand and implement the present invention, an embodiment 2 of a specific application scenario of the present invention is provided below: Researchers conducted a study on the optimization of a marine plankton collection system in a certain sea area. In this study, a mechanical bio-net orifice flowmeter microfluidic field measurement method including five parts: a microfluidic chip body, a net orifice body structure, inlet and outlet fluid channels, a pressure sensor, and a fluid drive system as shown in Figure 2 was used to accurately characterize the hydrodynamic characteristics at the collection net orifice, improve the collection efficiency, and reduce the damage to the target organisms. The bio-net orifice flowmeter adopts a conical design, with a diameter gradually shrinking from 12 mm at the inlet to 3 mm at the narrowest point, and then expanding to 8 mm at the outlet. The overall length is 36 mm, and the material is medical-grade transparent polymethyl methacrylate.

[0087] First, the researchers used a high-precision three-dimensional microscopic imaging system to perform a lateral scan of the bio-net orifice flowmeter. The system uses a Zeiss LSM 980 laser scanning microscope in Germany, with a beam diameter set to 6.5 microns, a focal length of 250 microns, a numerical aperture of 0.75, and a laser light source with a wavelength of 532 nm selected. The Gaussian beam scanning step is set to 15 microns to ensure a 35% overlapping area between adjacent scanning positions. In the depth direction, images are collected at intervals of 5 microns from the top to the bottom of the cavity, and a total of 53 optical slices are collected. Through the voxel reconstruction algorithm of the Markov random field theory, the two-dimensional slice sequence is converted into a three-dimensional digital model, and the reconstruction accuracy reaches 0.8 microns. Based on the three-dimensional digital model, the main measurement areas of the microfluidic field are identified, including three key parts: the net orifice inlet area (diameter 12 mm), the contraction area (length 18 mm), and the expansion area (length 12 mm).

[0088] Secondly, prepare the fluorescent particle tracer solution. Select polystyrene fluorescent particles with a diameter of 1.2 microns, a carboxylated surface to enhance hydrophilicity, a density of 1.05 g / cm³, which is close to the density of seawater, and a fluorescence wavelength of 550 nm. Mix the fluorescent particles with filtered artificial seawater (salinity 35‰), adjust the particle density to 22,500 per milliliter, add 0.01% Tween-20 non-ionic surfactant to prevent particle aggregation, and add 0.02% ascorbic acid as an anti-bleaching agent to extend the fluorescence lifetime. The tracer solution is ultrasonically treated at a frequency of 40 kHz for 18 minutes to fully disperse the fluorescent particles. Use a Harvard PHD Ultra micro-injection pump to inject the tracer solution into the oral cavity of the net, with the flow rate controlled at 0.45 μL / min to ensure no bubbles are generated. Stop injecting when the fluorescence signal intensity reaches 12 times the background signal.

[0089] Subsequently, adjust the Nikon A1R+ laser confocal system to the main measurement area of the microfluidic field. Use standard fluorescent microspheres with a diameter of 1 micron for three-dimensional spatial calibration, and control the spatial resolution error within 0.18 microns. For the 550 nm fluorescence wavelength, optimize the laser excitation wavelength to 488 nm. Set the laser power to 22 mW, the scanning frequency to 350 Hz, and the thickness of the acquired optical section to 7 microns. Set the gain coefficient to 1.35, and improve the signal-to-noise ratio to 32 dB through a 550 ± 15 nm band-pass filter.

[0090] Next, start the MK5+ Elveflow microfluidic drive system to generate a constant pressure gradient. Use a precision pressure sensor to measure the system output pressure, control the error within ±0.4%, set the pressure feedback cycle frequency to 120 Hz, and ensure that the pressure fluctuation is less than 0.9% of the set value. Calculate the required pressure difference according to the length of the oral cavity of the net, and set the pressure gradient to 0.1 Pa per micron. Start the constant flow mode, set the flow rate to 0.95 μL / min, and the system automatically adjusts the inlet and outlet pressures to maintain a constant flow rate. Trigger the synchronous signal of pressure recording and image acquisition, with a time synchronization accuracy of 0.8 ms, and continuously acquire for 95 s, during which pressure data is acquired every 4 s.

[0091] By analyzing the fluid motion trajectory data, the Gaussian filtering algorithm (kernel size 5×5 pixels, standard deviation 1.2) is used for the original image sequence to remove background noise, and then the adaptive threshold segmentation algorithm is applied to extract fluorescent particles, with the threshold set to 1.55 times the local background mean. The centroid coordinates of the particles are calculated using the centroid method, and the sub-pixel accuracy reaches 0.18 pixels. For overlapping particles, the ellipse fitting algorithm is used for separation, and the separation threshold is 0.72 times the particle diameter. The original data is processed using a trajectory optimization function based on the Kalman filtering principle, and the input parameters include the particle three-dimensional coordinate sequence, the time interval between adjacent frames (2.86 milliseconds), the background noise level (4.8%), the particle recognition signal-to-noise ratio threshold (22 dB), and the velocity constraint range (0 to 96 μm / s). The fourth-order Runge-Kutta numerical integration method is used, combined with the least squares fitting algorithm for trajectory smoothing, and the smoothing factor is set to 0.82. The velocity field is calculated through the first derivative of the displacement function, and the acceleration field is calculated through the second derivative.

[0092] In the grid construction stage, the boundaries of the computational domain are determined and extended to 3.5 times the diameter upstream of the flow inlet and 5.5 times the diameter downstream of the flow outlet. An orthogonal grid is generated, with the grid spacing set to 3.2 μm. The elliptical grid generation algorithm is used, and the grid is refined to 1.2 μm at the boundaries. The unstructured velocity field data is mapped to the orthogonal grid system through the radial basis function interpolation method, and the interpolation weight coefficient is determined based on the cubic exponential function of the distance from the particle to the grid point. The discretized Navier-Stokes equations are applied, with the finite volume method used for the discretization method and the SIMPLE algorithm used for the pressure-velocity coupling. The convergence criterion is set to a residual less than 8×10⁻ 5 . The flow field characteristics in the mesh cavity at different flow rates are shown in Table 1.

[0093] Table 1 Flow field characteristic parameters in the mesh cavity under different flow rate conditions

[0094] The microfluidic field analysis neural network model adopts a multi-level residual network structure. The network depth is 8 layers (the ratio of the optical slice thickness to the Gaussian beam scanning step size is 7 / 15≈0.47, rounded up to 8), the convolution kernel size is 3×3×3 (the ratio of the fluorescent particle diameter to the grid spacing is 1.2 / 3.2≈0.38, taking the lower integer 1 and setting it to 3 considering the receptive field requirements), and the width of the attention mechanism is 122 μm (1 / 4 of the minimum diameter of the main measurement area of the microfluidic field). In the data preprocessing stage, the velocity field and pressure field data are normalized. The normalization coefficient of the velocity field is the reciprocal of the maximum velocity value (96 μm / s), and the normalization coefficient of the pressure field is the reciprocal of the maximum pressure difference (12 Pascal). The model automatically identifies the laminar-turbulent transition region, and the judgment basis is the local Reynolds number threshold, with the critical value set to 105. The fluid shear stress distribution in different regions of the mesh cavity is shown in Table 2.

[0095] Table 2 Fluid shear stress distribution in different regions of the mesh orifice body (Pa)

[0096] The flow flux of the mesh orifice cross-section was calculated under different pressure conditions. First, the minimum cross-section of the mesh orifice was identified, which was located at the end of the contraction zone with a diameter of 3 mm. The velocity values of each point on this cross-section were extracted from the fluid velocity field, and trilinear interpolation was used to ensure data continuity. The area integral of the cross-section velocity distribution was performed using Gaussian integration, with the number of integration points being 4.2 times the number of cross-section grid points, and the integration error was controlled within 0.85%. The output pressure of the microfluidic driving system was adjusted, and 6 measurement points were set, with the pressure range covering 0.05 to 0.15 Pascal per micrometer. Each pressure point was measured 3 times to evaluate the repeatability error. The flow rate-pressure relationship curve was fitted using the least squares method, and the form of the fitting function was a second-order polynomial, with the goodness of fit R² value reaching 0.992. The flow rate measurement results under different pressures are shown in Table 3.

[0097] Table 3 Flow rate measurement results under different pressure conditions

[0098] Finally, the researchers compared the flow rate-pressure relationship curve with the theoretical model of the modified Hagen-Poiseuille formula. The theoretical model considered the influence of the geometric shapes of the contraction zone and the expansion zone, as well as the inlet effect and the outlet effect. The measurement deviation was calculated. The measured flow rate was generally less than the theoretical flow rate, and the deviation decreased with the increase of the pressure gradient, from -5.12% to -2.22%. A correction coefficient table was generated. The correction coefficient was defined as the ratio of the theoretical flow rate to the measured flow rate, and was calculated based on the deviation under different pressure conditions. The range of the correction coefficient was from 1.054 to 1.023, and it decreased with the increase of the pressure gradient. The finally completed microfluidic field measurement correction coefficient table was arranged according to the pressure gradient from 0.05 to 0.15 Pascal per micrometer, with an interval of 0.01 Pascal per micrometer, covering the actual ocean collection application range.

[0099] Traditional measurement of the flow rate at the opening of marine plankton collection nets mainly relies on mechanical flowmeters or acoustic Doppler current profilers. These methods have technical bottlenecks such as low measurement resolution, inability to obtain the microscopic flow field distribution, and difficulty in evaluating the shear damage to biological samples. The minimum measurable flux of a mechanical flowmeter is about 0.5 liters per minute, which is much larger than the requirements for the opening of a micro - biological net. Acoustic Doppler instruments cannot accurately characterize the complex flow field inside the net opening at a millimeter - level spatial resolution. The micro - flow field measurement method adopted in this invention overcomes these technical problems: Firstly, through high - precision three - dimensional microscopic imaging combined with fluorescence tracing technology, flow field measurement with a micron - level spatial resolution is achieved, which is about two orders of magnitude higher than traditional methods. Secondly, the micro - flow field analysis neural network model based on deep learning can automatically identify the laminar - to - turbulent transition region and accurately calculate the boundary shear stress, providing a quantitative basis for evaluating the damage to biological samples. Finally, the generated micro - flow field measurement correction coefficient table organically combines theoretical and measured results, improving the flow rate measurement accuracy. The relative error is reduced from ±10% of traditional methods to ±2.5%. These technological innovations make it possible to accurately measure the micro - flow rate conditions of the marine plankton collection system, providing a more reliable technical means for marine ecosystem monitoring.

[0100] The following provides a specific Example 3 of the present invention: Marine researchers designed a mechanical biological net - mouth flowmeter for deep - sea micro - plankton collection, which includes five parts: a microfluidic chip body, a net - mouth cavity structure, inlet and outlet fluid channels, a pressure sensor, and a fluid drive system as shown in Figure 3 order to ensure its metering accuracy under high - pressure environments, micro - flow field measurement and calibration were carried out on it. This biological net - mouth flowmeter adopts a bionic design, simulating the water - inlet system of sponges, with the characteristics of high efficiency and low resistance, and is suitable for quantitative collection of micro - plankton in deep - sea environments. When conducting micro - flow field measurement, first, a high - precision three - dimensional microscopic imaging system was used to perform a transverse scan on the biological net - mouth flowmeter. The researchers used a confocal microscopic system equipped with two - photon excitation function, set the Gaussian beam diameter to 6.5 microns, the focal length to 250 microns, the numerical aperture to 0.75, and the wavelength to 520 nanometers. The scanning step was set to 12 microns to ensure a 35% overlapping area between adjacent positions. The depth - direction scan was from the top to the bottom of the cavity, and 52 optical slices were collected at intervals of 4.8 microns. The voxel reconstruction algorithm was applied to construct a three - dimensional digital model of the net - mouth cavity, and the reconstruction accuracy reached 0.8 microns. Based on the three - dimensional model, the main micro - flow field measurement regions were identified, including three key parts: a spiral inlet region (diameter 75 microns), a variable - cross - section contraction region (length 125 microns), and an expansion region (cone angle 15°).

[0101] For flow field visualization, the researchers selected seawater environment-compatible fluorescent particles to prepare the tracer solution. They chose salt-tolerant polystyrene fluorescent particles with a diameter of 1.2 microns. The surface was treated with carboxylation and hydroxylation, which improved the stability in a high-salt environment. The density was 1.08 g / cm³, which was close to the density of standard seawater. The fluorescence wavelength was selected as 545 nm, which had the least attenuation in seawater. The fluorescent particles were mixed with filtered seawater (filtered through a 0.2-micron pore size filter membrane), and the particle density was adjusted to 25,000 per milliliter. 0.015% non-ionic surfactant and 0.025% anti-bleaching agent were added. Ultrasonic treatment (frequency 45 kHz, time 18 minutes) was used to fully disperse the fluorescent particles. The tracer solution was injected into the net oral cavity through a micro-injection pump, and the flow rate was controlled at 0.4 μL / min to ensure no bubbles were generated. The injection was stopped when the fluorescence signal intensity in the cavity reached 12 times the background signal.

[0102] Subsequently, the parameters of the laser confocal system were adjusted to adapt to microfluidic field measurement. A standard microsphere for deep-sea environment calibration was used for three-dimensional spatial calibration, and the spatial resolution error was controlled within 0.18 microns. For the 545-nm fluorescence wavelength, the laser excitation wavelength was optimized to 488 nm. The laser power was set at 22 mW, the scanning frequency was 420 Hz, and the thickness of the acquired optical section was 6.5 microns. The gain coefficient was set at 1.35, and the signal-to-noise ratio was increased to 35 dB through a narrow-band pass filter (passband width 20 nm).

[0103] The high-precision microfluidic driving system independently developed in the laboratory was started to simulate the pressure conditions of the deep-sea environment. The system pressure range could reach 100 MPa, and the accuracy was better than 0.01%. First, a high-precision pressure sensor (range 100 MPa, accuracy 0.05%) was used to calibrate the system, and the pressure feedback cycle frequency was set at 120 Hz. According to the length of the net oral cavity (850 microns), the required pressure gradient was calculated as 0.15 Pascal per micron. The constant flow mode was started, the flow rate was set at 1.2 μL / min, and it lasted for 98 seconds. Pressure data was collected every 4 seconds, and finally 25 data points were obtained.

[0104] Analyze and process the collected fluid motion trajectory data. First, apply the Gaussian filtering algorithm (kernel size 5×5 pixels, standard deviation 1.2) to remove background noise, and extract fluorescent particles through the adaptive threshold segmentation algorithm (threshold coefficient 1.65). Calculate the centroid coordinates of the particles using the centroid method, with the sub-pixel accuracy reaching 0.15 pixels. For overlapping particles, apply the ellipse fitting algorithm for separation, and the separation threshold is 0.65 times the particle diameter. Apply the trajectory optimization function based on the Kalman filtering principle to process the data. The input parameters include: the three-dimensional coordinate sequence of the particles (a total of 25,000 tracking points), the time interval between adjacent frames of 2.5 milliseconds, the background noise level of 4.5%, the particle recognition signal-to-noise ratio threshold of 22 dB, and the velocity constraint range of 0 to 125 μm / s. The fourth-order Runge-Kutta numerical integration method is used in the trajectory optimization process, combined with the least squares fitting algorithm for trajectory smoothing (smoothing factor 0.85), and the Hungarian algorithm is used for trajectory connection (connection threshold is 2.2 particle diameters). Calculate the velocity field through the first derivative of the displacement function and the acceleration field through the second derivative.

[0105] Construct an orthogonal grid system for flow field analysis. Extend the computational domain boundary to 3.5 times the diameter upstream of the flow inlet and 6 times the diameter downstream of the flow outlet. Generate orthogonal grids with a grid spacing of 2.8 μm, and the grids at the boundary are encrypted to 1.2 μm. Use the elliptical grid generation algorithm to create a total of approximately 1.2 million grid points. Map the unstructured velocity field data to the orthogonal grid through the radial basis function interpolation method (the kernel function is selected as the polynomial type with an order of 3). Apply the discretized Navier-Stokes equations, discretize them using the finite volume method, couple the pressure and velocity with the SIMPLE algorithm, and set the convergence criterion to a residual less than 8×10⁻ 5 。

[0106] Use a pre-trained microfluidic flow field analysis neural network model to process the flow field data. The network depth is calculated as 6.5 / 12 = 0.542, and after rounding, it is set to 9 layers. The convolution kernel size is calculated as 1.2 / 2.8 = 0.429, and considering the flow field characteristic scale, it is set to 3×3×3. The width of the attention mechanism is set to 1 / 4 of the diameter of the main measurement area of 75 μm, approximately 18.75 μm. The network structure parameters are shown in Table 4:

[0107] Table 4 Microfluidic Flow Field Analysis Neural Network Structure Parameter Table

[0108] Perform data preprocessing, normalize the velocity field and pressure field data. The normalization coefficient of the velocity field is the reciprocal of the maximum velocity value of 160 μm / s, and the normalization coefficient of the pressure field is the reciprocal of the maximum pressure difference of 127.5 Pa. The network automatically identifies the laminar-turbulent transition region, and the identification results are shown in Table 5:

[0109] Table 5 Flow Field Region Characteristic Identification Result Table

[0110] Calculate the fluid shear stress. Extract the velocity gradient along the boundary of the net orifice body and multiply it by the seawater dynamic viscosity coefficient (1.08×10⁻³ Pascal·seconds at 20 °C) to obtain the shear stress distribution. The maximum shear stress appears at the rear of the contraction zone, with a value of 0.38 Pascal.

[0111] Calculate the flow flux of the net orifice cross-section based on the fluid velocity field and pressure distribution. Identify that the position of the minimum cross-section of the net orifice is at the end of the contraction zone (x = 462 microns). The calculation results of the cross-section flow flux are shown in Table 6:

[0112] Table 6 Measurement Results of the Net Orifice Cross-Section Flow under Different Pressure Conditions

[0113] Compare with the standard biological net orifice theoretical model (modified Hagen-Poiseuille formula) to calculate the measurement deviation. The deviation first decreases and then slightly increases with the increase of pressure, and reaches a minimum value of 3.55% in the medium pressure region (0.14 - 0.16 Pascal per micron). Generate a correction coefficient table based on the deviation analysis. The correction coefficient is defined as the ratio of the theoretical flow rate to the measured flow rate, arranged from small to large according to the pressure gradient, with an interval of 0.01 Pascal per micron, covering the actual application range of 0.05 to 0.20 Pascal per micron. After applying the correction coefficient, the flow measurement accuracy is improved to within ±1.2%.

[0114] In the traditional measurement of the microfluidic field of a biological network port flowmeter in a marine environment, the particle image velocimetry (PIV) and laser Doppler velocimetry (LDV) are mainly used. These methods have obvious limitations in measurement accuracy, spatial resolution, and adaptability. The spatial resolution of the traditional PIV method is generally in the range of 10 - 20 micrometers, making it difficult to capture the local flow field characteristics in micrometer-sized flow channels. Although the traditional LDV method has a high time resolution, it can only obtain point measurement data and cannot provide a complete flow field distribution. In addition, the traditional methods have poor adaptability to the seawater environment. Under high-salt and high-pressure conditions, the optical system is easily interfered, and the stability of the measurement results is insufficient. The method of the present invention has made significant progress compared with the traditional means: First, by using a high-precision three-dimensional microscopic imaging system combined with an optimized fluorescent tracer liquid, the spatial resolution is improved to the sub-micrometer level (0.8 micrometers), enabling the precise capture of the flow field details in the microchannel. Second, the flow field analysis method based on neural network breaks through the limitations of the traditional flow field reconstruction algorithm, can accurately identify the laminar-turbulent transition region, and improves the measurement accuracy under complex flow conditions. Third, through the trajectory optimization function and orthogonal grid mapping technology, high-quality flow field data processing is achieved, reducing the influence of measurement noise. Finally, the generated microfluidic field measurement correction coefficient table takes into account the systematic errors under different pressure conditions, improving the flow rate measurement accuracy from ±5 - 8% of the traditional method to ±1.2%, meeting the requirements of precise quantitative collection of deep-sea microplankton.

[0115] It should be noted that the detailed explanations of the variables involved in the present invention are shown in Table 7.

[0116] Table 7 Variable Explanation Table

[0117] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention.

Claims

1. A microfluidic field measurement method for a mechanical biological network port flowmeter, characterized in that, Including: Construct a three-dimensional digital model of the mesh orifice body and determine the main measurement area of the microfluidic field; Inject a tracer fluid containing fluorescent particles into the mesh orifice body; Adjust the laser confocal system to the main measurement area of the microfluidic field; Start the microfluidic driving system to generate a constant pressure gradient, and synchronously record the fluid movement trajectory in the mesh orifice body; Analyze the fluid movement trajectory data, process the original data using a trajectory optimization function, and calculate the fluid velocity field and acceleration field; Construct an orthogonal grid system, map the fluid velocity field data to the orthogonal grid system, and apply the Navier-Stokes equation to solve the fluid pressure distribution; Use a pre-trained microfluidic field analysis neural network model to process the fluid velocity field and fluid pressure distribution data, automatically identify the laminar-turbulent transition region, and calculate the fluid shear stress; Based on the fluid velocity field and fluid pressure distribution, calculate the flow flux of the mesh orifice cross-section, and obtain the flow-pressure relationship curve through integration; Compare the flow-pressure relationship curve with the standard biological mesh orifice theoretical model to generate a microfluidic field measurement correction coefficient table.

2. The microfluidic field measurement method of the mechanical biological network port flowmeter according to claim 1, characterized in that The steps for constructing the three-dimensional digital model of the mesh orifice body are specifically to perform a transverse scan of the biological mesh orifice flowmeter using a three-dimensional microscopy imaging system, and set the Gaussian beam scan step size to 10 to 20 micrometers. The Gaussian beam scan step size refers to the distance between two adjacent scans when the laser scanning system performs a transverse scan of the sample, and the Gaussian beam scan step size determines the transverse resolution of the three-dimensional imaging.

3. The microfluidic field measurement method of the mechanical biological network port flowmeter according to claim 2, characterized in that, The main measurement area of the microfluidic field refers to the spatial range in the mesh orifice body where the fluid flow is the most significant and has the greatest impact on the measurement results, including three key parts: the inlet area, the contraction area, and the expansion area.

4. The microfluidic field measurement method of the mechanical biological network port flowmeter according to claim 3, characterized in that, In the step of injecting the tracer fluid containing fluorescent particles into the mesh orifice body, the diameter of the fluorescent particles in the tracer fluid is 0.5 to 2 micrometers, the density of the fluorescent particles is 10,000 to 30,000 per milliliter, and the fluorescence wavelength range is 520 to 580 nanometers.

5. The microfluidic field measurement method of the mechanical biological network port flowmeter according to claim 4, characterized in that, In the step of adjusting the laser confocal system to the main measurement area of the microfluidic field, the laser power is set to 15 to 25 milliwatts, the scanning frequency is 200 to 500 hertz, and the thickness of the collected optical section is 5 to 10 micrometers. The thickness of the optical section refers to the thickness of the sample collected by the confocal microscopy system in the depth direction, and the thickness of the optical section determines the longitudinal resolution of the three-dimensional imaging.

6. The microfluidic field measurement method of the mechanical biological network port flowmeter according to claim 5, characterized in that In the step of starting the microfluidic driving system to generate a constant pressure gradient, the pressure gradient range is 0.05 to 0.2 pascals per micrometer, and the duration is 60 to 120 seconds.

7. The microfluidic field measurement method of the mechanical biological network port flowmeter according to claim 6, characterized in that, In the step of using the trajectory optimization function to process the original data, the trajectory optimization function is used to perform noise reduction and smoothing processing on the originally collected fluid movement trajectory data, improve the accuracy and continuity of the fluid movement trajectory data. The input includes the three-dimensional coordinate sequence of the fluorescent particles at consecutive time points, the time interval between adjacent frames, the evaluation value of the flow field background noise level, the signal-to-noise ratio threshold for fluorescent particle identification, and the velocity constraint range. The output is a high-quality particle trajectory data set after noise filtering and trajectory connection processing.

8. The microfluidic field measurement method of the mechanical biological network port flowmeter according to claim 7, characterized in that, The microfluidic field analysis neural network model adopts a multi-level residual network structure, where the network depth is determined by the ratio of the optical section thickness to the Gaussian beam scanning step size, the convolutional kernel size is determined by the ratio of the fluorescent particle diameter to the grid spacing, and the width of the attention mechanism is determined by the size of the main measurement area of the microfluidic field.

9. The microfluidic field measurement method of the mechanical biological network port flowmeter according to claim 8, characterized in that, The specific structure of the microfluidic field analysis neural network model is a hybrid architecture that combines a multi-scale residual convolutional network and a graph attention network. The network input layer receives three-dimensional fluid velocity field and fluid pressure distribution data. The encoder part uses eight layers of three-dimensional convolutional layers to extract multi-scale flow field features, and residual connections are set every two layers to avoid the problem of gradient disappearance. The intermediate layer uses an adaptive graph attention mechanism to establish long-distance flow field correlations, and the weight assignment is automatically adjusted based on fluid physical constraints. The decoder part consists of six layers of deconvolutional layers for reconstructing the fluid shear stress distribution, and the final output layer is a fully connected layer that generates the fluid shear stress.

10. The microfluidic field measurement method of the mechanical biological network port flowmeter according to claim 9, characterized in that, The steps for establishing the training data set during the pre-training process of the microfluidic field analysis neural network model include collecting fluid numerical simulation data under various typical biological orifice geometric structures, covering both laminar and turbulent conditions. For each geometric structure, thirty sets of flow field data are generated at different Reynolds numbers, including complete information on the fluid velocity field, fluid pressure distribution, and fluid shear stress. All data are normalized to eliminate the influence of dimensions, and then randomly divided into a training set, a validation set, and a test set. The training set accounts for 70% of the total data volume, the validation set accounts for 20%, and the test set accounts for 10%.

Images