A sensitive nanoparticle normal temperature green concentration device and method

Abstract

The application discloses a sensitive nanoparticle normal-temperature green concentration device and method. The device comprises a sample bin, a continuous flow circulation pipeline, a membrane separation unit, a temperature regulation unit, a flow rate regulation unit, a concentration regulation unit and a main control unit. The sample bin and the membrane separation unit are arranged at different sections of the continuous flow circulation pipeline and are in communication with the pipeline. Nanoparticle dispersion liquid is transported from the sample bin to the membrane separation unit to retain the nanoparticles by using a selective permeation membrane and allow the solvent and small molecules to permeate. The main control unit controls the temperature regulation unit to maintain the temperature of the nanoparticle dispersion liquid within a preset normal temperature range, controls the concentration regulation unit and the flow rate regulation unit to regulate the concentration of the nanoparticle dispersion liquid and the flow rate of the nanoparticle dispersion liquid flowing through the membrane separation unit, respectively, and monitors the agglomeration condition based on a visual learning large model and actively warns and intervenes when an agglomeration sign is monitored, so that the sensitive nanoparticles are efficiently and stably concentrated at normal temperature, low energy consumption and without the need of adding chemical additives.

Description

Technical Field

[0001] This invention relates to the field of concentration technology for nanomaterial dispersion systems, specifically to a room-temperature green concentration device and method suitable for sensitive nanoparticle dispersions. Background Technology

[0002] Sensitive nanoparticles refer to nanoscale dispersion systems that are prone to aggregation, particle size increase, and decreased dispersion stability under conditions such as external thermal stimulation, shearing, interfacial concentration, or changes in solvent environment, leading to a decline in functional performance. Typical examples include bio-based or polymer-based nanoparticles and their dispersions, such as chitin nanocrystals, cellulose nanocrystals, and nanofibers. The preparation process of these materials is complex and costly, and the initial concentration of the prepared dispersions is generally low, making it difficult to directly meet the requirements of subsequent applications such as film formation, composites, coatings, and rheological control. Therefore, concentration under mild conditions is necessary to increase the solid content.

[0003] Existing concentration technologies mostly employ methods such as heating and drying, centrifugation, and rotary evaporation. However, these methods are generally accompanied by temperature increases, shearing, or pressure changes, which can easily induce irreversible aggregation of sensitive nanoparticles, leading to decreased product performance or even sample failure. For example, Chinese patent (CN104437089B) published on July 6, 2016, provides a juice concentration membrane distillation device that uses temperature differences to distill water and thus concentrate the sample. However, a large temperature difference can cause irreversible aggregation of nanoparticles, making it unsuitable for concentrating sensitive nanoparticles. Furthermore, although membrane separation concentration methods can be implemented under relatively mild conditions, problems such as concentration polarization at the membrane interface and gradual decay of the concentration gradient across the membrane are prone to occur during actual operation. This leads to a decrease in the mass transfer driving force, resulting in reduced flux and efficiency in the later stages of concentration. Moreover, the formation of local high-concentration regions can further exacerbate the risk of aggregation. For example, Chinese patent (CN216260049U) published on April 12, 2022, provides a membrane separation concentration device for protein production. This concentration device lacks real-time monitoring of temperature, flow state, and sample state. After controlling the reaction conditions, it may only be used for the concentration of a certain type of protein or a certain class of proteins, and is not suitable for the concentration of sensitive nanoparticles. Summary of the Invention

[0004] To overcome the shortcomings of existing sensitive nanoparticle concentration technologies, such as high energy consumption, complex operation, and easy agglomeration due to heating or strong shearing leading to sample loss, and to address the problems of concentration polarization on the membrane surface and attenuation of the concentration gradient across the membrane when using membrane separation, which ultimately leads to reduced throughput and decreased concentration efficiency, this invention provides a green room-temperature concentration device and method for sensitive nanoparticles, achieving stable operation near room temperature while simultaneously suppressing agglomeration and maintaining concentration efficiency.

[0005] According to one aspect of the present invention, a room-temperature green concentration device for sensitive nanoparticles is provided, comprising a sample chamber, a continuous flow circulation pipeline, a membrane separation unit, a temperature control unit, a flow rate control unit, a concentration control unit, and a main control unit; the continuous flow circulation pipeline is designed in a ring or spiral shape, and the sample chamber and the membrane separation unit are respectively disposed in different sections of the continuous flow circulation pipeline and connected to the continuous flow circulation pipeline; the main control unit is connected to the temperature control unit, the flow rate control unit, and the concentration control unit; the flow rate control unit includes a peristaltic pump, and the main control unit controls the peristaltic pump to deliver the nanoparticle dispersion in the sample chamber to the continuous flow circulation pipeline, driving the dispersion to circulate along the continuous flow circulation pipeline. The nanoparticles flow into the membrane separation unit; the membrane separation unit is equipped with a selectively permeable membrane, which allows solvents and small molecules in the nanoparticle dispersion to pass through while retaining nanoparticles; the main control unit also controls the temperature control unit to regulate the temperature of the nanoparticle dispersion so as to maintain the temperature of the nanoparticle dispersion within a preset ambient temperature range; the main control unit also controls the flow rate control unit and the concentration control unit to regulate the temperature of the nanoparticle dispersion, the flow rate of the nanoparticle dispersion through the membrane separation unit, and the concentration of the nanoparticle dispersion, respectively, and monitors the aggregation of nanoparticles in the nanoparticle dispersion based on a large visual learning model, so as to provide proactive warning and intervention when aggregation signs are detected.

[0006] Furthermore, the main control unit integrates an online monitoring system, which includes an image acquisition and preprocessing module, an aggregation state discrimination module, and an early warning and decision-making module. The image acquisition and preprocessing module is used to acquire and preprocess images of the nanoparticle dispersion flowing through the observation window in real time. The aggregation state discrimination module judges the dispersion state of the nanoparticles based on the preprocessed images and outputs aggregation indication information. The early warning and decision-making module triggers an early warning based on the aggregation indication information and adjusts the operating parameters of the flow rate control unit and the temperature control unit in conjunction to achieve proactive intervention in the risk of aggregation.

[0007] Furthermore, the clustering state discrimination module includes a large visual learning model, which comprises an up-dimensional pointwise convolutional layer, a depthwise separable convolutional layer, a down-dimensional pointwise convolutional layer, a residual connection layer, and a dual-head output classification head. The up-dimensional pointwise convolutional layer projects the low-dimensional input into a high-dimensional space to obtain an up-dimensional feature map; the depthwise separable convolutional layer performs spatial feature extraction and channel feature fusion on the up-dimensional feature map to obtain a depth feature map; and the down-dimensional pointwise convolutional layer compresses the depth feature map. The process returns to a low-dimensional space to obtain a dimensionality-reduced feature map. The residual connection layer is used to directly add the dimensionality-reduced feature map output by the dimensionality-reduced pointwise convolutional layer to the low-dimensional feature map input to the dimensionality-upgrading pointwise convolutional layer when the spatial dimensions and channel numbers are the same. This forms a residual connection. The dual-head output classification head is used to perform clustering level classification and clustering risk scoring to output clustering indication information, which includes classification probability and clustering risk coefficient.

[0008] Furthermore, the image acquisition and preprocessing module preprocesses the acquired nanoparticle dispersion image, including one or more combinations of denoising, illumination or white balance correction, background subtraction, region of interest cropping, scale or grayscale normalization, and contrast enhancement.

[0009] Furthermore, the concentration device also includes a turbulence structure located on the feed side of the membrane separation unit, which is used to change the flow field distribution near the membrane surface and reduce the boundary layer thickness; wherein, the feed side of the membrane separation unit refers to the side before the solvent and small molecules permeate through the membrane separation unit.

[0010] Furthermore, the concentration device also includes an outer medium circuit, which is located on the permeate side of the membrane separation unit. The outer medium circuit includes an outer medium flow path and a circulation and replacement path. The outer medium flow path is connected to the circulation and replacement path, and the outer medium flow path is filled with an outer medium. The circulation and replacement path is used to realize the circulation or periodic replacement of the outer medium in the outer medium flow path, thereby continuously maintaining the effective concentration gradient on both sides of the membrane separation unit. The permeate side flow path of the membrane separation unit refers to the side reached by the solvent and small molecules after passing through the membrane separation unit.

[0011] Furthermore, the concentration device also includes a fan control unit, which includes a heat sink, an integrated air plate, and a centrifugal fan. The heat sink and the integrated air plate form a convective heat exchange space, and the centrifugal fan is used to drive airflow through the convective heat exchange space to perform forced convective heat exchange on the membrane separation unit.

[0012] Furthermore, the temperature control unit includes a heater, a cooler, and a temperature sensor. The temperature sensor is used to monitor the temperature of the nanoparticle dispersion in real time and feed it back to the main control unit. The main control unit adjusts the heater and cooler according to the current temperature of the nanoparticle dispersion so that the temperature of the nanoparticle dispersion is maintained within a preset room temperature range.

[0013] According to another aspect of the present invention, a method for room-temperature green concentration of sensitive nanoparticles is provided, which is implemented using the aforementioned room-temperature green concentration device for sensitive nanoparticles. The concentration method includes: adding the nanoparticle dispersion to be concentrated into a sample chamber, and setting target operating parameters for the concentration device; the target operating parameters include target temperature, target flow rate, and target concentration; driving the nanoparticle dispersion to circulate along a continuous flow circulation pipe and enter a membrane separation unit, so as to use the selective permeable membrane pore size to retain nanoparticles in the nanoparticle dispersion while allowing solvents and small molecules to permeate; acquiring the current operating parameters in real time, and controlling the temperature control unit, flow rate control unit, and concentration control unit respectively based on the current operating parameters and the target operating parameters, so as to achieve dynamic adjustment of the temperature, flow rate, and concentration of the nanoparticle dispersion; acquiring images of the nanoparticle dispersion flowing through an observation window in real time, monitoring the aggregation of nanoparticles in the nanoparticle dispersion based on a large visual learning model, and triggering an early warning and adjusting the current operating parameters in conjunction when aggregation signs are detected.

[0014] Furthermore, driving the nanoparticle dispersion to circulate along a continuous flow circulation pipe and enter the membrane separation unit includes: driving the nanoparticle dispersion to form a circulating flow on the feed side and pass through the membrane separation unit by a peristaltic pump; generating secondary flow or local eddies in the nanoparticle dispersion through a turbulence structure to change the flow field distribution on the membrane surface and weaken the concentration boundary layer thickness; and circulating or periodically replacing the medium on the permeate side through circulation and replacement channels.

[0015] In the above technical solution, the continuous flow circulation pipeline is annular or spiral in shape. The sample chamber and the membrane separation unit are located at different sections of the continuous flow circulation pipeline and connected to it, thus forming a closed loop. This structure, combined with a peristaltic pump, enables continuous and stable circulation of the nanoparticle dispersion, effectively improving membrane separation efficiency. The main control unit drives the flow of the nanoparticle dispersion through the peristaltic pump, enabling a stable mass transfer process at the membrane separation unit, achieving solvent and small molecule permeation and nanoparticle retention, resulting in a mild concentration effect. Through the synergistic effect of turbulence and flow rate closed-loop control, concentration polarization on the membrane surface can be effectively suppressed. At the same time, by circulating or replacing the outer medium, an effective concentration gradient across the membrane is maintained, reducing the risk of flux decline caused by the decay of mass transfer driving force in the later stages of concentration. The concentration device operates near room temperature throughout the process, and the temperature control unit can suppress temperature rise and local overheating during operation, thereby reducing the risk of nanoparticle aggregation. In addition, the main control unit is also equipped with an online monitoring module to realize in-situ monitoring of the dispersion state of nanoparticles. When signs of aggregation are detected, an early warning is automatically triggered, and operating parameters such as flow rate and temperature are adjusted in conjunction with the control, ultimately achieving efficient, green and stable concentration of sensitive nanoparticles.

[0016] Compared with the prior art, the present invention has at least the following beneficial effects:

[0017] (1) Stable operation at room temperature: The device operates stably within the preset room temperature range, which can effectively suppress the temperature rise and local overheating, avoid the temperature and pressure rise problems caused by traditional drying, rotary evaporation and other processes, and significantly reduce the risk of agglomeration of sensitive nanoparticles.

[0018] (2) Continuous and stable concentration: The dispersion is driven to circulate by a peristaltic pump, and the solvent and small molecules are continuously removed by a selectively permeable membrane. The concentration process is stable and controllable and has good potential for scale-up applications.

[0019] (3) Suppressing flux decline: The feed-side turbulence structure weakens the concentration polarization on the membrane surface. Through the circulation and replacement of the medium on the side and outside, the effective concentration gradient on both sides of the membrane can be maintained, which effectively improves the problem of sudden drop in flux in the later stage of concentration and ensures concentration efficiency.

[0020] (4) Closed-loop control ensures quality: The main control unit adjusts the temperature, flow rate and concentration in a coordinated manner, and automatically terminates or reduces the flow rate after the target concentration is reached; and the online monitoring system can provide early warning and timely intervention for early signs of aggregation, thereby improving the controllability of the concentration process and the consistency of product batches.

[0021] (5) Green and easy to operate: The entire concentration process does not require the introduction of additional chemical additives, has low energy consumption, and the operation process is clear and simple. It is suitable for the mild concentration treatment of various sensitive nanoparticle dispersions. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of a room-temperature green concentration device for sensitive nanoparticles provided in an embodiment of the present invention.

[0023] Figure 2 This is a schematic diagram of the main control unit provided in an embodiment of the present invention.

[0024] Figure 3 A schematic diagram of the temperature control unit provided in an embodiment of the present invention.

[0025] Figure 4 A closed-loop control flowchart of the temperature control unit provided in an embodiment of the present invention (which may be implemented using a PID control algorithm).

[0026] Figure 5 This is a schematic diagram of the flow rate control unit provided in an embodiment of the present invention.

[0027] Figure 6 The following is a logic diagram of the online monitoring and early warning operation for family reunion provided in an embodiment of the present invention.

[0028] Figure 7 The logic diagram of the aggregation state discrimination module and the early warning decision module provided in the embodiments of the present invention.

[0029] Figure 8 A flowchart illustrating the operation of the device provided in an embodiment of the present invention.

[0030] In the diagram: 1. Peristaltic pump; 2. Heat sink; 3. Drive motor; 4. 4*4 matrix keypad; 5. Sample chamber; 6. Liquid level sensor; 7. LCD display screen; 8. Heating integrated air plate; 9. Membrane separation unit; 10. Flow sensor; 11. Turbulence structure; 12. Camera. Detailed Implementation

[0031] The technical solutions of various embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Please refer to Figures 1 to 5 This embodiment provides a room-temperature green concentration device for sensitive nanoparticles. The device includes a sample chamber 5, a continuous flow circulation pipeline, a membrane separation unit 9, a temperature control unit, a flow rate control unit, a concentration control unit, and a main control unit; wherein, the continuous flow circulation pipeline is annular or spiral, and the sample chamber 5 and the membrane separation unit are respectively located in different sections of the continuous flow circulation pipeline and connected to the continuous flow circulation pipeline to form a closed circulation path.

[0033] Under the control of the main control unit, the peristaltic pump 1 in the flow rate regulation unit drives the nanoparticle dispersion to circulate smoothly in the continuous flow circulation pipeline and enter the membrane separation unit. The membrane separation unit is equipped with a selectively permeable membrane, which allows solvents and small molecules in the dispersion to pass through while efficiently retaining nanoparticles, thereby achieving a mild concentration effect. Simultaneously, under the control of the main control unit, the temperature regulation unit regulates the temperature of the nanoparticle dispersion to maintain it within a preset ambient temperature range. The flow rate regulation unit and the concentration regulation unit respectively regulate the temperature, flow rate of the nanoparticle dispersion through the membrane separation unit, and concentration of the nanoparticle dispersion. The main control unit also monitors the aggregation of nanoparticles in the dispersion based on a large visual learning model, providing proactive warnings and interventions when aggregation signs are detected.

[0034] (1) Main control unit (see Figure 2 )

[0035] The core function of the main control unit is to collect and process key operating parameters such as temperature, flow rate / velocity, and liquid level / mass, and send precise control signals to the temperature control unit, flow rate control unit, and concentration control unit to achieve closed-loop control of the concentration unit within a preset ambient temperature range. The main control unit can be implemented based on a microprocessor and can also be used via a control panel (e.g., a 4×4 matrix keyboard, an LCD display, or a touchscreen) to set target parameters and display the real-time operating status of the equipment, facilitating monitoring and adjustment by operators.

[0036] (2) Sample chamber 5 and concentration control unit

[0037] Sample chamber 5 is used to store the nanoparticle dispersion to be concentrated. Its material can be stainless steel or corrosion-resistant materials (such as PTFE), and it is equipped with an inlet, outlet, and circulation pipeline interface for convenient feeding, discharging, and circulation. The concentration control unit includes a level sensor 6, a weighing sensor (not shown), and a concentration calculation module. The main control unit estimates the current concentration of the dispersion based on the level change data detected by the level sensor 6 and the mass change data detected by the weighing sensor, combined with the principle of mass conservation. When the estimated concentration reaches the preset target concentration, the main control unit automatically controls the peristaltic pump to stop running, reduce the flow rate, or switch to a conservative circulation mode to avoid over-concentration.

[0038] (3) Continuous flow circulation pipe, turbulence structure, and external medium circuit

[0039] The continuous flow circulation pipeline connects the sample chamber 5, the peristaltic pump 1, and the membrane separation unit sequentially to form a closed loop. The pipeline material is preferably corrosion-resistant PTFE or silicone tubing to accommodate nanoparticle dispersions of different properties. To reduce concentration polarization at the membrane interface and maintain an effective concentration gradient across the membrane, this embodiment includes a flow-disrupting structure 11 (e.g., a flow-disrupting pipe, a static mixer, or a baffle structure) and an outer medium loop in the continuous flow circulation pipeline. The flow-disrupting mechanism and the outer medium loop are located on opposite sides of the membrane separation unit, and their flow paths are independent, with mass transfer achieved solely through the membrane separation unit.

[0040] The turbulence structure 11 effectively suppresses concentration polarization and improves membrane separation efficiency by altering the flow field distribution near the membrane surface and reducing the boundary layer thickness. It should be noted that the membrane surface refers to the membrane-liquid interface on the feed side of the selectively permeable membrane. Areas near the membrane surface tend to have very low flow velocities, flat streamlines, and almost stationary flow (especially the layer closest to the membrane surface). "Altering the flow field distribution" means using the turbulence structure to make the liquid near the membrane surface flow more uniformly, faster, and with vortex components, reducing dead zones and localized stagnation areas. "Reducing the boundary layer thickness" means thinning the boundary layer by increasing shear and disturbance near the membrane surface, essentially shortening and weakening the "buffer layer for mass transfer." The boundary layer refers to the concentration boundary layer formed near the membrane surface due to reduced flow. This boundary layer lies between the membrane surface and the bulk fluid on the feed side. Within it, the concentration of trapped nanoparticles and large molecular solutes increases and accumulates, thereby increasing mass transfer resistance and inducing concentration polarization. This leads to two consequences: the actual effective concentration difference across the membrane decreases, reducing the mass transfer driving force and lowering the flux; and the localized high concentration at the membrane surface makes it easier for sensitive nanoparticles to aggregate / deposit / block the membrane. Therefore, by "changing the flow field distribution and reducing the boundary layer thickness," the high-concentration layer near the membrane surface is less likely to accumulate, and the concentration becomes closer to that of the bulk solution, thus suppressing concentration polarization.

[0041] The outer medium circuit includes an outer medium flow path (or outer medium cavity) and a circulation and replacement passage. The outer medium flow path is filled with outer medium, and the outer medium flow path is connected to the circulation and replacement passage. The outer medium can be deionized water or a solvent compatible with the dispersion system. The circulation and replacement passage is used to realize the circulation flow or periodic replacement of the outer medium in the outer medium flow path, thereby continuously maintaining the mass transfer driving force on both sides of the membrane separation unit and effectively improving the problem of flux decline in the later stage of concentration.

[0042] (4) Membrane separation unit

[0043] The membrane separation unit is the core component for concentration, used to achieve the permeation of solvents and small molecules while retaining nanoparticles. Membrane separation unit 9 employs a selectively permeable membrane. This membrane can be any type that allows permeability to the solvent and one or more small molecule components in the nanoparticle dispersion while retaining the nanoparticles; no limitation is made here. Furthermore, the retention capacity of the selectively permeable membrane can be characterized by its pore size range, molecular weight cutoff (MWCO), or equivalent retention performance. In this embodiment, the membrane material's MWCO ranges from 1 kDa to 500 kDa; other membrane specifications with equivalent retention performance within this MWCO range can also be selected.

[0044] The membrane separation unit includes a feed / retentate side and a permeate side. The feed side of the membrane separation unit refers to the side before the solvent and small molecules permeate through the membrane separation unit. The permeate side of the membrane separation unit refers to the side after the solvent and small molecules permeate through the membrane separation unit. The turbulence structure 11 is disposed on the feed side of the membrane separation unit, and the outer medium loop is disposed on the permeate side of the membrane separation unit.

[0045] (5) Temperature control unit and fan control unit (see Figures 3 to 4 )

[0046] The temperature control unit includes a temperature sensor, a cooler, and a heater. The temperature sensor monitors the temperature of the nanoparticle dispersion in real time and feeds the data back to the main control unit. The main control unit precisely adjusts the operation of the cooler and heater based on the deviation between the current temperature and the preset temperature, ensuring that the dispersion temperature is stably maintained within the preset ambient temperature range (e.g., 10°C to 35°C). In some preferred embodiments, the concentration device also integrates a fan control unit, which includes a heat sink 2, an integrated air plate 8, and a centrifugal fan (not shown). The heat sink and the integrated air plate 8 form a convective heat exchange space. The centrifugal fan drives the airflow through the convective heat exchange space, forming a stable convective airflow to force convective heat exchange on the membrane separation unit and its surrounding space, further suppressing temperature rise and local overheating during the operation of the concentration device, thereby reducing the risk of nanoparticle aggregation.

[0047] (6) Flow rate control unit (see Figure 5 )

[0048] The flow rate control unit consists of a peristaltic pump 1, a drive motor 3, and a flow sensor 10. The flow sensor 10 is installed on the continuous flow circulation pipeline to detect the flow rate or velocity of the dispersion in the pipeline in real time and feed the data back to the main control unit. The main control unit performs closed-loop adjustment of the speed of the drive motor 3 based on the feedback signal to ensure that the dispersion passes through the membrane separation unit stably at a preset flow rate, thereby obtaining stable mass transfer conditions and ensuring the consistency of the concentration effect.

[0049] (7) Online monitoring system (see Figures 6 to 7 )

[0050] The main control unit integrates an online monitoring system, which includes an image acquisition and preprocessing module, an aggregation state discrimination module, and an early warning and decision-making module. The image acquisition and preprocessing module captures images of the nanoparticle dispersion flowing through the observation window using camera 12 and preprocesses these images to improve image quality. Preprocessing includes one or more of the following: noise reduction, illumination or white balance correction, background subtraction, region of interest cropping, scale or grayscale normalization, and contrast enhancement. The aggregation state discrimination module extracts feature parameters such as turbidity, texture, and edge or cluster connectivity from the preprocessed nanoparticle dispersion image. Combined with threshold rules, it discriminates the nanoparticle dispersion state (i.e., aggregation state) and outputs aggregation indication information. This method is simple in structure, has a fast response speed, and is suitable for scenarios with high real-time requirements. The early warning and decision-making module determines whether to trigger an early warning based on the aggregation indication information. If aggregation signs are detected, it immediately adjusts the peristaltic pump speed and temperature setpoint, and if necessary, pauses the concentration process or switches to a conservative operating mode to reduce the risk of further aggregation.

[0051] Furthermore, the clustering state discrimination module includes a large visual learning model (i.e., MobileNetV3). This large visual learning model uses a linear bottleneck and inverse residual structure, including dimensionality-upgrading pointwise convolutional layers, depthwise separable convolutional layers, dimensionality-reducing pointwise convolutional layers, residual connection layers, and a dual-head output classification head. The dimensionality-upgrading pointwise convolutional layers project the low-dimensional input onto a high-dimensional space to obtain an upgradable feature map. The depthwise separable convolutional layers extract spatial features and fuse channel features from the upgradable feature map to obtain a depth feature map. The dimensionality-reducing pointwise convolutional layers compress the depth feature map back into a low-dimensional space to obtain a reduced-dimensional feature map. The residual connection layers, when the low-dimensional feature map input to the upgradable pointwise convolutional layer and the reduced-dimensional feature map output by the reduced-dimensional pointwise convolutional layer have the same spatial size and number of channels, directly add the reduced-dimensional feature map output by the reduced-dimensional pointwise convolutional layer to the low-dimensional feature map input to the upgradable pointwise convolutional layer, forming a residual connection. The dual-head output classification head is used to classify reunification levels and score reunification risks, outputting classification probabilities and reunification risk coefficients (i.e., reunification indication information).

[0052] ① Dimensional Upgrading Pointwise Convolutional Layer: A 1×1 pointwise convolution is used to transform the low-dimensional input (number of channels) into a higher dimension. Projected into a higher-dimensional space (number of channels) ×t, where t is the expansion factor), and after batch normalization and h-swish activation function, the features are obtained. The corresponding formula is:

[0053] ,

[0054] in, The low-dimensional feature map (i.e., the preprocessed image of the nanoparticle dispersion) is input to the pointwise convolutional layer for the dimension upscaling. H is the high-dimensional feature map output by the pointwise convolutional layer (i.e., the up-dimensional feature map); H is the low-dimensional feature map. The height of W is the low-dimensional feature map. width, For low-dimensional feature maps The number of channels; It is a 1×1 pointwise convolution kernel. ∈R1×1×Cin×tCin; For h-swish activation function, , For pointwise convolution kernels The output is ReLU6(x)=min(max(0,x),6).

[0055] ② Depthwise Separable Convolutional Layer: This layer splits the standard convolution into two parts (depthwise convolution and pointwise convolution), separating spatial feature extraction from channel feature fusion. This decomposition significantly reduces computation and the number of parameters (typically to 1 / 8 to 1 / 9 of the standard convolution) while only incurring a minor loss of accuracy. In nanoparticle detection, this structure is particularly suitable for processing images of nanoparticle dispersions with fine features, significant noise, and requiring real-time processing, providing key technical support for efficient and accurate nanoparticle state monitoring. Specifically, depthwise convolution performs convolution on each input channel individually, extracting only spatial features without channel mixing. Pointwise convolution performs a 1x1 convolution on the output of the depthwise convolution, fusing the channels.

[0056] Furthermore, for low-dimensional feature maps Perform a 3×3 depthwise convolution to extract spatial features, then perform batch normalization and h-swish activation to obtain the depth feature map. The corresponding formula is:

[0057] ,

[0058] in, For h-swish activation function, For upscaling feature maps height, For upscaling feature maps width, For upscaling feature maps The number of channels, This refers to the high-dimensional feature map output by the depth-separable convolutional layer, i.e., the depth feature map. For depthwise convolution kernels, .

[0059] ③ Dimensionality Reduction Pointwise Convolutional Layer: A 1×1 pointwise convolution is used to reduce the depth feature map. Compressed back to a lower output dimension And after batch normalization. The key innovation here lies in the application of linear bottleneck: when the condition is met... < When the network is deep, no non-linear activation function is used after the compression layer, preserving the linear transformation. Experimental results show that non-linear mapping in low-dimensional space destroys the manifold structure and leads to information loss, while linear transformation can retain the feature information learned from high-dimensional space to the greatest extent.

[0060] ,

[0061] in, The low-dimensional feature map output by the pointwise convolutional layer is called the dimensionality-reduced feature map. It is a 1×1 pointwise convolution kernel. For dimensionality reduction feature maps The number of channels.

[0062] ④ Residual Connection Layer: When the low-dimensional feature map X input to the up-dimensional pointwise convolutional layer is compared with the reduced-dimensional feature map output from the down-dimensional pointwise convolutional layer... The same spatial dimensions (s) =1) and the number of channels is equal ( = When ), the dimensionality-reduced feature map output by the pointwise convolutional layer is... The feature map X is directly added to the low-dimensional feature map X input to the pointwise convolutional layer of the higher-dimensional layer, forming a fast residual connection. This facilitates gradient flow and alleviates the gradient vanishing problem in the training of deep networks.

[0063] ,

[0064] in, This is the output feature map of the residual connection layer (i.e., the feature map with the same size as the input). The step size.

[0065] ⑤ Dual-head output classification head: This invention adopts the "multi-task learning" approach, designing a dual-head output custom classification head (i.e., including clustering level classification and clustering risk scoring) to provide richer state information: ① Discrete classification head, used for clustering level classification: Using the Softmax activation function, it outputs a discrete state classification probability P(Class), where Class is the clustering level. For example: ["Class=0: Normal dispersion", "Class=1: Slight risk", "Class=2: Severe clustering"]. ② Continuous regression head, used for clustering risk scoring: Using the Sigmoid activation function, it outputs a continuous value between 0 and 1 as the comprehensive clustering risk coefficient R. This value can more smoothly and sensitively reflect state changes, facilitating the setting of dynamic thresholds and trend analysis.

[0066] In addition, the large-scale visual learning model also employs neural network architecture search, a meta-design tool used to automatically search for the optimal module type / kernel size / channel number / activation function position, thereby finding the most optimized combination of network modules in terms of speed and accuracy for the large-scale visual learning model, which is the key to the leading performance of the large-scale visual learning model.

[0067] Early warning and decision-making module: Based on the classification probability P and clustering risk coefficient R output by the visual learning big model, a multi-level response is triggered, the visually abnormal area is highlighted on the LCD screen of the control panel, and an audible and visual alarm is issued. Specifically, for "Class=0: normal dispersion", in P(0) 0.6 and R At 0.3, the concentration unit is determined to be in normal condition, and the audible and visual alarm device is in a green state. Otherwise, a re-inspection is triggered, and the audible and visual alarm device is in a yellow warning state (meaning a re-inspection is required). For "Class=1: Minor Risk", in P(1) 0.4 or R At 0.5, it is confirmed as a minor risk, and the audible and visual alarm device will be in a yellow warning state (meaning that adjustment is needed); otherwise, it will be downgraded to a monitoring state, and the audible and visual alarm device will be in a blue monitoring state; for "Class=2: Severe Clustering", at P(1) 0.4 or R At a concentration of 0.5, severe aggregation is confirmed, triggering a red alarm on the audible and visual system, terminating the concentration process. Otherwise, the risk level is downgraded to minor, triggering a yellow alarm (indicating the need for adjustments, such as momentarily reducing the flow rate of peristaltic pump 1 to decrease shear force, fine-tuning the temperature control unit to ensure the temperature remains within its most stable range, temporarily pausing the concentration process, and awaiting operator inspection). Understandably, the thresholds for the classification probability P and aggregation risk coefficient R can be adjusted according to actual needs and are not limited here. Furthermore, the main control unit records data: image sequences and corresponding equipment parameters before and after all alarm events are automatically saved, forming a "black box" record for subsequent process optimization and retraining of the large visual learning model.

[0068] It should be noted that the online monitoring system of this invention upgrades the traditional control mode, which relies on a single physical sensor (temperature, flow rate), to a dual-modal intelligent control system combining physical sensing and visual perception. This endows the device with a "visual" capability, enabling it to directly "see" the essential state changes of the nanocrystalline solution sample, rather than relying solely on inferences from indirect parameters. This fundamentally solves the industry problem of "sensitive" nanoparticles agglomerating during concentration due to unpredictable minute disturbances, achieving a technological leap from "passive prevention" to "active early warning and intervention," ensuring the high quality and batch-to-batch consistency of the concentrated product.

[0069] Please refer to the appendix for details. Figure 8 Based on the same technical concept as the above-described device embodiments, the present invention also provides a room-temperature green concentration method for sensitive nanoparticles, which is implemented using the above-described room-temperature green concentration device for sensitive nanoparticles. The method specifically includes the following steps:

[0070] Step S1: Add the nanoparticle dispersion to be concentrated into the sample chamber and set the target operating parameters of the concentration device.

[0071] In step S1, the nanoparticle dispersion to be concentrated is added to sample chamber 5 through the injection port. Then, the target operating parameters are set through the control panel. The target operating parameters include, but are not limited to, target temperature, target flow rate and target concentration (for example, target temperature is 30°C, target flow rate is 0.5 L / min and target concentration is 0.2 mol / L).

[0072] Step S2 involves driving the nanoparticle dispersion to circulate along a continuous flow circulation pipe and enter the membrane separation unit to retain nanoparticles in the nanoparticle dispersion using the selectively permeable membrane pore size, while allowing solvents and small molecules to pass through.

[0073] In step S2, the peristaltic pump 1 is started, driving the nanoparticle dispersion to circulate along the continuous flow circulation pipe and smoothly enter the membrane separation unit. Utilizing the selective permeability of the membrane, solvents and small molecules permeate through the membrane structure, while the nanoparticles are retained, achieving gentle concentration. Driving the nanoparticle dispersion to circulate along the continuous flow circulation pipe and enter the membrane separation unit includes: using the peristaltic pump to drive the nanoparticle dispersion to form a circulating flow on the feed side and pass through the membrane separation unit, avoiding localized high-concentration accumulation on the membrane surface due to stagnation; using a turbulence structure to generate secondary flow or localized eddies in the nanoparticle dispersion to change the flow field distribution near the membrane surface and weaken the concentration boundary layer thickness, thereby suppressing concentration polarization; and setting an outer medium cavity and circulation and replacement pathway on the permeate side of the selective permeable membrane, using the circulation and replacement pathway to circulate or periodically replace the outer medium on the permeate side, maintaining the concentration gradient and mass transfer driving force across the membrane, and preventing a sudden drop in flux due to concentration decay in the later stages of concentration. In addition, the concentration unit operates at near room temperature. Optionally, if necessary, an external fan control unit can be added to exchange heat with the membrane separation unit and the surrounding space to suppress operating temperature rise and local overheating, thereby reducing the risk of agglomeration of sensitive nanoparticles.

[0074] Step S3: Real-time acquisition of current operating parameters, and control of temperature control unit, flow rate control unit and concentration control unit based on current operating parameters and target operating parameters, so as to achieve dynamic adjustment of temperature, flow rate and concentration of nanoparticle dispersion through membrane separation unit.

[0075] In step S3, the main control unit collects current operating parameters in real time, including but not limited to temperature, flow rate or velocity, liquid level or mass, and controls the temperature control unit, flow rate control unit and concentration control unit respectively to dynamically adjust the temperature of the nanoparticle dispersion, the flow rate through the membrane separation unit and the concentration based on the current operating parameters and the target operating parameters, so as to avoid over-concentration.

[0076] Step S4: Real-time image of the nanoparticle dispersion flowing through the observation window is acquired. The aggregation of nanoparticles in the nanoparticle dispersion is monitored based on a large visual learning model. When aggregation signs are detected, an early warning is triggered and the current operating parameters are adjusted accordingly.

[0077] In step S4, images of the dispersion flowing through the observation window are acquired in real time and input into the visual learning model. The model is then used to determine the aggregation state. Specifically, the visual learning model focuses on identifying two key visual features: ① Changes in particle size distribution: Bright spots or dark patches in the image that are significantly larger than individual nanoparticles; these are direct manifestations of micro-aggregates. ② Changes in texture features: When particles begin to aggregate, the overall optical texture of the solution changes from uniform and fine to coarse and uneven, exhibiting "flocculent" or "cloud-like" patterns. By learning these subtle changes in texture, the visual learning model can identify early aggregation that is difficult for the human eye to detect. Finally, when aggregation signs are detected, an early warning is triggered, and operating parameters such as flow rate and temperature are adjusted accordingly; if necessary, the concentration process is paused.

[0078] In summary, the continuous flow circulation pipeline of this invention is annular or spiral in shape. The sample chamber and the membrane separation unit are located at different sections of the continuous flow circulation pipeline and are connected to it, thus forming a closed loop. This structure, combined with a peristaltic pump, enables continuous and stable circulation of the nanoparticle dispersion, effectively improving membrane separation efficiency. The main control unit drives the flow of the nanoparticle dispersion through the peristaltic pump, enabling a stable mass transfer process at the membrane separation unit, achieving solvent and small molecule permeation and nanoparticle retention, resulting in a mild concentration effect. Through the synergistic effect of turbulence and flow rate closed-loop control, concentration polarization at the membrane surface can be effectively suppressed. At the same time, by circulating or replacing the outer medium, an effective concentration gradient across the membrane is maintained, reducing the risk of flux decline caused by the decay of mass transfer driving force in the later stages of concentration. The concentration device operates near room temperature throughout the process, and the temperature control unit can suppress temperature rise and local overheating during operation, thereby reducing the risk of nanoparticle aggregation. In addition, the main control unit is also equipped with an online monitoring module to realize in-situ monitoring of the dispersion state of nanoparticles. When signs of aggregation are detected, an early warning is automatically triggered, and operating parameters such as flow rate and temperature are adjusted in conjunction with the control, ultimately achieving efficient, green and stable concentration of sensitive nanoparticles.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the present invention. Those skilled in the art can make various modifications or equivalent substitutions to the above embodiments, or combine, add, or subtract some of the technical features, without departing from the spirit and principles of the present invention; all technical solutions falling within the scope defined by the claims of the present invention and their equivalents should be included within the protection scope of the present invention.

Claims

1. A room-temperature green concentration device for sensitive nanoparticles, characterized in that, The system includes a sample chamber, a continuous flow circulation pipeline, a membrane separation unit, a temperature control unit, a flow rate control unit, a concentration control unit, and a main control unit. The continuous flow circulation pipeline is designed in a ring or spiral shape, with the sample chamber and membrane separation unit located in different sections of the pipeline and connected to it. The main control unit is connected to the temperature control unit, the flow rate control unit, and the concentration control unit. The flow rate control unit includes a peristaltic pump, which the main control unit controls to deliver the nanoparticle dispersion from the sample chamber to the continuous flow circulation pipeline, driving the dispersion to circulate along the pipeline and enter the membrane separation unit. The unit is equipped with a selectively permeable membrane, which allows solvents and small molecules in the nanoparticle dispersion to pass through while retaining nanoparticles. The main control unit also controls the temperature control unit to regulate the temperature of the nanoparticle dispersion so that the temperature of the nanoparticle dispersion is maintained within a preset room temperature range. The main control unit also controls the flow rate control unit and the concentration control unit to regulate the temperature of the nanoparticle dispersion, the flow rate of the nanoparticle dispersion through the membrane separation unit, and the concentration of the nanoparticle dispersion, respectively. Based on a large visual learning model, the unit monitors the aggregation of nanoparticles in the nanoparticle dispersion to provide proactive warnings and interventions when aggregation signs are detected.

2. The room-temperature green concentration device for sensitive nanoparticles according to claim 1, characterized in that, The main control unit integrates an online monitoring system, which includes an image acquisition and preprocessing module, an aggregation state discrimination module, and an early warning and decision-making module. The image acquisition and preprocessing module is used to acquire and preprocess images of the nanoparticle dispersion flowing through the observation window in real time. The aggregation state discrimination module determines the dispersion state of nanoparticles based on the preprocessed image and outputs aggregation indication information; the early warning and decision-making module triggers an early warning based on the aggregation indication information and adjusts the operating parameters of the flow rate control unit and the temperature control unit in conjunction to achieve proactive intervention in aggregation risk.

3. The room-temperature green concentration device for sensitive nanoparticles according to claim 2, characterized in that, The clustering state discrimination module includes a large visual learning model, which comprises an up-dimensional pointwise convolutional layer, a depthwise separable convolutional layer, a down-dimensional pointwise convolutional layer, a residual connection layer, and a dual-head output classification head. The up-dimensional pointwise convolutional layer projects the low-dimensional input into a high-dimensional space to obtain an up-dimensional feature map. The depthwise separable convolutional layer extracts spatial features and fuses channel features from the up-dimensional feature map to obtain a depth feature map. The down-dimensional pointwise convolutional layer compresses the depth feature map back to a low dimension. The space is used to obtain a dimensionality-reduced feature map; the residual connection layer is used to directly add the dimensionality-reduced feature map output by the dimensionality-reduced pointwise convolutional layer to the low-dimensional feature map input to the dimensionality-upgrading pointwise convolutional layer when the spatial size and channel number are the same as those of the low-dimensional feature map input to the dimensionality-upgrading pointwise convolutional layer, forming a residual connection; the dual-head output classification head is used to perform clustering level classification and clustering risk scoring to output clustering indication information, which includes classification probability and clustering risk coefficient.

4. The room-temperature green concentration device for sensitive nanoparticles according to claim 2, characterized in that, The image acquisition and preprocessing module preprocesses the acquired nanoparticle dispersion image, including one or more combinations of noise reduction, illumination or white balance correction, background subtraction, region of interest cropping, scale or grayscale normalization, and contrast enhancement.

5. The room-temperature green concentration device for sensitive nanoparticles according to claim 1, characterized in that, The concentration device also includes a turbulence structure located on the feed side of the membrane separation unit, which is used to change the flow field distribution near the membrane surface and reduce the boundary layer thickness; wherein, the feed side of the membrane separation unit refers to the side before the solvent and small molecules permeate through the membrane separation unit.

6. The room-temperature green concentration device for sensitive nanoparticles according to claim 5, characterized in that, The concentration device further includes an outer medium circuit, which is located on the permeate side of the membrane separation unit. The outer medium circuit includes an outer medium flow path and a circulation and replacement path. The outer medium flow path is connected to the circulation and replacement path. The outer medium flow path is filled with an outer medium. The circulation and replacement path is used to realize the circulation or periodic replacement of the outer medium in the outer medium flow path, thereby continuously maintaining the effective concentration gradient on both sides of the membrane separation unit. The permeate side flow path of the membrane separation unit refers to the side reached by the solvent and small molecules after passing through the membrane separation unit.

7. The room-temperature green concentration device for sensitive nanoparticles according to claim 1, characterized in that, The concentration device also includes a fan control unit, which includes a heat sink, an integrated air plate, and a centrifugal fan. The heat sink and the integrated air plate form a convective heat exchange space. The centrifugal fan is used to drive airflow through the convective heat exchange space to perform forced convective heat exchange on the membrane separation unit.

8. The ambient temperature green concentration device for sensitive nanoparticles according to claim 1, characterized in that, The temperature control unit includes a heater, a cooler, and a temperature sensor. The temperature sensor is used to monitor the temperature of the nanoparticle dispersion in real time and feed it back to the main control unit. The main control unit adjusts the heater and cooler according to the current temperature of the nanoparticle dispersion to keep the temperature of the nanoparticle dispersion within a preset room temperature range.

9. A room-temperature green concentration method for sensitive nanoparticles, characterized in that, The concentration is achieved using a room-temperature green concentration device for sensitive nanoparticles as described in any one of claims 1 to 8, wherein the concentration method comprises: The nanoparticle dispersion to be concentrated is added to the sample chamber, and the target operating parameters of the concentration device are set; the target operating parameters include target temperature, target flow rate and target concentration. The nanoparticle dispersion is driven to circulate along a continuous flow circulation pipe and enter the membrane separation unit to retain the nanoparticles in the nanoparticle dispersion by utilizing the selective permeable membrane pore size, while allowing the solvent and small molecules to permeate. The system collects the current operating parameters during actual sampling and controls the temperature control unit, flow rate control unit, and concentration control unit based on the current operating parameters and the target operating parameters, so as to achieve dynamic adjustment of the temperature, flow rate, and concentration of the nanoparticle dispersion through the membrane separation unit. The system acquires real-time images of the nanoparticle dispersion flowing through the observation window, monitors the aggregation of nanoparticles in the dispersion based on a large visual learning model, and triggers an early warning and adjusts the current operating parameters in conjunction with the detection of aggregation signs.

10. The method for room-temperature green concentration of sensitive nanoparticles according to claim 9, characterized in that, The process of driving the nanoparticle dispersion to circulate along a continuous flow circulation pipe and enter the membrane separation unit includes: The nanoparticle dispersion is driven by a peristaltic pump to form a circulating flow on the feed side and then separated by a membrane separation unit; By using turbulence structures to generate secondary flow or local eddies in the nanoparticle dispersion, the flow field distribution near the membrane surface can be altered and the concentration boundary layer thickness weakened. The medium is circulated through the side, middle and outer sides or replaced periodically via circulation and replacement channels.

Images