A fully automated method for generating microdroplets

Abstract

This invention relates to the field of microfluidics and discloses a fully automated method for generating microdroplets, comprising the following steps: S1, acquiring a dataset including videos and images of droplet generation processes under various working environments; S2, establishing and training a scene recognition model; S3, establishing and training a droplet contour recognition model; S4, setting an initial state; S5, using the trained scene recognition model to identify the current scene state and adjusting the external phase pressure value according to the current scene state until droplet generation occurs; S6, using the trained droplet contour recognition model to detect the droplet contour, calculating the deviation between the current droplet size and the desired size, and adjusting the pressure pump parameters according to the deviation until the deviation is less than a set threshold, stopping the adjustment, and completing droplet generation. This invention can evaluate the effectiveness of the entire system environment, assess solution and chip design issues, and can also fully automatically achieve microdroplet generation from scratch.

Description

Technical Field

[0001] This invention relates to the field of microfluidics, and in particular to a fully automated method for generating microdroplets. Background Technology

[0002] Machine vision applications are a hot topic with a wide range of applications, such as facial recognition and autonomous driving. Effective image analysis using deep learning algorithms can save manpower and increase work efficiency. Furthermore, image temporal information contains rich temporal features, which can be fully utilized to solve problems in dynamic and complex scenarios. Microfluidic systems refer to systems that use microchannels to process or manipulate tiny fluids. Microfluidics can achieve a series of microfabrications and micromanipulations that are difficult to accomplish using conventional methods.

[0003] The generation of microdroplets requires operators to manually adjust the pump's flow rate or pressure to produce droplets of the appropriate size. For newly formulated solutions or designed microfluidic chips, it's uncertain whether droplet generation will occur, often requiring operators to repeatedly experiment with various parameters. This process is time-consuming, requires significant operator effort, and heavily relies on experience. Even if microdroplets have been successfully generated previously, changes in solution properties such as viscosity due to varying storage times, and repeated use of microfluidic chips can leave liquid or impurities in the microchannels, necessitating readjustment of previously effective droplet generation parameters, or even preventing droplet generation altogether. A solution that can generate droplets from scratch while automatically evaluating the chip's status would significantly reduce operator time spent on this process, making it a highly valuable tool.

[0004] Existing technology discloses a microfluidic droplet generation device with fast adjustment speed, low cost, and a small proportion of defective droplets. This microfluidic droplet generation device includes a microfluidic chip, a vibration mechanism, a photodetector, a storage mechanism, a human-machine interface mechanism, and a control mechanism. This microfluidic droplet generation device combines automatic feedback and pressure-controlled oscillation adjustment technology, enabling adaptive adjustment of the generated droplet size. It can generate stable and highly uniform droplets, and features fast adjustment and droplet generation speeds. Furthermore, the adjustment process does not affect droplet generation, allowing for continuous adjustment. This patent allows users to input required droplet size, continuous phase, and dispersed phase information via a human-machine interface. The control mechanism then searches the storage mechanism for the relationship between the input and the corresponding droplet size, obtaining the amplitude and / or frequency of the electrical signal driving the vibration mechanism. This amplitude and frequency are adjusted accordingly. The size of the droplets generated in the output channel is monitored and fed back to the control mechanism. The control mechanism then controls the human-machine interface to display the current droplet generation information and records the relationship between the current amplitude and / or frequency of the electrical signal driving the vibration mechanism, as well as the composition of the continuous and / or dispersed phases, and the droplet size. This information is compared with the stored information. If no corresponding information is found, or if the detected droplet size information is inconsistent with the found information, the relevant information in the storage mechanism is updated. Simultaneously, the control mechanism compares the detected droplet size with the desired size and adjusts the amplitude and / or frequency of the electrical signal driving the vibration mechanism based on the comparison result. However, this patent can only adjust the amplitude and / or frequency of the electrical signal driving the vibration mechanism based on already generated droplets. It cannot control the process when there are no droplets or when droplets are being generated, and it cannot monitor the process from no droplets to droplet generation. Summary of the Invention

[0005] The purpose of this invention is to provide a fully automated method for generating microdroplets that is highly adaptable and has a higher degree of automation.

[0006] To achieve the above objectives, the present invention provides a fully automated method for generating microdroplets, comprising the following steps:

[0007] S1. Obtain the dataset, which includes videos and images of droplet generation processes under various working environments;

[0008] S2. Establish a scene recognition model. Use the video of the droplet generation process in the dataset of step S1 to train the scene recognition model and obtain a trained scene recognition model.

[0009] S3. Establish a droplet contour recognition model. Use the droplet images in the dataset from step S1 to train the droplet contour recognition model and obtain the trained droplet contour recognition model.

[0010] S4. Set the initial state;

[0011] S5. Use the scene recognition model trained in step S2 to identify the current scene state, and adjust the external phase pressure value according to the current scene state until droplets are generated.

[0012] S6. Use the droplet contour recognition model trained in step S3 to detect the droplet contour, calculate the deviation between the current droplet size and the desired droplet size, and adjust the pressure pump parameters according to the deviation until the deviation is less than the set threshold, then stop adjusting and complete the droplet generation.

[0013] As a preferred embodiment, in step S1, the videos in the dataset are videos of normal droplet generation state, laminar flow state, internal phase reflux state, external phase reflux state, impurity state, and unknown state. Among them, the video segments of internal phase reflux and external phase reflux state need to include the entire process from the start of reflux to the completion of reflux.

[0014] As a preferred embodiment, in step S1, a portion of the droplets are extracted from the image where droplets are normally generated, and the extracted droplets are then transferred to the image where no droplets are generated.

[0015] As a preferred embodiment, in step S4, the initial state is that the pressure values ​​of the internal and external phases are the same.

[0016] As a preferred option, in step S5, a recognition time threshold is set. If the scene recognition model fails to recognize the scene state after the detection time exceeds the threshold, it is considered that the current solution or chip has been contaminated and cannot generate droplets.

[0017] As a preferred option, in step S5, when the scene is in a laminar flow state, the control step size is increased to increase the external phase pressure value; when the scene is in an external phase reflux state, the control step size is decreased to reduce the internal phase pressure.

[0018] As a preferred option, the video length of the droplet generation process is 10–20 seconds.

[0019] As a preferred option, the scene recognition model is a 3D convolutional network model.

[0020] As a preferred embodiment, the droplet contour recognition model is a segmentation network model.

[0021] As a preferred embodiment, the droplet contour recognition model is a Mask RCNN network model.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0023] This invention identifies scene states through a scene recognition model and adjusts accordingly based on the identified scenes. It can control the process from no droplets to droplet formation, demonstrating strong adaptability. Furthermore, after droplet formation, it adjusts according to the droplet's outline until a droplet that meets the requirements is formed. It provides full monitoring of the process from no droplets to droplet formation and controls it in real time based on the scene and liquid outline. This results in a high degree of automation, eliminating the need for operators to observe and control the process themselves, thus freeing up operators. Attached Figure Description

[0024] Figure 1 This is a flowchart of the fully automated microdroplet generation method according to an embodiment of the present invention.

[0025] Figure 2 This is a flowchart of the model training process according to an embodiment of the present invention.

[0026] Figure 3 This is a schematic diagram of the droplet generation process control according to an embodiment of the present invention. Detailed Implementation

[0027] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0028] Example 1

[0029] like Figure 1 , Figure 2 and Figure 3 As shown, a preferred embodiment of the present invention provides a fully automated microdroplet generation method, comprising the following steps:

[0030] S1. Obtain the dataset, which includes videos and images of droplet generation processes under various working environments;

[0031] S2. Establish a scene recognition model. Use the video of the droplet generation process in the dataset of step S1 to train the scene recognition model and obtain a trained scene recognition model.

[0032] S3. Establish a droplet contour recognition model. Use the droplet images in the dataset from step S1 to train the droplet contour recognition model and obtain the trained droplet contour recognition model.

[0033] S4. Set the initial state;

[0034] S5. Use the scene recognition model trained in step S2 to identify the current scene state, and adjust the external phase pressure value according to the current scene state until droplets are generated.

[0035] S6. Use the droplet contour recognition model trained in step S3 to detect the droplet contour, calculate the deviation between the current droplet size and the desired droplet size, and adjust the pressure pump parameters according to the deviation until the deviation is less than the set threshold, then stop adjusting and complete the droplet generation.

[0036] This embodiment identifies scene states through a scene recognition model and adjusts accordingly based on different scenes. It can control the process from no droplets to droplet formation, demonstrating strong adaptability. Furthermore, after droplet formation, it adjusts according to the droplet's outline until a droplet that meets the requirements is formed. The entire process from no droplets to droplet formation is monitored and controlled in real time based on the scene and liquid outline, resulting in a high degree of automation. This eliminates the need for operators to observe and control the process themselves, freeing up operators.

[0037] Specifically, in step S1 of this embodiment, the videos in the dataset are videos of normal droplet generation, laminar flow, internal phase reflux, external phase reflux, impurity, and unknown states. The video clips for internal and external phase reflux states need to include the entire process from the start of reflux to its completion. Optionally, the video length for the droplet generation process is 10–20 seconds.

[0038] Specifically, in step S1, a dataset of droplets under various working environments is acquired. Each dataset is a 10-20 second video clip containing videos of normal droplet generation, laminar flow, internal phase reflux, external phase reflux, impurity, and unknown states. For the video clips of internal and external phase reflux states, the entire process from the start to the completion of reflux needs to be included, because if the reflux process has already been completed, the image lacks obvious features to distinguish the specific state. In this embodiment, the dataset in step S1 contains images under various scenarios, which can be generated by glass tube chips or PDMS chips under different ambient light and different solutions.

[0039] Next, the dataset was manually labeled, and different video segments were categorized into: droplet state, laminar flow state, internal phase reflux state, external phase reflux state, impurity state, and unknown state. Droplet contours were also labeled for the videos that generated droplets.

[0040] Additionally, in step S1, a portion of the droplets is extracted from the image where droplets are normally generated, and then the extracted droplets are transferred to the image where no droplets are generated. That is, by extracting a portion of the droplets from the image where droplets are normally generated and then transferring the extracted droplets to the image where no droplets are generated, the algorithm's ability to learn about the background of the ungenerated droplets is enhanced, thereby reducing recognition errors.

[0041] The scene recognition model is trained by video clips to generate a scene recognition model that can recognize scenes; the droplet contour recognition model is trained by labeled droplet images to generate a droplet contour recognition model that can accurately recognize droplet contours.

[0042] Real-time control of droplet generation is defined in steps S4-S6. In step S4, the initial state is that the pressure values ​​of the internal and external phases are the same, that is, the initial state of the system is set to the same pressure value for the internal and external phases. If there are no impurities interfering in the scene, the microfluidic system will be in a laminar flow state at this time.

[0043] In step S5, a recognition time threshold is set. If the scene recognition model fails to recognize the scene state within the detection time exceeding the threshold, it is considered that the current solution or chip has been contaminated and droplets cannot be generated. Furthermore, when the scene is in a laminar flow state, the control step size is increased to increase the external phase pressure value; when the scene is in an external phase reflux state, the control step size is decreased to reduce the internal phase pressure.

[0044] Specifically, in step S5, the trained scene recognition model is run to determine the current scene state. If the scene is in an impurity state, monitoring continues for a period of time. If the monitoring time exceeds a threshold, the solution or chip is considered contaminated and droplet generation is impossible. After ruling out an impurity state, the scene should now be in a laminar flow state. In this state, the control step size should be larger. The control step size is updated, the external phase pressure is increased, and further adjustments are made based on the scene state. When the scene is in an internal phase reflux state, the control step size should be smaller. The control step size is updated, the external phase pressure is reduced, and further adjustments are made based on the scene state. When the scene is in an external phase reflux state, the control step size should be smaller. The control step size is updated, the internal phase pressure is reduced, and further adjustments are made based on the scene state. If the scene is in an unknown state, monitoring continues for a period of time. If the monitoring time exceeds a threshold, a problem is considered in the current system environment, possibly a pressure pump issue or insufficient solution, preventing droplet generation. If the scene is in a droplet state, the current stage of control ends, and droplet generation is successful. The control step size is the increment of the control pressure each time. For laminar flow conditions, the control step size should be set to a relatively large value. Specifically, the initial step size can be set to 0.5 psi, and the step size increases by 0.1 psi each time a laminar flow scenario is identified, with an upper limit of 1 psi. The external phase pressure value = current pressure value + step size, and the step size is determined by the above rules.

[0045] In step S6, after successful droplet formation, the droplet profile is detected using a droplet profile recognition model, and the deviation between the current droplet size and the desired droplet size is calculated. The pressure pump parameters are adjusted until the deviation is less than a set threshold, at which point the adjustment stops. The adjustment step size is the increment of pressure for each adjustment. For internal and external phase reflux states, the adjustment step size should be set to a small value for fine-tuning. Specifically, the initial step size can be set to 0.2 psi, increasing by 0.02 psi each time reflux is detected, with an upper limit of 0.3 psi.

[0046] Example 2

[0047] The difference between this embodiment and Embodiment 1 is that, based on Embodiment 1, this embodiment provides further explanation of the scene recognition model.

[0048] In this embodiment, the scene recognition model is a 3D convolutional network model, and it is a time-series-based scene recognition model. It should be noted that other time-series recognition models are also possible.

[0049] Other aspects of this embodiment are the same as those of Embodiment 1, and will not be repeated here.

[0050] Example 3

[0051] The difference between this embodiment and Embodiment 1 is that, based on Embodiment 1, this embodiment provides further explanation of the droplet contour recognition model.

[0052] In this embodiment, the droplet contour recognition model is a segmentation network model. Specifically, the droplet contour recognition model is a Mask RCNN network model.

[0053] Other aspects of this embodiment are the same as those of Embodiment 2, and will not be repeated here.

[0054] In summary, this invention provides a fully automated microdroplet generation method. First, a dataset is acquired, including videos and images of droplet generation processes under various working environments. Then, a scene recognition model is established, trained using the videos of droplet generation processes from the acquired dataset. Next, a droplet contour recognition model is established, trained using the droplet images from the acquired dataset. In practical use, an initial state is set. The trained scene recognition model identifies the current scene state, and the external phase pressure is adjusted according to the current scene state until droplet generation occurs. The droplet contour is detected using the droplet contour recognition model trained in step S3, and the deviation between the current droplet size and the desired droplet size is calculated. The pressure pump parameters are adjusted according to this deviation until the deviation is less than a set threshold, at which point the adjustment stops, completing droplet generation.

[0055] This invention utilizes a scene recognition model to identify scene states and adjusts the process based on the identified scenes. This allows for control over the entire process from no droplets to droplet generation, demonstrating strong adaptability. Furthermore, after droplet generation, adjustments are made based on the droplet's contour until a suitable droplet is formed. The entire process from no droplets to droplet formation is fully monitored and controlled in real-time based on the scene and liquid contour, resulting in a high degree of automation. This eliminates the need for manual observation and control by operators, freeing them up. Compared to other methods, this droplet generation method not only assesses the effectiveness of the entire system environment and evaluates solution and chip design issues, but also achieves fully automated microdroplet generation from scratch, which is of significant importance for scientific research and product commercialization.

[0056] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and substitutions can be made without departing from the technical principles of the present invention, and these improvements and substitutions should also be considered within the scope of protection of the present invention.

Claims

1. A fully automated method for generating microdroplets, characterized in that, Includes the following steps: S1. Obtain the dataset, which includes videos and images of droplet generation processes under various working environments; S2. Establish a scene recognition model. Use the video of the droplet generation process in the dataset of step S1 to train the scene recognition model and obtain a trained scene recognition model. S3. Establish a droplet contour recognition model. Use the droplet images in the dataset from step S1 to train the droplet contour recognition model and obtain the trained droplet contour recognition model. S4. Set the initial state; S5. Use the scene recognition model trained in step S2 to identify the current scene state, and adjust the external phase pressure value according to the current scene state until droplets are generated. S6. Use the droplet contour recognition model trained in step S3 to detect the droplet contour, calculate the deviation between the current droplet size and the desired droplet size, and adjust the pressure pump parameters according to the deviation until the deviation is less than the set threshold, then stop adjusting and complete the droplet generation.

2. The fully automated microdroplet generation method according to claim 1, characterized in that, In step S1, the videos in the dataset are videos of normal droplet generation state, laminar flow state, internal phase reflux state, external phase reflux state, impurity state, and unknown state. Among them, the video segments of internal phase reflux and external phase reflux state need to include the entire process from the start of reflux to the completion of reflux.

3. The fully automated microdroplet generation method according to claim 1, characterized in that, In step S1, a portion of the droplets are extracted from the image where droplets are normally generated, and the extracted droplets are then transferred to the image where no droplets are generated.

4. The fully automated microdroplet generation method according to claim 1, characterized in that, In step S4, the initial state is that the pressure values ​​of the internal and external phases are the same.

5. The fully automated microdroplet generation method according to claim 1, characterized in that, In step S5, a recognition time threshold is set. If the scene recognition model fails to recognize the scene state after the detection time exceeds the threshold, it is considered that the current solution or chip has been contaminated and cannot generate droplets.

6. The fully automated microdroplet generation method according to claim 1, characterized in that, In step S5, when the scene is in laminar flow, the control step size is increased to increase the external phase pressure value; when the scene is in external phase reflux, the control step size is decreased to reduce the internal phase pressure.

7. The fully automated microdroplet generation method according to claim 1, characterized in that, The video of the droplet generation process is 10–20 seconds long.

8. The fully automated microdroplet generation method according to claim 1, characterized in that, The scene recognition model is a 3D convolutional network model.

9. The fully automated microdroplet generation method according to claim 1, characterized in that, The droplet contour recognition model is a segmentation network model.

10. The fully automated microdroplet generation method according to claim 9, characterized in that, The droplet contour recognition model is a Mask RCNN network model.

Images