Calibration method and system of mass flow controller

Abstract

The invention provides a method and system for calibrating a mass flow controller, and the method comprises the steps: obtaining a first characteristic parameter of a to-be-calibrated mass flow controller, the first characteristic parameter comprising at least one of a valve starting voltage, a maximum valve voltage, a full scale, and a plurality of instantaneous flows; wherein the plurality of instantaneous flows are obtained when flow control is carried out on the mass flow controller to be calibrated through the initial flow control parameters; on the basis of the first characteristic parameter, predicting a flow control parameter of the to-be-calibrated mass flow controller through a preset machine learning algorithm; and performing flow control on the to-be-calibrated mass flow controller by using the flow control parameter, acquiring a second characteristic parameter of the to-be-calibrated mass flow controller, and writing the flow control parameter into the to-be-calibrated mass flow controller under the condition that the second characteristic parameter meets a preset standard. The flow control parameters of the quality flow controller are predicted through the machine learning algorithm, and the calibration efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of semiconductor process equipment, and specifically, to a calibration method for a mass flow controller and a calibration system for a mass flow controller. Background Art

[0002] A mass flow controller (MFC) can be used to quickly and stably control the magnitude of gas flow, and is widely used in industries such as semiconductors, photovoltaics, fuel cells, and solar energy that require precise control of gas flow. The PID algorithm is a closed-loop control algorithm and is one of the most widely used algorithms in industry. It can automatically correct the control system accurately and quickly. Currently, the control algorithms of most mass flow controllers are based on the PID algorithm. Therefore, the method of calibrating flow control parameters directly affects the control accuracy and production efficiency of the mass flow controller. It is extremely important to select a method with fast calibration speed and high accuracy.

[0003] Currently, the flow control parameter calibration of mass flow controllers mainly adopts the manual calibration method. According to past experience, the operator first gives an initial parameter, and then adjusts the parameter according to the morphological characteristics of the actual flow curve to achieve a flow curve that meets the product performance. This method relies on manual experience, and different operators may have different adjustment methods, which may lead to poor consistency of the mass flow controller, and the calibration process takes a long time, affecting production efficiency.

[0004] Therefore, how to provide a calibration method that can improve the calibration efficiency and calibration accuracy of the mass flow controller has become an urgent technical problem in this field. Summary of the Invention

[0005] The present invention aims to provide a calibration method for a mass flow controller and a calibration system for a mass flow controller, which can improve the calibration efficiency and calibration accuracy of the mass flow controller.

[0006] To achieve the above object, as one aspect of the present invention, there is provided a calibration method for a mass flow controller, the calibration method for the mass flow controller comprising:

[0007] Obtaining a first characteristic parameter of the mass flow controller to be calibrated, the first characteristic parameter including at least one of a valve start voltage, a maximum valve voltage, a full scale, and a plurality of instantaneous flows, wherein the plurality of instantaneous flows are obtained when controlling the flow of the mass flow controller to be calibrated by an initial flow control parameter;

[0008] Based on the first characteristic parameter, predicting the flow control parameter of the mass flow controller to be calibrated by a preset machine learning algorithm;

[0009] Use the flow control parameters to perform flow control on the mass flow controller to be calibrated, and obtain the second characteristic parameters of the mass flow controller to be calibrated. When the second characteristic parameters meet the preset criteria, write the flow control parameters into the mass flow controller to be calibrated.

[0010] Optionally, the initial flow control parameters include a proportionality coefficient, and the proportionality coefficient is less than or equal to 0.1.

[0011] Optionally, it further includes:

[0012] When the second characteristic parameters do not meet the preset criteria, adjust the flow control parameters until the second characteristic parameters meet the preset criteria.

[0013] Optionally, the second characteristic parameters include a response time and an overshoot ratio, and the preset criteria include that the response time is less than or equal to a preset time threshold and the overshoot ratio is less than a preset ratio.

[0014] Optionally, the flow control parameters include at least one of the PID control parameters of multiple flow segment set points and a filtering coefficient.

[0015] Optionally, the PID control parameters of multiple flow segment set points include a proportionality coefficient, an integral coefficient, and a derivative coefficient with a set point of 10% F.S. segment point, and proportionality coefficients and integral coefficients with set points of 20% F.S. segment point, 50% F.S. segment point, and 100% F.S. segment point respectively.

[0016] Optionally, it further includes:

[0017] Obtain the flow control parameters and the first characteristic parameters of a predetermined number of calibrated mass flow controllers;

[0018] Construct a training model, and use the first characteristic parameters as inputs and the flow control parameters as outputs to iteratively train the training model until the loss function of the training model converges to a preset value to obtain the preset machine learning algorithm.

[0019] Optionally, the obtaining the flow control parameters and the first characteristic parameters of a predetermined number of calibrated mass flow controllers includes:

[0020] Use the initial flow control parameters to perform flow control on the calibrated mass flow controller to obtain multiple instantaneous flows of the calibrated mass flow controller.

[0021] Optionally, before controlling the flow rate of the calibrated mass flow controller using the initial flow control parameters, initialize the calibrated mass flow controller and write the initial flow control parameters thereto.

[0022] And / or, after obtaining the first characteristic parameter of the calibrated mass flow controller, restart the calibrated mass flow controller to restore the original value of the flow control parameter.

[0023] Optionally, the calibration method further includes:

[0024] Verify the preset machine learning algorithm, and when the output result of the preset machine learning algorithm does not meet the production requirements, modify the parameters of the training model, and go to the step of iteratively training the training model based on the first characteristic parameters of a predetermined number of the mass flow controllers until the preset machine learning algorithm meets the production requirements.

[0025] As a second aspect of the present invention, there is provided a calibration system for a mass flow controller, including: a host computer, a calibration fluid passage, and a plurality of calibrated mass flow controllers. The calibration fluid passage has a mass flow controller installation position. The host computer is configured to sequentially obtain the first characteristic parameters of each of the calibrated mass flow controllers installed on the calibration fluid passage, and execute the calibration method of the mass flow controller described above.

[0026] The calibration method for a mass flow controller provided by the present invention predicts the flow control parameters of the mass flow controller through a machine learning algorithm, realizes automatic calibration of the mass flow controller, reduces the trial-and-error steps of manual operation, and improves the PID calibration efficiency of the mass flow controller. Moreover, the preset machine learning algorithm does not rely on manual experience, reduces the manual learning cost, is easy for operators to get started, and saves manpower. Description of the Drawings

[0027] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the present invention, but do not constitute a limitation to the present invention. In the drawings:

[0028] Figure 1 is a flowchart of the calibration method for a mass flow controller provided by an embodiment of the present invention;

[0029] Figure 2 is a partial flowchart of the calibration method for a mass flow controller provided by an embodiment of the present invention;

[0030] Figure 3 is a partial flowchart of the calibration method for a mass flow controller provided by an embodiment of the present invention;

[0031] Figure 4 It is a partial process schematic diagram of the calibration method for the mass flow controller provided by the embodiments of the present invention;

[0032] Figure 5 It is a schematic diagram of the training result without adding a penalty coefficient;

[0033] Figure 6 It is a schematic diagram of the training result with an added penalty coefficient;

[0034] Figure 7 It is a schematic diagram of the instantaneous flow rate data corresponding to different mass flow controllers. Detailed implementation manners

[0035] The following will describe the detailed implementation manners of the present invention with reference to the accompanying drawings. It should be understood that the detailed implementation manners described herein are only for explaining and illustrating the present invention, and are not used to limit the present invention.

[0036] To solve the above technical problems, as one aspect of the present invention, a calibration method for a mass flow controller is provided. As Figure 1 shown, the calibration method for the mass flow controller includes:

[0037] Step S10: Obtain the first characteristic parameters of the mass flow controller to be calibrated. The first characteristic parameters include at least one of the valve startup voltage, the maximum valve voltage, the full scale, and multiple instantaneous flow rates, where the multiple instantaneous flow rates are obtained when controlling the flow rate of the mass flow controller to be calibrated through the initial flow control parameters;

[0038] Step S20: Based on the first characteristic parameters, predict the flow control parameters of the mass flow controller to be calibrated through a preset machine learning algorithm;

[0039] Step S30: Use the flow control parameters to control the flow rate of the mass flow controller to be calibrated, and obtain the second characteristic parameters of the mass flow controller to be calibrated. When the second characteristic parameters meet the preset standards, write the flow control parameters into the mass flow controller to be calibrated.

[0040] The calibration method for the mass flow controller provided by the present invention predicts the flow control parameters of the mass flow controller through a machine learning algorithm, realizes the automatic calibration of the mass flow controller, reduces the trial-and-error steps of manual operation, and improves the PID calibration efficiency of the mass flow controller. And the preset machine learning algorithm does not rely on manual experience, reduces the manual learning cost, is easy for operators to get started, and saves manpower.

[0041] As a preferred implementation manner of the present invention, the calibration method further includes:

[0042] When the second characteristic parameter does not meet the preset standard, the flow control parameter is adjusted until the second characteristic parameter meets the preset standard. That is, when the mass flow controller fails the test verification, the flow control parameter is further fine-tuned until the mass flow controller passes the test verification. For example, it can be fine-tuned through the calibration software.

[0043] The calibration method of the mass flow controller provided by the embodiment of the present invention can also, after determining the coefficient of the mass flow controller to be calibrated by using the preset machine learning algorithm, perform test verification on the mass flow controller and automatically fine-tune its coefficient until the mass flow controller to be calibrated meets the preset standard, further ensuring the calibration accuracy of the mass flow controller.

[0044] As an optional implementation manner of the present invention, the second characteristic parameter includes the response time and the overshoot ratio, and the preset standard includes that the response time is less than or equal to the preset time threshold and the overshoot ratio is less than the preset ratio.

[0045] As a preferred implementation manner of the present invention, the initial flow control parameter includes the proportional coefficient Kp, and the proportional coefficient Kp is less than or equal to 0.1 to ensure the smoothness of the instantaneous flow curve and facilitate obtaining more eigenvalue in a shorter time.

[0046] Optionally, the proportional coefficient Kp in the initial flow control parameter is 0.01.

[0047] As an optional implementation manner of the present invention, the preset time threshold is 1 second, and the preset ratio is 0.3% F.S.

[0048] As an optional implementation manner of the present invention, the first characteristic parameter includes 300 instantaneous flows during the process of the opening of the mass flow controller changing from 0% F.S. to the preset opening, or as an optional implementation manner of the present invention, the first characteristic parameter includes a plurality of instantaneous flows collected within the preset duration, and the preset duration can enable most of the mass flow controllers to reach the full scale. For example, the preset duration is 15 s.

[0049] In some implementation manners of the present invention, the preset opening can be the full scale of the mass flow controller, that is, 100% F.S., or the preset opening can also be any value between 0% F.S. and 100% F.S., as long as the curve of the collected instantaneous flow can be used to distinguish the characteristics of different mass flow controllers.

[0050] As an optional implementation manner of the present invention, the calibration method of the mass flow controller further includes the step of training to obtain the preset machine learning algorithm. Specifically, as Figure 2 shown, the calibration method of the mass flow controller further includes:

[0051] Step S1: Obtain the flow control parameters and the first characteristic parameters of a predetermined number of calibrated mass flow controllers;

[0052] Step S2: Construct a training model, and use the first characteristic parameters as the input and the flow control parameters as the output to iteratively train the training model until the loss function of the training model converges to a preset value, obtaining a preset machine learning algorithm.

[0053] As an alternative implementation manner of the present invention, as Figure 2 shown, the calibration method of the mass flow controller further includes:

[0054] Step S3: Verify the preset machine learning algorithm, and when the output result of the preset machine learning algorithm does not meet the production requirements, modify the parameters of the training model, and go back to execute the step of iteratively training the training model based on the first characteristic parameters of a predetermined number of mass flow controllers until the preset machine learning algorithm meets the production requirements.

[0055] As an alternative implementation manner of the present invention, as Figure 3 shown, verifying the preset machine learning algorithm in Step S3 specifically includes:

[0056] Step S31: Obtain the first characteristic parameters of the mass flow controllers that did not participate in the model training;

[0057] Step S32: Determine the flow control parameters corresponding to the mass flow controllers through the preset machine learning algorithm;

[0058] Step S33: Use the flow control parameters to control the flow of the mass flow controller, and obtain the second characteristic parameters of the mass flow controller. When the second characteristic parameters meet the preset standard, it is determined that the preset machine learning algorithm meets the production requirements; otherwise, it is determined that the preset machine learning algorithm does not meet the production requirements.

[0059] In the embodiment of the present invention, the step of verifying the preset machine learning algorithm is to directly apply the currently trained preset machine learning algorithm to the calibration scenario, calibrate a new mass flow controller that did not participate in the previous data collection step, and verify whether the mass flow controller written with the flow control parameters predicted by the algorithm can meet the preset standard corresponding to the production requirements, so as to verify whether the currently obtained preset machine learning algorithm meets the production requirements.

[0060] As an alternative implementation manner of the present invention, as Figure 2 shown, the calibration method of the mass flow controller further includes:

[0061] In an embodiment of the present invention, first obtain the flow control parameters and the first characteristic parameters of the calibrated mass flow controller, so that the first characteristic parameters are used as the input values of the training model, and the flow control parameters are used as the output values of the training model, and a preset machine learning algorithm capable of obtaining the flow control parameters of the mass flow controller from the first characteristic parameters of the mass flow controller is trained. The preset machine learning algorithm in the calibration method provided by the embodiment of the present invention is trained by a large amount of calibration data of the mass flow controller (that is, the flow control parameters calibrated manually), and has high accuracy. The consistency of calibrating the mass flow controller using the preset machine learning algorithm trained by the method of the present invention can achieve better results and higher calibration efficiency than manual calibration.

[0062] As an alternative embodiment of the present invention, the flow control parameters include at least one of the PID control parameters of multiple flow segmentation set points and the filtering coefficient.

[0063] As an alternative embodiment of the present invention, the PID control parameters of multiple flow segmentation set points include the proportional coefficient (Kp), integral coefficient (Ki), and differential coefficient (Kd) at the set point of the 10% F.S. segmentation point, and the proportional coefficient (Kp) and integral coefficient (Ki) at the set points of the 20% F.S. segmentation point, 50% F.S. segmentation point, and 100% F.S. segmentation point.

[0064] As an alternative embodiment of the present invention, step S1 of obtaining the flow control parameters and the first characteristic parameters of a predetermined number of calibrated mass flow controllers specifically includes:

[0065] Use the initial flow control parameters to control the flow of the calibrated mass flow controller to obtain multiple instantaneous flows of the calibrated mass flow controller.

[0066] As an alternative embodiment of the present invention, before using the initial flow control parameters to control the flow of the calibrated mass flow controller, initialize the calibrated mass flow controller and write the initial flow control parameters;

[0067] And / or, after obtaining the first characteristic parameters of the calibrated mass flow controller, restart the calibrated mass flow controller to restore the original value of the flow control parameters.

[0068] That is, in the embodiment of the present invention, first write the initial flow control parameters to different calibrated mass flow controllers, so that they operate under the same conditions as the mass flow controller to be calibrated, and collect their first characteristic parameters; after the collection is completed, the original calibrated flow control parameters can also be restored so that these mass flow controllers can continue to be used.

[0069] As an alternative embodiment of the present invention, step S1 of obtaining the first characteristic parameters of a predetermined number of calibrated mass flow controllers specifically includes:

[0070] Step S11: Set up the gas path;

[0071] Step S12: Install the calibrated mass flow controller and run the data collection program to obtain the flow control parameters and the first characteristic parameters of the calibrated mass flow controller;

[0072] Step S13: Repeat step S12 until the data collection programs for a predetermined number of mass flow controllers are completed.

[0073] As an alternative embodiment of the present invention, the inlet pressure of the mass flow controller in step S11 is 0.2 Mpa.

[0074] To ensure the adaptability of the preset machine learning algorithm to mass flow controllers with different ranges, as a preferred embodiment of the present invention, at least 5 mass flow controllers need to be selected for each range in step S1, and the ranges of the mass flow controllers can be selected from 1 sccm to 50000 sccm.

[0075] Optionally, the predetermined number can be 600, that is, the data of 600 mass flow controllers need to be collected.

[0076] As an alternative embodiment of the present invention, the data collection program run in step S12 includes the following steps:

[0077] Step S121: Obtain and back up the mass flow controller data, that is, back up and save the flow control parameters of the calibrated mass flow controller;

[0078] Step S122: Write the initial flow control parameters into the mass flow controller;

[0079] Step S123: Adjust the set point of the mass flow controller from 0% F.S. (i.e., the valve closed state) to a preset opening (for example, 100% F.S.), and collect the instantaneous flow rate, and save the collected flow rate data (for example, it can be 300 instantaneous flow rates);

[0080] Step S124: Set the set point of the mass flow controller to 0% F.S. (i.e., close the control valve of the mass flow controller), and restart the mass flow controller to restore the PID data (flow control parameters) of the mass flow controller to the original value (the initialization parameters written into the mass flow controller before data collection are not written into the EEPROM (saved) of the mass flow controller, and the initialization parameters will be lost after restart, and the mass flow controller still retains the originally calibrated parameters).

[0081] As an alternative embodiment of the present invention, in step S12, the data collection program of the host computer saves and backs up the calibration data of the calibrated mass flow controller, then uniformly writes a set of initial flow control parameters, and then adjusts the opening of the mass flow controller from 0% F.S. (i.e., the valve closed state) to a preset opening (for example, 100% F.S.). At the same time, the instantaneous flow rate is collected. Finally, the valve is closed, and the host computer obtains and saves the valve startup voltage, maximum valve voltage, MFC full scale, 10 flow control parameters (i.e., the proportional coefficient (Kp), integral coefficient (Ki), and derivative coefficient (Kd) of the set point at the 10% F.S. segmentation point, a total of 3 coefficients, the proportional coefficient (Kp) and integral coefficient (Ki) of the set point at the 20% F.S. segmentation point, 50% F.S. segmentation point, and 100% F.S. segmentation point, a total of 6 coefficients, and 1 filtering coefficient), and the instantaneous flow rate (300).

[0082] After the data collection program collects the data, a prompt window will pop up and an instantaneous flow rate curve will be generated (the purpose of generating the instantaneous curve is to observe whether the instantaneous curve of this MFC is smooth. If there are mutation points, there may be valve assembly problems or communication interruption problems. If these problems are introduced into the learned data, it will affect the accuracy of the model).

[0083] As Figure 7 shown in the figure is a schematic diagram of the instantaneous flow rate data corresponding to different mass flow controllers. It can be seen that under the same flow control parameters, the instantaneous flow rates of different mass flow controllers are different. Even for mass flow controllers with the same range, due to the different valve characteristics of each mass flow controller (reasons such as assembly and mechanical part size errors), there are differences in the obtained instantaneous flow rates.

[0084] As an alternative embodiment of the present invention, as Figure 4 shown, in step S2, a training model is constructed, and the training model is iteratively trained based on the flow control parameters and the first characteristic parameters, including:

[0085] Step S21: Construct a training model of a preset machine learning algorithm using a neural network algorithm;

[0086] Step S24: Based on the first characteristic parameters, use the pytorch machine learning library to iteratively train the training model.

[0087] In the embodiments of the present invention, a machine learning model is constructed using a neural network algorithm. The neural network algorithm has advantages in processing complex data. Moreover, predicting traffic control parameters is not a classification problem (such as the cat-dog battle). In addition, the predicted traffic control parameters are interrelated. Therefore, the traffic control parameters should be taken as a whole as the output, and the loss function should also be obtained as a whole when calculating the loss function. Considering the above reasons, a traditional machine learning algorithm can only predict one coefficient when training a model, which cannot meet the requirements of predicting traffic control parameters. Therefore, the present invention uses a neural network algorithm to construct a training model for a preset machine learning algorithm, which can improve the training efficiency of the preset machine learning algorithm.

[0088] As an alternative embodiment of the present invention, the training model is a linear neural network, which includes an input layer, three hidden layers, and an output layer.

[0089] As an alternative embodiment of the present invention, the linear neural network is a 303*512*1024*512*10 linear neural network, that is, it includes an input layer (303 data), three hidden layers (512*1024*512), and an output layer (10 traffic control parameters, that is, the proportional coefficient (Kp), integral coefficient (Ki), and derivative coefficient (Kd) corresponding to the 10% F.S. ratio coefficient, a total of 3 coefficients, the proportional coefficient (Kp) and integral coefficient (Ki) corresponding to three segmentation points (i.e., 20% F.S., 50% F.S., 100% F.S.), a total of 6 coefficients, and 1 filter coefficient).

[0090] As a preferred embodiment of the present invention, as Figure 4 shown, constructing a training model for a preset machine learning algorithm further includes:

[0091] Step S22: Set the learning rate of the training model to be greater than 0 and less than or equal to 0.000001, and use the Adam optimizer as the optimizer for the training model. Through experimental verification, using the Adam optimizer as the optimizer for the training model has better effects than the SGD optimizer and the AdamW optimizer.

[0092] As a preferred embodiment of the present invention, the learning rate of the training model is set to 0.000001.

[0093] To improve the prediction effect of the preset machine learning algorithm, as Figure 4 shown, as a preferred embodiment of the present invention, constructing a training model for a preset machine learning algorithm further includes:

[0094] Step S23: Use the mean squared error loss (MSELoss) as the loss function for training the model, use RELU as the activation function for training the model, and set the penalty coefficient (Dropout) P of the training model to be greater than or equal to 0.1 and less than or equal to 0.3. Preferably, the penalty coefficient p of the training model is 0.2;

[0095] The calibration method of the mass flow controller further includes:

[0096] When the loss value of the test set of the training model rises, end the iterative training in advance.

[0097] The model obtained by machine training usually has a good prediction effect on the data participating in the training. However, after inputting the data that has not participated in the training (i.e., the input validation set), the prediction effect is often very poor. In the embodiments of the present invention, the penalty coefficient is increased, so that the verification effect is closer to the actual test effect.

[0098] Specifically, after setting all the above parameters (learning rate, optimizer, loss function, activation function, penalty coefficient) and preparing the input and output data, the training program can be run for training. During training, the data is divided into a training set and a validation set (500 groups of data in the training set and 100 groups of data in the validation set). Each time an iterative training is performed, the system will calculate the loss values of the training set and the validation set (i.e., the error between the predicted value and the true value) (the TensorBoard tool can be used to view the current training situation in real time during the training process). When the loss value of the test set rises, the training can be ended in advance to reduce the occurrence of overfitting.

[0099] Figure 5 、 Figure 6 Figures respectively show the images representing the training results of two trainings drawn by the TensorBoard tool. Figure 5 is the training result when the penalty coefficient is not increased and the learning rate is relatively large. In the figure, the upper curve is the loss of the test set, and the lower curve is the loss of the training set. It can be seen that the loss of the test set has a significant increase after 10K iterations, but the loss of the training set continues to decrease. This situation indicates the existence of overfitting at the beginning. Figure 6 is the training result when the penalty coefficient is increased and the learning rate is 0.000001. It can be seen that the loss of the test set gradually converges and tends to be stable, and no overfitting occurs.

[0100] As an alternative embodiment of the present invention, as Figure 4 shown, the calibration method of the mass flow controller further includes:

[0101] End the iterative training when the number of iterative training times is greater than or equal to the preset maximum number of iterative training times.

[0102] As an alternative embodiment of the present invention, as Figure 4 shown, the calibration method of the mass flow controller further includes:

[0103] After the iterative training is completed, it is judged whether the loss converges. If not, the number of iterations is increased and the iterative training is carried out again.

[0104] As an alternative embodiment of the present invention, as Figure 4 shown, step S2 further includes:

[0105] Step S201: Normalize the first characteristic parameter.

[0106] In the embodiment of the present invention, the input data for model training (i.e., the first characteristic parameter) includes information of the mass flow controller (valve start voltage, maximum valve voltage, range) and instantaneous flow rate (taking the first 300 data), a total of 303 data (eigenvalues). Since the orders of magnitude of these two types of data are quite different, the data needs to be normalized before use.

[0107] As a second aspect of the present invention, there is provided a calibration system for a mass flow controller, including: a host computer and a calibration fluid passage. The calibration fluid passage has a mounting position for the mass flow controller. The host computer is used to obtain the first characteristic parameter of the mass flow controller installed on the calibration fluid passage and execute the calibration method of the mass flow controller provided by the embodiment of the present invention.

[0108] In the calibration system for the mass flow controller provided by the present invention, the host computer predicts the flow control parameter of the mass flow controller through a machine learning algorithm, realizes automatic calibration of the mass flow controller, reduces the trial-and-error steps of manual operation, and improves the PID calibration efficiency of the mass flow controller. And the preset machine learning algorithm does not rely on manual experience, reduces the manual learning cost, is easy for operators to get started, and saves manpower.

[0109] It can be understood that the above embodiments are merely exemplary embodiments adopted to illustrate the principle of the present invention. However, the present invention is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also regarded as the protection scope of the present invention.

Claims

1. A calibration method for a mass flow controller, characterized in that, the calibration method includes: obtaining a first characteristic parameter of the mass flow controller to be calibrated, where the first characteristic parameter includes at least one of a valve start voltage, a maximum valve voltage, a full scale, and a plurality of instantaneous flows, and the plurality of instantaneous flows are obtained when the mass flow controller to be calibrated is controlled for flow by an initial flow control parameter; predicting a flow control parameter of the mass flow controller to be calibrated based on the first characteristic parameter through a preset machine learning algorithm; using the flow control parameter to control the flow of the mass flow controller to be calibrated, and obtaining a second characteristic parameter of the mass flow controller to be calibrated. When the second characteristic parameter meets a preset standard, writing the flow control parameter into the mass flow controller to be calibrated.

2. The calibration method for a mass flow controller according to claim 1, characterized in that, the initial flow control parameter includes a proportionality coefficient, and the proportionality coefficient is less than or equal to 0.

1.

3. The calibration method for a mass flow controller according to claim 1, characterized in that, further includes: when the second characteristic parameter does not meet the preset standard, adjusting the flow control parameter until the second characteristic parameter meets the preset standard.

4. The calibration method for a mass flow controller according to claim 1, characterized in that, the second characteristic parameter includes a response time and an overshoot ratio, and the preset standard includes that the response time is less than or equal to a preset time threshold and the overshoot ratio is less than a preset ratio.

5. The calibration method for a mass flow controller according to claim 1, characterized in that, the flow control parameter includes at least one of PID control parameters of a plurality of flow segment set points and a filtering coefficient.

6. The calibration method for a mass flow controller according to claim 5, characterized in that, the PID control parameters of a plurality of flow segment set points include a proportionality coefficient, an integral coefficient, and a differential coefficient at a set point of 10% F.S. segment point, and proportionality coefficients and integral coefficients at set points of 20% F.S. segment point, 50% F.S. segment point, and 100% F.S. segment point respectively.

7. The calibration method for a mass flow controller according to any one of claims 1 to 6, characterized in that, further includes: obtaining flow control parameters and first characteristic parameters of a predetermined number of calibrated mass flow controllers; constructing a training model, and iteratively training the training model with the first characteristic parameter as an input and the flow control parameter as an output until a loss function of the training model converges to a preset value to obtain the preset machine learning algorithm.

8. The calibration method for a mass flow controller according to claim 7, characterized in that, the obtaining flow control parameters and the first characteristic parameters of a predetermined number of calibrated mass flow controllers includes: Use the initial flow control parameters to perform flow control on the calibrated mass flow controller to obtain multiple instantaneous flows of the calibrated mass flow controller.

9. The calibration method of the mass flow controller according to claim 8, wherein, before using the initial flow control parameters to perform flow control on the calibrated mass flow controller, initialize the calibrated mass flow controller and write the initial flow control parameters; and / or, after obtaining the first characteristic parameter of the calibrated mass flow controller, restart the calibrated mass flow controller to restore the original value of the flow control parameter.

10. The calibration method of the mass flow controller according to claim 7, wherein, the calibration method further includes: verifying the preset machine learning algorithm, and when the output result of the preset machine learning algorithm does not meet the production requirements, modifying the parameters of the training model and going to the step of performing iterative training on the training model based on the first characteristic parameters of a predetermined number of the mass flow controllers until the preset machine learning algorithm meets the production requirements.

11. A calibration system for a mass flow controller, wherein, it includes: a host computer and a calibration fluid passage, the calibration fluid passage has a mass flow controller installation position, and the host computer is used to obtain the first characteristic parameter of the mass flow controller installed on the calibration fluid passage and execute the calibration method of the mass flow controller according to any one of claims 1 to 10.

Images