Flow rate estimation for mass flow controllers

Abstract

To accurately determine a flow rate of fluid flowing through a mass flow controller.SOLUTION: A mass flow controller includes: a block body having a flow path; a valve provided at least in part within the flow path; a valve position sensor; a first temperature sensor configured to detect the temperature of a fluid flowing through the flow path; a first pressure sensor located on an upstream side of the flow path from the valve; and a second pressure sensor located on a downstream side of the flow path from the valve. The mass flow controller is configured to receive valve position data, first temperature data, first pressure data, and second pressure data and estimate a flow rate of fluid in the flow path through the application of an already trained machine learning model.SELECTED DRAWING: Figure 3

Description

[Background technology]

[0001] Mass flow controllers are frequently used in applications such as semiconductor device manufacturing where it is desirable to determine with high accuracy the flow rate of a fluid through the mass flow controller, and the flow rate is typically calculated based on measured properties of the fluid and the valve, rather than being measured directly. However, the relationship between such measured properties and the flow rate can be complex. For example, there may be regions of parameter space where a transition between laminar and turbulent flow conditions occurs. Thus, the flow rate of a fluid through a mass flow controller can be difficult to accurately estimate using conventional methods. Summary of the Invention

[0002] According to an aspect of the present disclosure, there is provided a mass flow controller including a block body having a flow path. The mass flow controller may further include a valve provided in at least a portion of the flow path of the block body. The mass flow controller may further include a valve position sensor, a first temperature sensor configured to detect a temperature of a fluid flowing in the flow path, a first pressure sensor disposed upstream of the valve in the flow path, and a second pressure sensor disposed downstream of the valve in the flow path. The mass flow controller may further include a processor configured to receive valve position data from the valve position sensor and receive first temperature data from the first temperature sensor. The processor may be further configured to receive first pressure data of the fluid in the flow path from the first pressure sensor and second pressure data of the fluid in the flow path from a second pressure sensor. The processor may be further configured to estimate a flow rate of the fluid in the flow path by at least partially applying a trained machine learning model to the valve position data, the first temperature data, the first pressure data, and the second pressure data.

[0003] This Summary is intended to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used to limit the scope of the claimed subject matter. Moreover, the claimed subject matter is not limited to embodiments that solve any or all of the disadvantages noted in any part of the present disclosure. [Brief description of the drawings]

[0004] [Figure 1] FIG. 1 illustrates a conventional mass flow controller having a valve and a restrictor. [Diagram 2] FIG. 1 illustrates an example of a mass flow controller having a valve and a restrictor in accordance with an embodiment of the present disclosure. [Diagram 3] FIG. 3 is a schematic diagram illustrating a processor of a mass flow controller when a trained machine learning model is executed according to the embodiment of FIG. [Figure 4] FIG. 3 is a schematic diagram illustrating an example of a trained machine learning model according to the embodiment of FIG. 2. [Diagram 5] FIG. 3 is a schematic diagram illustrating the training process of a trained machine learning model according to the embodiment of FIG. 2. [Figure 6] FIG. 3 is a diagram showing an example of an arrangement during calibration of the mass flow controller according to the embodiment of FIG. 2. [Figure 7A] 3 is a flow chart illustrating an example of a method that may be implemented in the mass flow controller of FIG. 2. [Figure 7B] 7B illustrates additional steps that may be performed in the method of FIG. 7A in some embodiments. [Figure 7C] 7B illustrates additional steps that may be performed in the method of FIG. 7A in some embodiments. [Figure 7D] 7B illustrates additional steps that may be performed in the method of FIG. 7A in some embodiments. [Figure 7E] 7B illustrates additional steps that may be performed in the method of FIG. 7A in some embodiments. [Figure 8] FIG. 3 is a schematic diagram illustrating an example of a computing environment in which the method used in the mass flow controller of FIG. 2 may be implemented. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0005] An example of the configuration of a conventional mass flow controller 110 is shown in FIG. 1. The mass flow controller 110 of FIG. 1 includes a block body 112 that forms a flow path 114. In the example of FIG. 1, a fluid in the mass flow controller 110 flows from left to right along the flow path 114. The conventional mass flow controller 110 further includes a valve 116 configured to be opened and closed to control the flow rate of the fluid through the mass flow controller 110. The valve 116 may be opened and closed by an actuator 134. In addition, the mass flow controller 110 includes a restrictor 118 configured to reduce the flow rate of the fluid. For example, the restrictor 118 may be configured to impose an upper limit on the flow rate. In the example of FIG. 1, the restrictor 118 is downstream of the valve 116.

[0006] 1 further includes a first pressure sensor 120, a second pressure sensor 122, and a third pressure sensor 124. The first pressure sensor 120 is disposed upstream of the valve 116 on the flow path 114. The second pressure sensor 122 is disposed between the valve 116 and the restrictor 118, and the third pressure sensor 124 is disposed downstream of the restrictor 118. The first pressure sensor 120, the second pressure sensor 122, and the third pressure sensor 124 may each transmit their respective pressure measurements to the processor 130.

[0007] The processor 130 is configured to calculate an estimate of the flow rate of the fluid based on the pressure measurements received from the first pressure sensor 120, the second pressure sensor 122, and the third pressure sensor 124. According to one conventional method for determining the flow rate, the processor 130 may input the pressure measurements into a polynomial regression equation having predetermined coefficients.

[0008] However, the conventional configuration of the mass flow controller 110 shown in FIG. 1 has potential drawbacks that may adversely affect performance in some applications, such as those requiring very high precision, such as certain next-generation semiconductor manufacturing processes. The restrictor 118 of the mass flow controller 110 adds volume and dead space to the mass flow controller 110, where fluid may become trapped. Furthermore, the restrictor 118 may reduce the response time of the mass flow controller 110 when changing the flow rate at high speeds, since the fluid must pass through the additional volume of the restrictor 118 and the flow passage 114 that contains the restrictor 118. Also, the introduction of the restrictor 118 into the mass flow controller 110 may increase the manufacturing complexity and cost of the mass flow controller 110, and may introduce additional risks of aging, failure and / or performance degradation during operation of the mass flow controller.

[0009] In place of the mass flow controller 110 of FIG. 1, a mass flow controller 10 according to one embodiment of the present disclosure is shown in FIG. 2. The mass flow controller 10 of FIG. 2 includes a block body 12 having a flow passage 14 passing through the block body 12. In the example of FIG. 2, the flow passage 14 is shown to flow from left to right, similar to FIG. 1. The mass flow controller 10 further includes a valve 16 disposed at least partially within the flow passage 14 of the block body 12. The valve 16 may be opened or closed to control the flow of fluid along the flow passage 14. However, unlike the conventional mass flow controller 110 of FIG. 1, the example mass flow controller 10 of FIG. 2 does not include a restrictor 118. Thus, the above-mentioned problems caused by using a restrictor may be avoided.

[0010] 2 may further include a valve position sensor 20 configured to measure the degree to which the valve 16 is open or closed. The valve position sensor 20 may measure the valve position within a range between a fully closed condition and a fully open condition, for example.

[0011] The mass flow controller 10 may further include a first pressure sensor 26 disposed upstream of the valve 16 in the flow path 14. Additionally, the mass flow controller 10 may further include a second pressure sensor 28 disposed downstream of the valve 16 in the flow path 14. Thus, the first pressure sensor 26 and the second pressure sensor 28 may be configured to measure a pressure differential across the valve 16. In some embodiments, multiple mass flow controllers sharing a common manifold may share a common downstream second pressure sensor 28.

[0012] The mass flow controller 10 may further include a first temperature sensor 22 configured to detect a temperature of the fluid flowing in the flow path 14. As shown in FIG. 2, the first temperature sensor 22 may be disposed upstream of the valve 16 in the flow path 14. In some embodiments, the mass flow controller 10 may further include a second temperature sensor 24. The second temperature sensor 24 may be disposed downstream of the valve 16 in the flow path 14. Thus, the first temperature sensor 22 and the second temperature sensor 24 may be configured to measure a temperature difference occurring across the valve 16. The first temperature sensor 22 and the second temperature sensor 24 may each be disposed at least partially within the block body 12.

[0013] The mass flow controller 10 of FIG. 2 further includes a processor 30. In addition, the mass flow controller 10 may include a memory 32 operably coupled to the processor 30. In some embodiments, the processor 30 and the memory 32 may be physically incorporated into the mass flow controller 10. Alternatively, the processor 30 and / or the memory 32 may be included in a separate physical computing device configured to communicate with the components of the mass flow controller 10 via wired and / or wireless signals. The functions of the processor 30 and the memory 32 may in some embodiments be distributed among multiple communicatively connected computing devices, which may include one or more client computing devices and / or one or more server computing devices.

[0014] Each sensor included in the mass flow controller 10 may be configured to transmit its measurements to a processor 30, as shown in FIG. 3. The processor 30 may be configured to receive valve position data 40 from the valve position sensor 20. Additionally, the processor 30 may be configured to receive first temperature data 42 from the first temperature sensor 22. In embodiments in which the mass flow controller 10 includes a second temperature sensor 24, the processor 30 may be further configured to receive second temperature data 44 from the second temperature sensor 24. Additionally, the processor 30 may be further configured to receive first pressure data 46 from the first pressure sensor 26 and second pressure data 48 from the second pressure sensor 28 for the fluid in the flow path 14.

[0015] The processor 30 may be further configured to estimate the flow rate of the fluid in the flow path 14. Because the mass flow controller 10 does not include the restrictor 118 of the mass flow controller 110 of FIG. 1, the fluid flowing through the mass flow controller 10 of FIG. 2 may have multiple potential flow regimes. Thus, the flow dynamics of the fluid flowing through the mass flow controller 10 of FIG. 2 may be more complex than that of the mass flow controller 110 of FIG. 1. Modeling the flow rate of the fluid flowing through the mass flow controller 10 using only polynomial regression may result in inaccurate flow rate estimates.

[0016] Instead of using only polynomial regression for flow rate estimation, the processor 30 may be configured to estimate the flow rate of the fluid in the flow path 14 by, at least in part, applying a trained machine learning model 50 to the valve position data 40, the first temperature data 42, the first pressure data 46, and the second pressure data 48. In embodiments in which the mass flow controller 10 includes a second temperature sensor 24, the processor 30 may be further configured to estimate the flow rate of the fluid in the flow path 14 by, at least in part, applying the trained machine learning model 50 to the second temperature data 44.

[0017] As shown in the example of FIG. 4, the trained machine learning model 50 may be a deep neural network including an input layer 53, one or more hidden layers 55, and an output layer 57. For example, the trained machine learning model 50 may be a cascade forward neural network. As another example, the trained machine learning model 50 may be a convolutional neural network. Other types of machine learning models may alternatively be used. As shown in FIG. 4, the input layer 53, the output layer 57, and the one or more hidden layers 55 each include one or more neurons 51. Each neuron 51 of the trained machine learning model 50 may have connections with neurons of other layers, and each connection between a pair of neurons of different layers may have an associated weight. Furthermore, each neuron 51 may have a respective bias value. Furthermore, each neuron 51 of the trained machine learning model 50 may have a respective transfer function. The transfer function for the neuron 51 of the trained machine learning model 50 may be, for example, a linear transfer function or a log-sigmoid transfer function. The transfer function determines whether and how much of a signal a neuron will output, given the sum of the input signals to each neuron.

[0018] Returning to FIG. 3 , the trained machine learning model 50 may be a regression model configured to output at least one scalar value. In some embodiments, the scalar value may be an estimated flow rate 52. In other embodiments, the scalar value may be an intermediate value to which the processor 30 can apply further processing to generate the estimated flow rate 52. For example, in the trained machine learning model 50, the processor 30 may be configured to determine normalized valve position data 60, normalized first temperature data 62, normalized first pressure data 66, and normalized second pressure data 68 from the valve position data 40, the first temperature data 42, the first pressure data 46, and the second pressure data 48. In embodiments in which the mass flow controller 10 includes a second temperature sensor 24, the processor 30 may be further configured to determine normalized second temperature data 64. The normalized valve position data 60, the first normalized temperature data 62, the second normalized temperature data 64, the first normalized pressure data 66, and the second normalized pressure data 68 may each be expressed as a dimensionless quantity such that operations performed on the normalized data do not result in equations containing terms with mismatched units. In one example, the values ​​may be normalized to be between 0 and 1.

[0019] In embodiments in which the trained machine learning model 50 is configured to generate normalized data, the processor 30 may be further configured to at least partially apply polynomial regression 58 to the normalized valve position data 60, the first normalized temperature data 62, the first normalized pressure data 66, and the second normalized pressure data 68 to estimate a flow rate of the fluid in the flow path 14. In embodiments in which the processor 30 is further configured to generate normalized second temperature data 64, the processor 30 may be further configured to apply polynomial regression 58 to the second normalized temperature data 64 when estimating the flow rate.

[0020] In some embodiments, memory 32 may store one or more polynomial regression coefficients 72 of polynomial regression 58. In such embodiments, processor 30 may be further configured to estimate a flow rate of the fluid based at least in part on one or more polynomial regression coefficients 72. For example, one or more polynomial regression coefficients 72 may be coefficients of a Taylor series expansion. Additionally or alternatively, if trained machine learning model 50 is configured as a convolutional neural network, memory 32 may store neural network coefficients.

[0021] In some embodiments, the trained machine learning model 50 may be configured to identify a flow regime 74 (such as laminar or turbulent flow) of the fluid in the flow path 14. The flow regime 74 may be selected from among a plurality of predetermined flow regimes 74. In such embodiments, the processor 30 may be further configured to select one or more polynomial regression coefficients 72 corresponding to the selected flow regime 74 from a plurality of potential sets of polynomial regression coefficients 72 corresponding to the plurality of flow regimes 74. Thus, in such embodiments, the trained machine learning model 50 may be a classification model configured to output a classification of the flow regime 74, rather than a regression model configured to output one or more scalar values ​​corresponding to the normalized sensor data.

[0022] In some embodiments, the time elapsed 56 since calibration of the first pressure sensor 26 and / or the second pressure sensor 28 may be used as an additional input to the trained machine learning model 50. Thus, the processor 30 may be further configured to estimate the flow rate based at least in part on the time elapsed 56 since calibration of at least one of the first pressure sensor 26 and the second pressure sensor 28. For example, some types of pressure sensors tend to drift in the same direction (towards higher or lower pressure readings) after calibration. When such pressure sensors are used as the first pressure sensor 26 and / or the second pressure sensor 28, the trained machine learning model 50 may be configured to adjust the first pressure data 46 and / or the second pressure data 48 based at least in part on the time elapsed 56 since calibration of at least one or both of the first pressure sensor 26 and the second pressure sensor 28.

[0023] As shown in Figure 2, the valve 16 may further include an actuator 34 configured to change the position of the valve 16. Returning to Figure 3, the processor 30 may be configured to send a drive signal 54 to the actuator 34 based at least in part on the estimated flow rate 52. The actuator 34 may be configured to change the position of the valve 16 in response to receiving the drive signal 54. In this manner, the processor 30 may be configured to control the position of the valve 16 based on the estimated flow rate 52.

[0024] Turning now to FIG. 5, the training process of the trained machine learning model 50 is depicted. The trained machine learning model 50 may be trained using a training data set 76 including training valve position data 80, training first temperature data 82, training first pressure data 86, training second pressure data 88, and training flow rate data 90. In an embodiment in which the mass flow controller 10 includes a second temperature sensor 24, the training data set 76 may further include training second temperature data 84. The processor 30 may be configured to train the trained machine learning model 50 starting from a pre-trained machine learning model 70 that has not yet been trained. The pre-trained machine learning model 70 may be initialized with, for example, randomly selected weights. As another example, the initial weights may be normally distributed (follow a normal distribution).

[0025] The processor 30 may be configured to use the training valve position data 80, the training first temperature data 82, the training first pressure data 86, and the training second pressure data 88 as exemplary inputs to an input layer of the pre-training machine learning model 70. Also, in an embodiment in which the mass flow controller 10 includes a second temperature sensor 24, the training second temperature data 84 may be used as an exemplary input. The training valve position data 80, the training first temperature data 82, the training second temperature data 84, the training first pressure data 86, and the training second pressure data 88 may be received by each neuron 71 included in the input layer 73 of the pre-training machine learning model 70. Each neuron 71 included in the input layer 73 may send an output value to a hidden layer 75 of the pre-training machine learning model 70, and each neuron 71 included in the hidden layer 75 may send an output value to another hidden layer 75 or to the output layer 77.

[0026] In embodiments in which the trained machine learning model 50 is configured to use as an input the elapsed time since calibration of the first pressure sensor 26 and / or the second pressure sensor 28, the training data set 76 may further include a plurality of training times 94 since calibration of the first pressure sensor 26 and / or the second pressure sensor 28. In some embodiments, rather than directly inputting the plurality of training times 94 since calibration to one or more neurons 71 in the input layer 73, the processor 30 may be configured to modify the first training pressure data 86 and / or the second training pressure data 88 using a predetermined correction function based on the training time 94 since calibration prior to inputting the first training pressure data 86 and / or the second training pressure data 88 to the pre-training machine learning model 70. For example, the correction function may be linear as a function of the training time 94 since calibration.

[0027] The training flow data 90 may be used as an example output data to which the processor 30 is further configured to compare the output of the pre-trained machine learning model 70. In one example, backpropagation may be used to match the output of the pre-trained machine learning model 70 to the training flow data 90. In such an embodiment, an error function, such as a mean squared error function, may be used to compare the output of the pre-trained machine learning model 70 to the training flow data 90. Additionally, a pre-set learning rate may be used when training the pre-trained machine learning model 70 using backpropagation. In other embodiments, the processor 30 may be configured to use a Bayesian normalized backpropagation method. In an embodiment in which the trained machine learning model 50 is configured to output the normalized valve position data 60, the normalized first temperature data 62, the normalized first pressure data 66, and the normalized second pressure data 68, the processor 30 may be further configured to apply a polynomial regression 58 to the output values ​​of the pre-trained machine learning model 70 to generate the estimated flow rate 52.

[0028] Alternatively, in embodiments in which the trained machine learning model 50 is configured to output a flow regime 74, the processor 30 may be further configured to apply a polynomial regression 58, including one or more polynomial regression coefficients 72, to example inputs included in the training data set 76 to generate an estimated flow rate 52. In such embodiments, the processor 30 may select one or more polynomial regression coefficients 72 that correspond to the identified flow regime 74. The processor 30 may be further configured to compare the estimated flow rate 52 to the training flow rate data 90 and modify neuron weights of one or more neurons 71 included in the pre-trained machine learning model 70 based on the comparison.

[0029] In this manner, the processor 30 may be configured to generate the trained machine learning model 50 by training the pre-trained machine learning model 70 using a plurality of example inputs and outputs included in the training data set 76. Although Fig. 5 illustrates the training of the trained machine learning model 50 occurring in the processor 30 of the mass flow controller 10, the trained machine learning model 50 may instead be generated in one or more other processors included in one or more additional computing devices. In embodiments in which the trained machine learning model 50 is trained in one or more other computing devices, the trained machine learning model 50 may be loaded into the memory 32 of the mass flow controller 10 from the one or more other computing devices after training.

[0030] When multiple mass flow controllers 10 are manufactured, there may be small differences between the mass flow controllers 10 that would cause the estimated flow rates 52 of those mass flow controllers 10 to be inaccurate if the same trained machine learning model 50 and polynomial regression coefficients 72 were used for each mass flow controller 10. Therefore, after the mass flow controller 10 is manufactured, the polynomial regression coefficients 72 of that mass flow controller 10 may be calibrated. Alternatively, additional training may be performed on the output layer 77 of the pre-trained machine learning model 70. In this way, the manufacturer of the mass flow controller 10 may take into account the differences between the components of the individual mass flow controllers 10.

[0031] FIG. 6 shows an example of the arrangement of the mass flow controller 10 during calibration. In the example of FIG. 6, the mass flow controller 10 is arranged between an upstream mass flow controller 10A and a downstream mass flow controller 10B that have already been calibrated. Calibrating the polynomial regression coefficients 72 or the output layer 77 may include, for example, flowing a fluid with a known flow rate through the flow path 14 of the mass flow controller 10. The fluid may be flowed through the mass flow controller 10 by the upstream mass flow controller 10A arranged at the upstream end of the flow path 14. The pressure of the fluid in the flow path 14 may be set by the downstream mass flow controller 10B arranged at the downstream end of the flow path 14. The upstream mass flow controller 10A and the downstream mass flow controller 10B may measure the pressure and / or temperature upstream and downstream of the mass flow controller 10, respectively. In this manner, the flow rate of the fluid and the pressure difference across the valve 16 may be measured. The estimated flow rate 52, the first pressure data 46, and the second pressure data 48 may then be adjusted to match the known flow rate and pressure difference values.

[0032] FIG. 7A shows a flow chart of an exemplary method 200 for use in a mass flow controller. The method 200 may be used in the mass flow controller 10 of FIG. 2 or alternatively in some other mass flow controllers. In step 202, the method 200 may include measuring valve position data of a valve disposed at least partially in a flow path through a block body of the mass flow controller with a valve position sensor. In step 204, the method 200 may further include measuring first temperature data of a fluid flowing in the flow path with a first temperature sensor. In step 206, the method 200 may further include measuring first pressure data for the fluid with a first pressure sensor disposed in the flow path upstream of the valve. In step 208, the method 200 may further include measuring second pressure data of the fluid with a second pressure sensor located in the flow path downstream of the valve.

[0033] At step 210, the method 200 may further include receiving the valve position data, the first temperature data, the first pressure data, and the second pressure data at a processor. The processor may be physically embedded in the mass flow controller or may be located in a separate physical device communicatively coupled to the valve position sensor, the first temperature sensor, the first pressure sensor, and the second pressure sensor.

[0034] At step 212, the method 200 may further include estimating a flow rate of the fluid in the flow path by at least partially applying the trained machine learning model to the valve position data, the first temperature data, the first pressure data, and the second pressure data. In some embodiments, the estimation of the flow rate may be further based on an elapsed time since calibration of at least one of the first pressure sensor and the second pressure sensor. In some embodiments, the trained machine learning model may be a regression model configured to output at least one scalar value. For example, the at least one scalar value may be an estimated flow rate. In other embodiments, the trained machine learning model may be a classification model.

[0035] 7B illustrates further steps of method 200 that may be performed to estimate a flow rate of a fluid when the trained machine learning model is a regression model configured to output one or more scalar values. Method 200 may include step 241, which is determining normalized valve position data, normalized first temperature data, normalized first pressure data, and normalized second pressure data from the valve position data, the first temperature data, the first pressure data, and the second pressure data in the trained machine learning model. Method 200 may further include applying a polynomial regression to the normalized valve position data, the normalized first temperature data, the normalized first pressure data, and the normalized second pressure data in step 216. Normalizing the valve position data, the first temperature data, the first pressure data, and the second pressure data allows them to be used as inputs for a polynomial regression that does not include terms of a best-fit polynomial with improper units.

[0036] 7C illustrates additional steps of method 200 that may be performed when the trained machine learning model is a classification model. In step 218, the trained machine learning model may be configured to identify a flow regime of the fluid in the flow path. The flow regime may be selected from a plurality of flow regimes. In step 220, method 200 may further include applying a polynomial regression to the valve position data, the first temperature data, the first pressure data, and the second pressure data. The polynomial regression may have one or more polynomial regression coefficients selected based at least in part on the identified flow regime.

[0037] FIG. 7D illustrates further steps of the method 200 that may be performed in some embodiments. In step 222, the method 200 may further include measuring second temperature data of the fluid flowing in the flow path with a second temperature sensor. The first temperature sensor may be located upstream of the valve in the flow path, and the second temperature sensor may be located downstream of the valve in the flow path. In this manner, the first temperature sensor and the second temperature sensor may measure a temperature difference across the valve. In step 224, the method 200 may further include estimating, by the processor, a flow rate of the fluid in the flow path by at least partially applying a trained machine learning model to the second temperature data. In embodiments in which the steps shown in FIG. 7B are performed, the normalized second temperature data may further be determined with the trained machine learning model. In such embodiments, a polynomial regression may further be applied to the normalized second temperature data. Alternatively, in embodiments in which the steps of FIG. 7C are performed, a polynomial regression may further be applied to the second temperature data.

[0038] 7E illustrates additional steps that may be performed in performing method 200. At step 226, method 200 may further include transmitting a drive signal to an actuator included within the valve based at least in part on the estimated flow rate. At step 228, method 200 may further include altering a position of the valve in response to receiving the drive signal, via the actuator. In this manner, the position of the valve may be adjusted based on the estimated flow rate.

[0039] The above-described systems and methods can be used to more accurately estimate the flow rate of a fluid through a mass flow controller when an adjustable valve is used as a restrictor than when the mass flow controller includes a separate restrictor. The lack of a separate restrictor can reduce the cost and complexity of manufacturing the mass flow controller. Additionally, the use of a valve as a restrictor allows adjustments to the flow rate of the fluid to be made more quickly than is possible with conventional mass flow controllers. When estimating the flow rate, trained machine learning models can be used in addition to or instead of polynomial regression to model the complex fluid dynamics inherent in this simplified mass flow controller configuration.

[0040] In some embodiments, the methods and processes described in this disclosure may be associated with a computing system of one or more computing devices. In particular, such methods and processes may be implemented as computer application programs or services, application programming interfaces (APIs), libraries and / or other computer program products.

[0041] 8 illustrates a schematic of a non-limiting embodiment of a computing system 300 capable of implementing one or more of the above-described methods and processes. The computing system 300 is illustrated in simplified form. The computing system 300 may embody the above-described mass flow controller 10 illustrated in FIG. 2. The computing system 300 may take the form of one or more personal computers, server computers, tablet computers, home computers, network computing devices, gaming devices, mobile computing devices, mobile communication devices (e.g., smartphones) and / or other computing devices, as well as wearable computing devices such as smart wristwatches and head-mounted augmented reality devices.

[0042] Computing system 300 includes a logical processor 302, a volatile memory 304, and a non-volatile storage device 306. Computing system 300 may optionally include a display subsystem 308, an input subsystem 310, a communication subsystem 312, and / or other components not shown in FIG.

[0043] Logical processor 302 includes one or more physical devices configured to execute instructions. For example, a logical processor may be configured to execute instructions that are part of one or more applications, programs, routines, libraries, objects, components, data structures, or other logical constructs. Such instructions may be implemented to perform a task, implement a data type, transform the state of one or more components, achieve a technical effect, or arrive at a desired result.

[0044] A logical processor may include one or more physical processors (hardware) configured to execute software instructions. Additionally or alternatively, a logical processor may include one or more hardware logic circuits or firmware devices configured to execute hardware-implemented logic or firmware instructions. The processors of logical processor 302 may be single-core or multi-core, and the instructions executed therein may be configured for sequential, parallel and / or distributed processing. Individual components of a logical processor may be distributed among two or more separate devices as needed, which may be remotely located and / or configured for cooperative processing. Aspects of a logical processor may be virtualized and executed by networked computing devices configured in a cloud computing configuration and accessible remotely. In such cases, it will be understood that these virtualized aspects are executed on different physical logical processors of various different machines.

[0045] Non-volatile storage 306 includes one or more physical devices configured to recognize, by a logical processor, instructions executable to implement the methods and processes described in this disclosure. When such methods and processes are executed, the state of non-volatile storage 306 may be transformed, for example, to hold different data.

[0046] The non-volatile storage 306 may include removable and / or built-in physical devices. The non-volatile storage 306 may include optical memory (e.g., CD, DVD, HD-DVD, Blu-ray disc, etc.), semiconductor memory (e.g., ROM, EPROM, EEEPROM, FLASH memory, etc.) and / or magnetic memory (e.g., hard disk drive, floppy disk drive, tape drive, MRAM, etc.) or other mass memory technology. The non-volatile storage 306 may include non-volatile, dynamic, static, read / write, read-only, sequential access, addressable, file addressable and / or content addressable devices. It will be appreciated that the non-volatile storage 306 is configured to retain instructions even when power is cut off.

[0047] Volatile memory 304 may include physical devices including random access memory. Volatile memory 304 is typically utilized by logic processor 302 to temporarily store information during the processing of software instructions. It will be understood that volatile memory 304 does not continue to store instructions when power consumption is cut off.

[0048] Aspects of the logic processor 302, volatile memory 304, and non-volatile storage 306 may be implemented together in one or more hardware logic components, which may include, for example, field programmable gate arrays (FPGAs), program and application specific integrated circuits (PASICs / ASICs), program and application specific standard products (PSSPs / ASSPs), systems on a chip (SOCs), and complex programmable logic devices (CPLDs).

[0049] The terms "module," "program," and "engine" may be used to describe an aspect of the computing system 300, which is typically implemented in software by a processor, using a portion of volatile memory to perform a particular function, including transformation operations that specifically configure the processor to perform the function. Thus, a module, program, or engine may be instantiated via the logical processor 302, which uses a portion of volatile memory 304 to execute instructions held by the non-volatile storage 306. It will be understood that different modules, programs, and / or engines may be instantiated from the same application, service, code block, object, library, routine, API, function, etc. Similarly, the same module, program, and / or engine may be instantiated from different applications, services, code blocks, objects, routines, APIs, functions, etc. The terms "module," "program," and "engine" may encompass individual or groups of executable files, data files, libraries, drivers, scripts, database records, etc.

[0050] If included, the display subsystem 308 may be used to present a visual representation of the data held by the non-volatile storage 306. The visual representation may take the form of a graphical user interface (GUI). Just as the methods and processes described in this disclosure change the data held by the non-volatile storage, thereby transforming the state of the non-volatile storage, the state of the display subsystem 308 may also be transformed to visually represent the changes in the underlying data. The display subsystem 308 may include one or more display devices utilizing virtually any type of technology. Such display devices may be coupled with the logic processor 302, the volatile memory 304, and / or the non-volatile memory 306 in a shared enclosure, or such display devices may be peripheral display devices.

[0051] If included, the input subsystem 310 may consist of or be coupled to one or more user input devices, such as a keyboard, mouse, touch screen, or game controller. In some embodiments, the input subsystem may consist of or be coupled to selected natural user interface (NUI) components. Such components may be integrated or peripheral, and the transmission and / or processing of input actions may be handled on-board or off-board. Exemplary NUI components may include microphones for speech and / or voice recognition; infrared, color, stereoscopic, and / or depth cameras for machine vision and / or gesture recognition; head trackers, eye trackers, accelerometers, and / or gyroscopes for motion detection and / or intent recognition; and electric field sensing components for assessment of brain activity; and / or any other suitable sensors.

[0052] If included, the communications subsystem 312 may be configured to communicatively connect the various computing devices described in this disclosure to each other and to other devices. The communications subsystem 312 may include wired and / or wireless communication devices compatible with one or more different communications protocols. As a non-limiting example, the communications subsystem may be configured to communicate over a wireless telephone network or over a wired or wireless local or wide area network, such as an HDMI over a Wi-Fi connection. In some embodiments, the communications subsystem may enable the computing system 300 to send and / or receive messages to other devices over a network, such as the Internet.

[0053] It will be understood that the configurations and / or approaches described in this disclosure are exemplary in nature and that numerous variations are possible, and therefore these specific embodiments or examples should not be considered in a limiting sense. The specific routines or methods described in this disclosure may represent one or more of any processing strategies. As such, the various acts shown and / or described may be performed in the order shown and / or described, in other orders, in parallel, or omitted. Similarly, the order of the processes described above may be changed.

[0054] The subject matter of the present disclosure includes all novel and non-obvious combinations and subcombinations of the various processes, systems, and configurations, as well as other features, functions, acts and / or properties disclosed in the present disclosure, and any and all equivalents thereof.

Claims

1. A block body having a flow path; a valve provided in at least a portion of the flow passage of the block body; a valve position sensor that measures the degree to which the valve is open or closed; a first temperature sensor configured to detect a temperature of a fluid flowing through the flow path; a first pressure sensor disposed in the flow path upstream of the valve; a mass flow controller including a second pressure sensor disposed downstream of the valve in the flow path; a processor; The processor, receiving valve position data from the valve position sensor; receiving first temperature data from the first temperature sensor; receiving first pressure data for the fluid in the flow path from the first pressure sensor; receiving second pressure data for the fluid in the flow path from the second pressure sensor; estimating a flow rate of the fluid in the flow path by at least partially applying a trained machine learning model to the valve position data, the first temperature data, the first pressure data, and the second pressure data; the trained machine learning model is trained using a training data set including training valve position data, first training temperature data, first training pressure data, second training pressure data, and training flow rate data, and estimates the flow rate based on at least a flow regime of the fluid in the flow path; A mass flow controller, wherein the flow path is free of a restrictor, and the fluid flowing through the flow path has a plurality of potential flow patterns.

2. 2. The mass flow controller of claim 1, wherein the trained machine learning model is a regression model configured to output at least one scalar value.

3. The processor further determines, in the trained machine learning model, normalized valve position data, normalized first temperature data, normalized first pressure data, and normalized second pressure data from the valve position data, the first temperature data, the first pressure data, and the second pressure data; 3. The mass flow controller of claim 2, configured to estimate a flow rate of the fluid in the flow path by at least partially applying a polynomial regression to the normalized valve position data, the normalized first temperature data, the normalized first pressure data, and the normalized second pressure data.

4. a memory for storing one or more coefficients of the polynomial regression; The mass flow controller of claim 3 , wherein the processor is configured to estimate a flow rate of the fluid based at least in part on the one or more coefficients.

5. a second temperature sensor provided in the block body and downstream of the valve in the flow path; the first temperature sensor is disposed in the flow path upstream of the valve; The processor, receiving second temperature data from the second temperature sensor; 10. The mass flow controller of claim 1, further configured to estimate a flow rate of the fluid in the flow path by at least partially applying the trained machine learning model to the second temperature data.

6. The processor, determining normalized second temperature data in the trained machine learning model; The mass flow controller of claim 5 , further configured to determine a flow rate of the fluid in the flow path by at least partially applying a polynomial regression to the second temperature data.

7. 2. The mass flow controller of claim 1, wherein the trained machine learning model is a cascade forward neural network or a convolutional neural network.

8. The mass flow controller of claim 1 , wherein the trained machine learning model is configured to identify a flow regime of the fluid in the flow path.

9. the processor is further configured to send a drive signal to an actuator included in the valve based at least in part on the estimated flow rate; The mass flow controller of claim 1 , wherein the actuator is configured to change a position of the valve in response to receiving the drive signal.

10. 2. The mass flow controller of claim 1, wherein the processor is further configured to estimate the flow rate based at least in part on an elapsed time since calibration of at least one of the first pressure sensor and the second pressure sensor.

11. 2. The mass flow controller according to claim 1, wherein the first temperature sensor is disposed within a wall of the block body.

12. A method of using a mass flow controller in which a fluid flow path is not provided with a restrictor, and the fluid flowing through the path has a plurality of potential flow patterns, comprising: measuring valve position data of a valve provided in at least a portion of the flow path passing through the block body, using a valve position sensor that measures the degree of opening and closing of the valve; measuring a first temperature data of a fluid flowing through the flow passage with a first temperature sensor; measuring first pressure data of the fluid at a first pressure sensor disposed in the flow path upstream of the valve; measuring second pressure data of the fluid at a second pressure sensor disposed in the flow path downstream of the valve; In the processor, receiving the valve position data, the first temperature data, the first pressure data, and the second pressure data; and estimating a flow rate of the fluid in the flow path by at least partially applying a trained machine learning model to the valve position data, the first temperature data, the first pressure data, and the second pressure data; the trained machine learning model is trained using a training data set including training valve position data, training first temperature data, training first pressure data, training second pressure data, and training flow rate data; and wherein the trained machine learning model estimates the flow rate based on at least a flow regime of the fluid in the flow path.

13. The method of claim 12 , wherein the trained machine learning model is a regression model configured to output at least one scalar value.

14. estimating a flow rate of the fluid in the flow path, determining, in the trained machine learning model, normalized valve position data, normalized first temperature data, normalized first pressure data, and normalized second pressure data from the valve position data, the first temperature data, the first pressure data, and the second pressure data; 14. The method of claim 13, comprising applying a polynomial regression to the normalized valve position data, the normalized first temperature data, the normalized first pressure data, and the normalized second pressure data.

15. measuring a second temperature of the fluid flowing through the flow passage with a second temperature sensor; and estimating, in the processor, a flow rate of the fluid in the flow path by at least partially applying the trained machine learning model to the second temperature data.

16. the first temperature sensor is disposed in the flow path upstream of the valve; 16. The method of claim 15, wherein the second temperature sensor is disposed in the flow path downstream of the valve.

17. sending a drive signal to an actuator included in the valve based at least in part on the estimated flow rate; The method of claim 12, further comprising the step of altering a position of the valve in response to receiving the drive signal via the actuator.

18. A block body having a flow path; a valve provided in at least a portion of the flow passage of the block body; a valve position sensor that measures the degree to which the valve is open or closed; a first temperature sensor configured to sense a temperature of a fluid flowing through the flow path; a first pressure sensor disposed in the flow path upstream of the valve; a second pressure sensor disposed downstream of the valve in the flow path; a memory for storing one or more polynomial regression coefficients; a processor; The processor, receiving valve position data from the valve position sensor; receiving first temperature data from the first temperature sensor; receiving first pressure data of the fluid in the flow path from the first pressure sensor; receiving second pressure data of the fluid in the flow path from the second pressure sensor; determining, in a trained machine learning model, normalized valve position data, normalized first temperature data, normalized first pressure data, and normalized second pressure data from the valve position data, the first temperature data, the first pressure data, and the second pressure data; and configured to at least partially apply a polynomial regression having one or more polynomial regression coefficients to the normalized valve position data, the normalized first temperature data, the normalized first pressure data, and the normalized second pressure data, and further determine a flow rate of the fluid in the flow path based at least on a flow regime of the fluid in the flow path; A mass flow controller in which no restrictor is provided in the flow path, and the fluid flowing through the flow path has a plurality of potential flow patterns.

Images