Computing system, computing method, and gas delivery device

Abstract

The present invention provides a computing system, a computing method and a gas delivery apparatus, the computing system comprising: a processor and a non-volatile memory storing executable instructions that, in response to execution by the processor, cause the processor to: receive monitored gas delivery process data comprising sensor information and / or valve position information of a monitored gas delivery process; receive optimal gas delivery process data comprising sensor information and / or valve position information of an optimal gas delivery process; and execute a discriminative classifier comprising at least one artificial intelligence model, the discriminative classifier being configured to: extract features of the monitored gas delivery process data; extract features of the optimal gas delivery process data; calculate a similarity value of the monitored gas delivery process based on the extracted features of the monitored gas delivery process data and the extracted features of the optimal gas delivery process data; and output the similarity value.

Description

Technical Field

[0001] The present invention relates to a computational system, computational method, and gas delivery device having a discriminative classifier for determining the similarity of monitored gas delivery processes. Background Technology

[0002] In a gas delivery system used to supply various gases in semiconductor manufacturing, multiple gas supply channels can carry different gases to mix them and supply them for various manufacturing processes. A mass flow controller is provided for each gas supply channel to regulate the gas flow rate.

[0003] In some gas delivery systems, mass flow controllers can all be networked to a central control unit (CCU), which remotely sends commands and requests process data to the mass flow controllers via the network. Each mass flow controller maintains its own control loop in sync with the CCU, synchronized with the commands sent from the CCU via the network. A predefined gas delivery process, involving the sequential delivery of one or more gases at various flow rates and pressures to downstream equipment (e.g., a semiconductor process chamber), is accomplished by sending sequential flow commands and settings to the mass flow controllers via a local computer network. The CCU receives feedback from each mass flow controller, uses this feedback for feedback control of each controller, and records the actual gas delivery process to ensure quality. However, for high-precision control, with increasing sensor resolution and data streams, the CCU may struggle to acquire data in real time from the sensors and processors on individual mass flow controllers, especially when local computer network capacity and bandwidth are limited. Furthermore, even if the CCU acquires this data, it is difficult to monitor subtle deviations between the target process of the gas delivery system (i.e., the current process occurring in real time or a process executed in the past and subsequently analyzed) and the idealized process definition. Summary of the Invention

[0004] According to one aspect of this disclosure, a gas delivery device is provided, comprising: a plurality of valves and sensors; a processor operatively connected to the plurality of valves and sensors; and a non-volatile memory operatively connected to the processor and storing monitored gas delivery process data, optimal gas delivery process data, and a discriminative classifier. The non-volatile memory stores executable instructions that, in response to execution by the processor, cause the processor to: receive the monitored gas delivery process data, the monitored gas delivery process data including sensor information and / or valve position information of the monitored gas delivery process; receive the optimal gas delivery process data, the optimal gas delivery process data including sensor information and / or valve position information of the optimal gas delivery process; and execute the discriminative classifier including at least one artificial intelligence model, the discriminative classifier being configured to: extract features from the monitored gas delivery process data; extract features from the optimal gas delivery process data; calculate a similarity value of the monitored gas delivery process based on the extracted features of the monitored gas delivery process data and the extracted features of the optimal gas delivery process data; and output the similarity value.

[0005] This summary is provided to present, in a simplified form, a series of concepts that will be further described in the detailed description section below. This summary is neither intended to identify key or essential features of the claimed subject matter nor to limit the scope of the claimed subject matter. Furthermore, the claimed subject matter is not limited to embodiments that address any or all of the shortcomings pointed out in any part of this disclosure. Attached Figure Description

[0006] Figure 1 A schematic diagram of a gas delivery system according to an example of this disclosure is shown.

[0007] Figure 2 It shows that according to Figure 1 A schematic diagram illustrating the flow of data and control signals from a flow control circuit in an example gas delivery system.

[0008] Figure 3 It shows that according to Figure 1 An example of a gas delivery system, showing monitored gas delivery process data and examples of optimal gas delivery process data.

[0009] Figure 4 It shows that according to Figure 1 An example of a gas delivery system is embodied in a three-dimensional array monitoring gas delivery process data.

[0010] Figure 5A It shows that during the training phase, according to Figure 1A detailed schematic diagram of an example gas delivery system, including a central data repository, monitored gas delivery process data, optimal gas delivery process data, and a discriminative classifier.

[0011] Figure 5B It shows the runtime phase based on Figure 1 A detailed schematic diagram of an example gas delivery system, including a central data repository, monitored gas delivery process data, optimal gas delivery process data, and a discriminative classifier.

[0012] Figure 5C It shows according to Figure 1 A detailed schematic diagram of another example of a gas delivery system, including a central data repository, monitored gas delivery process data, optimal gas delivery process data, and a discriminative classifier.

[0013] Figure 6 It shows according to Figure 1 An example of a gas delivery system; an example of a similarity index for monitoring the gas delivery process in a gas delivery system.

[0014] Figure 7A It shows according to Figure 1 An example of a gas delivery system is a first method for outputting an alarm when the similarity index and / or repeatability confidence value of the monitored gas delivery process in the gas delivery system falls below a predetermined threshold.

[0015] Figure 7B It shows according to Figure 1 A second method involves analyzing the similarity index and repeatability confidence value of the monitored gas delivery process of an example gas delivery system and training an artificial intelligence model.

[0016] Figure 7C It shows according to Figure 1 Another example of a gas delivery system is the third method of monitoring the similarity index and / or repeatability confidence value of the gas delivery process of the output gas delivery system.

[0017] Figure 8 It shows where it is possible to formulate with Figure 1 and Figure 2 A schematic diagram of an example computational environment for a method used in conjunction with a gas delivery system.

[0018] Figure 9 It shows Figure 1 An exemplary graphical user interface for a gas delivery system indicates an alarm used to notify the user of repeatability deviations or offsets in the monitored gas delivery process from the optimal gas delivery process. Detailed Implementation

[0019] In view of the above issues, refer to Figure 1 A gas delivery system 10 is provided, comprising a unified high-speed electrical backplane 18 operatively connected to a system controller 12 via a first memory interface 22A, to multiple mass flow control circuits 30A-30P via a second memory interface 22B, to an input / output module 20 via a third memory interface 22C, to multiple pressure controllers 24A-24D via a fourth memory interface 22D, and to multiple flow ratio control circuits 26A-26D via a fifth memory interface 22E. It should be understood that the centralization of the gas delivery system 10 is achieved by connecting all sensors and control circuitry of the gas delivery system 10 to a single system controller 12 and a central data repository 15. Typically, the electrical backplane 18 comprises a printed circuit board, and operative connections between the electrical backplane and these system components are achieved through electrical connections between these components via electrical connectors. The mass flow controllers 30A-30P each include their own mass flow control circuitry, such as a printed circuit board, with each circuitry directly mounted to a corresponding electrical connector on the printed circuit board of the electrical backplane 18. The gas delivery system 10 can be configured as a gas delivery device that can be enclosed within a housing 11. The electrical backplane 18 is typically configured for high-speed communication between its electrically connected components. Furthermore, a client computing device 110 can be operatively connected to the gas delivery system 10 via a system controller 12 to exchange data communications with the system controller 12. These data communications may include, for example, instructions or commands from the gas delivery system 10, flow diagnostic information, and flow monitoring information.

[0020] System controller 12 includes processor 12A and volatile memory 12B. In some embodiments, system controller 12 may be configured as a system on module (SOM). For example, processor 12A may include a multi-core processor. For example, system controller 12 may also incorporate circuitry of a field-programmable gate array (FPGA) or application-specific integrated circuit (ASIC). System controller 12 is operatively connected to non-volatile memory 14, which stores a central data repository 15, which in turn stores monitored gas delivery process data 40, optimal gas delivery process data 42, and a discriminative classifier 44. Mass flow control circuits 30A-30P, flow ratio control circuits 26A-26D, and pressure controllers 24A-24D may each include at least one processor and at least one memory independent of system controller 12 and non-volatile memory 14. The input / output module 20, as well as the pressure controllers 24A-24D, the flow ratio control circuits 26A-26D, and the mass flow control circuits 30A-30P, can alternatively include FPGA circuitry without requiring an additional processor.

[0021] To reduce flow path and cable lengths, the circuit boards for system controller 12, mass flow control circuits 30A-30P, flow ratio control circuits 26A-26D, and pressure controllers 24A-24D are physically and electrically mounted on the printed circuit board of the electrical backplane 18 via corresponding electrical connectors. Therefore, all controllers are centralized within the housing 11 of the gas delivery system 10 to facilitate data collection at a single location. This has the advantage of reducing latency and bandwidth requirements for transmitting data from these controllers to remote locations for remote processing and storage.

[0022] The electrical backplane 18 can be further operatively connected to a power supply 16, which is configured to supply power to all modules and controllers physically mounted on the electrical backplane 18, including system controller 12, input / output modules 20, pressure controllers 24A-24D, flow ratio control circuits 26A-26D, and mass flow control circuits 30A-30P. This significantly reduces the length of power cables required in the gas delivery system 10. For example, the power supply 16 could comprise a single 24VDC, 150V power supply shared by all modules in the system.

[0023] The electrical backplane 18 forms the backbone of the entire gas delivery system 10. Since all communication data, control signals, and power are embedded in the electrical backplane 18, this allows the system controller 12 to access each sensor and actuator within the airflow path in real time. Furthermore, because the system controller 12 is operatively connected to the non-volatile memory 14 (a large, high-speed data storage device storing a central data repository 15), a large amount of sensor and actuator data can be stored in real time in the monitored gas delivery process data 40 for post-processing analysis and long-term storage.

[0024] The electrical backplane 18 may include a main printed circuit board (PCB) with high-speed low-voltage differential signaling (LVDS) interface elements for serial communication and control lines. However, it should be understood that the PCB is not particularly limited to LVDS elements, and other standards and interface elements may be implemented in the PCB alternatively to achieve high-speed data rates on the electrical backplane 18.

[0025] The electrical backplane 18 may contain two independent serial communication systems: one for configuration (configuration bus 18A) and the other for control (control bus 18B). They can operate together, operate completely independently, or operate simultaneously. During system power-up and module and system configuration, the control bus 18B can be in standby mode. Once configuration and system checks are complete, the control bus 18B can switch to operating mode. Appropriate configuration data, module identification data, calibration data, etc., are transmitted via the configuration bus 18A to each individual module 20, 24A-24D, 26A-26D, 30A-30P. Additionally, during operation, historical data containing operation logs of all modules connected to the electrical backplane 18 is transmitted via the configuration bus 18A. It should be understood that the configuration bus 18A operates independently of the control bus 18B and memory interface 22A of the electrical backplane 18.

[0026] In the gas delivery system 10, the system controller 12 can be designated as controller module 12, while the mass flow control circuits 30A-30P, flow ratio control circuits 26A-26D, pressure controllers 24A-24D, and input / output module 20 are designated as controlled modules. As a fully asynchronous system, controller module 12, controlled modules 20, 24A-24D, 26A-26D, 30A-30P, and electrical backplane 18 all operate independently and asynchronously, without waiting states or forced interrupts on the local processor. Electrical backplane 18 serves as a full-duplex serial bus that handles all communication between controller module 12 and all controlled modules 20, 24A-24D, 26A-26D, and 30A-30P, while simultaneously transmitting read and write data in each cycle. A coprocessor 13, operably connected to the first memory interface 22A, arbitrates or coordinates data exchanges between the system controller 12 and the controlled modules 20, 24A-24D, 26A-26D, and 30A-30P. In an alternative embodiment, the coprocessor 13 may be integrated into the first memory interface 22A.

[0027] Figure 2 This is an exemplary schematic diagram illustrating the flow of data and control signals with respect to the first mass flow control circuit 30A and the first flow ratio control circuit 26A. For simplicity, the flow of data and control signals with respect to other controlled modules is not shown in this schematic diagram.

[0028] refer to Figure 2An example of asynchronous flow of data signals from flow sensor 38 will be illustrated. Flow sensor 38 senses the flow rate in the flow path corresponding to the first mass flow control circuit 30A and sends the flow rate signal to the first mass flow control circuit 30A. The first mass flow control circuit 30A then embeds the address corresponding to the first mass flow control circuit 30A and flow sensor 38 into a data stream and sends the data stream containing the embedded address and flow rate signal to the second memory interface 22B. The second memory interface 22B directs the data stream to the electrical backplane 18, which in turn sends the data stream to the first memory interface 22A. Coprocessor 13 parallelizes the data stream and stores it in the read data register of the first memory interface 22A corresponding to the first mass flow control circuit 30A. This data stream transfer from the first mass flow control circuit 30A to the first memory interface 22A can be triggered by a request from coprocessor 13. Then, the system controller 12 accesses the read data register of the first memory interface 22A corresponding to the first mass flow control circuit 30A, and the system controller 12 identifies the flow signal as originating from the flow sensor 38 corresponding to the first mass flow control circuit 30A. This data is then stored in the monitored gas delivery process data 40 in the central data repository 15 of the non-volatile memory 14, so that the monitored gas delivery process data 40 can be used for subsequent analysis by the discriminative classifier 44. It should be understood that the temperature signal from the temperature sensor 36 and the pressure signal from the pressure sensor 34 of the first flow ratio control circuit 26A are similarly processed by the flow ratio control circuit 26A, the fifth memory interface 22E, the electrical backplane 18, the first memory interface 22A, and the system controller 12 to store the temperature signal from the temperature sensor 36 and the pressure signal from the pressure sensor 34 into the monitored gas delivery process data 40. Therefore, the system controller 12 can collect valve position information and sensor information from at least a plurality of sensors and valves operably connected to a plurality of mass flow control circuits 30A-30P and a plurality of flow ratio control circuits 26A-26D, and store the valve position information and sensor information in the monitored gas delivery process data 40 stored in the non-volatile memory 14.

[0029] System controller 12 is not specifically limited to information collection and storage, and is also configured to perform calculations based on stored valve position information and sensor information. In this example, system controller 12 is also configured to calculate flow rate and control values ​​based on valve position information and sensor information from active mass flow control circuitry and active flow ratio control circuitry. For example, when system controller 12 obtains pressure information, temperature information, and valve position information of first flow control valve 32A from first mass flow control circuitry 30A, system controller 12 can calculate flow rate and determine a new appropriate valve position accordingly, then store the new valve position in first memory interface 22A. Coprocessor 13 can send the new valve position back to first mass flow control circuitry 30A, and then first mass flow control circuitry 30A adjusts first flow control valve 32A to the new valve position. In other words, system controller 12 can handle data and control calculations for gas delivery system 10.

[0030] like Figure 2 As shown, in this example, a first mass flow control circuit 30A is operatively connected to valve 32A. The first mass flow control circuit 30A can send control signals to valve 32A to control the opening position of valve 32A. Conversely, valve 32A can send a data signal indicating its opening position to the first mass flow control circuit 30A. Similarly, a first flow ratio control circuit 26A is operatively connected to valve 32B. The first flow ratio control circuit 26A can send control signals to valve 32B to control its opening position. Conversely, valve 32B can send a data signal indicating its opening position to the first flow ratio control circuit 26A. When configured as a shutdown valve, the opening position can be either on or off. When configured as a linear valve, the opening position can be a linear opening position among a plurality of possible opening positions.

[0031] For example, when system controller 12 closes shut-off valve 32A asynchronously, system controller 12 sends a data stream containing the addresses of the first upstream shut-off valve 32A and the first mass flow control circuit 30A. The data stream enters the first memory interface 22A to the write data register corresponding to the first mass flow control circuit 30A. Then, coprocessor 13 serializes the data stream and sends it to electrical backplane 18, where the data stream is directed to second memory interface 22B. Second memory interface 22B reads the address embedded in the data stream and forwards the data stream to the first mass flow control circuit 30A. Then, the first mass flow control circuit 30A sends a control signal to the first upstream shut-off valve 32A according to the instructions from system controller 12 in the data stream and controls the opening of the first upstream shut-off valve 32A. It should be understood that system controller 12 can control other valves in the gas delivery system 10 in a similar, completely asynchronous manner.

[0032] As further detailed below, during the manufacturing process, the gas delivery system 10 trains a discriminative classifier 44 based on sensor information from multiple sensors and valve position information from multiple valves. The system controller 12 then executes the discriminative classifier 44 to monitor performance differences in the manufacturing process relative to an optimal manufacturing process, or to perform feedback training on the discriminative classifier 44 based on the sensor information and valve position information. The system controller 12 calculates a repeatability confidence value and can send an alarm when the repeatability confidence value exceeds a predetermined threshold.

[0033] refer to Figure 3 Examples of monitored gas delivery process data 40 and optimal gas delivery process data 42 are shown. Monitored gas delivery process data 40 includes sensor information and / or valve position information for the monitored gas delivery process. In these examples, pressure readings (P sensor readings), temperature readings (T sensor readings), and flow rate readings (Q sensor readings) are recorded at fixed time intervals. Figure 3 In the example, sensor readings are recorded every 0.1 seconds.

[0034] Optimal gas delivery process data 42 includes sensor information and / or valve position information for the optimal gas delivery process. In other words, optimal gas delivery process data 42 is a record of sensor readings that are designated as baseline sensor readings for comparison and contrast with the sensor readings of monitored gas delivery process data 40. It should be understood that the optimal gas delivery process is a model process established by operators under conditions of close monitoring on the gas delivery system during the pre-production phase, followed by the mass production process. The sensor readings of optimal gas delivery process data 42 may reflect the normal or ideal conditions of the process before process components (valves, sensors, mass flow controllers, electronic circuitry) are replaced, modified, or otherwise altered through intentional changes or unintentional wear (which may cause the various aspects of implementing the optimal gas delivery process to differ across systems). As these examples illustrate, each sensor reading is normalized to a value between 0 and 1.

[0035] refer to Figure 4 The diagram illustrates an example of monitored gas delivery process data 40 being converted into a three-dimensional array or as an input image for an RGB image with three channels (red, green, and blue). It should be understood that optimal gas delivery process data 42 undergoes a similar conversion. The input image comprises pixels with multiple channels, each pixel corresponding to a time point. Each channel is dedicated to a sensor reading from a single sensor or valve. In other words, each channel corresponds to a separate sensor or valve. In this example, channel 1 is dedicated to flow sensor data, channel 2 to temperature sensor data, and channel 3 to pressure sensor data. Each pixel represents a data point at a specific time point. Therefore, each pixel can represent a row in the monitored gas delivery process data 40. Each pixel stores information about the intensity of each channel.

[0036] The intensity of each pixel at each channel can indicate a normalized sensor reading or a normalized valve position value. In this example, the pixel in the top left corner of the image has an intensity of 11 at channel 1 for the flow sensor reading at coded 0s, an intensity of 102 at channel 2 for the temperature sensor reading at coded 0s, and an intensity of 35 at channel 3 for the pressure sensor reading at coded 0s. Although three channels are shown in this image, it should be understood that there is no particular limitation on the number of channels, and additional channels can be added to the image as needed to accommodate the number of sensors monitored in the gas delivery process data being monitored.

[0037] refer to Figure 5ADetailed schematic diagrams of monitored gas delivery process data 40 and optimal gas delivery process data 42 are shown. Two artificial intelligence models, namely a first artificial intelligence model 50 and a second artificial intelligence model 52, are provided as functions G with equal weights. w The same artificial intelligence model. In other words, the weights of the first artificial intelligence model 50 are equal to the weights of the second artificial intelligence model 52. The first artificial intelligence model 50 and the second artificial intelligence model 52 can be configured as convolutional neural networks.

[0038] Non-volatile memory 14 stores instructions that, in response to execution by processor 12A, cause processor 12A to: receive acceptable monitored gas delivery process data 40, the monitored gas delivery process data 40 including sensor information and / or valve position information of the monitored gas delivery process; receive optimal gas delivery process data 42, the optimal gas delivery process data 42 including sensor information and / or valve position information of the optimal gas delivery process; and execute discriminative classifier 44. In this schematic diagram, discriminative classifier 44 is in the training phase of training a first artificial intelligence model 50 and a second artificial intelligence model 52.

[0039] The acceptable monitored gas delivery process data 40, including pressure sensor data 40A, temperature sensor data 40B, and flow sensor data 40C, is fed as a first input image 46 into a first artificial intelligence model 50 of a discriminative classifier 44 to extract features from the first input image 46 and obtain a first output. The pressure sensor data 42A, temperature sensor data 42B, and flow sensor data 42C of the optimal gas delivery process data 42 are fed as the second input image 48 into the second artificial intelligence model 52 of the discriminative classifier 44 to extract features from the second input image 48 and obtain the second output.

[0040] The outputs of the first AI model 50 and the second AI model 52 are compared using the Euclidean distance function 54. An evaluation is performed to calculate the Euclidean distance between the two outputs. As a similarity value. Subsequently, the Euclidean distance function 54 outputs a similarity index 58 representing the similarity or dissimilarity between two outputs (see...). Figure 6 The Euclidean distance is defined as:

[0041]

[0042] Among them, G wThis is a function of one of the artificial intelligence models. X1 and X2 are the input data pairs. Alternatively, other suitable distance functions can be used, such as Minkowski distance, Manhattan distance, cosine distance, etc.

[0043] Furthermore, the output layers of the first AI model 50 and the second AI model 52 are fed into a contrast loss function 56 to calculate the contrast loss between the two images as a similarity value based on the calculated similarity index. The contrast loss function is as follows:

[0044]

[0045] Among them, D w The distance is defined as the Euclidean distance between the outputs of the first AI model 50 and the second AI model 52. Subsequently, the contrastive loss function 56 outputs a repeatability confidence value 60 based on the contrastive loss, which represents the level of confidence that the monitored gas delivery process is the same as the optimal gas delivery process. For example, the repeatability confidence value 60 can be a value from 0 to 1, where a value of 1 represents a 100% confidence level that the monitored gas delivery process is the same as the optimal gas delivery process. The similarity index 58 and the repeatability confidence value 60 are then output to the graphical user interface of the display device. The discriminative classifier 44 includes a threshold estimator 63 configured to determine whether the similarity index 58 and / or the repeatability confidence value 60 are below a predetermined threshold. When the discriminative classifier 44 determines that the similarity index 58 and / or the repeatability confidence value 60 are below the predetermined threshold, an alarm 64 indicating that the threshold has been exceeded and that the monitored gas delivery process does not match the optimal gas delivery process is output. When the discriminant classifier 44 determines that the similarity index 58 and / or the repeatability confidence value 60 are not lower than a predetermined threshold, it outputs an indicator 66 indicating that the monitored gas delivery process is acceptablely matched with the optimal gas delivery process.

[0046] During the training phase of the discriminative classifier 44, the output repeatability confidence value 60 is used to calculate the contrastive loss, Euclidean distance, and function G. w The gradient is computed by backpropagation of the two instances. Based on the computed gradient, the weights of the first AI model 50 and the second AI model 52 are updated using optimizer 62. Thus, the accuracy of the discriminative classifier 44 is improved through the training process.

[0047] While the above examples disclose a first input image processed by a first artificial intelligence model and a second input image processed by a second artificial intelligence model, along with the outputs of the first and second artificial intelligence models evaluated by an Euclidean distance function to subsequently output a similarity index, it should be understood that in other examples, as described below, the process data can alternatively be processed by a structural similarity assessment module that outputs a similarity index between the current process data and the optimal process data based on the similarity of structural features, such as brightness and contrast (i.e., gas delivery process data encoded as brightness and contrast in the pixel data of the image). The output of the structural similarity assessment module can then be processed by an artificial intelligence model to output a repeatability confidence value representing a level of confidence that the monitored gas delivery process is identical to the optimal gas delivery process (and optionally identical within a threshold variance).

[0048] refer to Figure 5B This diagram illustrates a detailed representation of the currently monitored gas delivery process data 41, the optimal gas delivery process data 42, and the discriminative classifier 44 during the runtime phase. Typically, during the runtime phase, while real-time feedback training based on in-situ process results during manufacturing can be implemented, the first and second AI models are not trained. In the example shown, instead of acceptable monitored gas delivery process data 40 from the training dataset, for example, currently monitored gas delivery process data 41 from the production run is fed to the discriminative classifier 44. This can be performed in real-time as the process occurs, or asynchronously in batches. Because... Figure 5B Examples and Figure 5A The example is essentially similar, except that the acceptable monitored gas transport process data 40, optimizer 62, gradient calculation via backpropagation, and the use of optimizer 62 to update the weights of the first AI model 50 and the second AI model 52 based on the calculated gradients are omitted. Therefore, for the sake of brevity, these are abbreviated here. Figure 5B A detailed explanation of the example.

[0049] exist Figure 5B In this process, the pressure sensor data 41A, temperature sensor data 41B, and flow sensor data 41C of the currently monitored gas transport process data 41 are fed as the first input image 46 into the first artificial intelligence model 50 of the discriminative classifier 44 to extract features from the first input image 46 and obtain the first output. The pressure sensor data 42A, temperature sensor data 42B, and flow sensor data 42C of the optimal gas delivery process data 42 are fed as the second input image 48 into the second artificial intelligence model 52 of the discriminative classifier 44 to extract features from the second input image 48 and obtain the second output. When the discriminant classifier 44 determines that the similarity index 58 and / or the repeatability confidence value 60 are below a predetermined threshold, it outputs an alarm 64 indicating that the threshold has been exceeded and the monitored gas delivery process does not match the optimal gas delivery process. When the discriminant classifier 44 determines that the similarity index 58 and / or the repeatability confidence value 60 are not below the predetermined threshold, it outputs an indicator 66 indicating that the monitored gas delivery process is acceptablely matched with the optimal gas delivery process.

[0050] refer to Figure 5C As another example of this disclosure, a detailed schematic diagram of monitored gas delivery process data 140 and optimal gas delivery process data 142 is shown. In this example, instead of Figure 5A The example of the first artificial intelligence model 50 and the second artificial intelligence model 52 includes a structural similarity evaluation module 150. It should be understood that in... Figures 5A-5C In the example, discriminative classifiers 44 and 144 each include at least one artificial intelligence model with different configurations.

[0051] The central data repository 115 of the non-volatile memory 14 stores instructions that, in response to execution by the processor 12A, cause the processor 12A to: receive monitored gas delivery process data 140, the monitored gas delivery process data 140 including sensor information and / or valve position information of the monitored gas delivery process; receive optimal gas delivery process data 142, the optimal gas delivery process data 142 including sensor information and / or valve position information of the optimal gas delivery process; and execute a discriminative classifier 144.

[0052] The pressure sensor data 140A, temperature sensor data 140B, and flow sensor data 140C of the monitored gas delivery process data 140 are fed into the structural similarity evaluation module 150 of the discriminant classifier 144. The pressure sensor data 142A, temperature sensor data 142B, and flow sensor data 142C of the optimal gas delivery process data 142 are also fed into the structural similarity evaluation module 150 of the discriminant classifier 144.

[0053] The structural similarity assessment module 150 then compares data from the monitored gas delivery process data 140 with data from the optimal gas delivery process data 142 to assess the similarity of structural features, such as brightness and contrast (i.e., gas delivery process data encoded as brightness and contrast in the pixels of the image). Therefore, the structural similarity assessment module 150 can be configured to receive a first input image 46 and a second input image 48 having channels containing encoded pressure data 140A, 142A, temperature data 140B, 142B, and flow data 140C, 142C, and calculate a structural similarity index (SSIM) 152 based on a comparison of the entire image. A similarity index 158 can then be calculated, which can be the raw value of SSIM 152, or a normalized, scaled, or other value derived from SSIM 152. The similarity index 158 indicates the similarity or dissimilarity between the two images. After calculation, the structural similarity assessment module 150 outputs a similarity index 158 for downstream calculation and / or display.

[0054] As an alternative or supplement to calculating SSIM values ​​for the comparison of the entire first input image 46 and the second input image 48, the structural similarity assessment module 150 can calculate and output a similarity index for a dedicated channel of each sensor dataset. In this example, the structural similarity assessment module 150 outputs: a first structural similarity index measurement (SSIM) value 152A, which calculates a similarity value for evaluating the similarity between pressure sensor data 140A of the monitored gas delivery process data 140 and pressure sensor data 142A of the optimal gas delivery process data 142; a second SSIM value 152B, which serves as a similarity value for evaluating the similarity between temperature sensor data 140B of the monitored gas delivery process data 140 and temperature sensor data 142B of the optimal gas delivery process data 142; and a third SSIM value 152C, which serves as a similarity value for evaluating the similarity between flow sensor data 140C of the monitored gas delivery process data 140 and flow sensor data 142C of the optimal gas delivery process data 142.

[0055] Furthermore, the SSIM value 152, along with additionally or alternatively, the first SSIM value 152A, the second SSIM value 152B, and the third SSIM value 152C, are fed into the convolutional neural network 156 of the artificial intelligence model 155 to calculate a repeatability confidence value between the two sets of data as a similarity value based on the calculated similarity indices 152A-152C. Subsequently, the artificial intelligence model 155 outputs a repeatability confidence value 160 based on the first SSIM value 152A, the second SSIM value 152B, and the third SSIM value 152C, which represents the level of confidence that the monitored gas delivery process is the same as the optimal gas delivery process. For example, the repeatability confidence value 160 can be a value from 0 to 1, where a value of 1 represents a 100% confidence level that the monitored gas delivery process is the same as the optimal gas delivery process (or, optionally, the same within a threshold deviation). The convolutional neural network 156 can be trained via supervised learning input 168, which may include ground truth labels from in-situ training. Subsequently, as referenced... Figure 9 The similarity index 158 and the repeatability confidence value 160 are output to the graphical user interface of the display device.

[0056] The discriminant classifier 144 includes a threshold estimator 163 configured to determine whether the similarity index 158 and / or the repeatability confidence value 160 are below a predetermined threshold. When the discriminant classifier 144 determines that the similarity index 158 and / or the repeatability confidence value 160 are below the predetermined threshold, it outputs an alarm 164 indicating that the threshold has been exceeded and the monitored gas delivery process does not match the optimal gas delivery process. When the discriminant classifier 144 determines that the similarity index 158 and / or the repeatability confidence value 160 are not below the predetermined threshold, it outputs an indicator 166 indicating that the monitored gas delivery process acceptablely matches the optimal gas delivery process.

[0057] refer to Figure 6Example of a similarity index 58 output by the Euclidean distance function 54 is shown, where the similarity index is 1 at the start of the process when the monitored process has not deviated from the optimal gas delivery process (in time periods 1 to 11), and less than 1 at the end of the process when the monitored process has significantly deviated from the optimal gas delivery process (in time periods 81 to 91). In this example, the similarity index is consistently less than 1 for the monitored gas delivery process (see the table below), indicating some dissimilarity between the monitored gas delivery process and the optimal gas delivery process. The table above shows the baseline similarity index results assuming a perfect match between the monitored gas delivery process and the optimal gas delivery process, where all similarity indices indicate a value of 1.

[0058] Figure 7A A flowchart illustrating an example configuration of a first method 200 according to one aspect of this disclosure is shown. (Reference) Figure 1 , Figure 2 and Figure 5B The software and hardware components shown above provide the following description of the first method 200. This flowchart of the first method 200 illustrates the runtime phases of a process for outputting an alarm when the similarity index and / or repeatability confidence value of the monitored gas delivery process in the gas delivery system falls below a predetermined threshold.

[0059] In step 202, monitored gas delivery process data is received. In step 204, optimal gas delivery process data is received. In step 206, features of the monitored gas delivery process data are extracted. This feature extraction can be performed using a first artificial intelligence model or a structural similarity assessment module. In step 208, features of the optimal gas delivery process data are extracted. This feature extraction can be performed using a second artificial intelligence model or a structural similarity assessment module.

[0060] In step 210, a repeatability confidence value is calculated based on the extracted features of the monitored gas delivery process data and the extracted features of the optimal gas delivery process data. The calculation of the repeatability confidence value can be performed by an artificial intelligence model, such as a convolutional neural network, based on a similarity index calculated by the structural similarity assessment module. Alternatively, the calculation of the repeatability confidence value can be performed via a contrastive loss function based on a similarity index calculated by a distance function. In step 212, the repeatability confidence value is output. In step 214, it is determined whether the similarity index and / or the repeatability confidence value are below a predetermined threshold. In step 216, an alarm is output in response to the determination that the similarity index and / or the repeatability confidence value are below the predetermined threshold.

[0061] Figure 7BA flowchart illustrating an example configuration of the second method 300 according to one aspect of this disclosure is shown. (Reference) Figure 1 , Figure 2 and Figure 5A The software and hardware components shown above provide the following description of the second method 300. This flowchart of the second method 300 illustrates the training phase of a process for outputting an alarm when the similarity index and / or repeatability confidence value of the monitored gas delivery process in the gas delivery system falls below a predetermined threshold.

[0062] In step 302, data on the monitored gas delivery process of the gas delivery system is received as the first input image, and data on the optimal gas delivery process of the gas delivery system is received as the second input image. In step 304, the function G... w The first artificial intelligence model extracts features from the first input image of the monitored gas transport process data to obtain the first output. In step 306, through function G w The second artificial intelligence model extracts features from the second input image of the optimal gas transport process data to obtain the second output. In step 308, a repeatability confidence value is calculated based on the similarity index, wherein, in step 308A, the Euclidean distance between the first output and the second output is used. Calculate the similarity index, and in step 308B, based on the first output Second output The calculated similarity index is then subjected to contrast loss. The repeatability confidence score is calculated. In step 310, the repeatability confidence score is output to the graphical user interface of the display device. In step 312, the first artificial intelligence model and the second artificial intelligence model are trained, wherein, in step 312A, the comparison loss, Euclidean distance, and function G are used. w The two instances use backpropagation to compute gradients, and in step 312B, the optimizer is used to update the weights of the first and second AI models based on the computed gradients. During the training phase, the first method 300 returns to steps 304 and 306 to recover the extracted features of the input image.

[0063] Figure 7C A flowchart illustrating an example configuration of a third method 400 according to one aspect of this disclosure is shown. (Reference) Figure 1 , Figure 2 and Figure 5C The software and hardware components shown above provide the following description of the third method 400.

[0064] In step 402, the monitored gas delivery process data and the optimal gas delivery process data of the gas delivery system are received through the structural similarity assessment module. In step 404, the similarity of structural features is assessed between the monitored gas delivery process data and the optimal gas delivery process data. Specifically, in step 404A, a similarity index is calculated based on the two sets of data, and in step 404B, SSIM values ​​are calculated for each sensor dataset in the dataset. In step 406, a repeatability confidence score is calculated based on the SSIM values ​​using an artificial intelligence model. In step 408, the repeatability confidence score is output to the graphical user interface of the display device. In step 410, the similarity index is output to the graphical user interface of the display device.

[0065] According to this disclosure, the integrated real-time central control can process all sensor data at a single location, control the entire gas delivery process, and record all data in real time to a central data repository. Furthermore, by storing all real-time sensor data in one location, complex performance analysis can be performed through machine learning and real-time adjustment of flow parameters to improve the performance and repeatability of the gas delivery system. Using a discriminative classifier, the monitored gas delivery process can be compared in real time with the optimal gas delivery process, and deviations from the optimal process can be detected immediately in real time, thereby effectively monitoring subtle changes in parameters that deviate from the normal parameters of the monitored gas delivery process.

[0066] In some implementations, the methods and processes described herein may be associated with a computing system of one or more computing devices. In particular, the methods and processes may be implemented as computer applications or services, application programming interfaces (APIs), libraries, and / or other computer program products.

[0067] Figure 8 A non-limiting embodiment of a computing system 500 capable of implementing one or more of the above processes is illustrated schematically. The computing system 500 is shown in a simplified form. The computing system 500 can be specifically implemented as follows: Figure 1 and Figure 2 The controller module 12 or controlled modules 20, 24A-24D, 26A-26D, 30A-30P shown above. The computing system 500 may take the form of one or more personal computers, server computers, tablet computers, home entertainment 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 smartwatches and head-mounted augmented reality devices.

[0068] The computing system 500 includes a logic processor 502, volatile memory 504, and non-volatile storage device 506. Optionally, the computing system 500 may include a display subsystem 508, an input subsystem 510, a communication subsystem 512, and / or Figure 8 Other components not shown.

[0069] The logic processor 502 includes one or more physical devices configured to execute instructions. For example, the logic 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 structures. These instructions can be executed to perform tasks, implement data types, transition the state of one or more components, achieve technical effects, or otherwise achieve desired results.

[0070] A logic processor may include one or more physical processors (hardware) configured to execute software instructions. Additionally or alternatively, a logic processor may include one or more hardware logic circuits or firmware devices configured to execute hardware-implemented logic or firmware instructions. The processor of logic processor 502 may be single-core or multi-core, and the instructions executed thereon may be configured for sequential, parallel, and / or distributed processing. Optionally, the various components of the logic processor may be distributed across two or more separate devices that may be remotely located and / or configured for coordinated processing. Various aspects of the logic processor may be virtualized and executed by remotely accessible networked computing devices configured for cloud computing. In this context, it should be understood that these virtualized aspects run on different physical logic processors on various different machines.

[0071] Non-volatile storage device 506 includes one or more physical devices configured to retain instructions executable by a logic processor to implement the methods and processes described herein. When these methods and processes are implemented, for example, the state of non-volatile storage device 506 may change to retain different data.

[0072] Non-volatile storage device 506 may include removable and / or built-in physical devices. Non-volatile storage device 506 may include optical storage (e.g., CD, DVD, HD-DVD, Blu-ray disc, etc.), semiconductor storage (e.g., ROM, EPROM, EEPROM, flash memory, etc.), and / or magnetic storage (e.g., hard disk drive, floppy disk drive, magnetic tape drive, MRAM, etc.) or other high-capacity storage technologies. Non-volatile storage device 506 may include non-volatile, dynamic, static, read / write, read-only, sequential access, location-addressable, file-addressable, and / or content-addressable devices. It should be understood that non-volatile storage device 506 is configured to retain instructions even when the non-volatile storage device 506 is powered off.

[0073] Volatile memory 504 may have a physical device including random access memory. Volatile memory 504 is typically used by logic processor 502 to temporarily store information during the processing of software instructions. It should be understood that when volatile memory 504 is powered off, volatile memory 504 typically does not continue storing instructions.

[0074] The logic processor 502, volatile memory 504, and non-volatile storage device 506 can be integrated together into one or more hardware logic components. For example, such hardware logic components may include field-programmable gate arrays (FPGAs), program-and-application-specific integrated circuits (PASICs / ASICs), program-and-application-specific standard products (PSSPs / ASSPs), system-on-a-chip (SoCs), and complex programmable logic devices (CPLDs).

[0075] The terms "module," "program," and "engine" can be used to describe aspects of a computing system 500, typically implemented in software by a processor, to perform specific functions using a portion of volatile memory. These functions involve translational processing specifically configured in the processor to perform the functions. Therefore, a module, program, or engine can be instantiated using a portion of volatile memory 504 by a logical processor 502 executing instructions held by non-volatile storage device 506. It should be understood that different modules, programs, and / or engines can be instantiated from the same application, service, code block, object, library, routine, API, function, etc. Similarly, the same module, program, and / or engine can be instantiated from different applications, services, code blocks, objects, routines, APIs, functions, etc. The terms "module," "program," and "engine" can include single or grouped executable files, data files, libraries, drivers, scripts, database records, etc.

[0076] When a display subsystem 508 is included, it can be used to present a visual representation of data held by the non-volatile storage device 506. The visual representation can take the form of a graphical user interface (GUI). As the methods and processes described herein change the data held by the non-volatile storage device, thereby changing the state of the non-volatile storage device, the state of the display subsystem 508 can also be similarly changed to visually represent changes in the underlying data. The display subsystem 508 can include one or more display devices utilizing virtually any type of technology. Such display devices can be combined with the logic processor 502, the volatile memory 504, and / or the non-volatile storage device 506 in a common housing, or such display devices can be peripheral display devices.

[0077] When the input subsystem 510 is included, it may include one or more user input devices (e.g., a keyboard, mouse, touchscreen, or game controller) or interact with such devices. In some embodiments, the input subsystem may include or interact with selected natural user input (NUI) components. Such components may be integrated or peripheral, and the transduction and / or processing of input actions may be handled on-board or off-board. Example NUI components may include microphones for speech and / or voice recognition; infrared, color, stereo, 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; electric field sensing components for assessing brain activity; and / or any other suitable sensors.

[0078] When a communication subsystem 512 is included, the communication subsystem 512 can be configured to communicatively connect the various computing devices described herein to each other, and to communicate with other devices. The communication subsystem 512 may include wired and / or wireless communication devices compatible with one or more different communication protocols. As a non-limiting example, the communication subsystem may be configured to communicate via a wireless telephone network or a wired or wireless local area network or wide area network (e.g., via an HDMI connection via Wi-Fi). In some embodiments, the communication subsystem may allow the computing system 500 to send and / or receive messages to and / or from other devices via a network such as the Internet.

[0079] refer to Figure 9 The diagram illustrates an example GUI 600 according to this disclosure. In this example, alarms 602 and 604 indicated on the GUI 600 displayed on the display subsystem 508 notify the user of any repeatability deviation or offset of the monitored gas delivery process from the optimal gas delivery process. As in this example, the GUI 600 can be displayed on a display device dedicated to the gas delivery system 10, or on a display device of a distributed control system (DCS) serving multiple gas delivery systems. In this example, the gas delivery system is simply a gas supply unit that allows gas to flow into chamber M5. Pressure and flow sensors are provided on valve XY, which is used to regulate the gas flow from the gas supply unit to chamber M5. On the GUI 600, the user can see a similarity index 58 and / or repeatability confidence value 60 of the gas delivery process in a graphical form as a linear graph, allowing the user to visually assess the changes of the similarity index 58 and / or repeatability confidence value 60 over time. Users can operate the "Select Optimal Process" button to access options for setting the optimal gas delivery process, comparing it with the current gas delivery process, and setting a predetermined threshold. When the similarity index 58 or the repeatability confidence value 60 exceeds the predetermined threshold, the system controller 12 sends a first alarm 602 and / or a second alarm 604 indicating that an offset or deviation has been detected. In this example, an offset in the similarity index 58 and the repeatability confidence value 60 that has exceeded the predetermined threshold has been detected. Therefore, a warning box indicating "Alarm! Similarity Index Offset" is displayed on the GUI 600 as the first alarm 602, and a warning box indicating "Alarm! Process Repeatability Offset" is displayed on the GUI 600 as the second alarm 604. Users can operate the "Select Classifier Type" button to select between a discriminative classifier using an artificial intelligence model and a discriminative classifier using a structural similarity evaluation module that does not contain an artificial intelligence model.

[0080] It should be understood that the predetermined threshold can be multiple predetermined thresholds or limit values. For example, multiple alarm levels can be provided: for an offset below a first lower threshold, a LO-level alarm can be activated, and for an offset below a second lower threshold which is smaller than the first lower threshold, a LO-LO-level alarm can be activated.

[0081] It should be understood that the configurations and / or methods described herein are exemplary in nature, and these specific implementations or examples should not be considered limiting, as many variations are possible. The particular routines or methods described herein may represent one or more of any number of processing strategies. Therefore, the various actions shown and / or described may be performed in the shown and / or described sequence, in other sequences, in parallel, or omitted. Similarly, the order of the above processing may be changed.

[0082] This disclosure includes all novel and non-obvious combinations and sub-combinations of the various features and techniques disclosed herein. The various features and techniques disclosed herein are not necessarily necessary for all examples of this disclosure. Furthermore, the various features and techniques disclosed herein may define patentable subject matter beyond the disclosed examples and may find utility in other implementations not expressly disclosed herein.

[0083] It should be understood that the “and / or” used in this article refers to the logical disjunction operation, and therefore A and / or B have the following truth table.

[0084] A B A and / or B T T T T F T F T T F F F

[0085] The terms “contains,” “has,” “includes,” and their variations are used in this document and are intended to be used as open-ended terms in a manner similar to the term “includes,” without excluding any additional or other elements.

[0086] Cross-references to related applications

[0087] This application claims priority to U.S. Provisional Patent Application Serial No. 63 / 042,723, filed June 23, 2020, the entire contents of which are incorporated herein by reference for all purposes.

Claims

1. A computing system comprising: A processor and non-volatile memory storing executable instructions, the instructions causing the processor to: Receive monitored gas delivery process data, which includes sensor information and / or valve position information of the monitored gas delivery process; Receive optimal gas delivery process data, which includes sensor information and / or valve position information of the optimal gas delivery process; as well as Execute a discriminative classifier comprising at least one artificial intelligence model, wherein the discriminative classifier is configured to: Extract features from the monitored gas transport process data; Extract features from the optimal gas delivery process data; The similarity value of the monitored gas transport process is calculated based on the extracted features of the monitored gas transport process data and the extracted features of the optimal gas transport process data; and Output the similarity value.

2. The computing system according to claim 1, in, The similarity value is a structural similarity value; The extracted feature of the monitored gas transport process data is a structural similarity feature; The extracted feature of the optimal gas delivery process data is a structural similarity feature; The discriminative classifier further includes a structural similarity evaluation module, which is configured to extract structural similarity features from the monitored gas transport process data and extract structural similarity features from the optimal gas transport process data, and output the structural similarity value based thereon; and The at least one artificial intelligence model is a convolutional neural network, which is configured to receive the structural similarity value as input and calculate the repeatability confidence value.

3. The computing system according to claim 1, wherein, The at least one artificial intelligence model is a first artificial intelligence model, which is configured to extract features from the monitored gas transport process data from the first input image, and the discriminative classifier further includes: A second artificial intelligence model is configured to extract features of the optimal gas delivery process data from a second input image; A Euclidean distance function module, configured to apply the Euclidean distance function to calculate a similarity index between the first input image and the second input image based on the outputs of the first artificial intelligence model and the second artificial intelligence model; and A contrastive loss function module is configured to calculate a repeatability confidence value based on the similarity index by applying a contrastive loss function; and The similarity value is at least one of the similarity index and the repeatability confidence value.

4. The computing system according to claim 3, wherein, The first input image and the second input image each include pixels with multiple channels, and each pixel corresponds to a time point.

5. The computing system according to claim 4, wherein, Each channel corresponds to a separate sensor or valve. The intensity of each pixel at each channel represents a normalized sensor reading or a normalized valve position value.

6. The computing system according to claim 3, further comprising: An optimizer configured to update the weights of the first AI model and the second AI model.

7. The computing system according to claim 3, wherein, The first artificial intelligence model and the second artificial intelligence model are convolutional neural networks; and The weights of the first artificial intelligence model are equal to the weights of the second artificial intelligence model.

8. The computing system according to claim 3, wherein, The discriminative classifier is further configured to: Determine whether the repeatability confidence value or the similarity index used to calculate the repeatability confidence value exceeds a predetermined threshold; and An alarm is output when the repeatability confidence value or the similarity index exceeds the predetermined threshold.

9. A calculation method, comprising: Receive monitored gas delivery process data, which includes sensor information and / or valve position information of the monitored gas delivery process; Receive optimal gas delivery process data, which includes sensor information and / or valve position information of the optimal gas delivery process; Execute a discriminative classifier comprising at least one artificial intelligence model, wherein the discriminative classifier is configured to: Extract features from the monitored gas transport process data; Extract features from the optimal gas delivery process data; The similarity value of the monitored gas transport process is calculated based on the extracted features of the monitored gas transport process data and the extracted features of the optimal gas transport process data; and Output the similarity value.

10. The calculation method according to claim 9, in, The similarity value is a structural similarity value; The extracted feature of the monitored gas transport process data is a structural similarity feature; The extracted feature of the optimal gas delivery process data is a structural similarity feature; The discriminative classifier further includes a structural similarity evaluation module, which is configured to extract structural similarity features from the monitored gas transport process data and extract structural similarity features from the optimal gas transport process data, and output the structural similarity value based thereon; and The at least one artificial intelligence model is a convolutional neural network, which is configured to receive the structural similarity value as input and calculate the repeatability confidence value.

11. The calculation method according to claim 9, in, The at least one artificial intelligence model is a first artificial intelligence model, which is configured to extract features of the monitored gas delivery process data from the first input image. The discriminative classifier also includes a second artificial intelligence model, which is configured to extract features of the optimal gas delivery process data from the second input image. The calculation of the similarity value includes: The Euclidean distance function is applied to calculate a similarity index between the first input image and the second input image based on the outputs of the first artificial intelligence model and the second artificial intelligence model; and The repeatability confidence value is calculated based on the similarity index by applying a contrastive loss function; and The similarity value is at least one of the similarity index and the repeatability confidence value.

12. The calculation method according to claim 11, wherein, The first input image and the second input image each include pixels with multiple channels, and each pixel corresponds to a time point; Each channel corresponds to a separate sensor or valve; as well as The intensity of each pixel at each channel represents a normalized sensor reading or a normalized valve position value.

13. The calculation method according to claim 11, further comprising: Update the weights of the first AI model and the second AI model. The weight of the first artificial intelligence model is equal to the weight of the second artificial intelligence model.

14. The calculation method according to claim 9, further comprising: Determine whether the repeatability confidence value or the similarity index used to calculate the repeatability confidence value exceeds a predetermined threshold; and An alarm is output when the repeatability confidence value or the similarity index exceeds the predetermined threshold. The similarity value is at least one of the similarity index and the repeatability confidence value.

15. A gas conveying device, comprising: Multiple valves and sensors; A processor operatively connected to the plurality of valves and sensors; and A non-volatile memory, operatively connected to the processor, stores monitored gas delivery process data, optimal gas delivery process data, and a discriminative classifier, wherein... The non-volatile memory stores executable instructions that, in response to execution by the processor, cause the processor to: Receive the monitored gas delivery process data, which includes sensor information and / or valve position information of the monitored gas delivery process; Receive the optimal gas delivery process data, which includes sensor information and / or valve position information of the optimal gas delivery process; as well as Execute the discriminative classifier, which includes at least one artificial intelligence model, and the discriminative classifier is configured to: Extract features from the monitored gas transport process data; Extract features from the optimal gas delivery process data; The similarity value of the monitored gas transport process is calculated based on the extracted features of the monitored gas transport process data and the extracted features of the optimal gas transport process data; and Output the similarity value.

16. The gas conveying device according to claim 15, in, The similarity value is a structural similarity value; The extracted feature of the monitored gas transport process data is a structural similarity feature; The extracted feature of the optimal gas delivery process data is a structural similarity feature; The discriminative classifier further includes a structural similarity evaluation module, which is configured to extract structural similarity features from the monitored gas transport process data and extract structural similarity features from the optimal gas transport process data, and output the structural similarity value based thereon; and The at least one artificial intelligence model is a convolutional neural network, which is configured to receive the structural similarity value as input and calculate the repeatability confidence value.

17. The gas conveying device according to claim 15, wherein, The at least one artificial intelligence model is a first artificial intelligence model, which is configured to extract features from the monitored gas transport process data from the first input image, and the discriminative classifier further includes: A second artificial intelligence model is configured to extract features of the optimal gas delivery process data from a second input image, wherein the weights of the first artificial intelligence model are equal to the weights of the second artificial intelligence model. A Euclidean distance function module, configured to apply the Euclidean distance function to calculate a similarity index between the first input image and the second input image based on the outputs of the first artificial intelligence model and the second artificial intelligence model; and A contrastive loss function module is configured to calculate a repeatability confidence value based on the similarity index by applying a contrastive loss function; and The similarity value is at least one of the similarity index and the repeatability confidence value.

18. The gas conveying device according to claim 17, wherein, The first input image and the second input image each include pixels with multiple channels, and each pixel corresponds to a time point; Each channel corresponds to a separate sensor or valve, and The intensity of each pixel at each channel represents a normalized sensor reading or a normalized valve position value.

19. The gas conveying device according to claim 17, wherein, The first artificial intelligence model and the second artificial intelligence model are convolutional neural networks; and The weights of the first artificial intelligence model are equal to the weights of the second artificial intelligence model.

20. The gas conveying device according to claim 15, wherein, The processor is also configured to: Determine whether the repeatability confidence value or the similarity index used to calculate the repeatability confidence value exceeds a predetermined threshold; as well as When the repeatability confidence value or the similarity index exceeds the predetermined threshold, an alarm is output, and The similarity value is at least one of the similarity index and the repeatability confidence value.

Images