A fabrication process control system and method based on a dynamic computing engine

Abstract

This application relates to the field of gas preparation process control technology, and discloses a preparation process control system and method based on a dynamic computing engine, including: responding to a gas formula table input by a user, analyzing the reaction between each component in the gas formula table through a preset gas interaction influence library, calculating the filling amount and filling order of each component to reach the target concentration by combining real-time gas cylinder environmental data, and generating a filling list; analyzing the concentration changes of each component in the gas formula table, calculating the mass ratio of the corresponding dilution gas, and constructing a dilution tree; identifying components in the dilution tree whose gas concentration deviates from the corresponding target concentration by more than a preset deviation threshold, calculating the gas replenishment amount of the corresponding component, and obtaining the gas replenishment task; and dynamically controlling the opening of the mass flow controller through a PID controller to control the gas formula preparation process; this application can improve the proportioning accuracy of the final gas product and reduce the waiting and adjustment time during the filling process.

Description

Technical Field

[0001] This application relates to the field of gas preparation process control technology, and more specifically to a preparation process control system and method based on a dynamic computing engine. Background Technology

[0002] Gas formulations typically involve multiple active or inert components that need to be mixed in precise proportions to achieve specific physical or chemical properties. The preparation process involves uniformly mixing various raw material gases in predetermined ratios within specialized gas cylinders or reactors. Traditional gas formulation preparation methods rely on empirical formulas and offline analysis. Operators calculate the target partial pressures of each component based on the target formulation, using the ideal gas law and simple partial pressure methods, and then sequentially fill the gas cylinders. During filling, the valves are manually adjusted based on the operator's experience. After filling, a long period of static mixing is allowed, and finally, sampling analysis is used to verify the accuracy of the proportions. If the accuracy is not up to standard, cumbersome gas replenishment or purging operations are required, or even re-formulation.

[0003] The existing technology has the following problems: it ignores the chemical reactions between gases, the adsorption on the inner wall of the gas cylinder, and the mutual influence of gas compressibility, which leads to the calculated filling amount and sequence failing to achieve the target concentration in actual operation, resulting in a large deviation; once filling begins, it is impossible to adjust according to the real-time changes in pressure, temperature, and dynamic deviation of gas concentration inside the gas cylinder. For concentration deviation, after all steps are completed, analysis is performed and a manual decision is made on whether to add gas, which is a slow and inefficient process. To solve at least one of the above problems, this application proposes a preparation process control system and method based on a dynamic computing engine. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the purpose of this application is to provide a preparation process control system and method based on a dynamic computing engine, which can effectively solve the problems in the background technology. The specific technical solution of this application is as follows:

[0005] The preparation process control method based on dynamic computing engine includes:

[0006] In response to the gas recipe table input by the user, the system analyzes the reaction between the components in the gas recipe table through a preset gas interaction influence library, and calculates the filling amount and filling order of each component to achieve the target concentration by combining real-time gas cylinder environmental data, and generates a filling list.

[0007] Based on the filled list, analyze the concentration changes of each component in the gas formulation table, calculate the mass ratio of the corresponding dilution gas, and construct a dilution tree;

[0008] Identify components in the dilution tree whose gas concentration deviates from the corresponding target concentration by more than a preset deviation threshold, calculate the gas replenishment amount for the corresponding components, and obtain the gas replenishment task.

[0009] By combining the real-time compensated gas supply with the target concentration in conjunction with the gas supply task, the opening degree of the mass flow controller is dynamically controlled by the PID controller to control the gas formulation preparation process.

[0010] Specifically, in response to the user-input gas formula table, the system analyzes the reactions between the components in the gas formula table using a preset gas interaction influence library, and calculates the filling amount and filling order of each component to achieve the target concentration by combining real-time gas cylinder environmental data, generating a filling list, including:

[0011] In response to the gas recipe table input by the user, the reaction between the components in the gas recipe table is analyzed through a preset gas interaction influence library to obtain a set of reaction parameters for each component;

[0012] Based on the set of reaction parameters and real-time gas cylinder environmental data, the filling amount and filling order of each component to achieve the target concentration are calculated, and a filling list is generated.

[0013] Specifically, based on the set of reaction parameters and real-time gas cylinder environmental data, the filling amount and filling order of each component to achieve the target concentration are calculated, and a filling list is generated, including:

[0014] Based on the set of reaction parameters, calculate the reaction heat values ​​and product types between components, and construct the reaction matrix between component reactions;

[0015] Based on the reaction matrix and combined with real-time gas cylinder environmental data, the adsorption rate and desorption rate of each component on the inner wall of the gas cylinder are calculated to obtain the corresponding adsorption coefficient.

[0016] Based on the reaction matrix and adsorption coefficient, the filling amount and filling order of each component to achieve the target concentration are calculated, and a filling list is generated.

[0017] Specifically, the process of analyzing the concentration changes of each component in the gas formulation table based on the filled list, calculating the mass ratio of the corresponding dilution gas, and constructing a dilution tree includes:

[0018] Based on the filled list, the concentration changes of each component in the gas formulation table are analyzed, the subordinate relationship between the dilution gas and the corresponding component is identified, and a dilution relationship topology diagram is obtained.

[0019] Based on the dilution relationship topology, calculate the mass ratio of the corresponding dilution gas and construct a dilution tree.

[0020] Specifically, the step of calculating the mass ratio of the corresponding dilution gas and constructing a dilution tree based on the dilution relationship topology graph includes:

[0021] The dilution relationship topology is matched with the corresponding filling order. Component nodes whose order in the dilution relationship topology does not conform to the filling order are identified and their positions are optimized to obtain the first dilution tree.

[0022] For each component node in the first dilution tree, calculate the mass ratio of the corresponding dilution gas and construct the corresponding dilution mass ratio matrix;

[0023] Based on the dilution mass ratio matrix, the cylinder pressure is analyzed, and pressure balance nodes are inserted between component nodes whose cylinder pressure is greater than a preset pressure threshold during the dilution process to obtain a dilution tree.

[0024] Specifically, the step of identifying components in the dilution tree whose gas concentration deviates from the corresponding target concentration by more than a preset deviation threshold, calculating the gas replenishment amount for the corresponding components, and obtaining the gas replenishment task includes:

[0025] Identify components in the dilution tree whose gas concentration deviates from the corresponding target concentration by a preset deviation threshold, and obtain the first node set;

[0026] Based on the first node set, calculate the gas replenishment amount of the corresponding component to obtain the gas replenishment task.

[0027] Specifically, the step of calculating the gas replenishment amount of the corresponding component based on the first node set to obtain the gas replenishment task includes:

[0028] Based on the first node set, analyze the mass ratio of the corresponding dilution gas for each component and calculate the deviation value of the component;

[0029] Based on the deviation value, the first gas replenishment amount of the corresponding component is calculated. The first deviation between the component concentration after gas replenishment and the corresponding target concentration is compared. The first gas replenishment amount is optimized by the least squares method until the first deviation is less than the preset deviation threshold, and then the second gas replenishment amount is obtained.

[0030] Analyze the reaction risk of each component to determine the corresponding gas replenishment sequence;

[0031] By combining the second replenishment amount and the corresponding replenishment sequence, the replenishment task is obtained.

[0032] Specifically, the step of comparing the real-time compensated gas supply with the target concentration in conjunction with the gas supply task, and dynamically controlling the opening of the mass flow controller through a PID controller, includes:

[0033] By combining the real-time compensated gas supply with the target concentration in conjunction with the gas supply task, the pressure and temperature data inside the gas cylinder are analyzed, and the initial parameter set of the PID controller for the corresponding component is matched.

[0034] Based on the initial parameter set, the opening degree of the mass flow controller is dynamically controlled by a PID controller.

[0035] Specifically, the dynamic control of the mass flow controller's opening degree using a PID controller based on an initial parameter set includes:

[0036] Based on the initial parameter set, the trajectory curve of concentration change over time during the gas replenishment process is analyzed, and the deviation value and the rate of change of the deviation from the target concentration are calculated to obtain the deviation signal and the rate of change of the deviation signal.

[0037] Based on the deviation signal and the deviation change rate signal, the parameter adjustment of the PID controller is calculated, and the opening degree of the mass flow controller is dynamically controlled by the adjusted PID controller.

[0038] A fabrication process control system based on a dynamic computing engine, used to implement the aforementioned fabrication process control method based on a dynamic computing engine, includes:

[0039] The fill list construction module responds to the gas formula table input by the user, analyzes the reaction between the components in the gas formula table through a preset gas interaction influence library, calculates the filling amount and filling order of each component to achieve the target concentration by combining real-time gas cylinder environmental data, and generates a fill list.

[0040] The dilution tree construction module, based on a filled list, analyzes the concentration changes of each component in the gas formulation table, calculates the mass ratio of the corresponding dilution gas, and constructs a dilution tree;

[0041] The gas replenishment analysis module identifies components in the dilution tree whose gas concentration deviates from the corresponding target concentration by more than a preset deviation threshold, calculates the gas replenishment amount for the corresponding components, and obtains the gas replenishment task.

[0042] The preparation process control module compares the real-time compensated gas supply with the target concentration in conjunction with the gas supply task, and dynamically controls the opening of the mass flow controller through a PID controller to control the gas formulation preparation process.

[0043] The beneficial effects of this application are as follows: By analyzing the gas formulation in real time, the chemical reactions and adsorption between components are dynamically evaluated, providing corresponding constraints and correction parameters for filling calculations; by combining the gas interaction library and real-time gas cylinder environmental data, the optimal filling sequence and filling volume are dynamically calculated; by constructing a dilution tree and optimizing the component node sequence, pressure balancing nodes are inserted between nodes with excessive pressure, thereby controlling the pressure and concentration uniformity during the dilution process; based on the dilution tree, nodes with excessive deviations are identified, and the optimal replenishment volume and safe replenishment sequence are calculated to obtain the replenishment task; during the replenishment stage, the initial PID parameters are matched according to the specific situation of the replenishment task, the concentration changes are analyzed in real time, and the PID control parameters are dynamically adjusted according to the deviation and the rate of change of deviation, enabling adaptive control of the mass flow controller opening; this can improve the proportioning accuracy of the final gas product and reduce the waiting and adjustment time during the filling process. Attached Figure Description

[0044] Figure 1 This is a flowchart illustrating the process control method for preparation based on a dynamic computing engine in the embodiments of this application.

[0045] Figure 2 This is a flowchart illustrating the dilution tree construction process in the embodiments of this application;

[0046] Figure 3 This is a schematic diagram of the fabrication process control system based on a dynamic computing engine in an embodiment of this application. Detailed Implementation

[0047] The present application will be further described in detail below with reference to the accompanying drawings and embodiments.

[0048] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0049] Hereinafter, the terms "first," "second," and other generic terms are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0050] refer to Figure 1 The diagram illustrates a specific implementation of the preparation process control method based on a dynamic computing engine, as described in this application, including:

[0051] S101. In response to the gas formula table input by the user, the reaction between each component in the gas formula table is analyzed through the preset gas interaction influence library, and the filling amount and filling order of each component to achieve the target concentration are calculated in combination with real-time gas cylinder environmental data, and a filling list is generated.

[0052] S102. Based on the filled list, analyze the concentration changes of each component in the gas formulation table, calculate the mass ratio of the corresponding dilution gas, and construct a dilution tree;

[0053] S103. Identify components in the dilution tree whose gas concentration deviates from the corresponding target concentration by more than a preset deviation threshold, calculate the gas replenishment amount for the corresponding components, and obtain the gas replenishment task.

[0054] S104. By comparing the real-time compensated gas supply with the target concentration in conjunction with the gas supply task, the opening degree of the mass flow controller is dynamically controlled by the PID controller to control the gas formulation preparation process.

[0055] In this embodiment, in response to a gas formulation table input by the operator on the interactive interface, which includes information such as the names of each component, target concentration values, and allowable error ranges, a preset gas interaction influence library is invoked to analyze the reaction. This library is a pre-built knowledge database based on thermodynamic calculations and experimental measurements, storing parameters such as the calorific value of gas combinations, reaction rate constants, types of products, adsorption coefficients, and risk levels. By traversing all component pairs in the gas formulation table and querying the gas interaction influence library, the reaction parameters for each component pair are obtained, resulting in a set of reaction parameters.

[0056] Specifically, pressure and temperature sensors installed on the gas cylinders collect real-time pressure and temperature values ​​to obtain cylinder environmental data. After acquiring the reaction parameter set and real-time environmental data, the required filling amount and filling order for each component to reach the target concentration are calculated. The filling amount calculation is based on the corresponding gas law, comprehensively considering the non-ideal nature of the gas under high pressure and the adsorption effect of the cylinder's inner wall. An iterative method is used to solve the material balance equation to obtain the number of moles required for each component, which is then converted into flow commands executable by the mass flow controller. The filling order is determined by comprehensively considering the reaction risk level, adsorption competition, and pressure balance requirements. By avoiding continuous filling of high-risk components, components with strong adsorption capacity are filled first, and the pressure after each filling step is predicted. If the pressure exceeds a safety threshold, a pressure balance waiting time is inserted. The filling list includes information such as step number, component name, raw material cylinder identification, set mass flow rate, estimated filling time, estimated waiting time, and estimated end pressure. By constructing a filling list, static formulations are converted into corresponding dynamic execution instructions. Calculations are then performed by combining gas interaction effects and real-time operating conditions. This avoids the risks of chemical reactions, eliminates concentration deviations caused by adsorption effects, and provides an accurate data foundation for subsequent filling analysis.

[0057] Furthermore, concentration change simulation is first performed based on the filling list. The evolution of the concentration of each component in the gas cylinder is simulated according to the filling steps, obtaining the trajectory curve of concentration change over time. By analyzing the shape and interrelationships of the trajectory curves, the gas that plays a dilution role and the logical subordination relationship between the dilution gas and the diluted component are identified. The dilution gas is a gaseous component in the formulation used to reduce the concentration of other components, which does not participate in the core function or only serves as a balance carrier, including but not limited to argon and nitrogen. Using hysteresis correlation analysis, the correlation coefficient between the concentration change of each component and the filling events of other components is calculated. Combined with the risk information and adsorption coefficient in the reaction matrix, the existence of dilution relationships is determined through Bayesian evidence fusion, constructing a dilution relationship topology graph. The dilution relationship topology graph is a directed graph where nodes represent gas components and directed edges point from the dilution gas to the diluted component.

[0058] Specifically, after obtaining the dilution relationship topology diagram, it is matched with the filling order in the filling list. Constraint optimization is used to adjust the filling order or dilution relationship, ensuring consistency between the dilution logic and the operation order, resulting in the first dilution tree. The dilution mass ratio is calculated for each node in the first dilution tree using a backpropagation iterative solution method, starting from the final product, to solve for the ratio of the mass of diluting gas added to the mass of the gas being diluted at each step. The calculation results are integrated to obtain the dilution mass ratio matrix. Based on the dilution mass ratio matrix, the cylinder pressure after each step in the dilution process is analyzed. The corresponding gas state equation is used to predict the pressure value. If the predicted pressure value exceeds a preset pressure threshold, a pressure balance node is inserted after that step, waiting for the gas to fully mix and reach temperature equilibrium before proceeding to the next step, thus obtaining the dilution tree. By constructing the dilution tree, the multi-level dilution process is represented as a clear tree structure, providing accurate structural and sequential support for subsequent concentration deviation analysis and gas replenishment calculations.

[0059] Specifically, the system obtains the actual concentration value of the gas mixture represented by each node in the dilution tree at the current moment through real-time measurement using an online gas analyzer. The actual concentration of each node is compared with the target concentration stored at that node to calculate the absolute deviation value. This absolute deviation value is then compared with a preset deviation threshold, which is dynamically set according to the accuracy requirements of the gas formulation. For semiconductor-grade high-purity gases, it can be set to 0.5%, and for general industrial gases, it can be set to 2%. For components whose deviation exceeds the deviation threshold, the system marks them as exceeding the deviation limit and traces upwards along the path of the dilution tree to locate the source node where the deviation first occurs. These source nodes are then collected to form the first node set.

[0060] For each source node in the first node set, the equivalent deviation mass ratio is calculated. This ratio, assuming other component concentrations remain constant, represents the ratio of the mass required to replenish the target concentration of that component to the current total mass. This initial estimate is used as the initial value. An optimization model is constructed based on this initial estimate, using the replenishment gas volume as the optimization variable and the sum of squared deviations between the predicted and target concentrations of all nodes along the path from the source node to the final product node as the objective function. Iterative optimization is performed using the least squares method. During the optimization process, the system calls the dilution mass ratio matrix to calculate the cascading effect after replenishment gas. The optimal replenishment gas volume that minimizes the objective function is then determined using the least squares method, yielding the second replenishment gas volume. After obtaining the second replenishment volume, the replenishment order is determined. A gas interaction influence library is used to analyze the reaction risks between the components to be replenished and between them and the existing gas in the cylinder. A risk matrix is ​​constructed, and inert gases are inserted to isolate high-risk components. The order within the same group is adjusted based on the adsorption coefficient. The second replenishment volume and replenishment order are integrated to obtain the replenishment task, which includes information such as task identifier, step number, component name, replenishment mass, set flow rate, estimated time, and waiting time. By constructing the replenishment task, quality deviations can be automatically identified and the corresponding replenishment volume calculated, avoiding mutual interference problems during multi-component coupled replenishment and improving the safety of the replenishment process.

[0061] Specifically, the system reads the current step of the gas replenishment task, obtains the component name, target gas replenishment mass, and set flow rate for this gas replenishment, and collects the current pressure and temperature data inside the gas cylinder from real-time sensors. The system pre-builds a PID parameter initial value knowledge base, which stores PID parameters for different gas types and different operating conditions. Using the current gas replenishment component, current pressure, and temperature as query conditions, the system matches the initial proportional coefficient, integral coefficient, and derivative coefficient in the PID parameter initial value knowledge base to obtain the initial parameter set for the PID controller. After entering the control cycle, the actual flow feedback value of the mass flow controller is collected in each control cycle. The target flow is compensated in real time according to the current gas cylinder pressure to obtain the set value. The deviation signal between the set value and the feedback value is calculated, and the rate of change of the deviation is calculated based on the concentration change trajectory. Using the fuzzy adaptive PID control algorithm, the deviation signal and the rate of change of the deviation signal are used as inputs. The adjustment amount of the PID parameters is obtained through fuzzy inference and defuzzification. The adjustment amount is superimposed on the current parameters to obtain the updated PID parameters. The incremental PID algorithm is used to calculate the increment of the control quantity, which is accumulated to the control output of the previous cycle to obtain the valve opening command of the current cycle, and sent to the mass flow controller to adjust the valve opening.

[0062] This process is repeated in each control cycle, performing online self-tuning of the PID parameters to enable the mass flow controller to adapt to the increase in cylinder pressure and changes in gas flow characteristics during the replenishment process. Simultaneously, the actual flow rate is integrated in real time to accumulate the replenished gas mass. When 99.9% of the target replenished gas mass is reached, preparations begin to close the valve; upon reaching 100%, a closing command is immediately issued, and the system enters the waiting time set in the replenishment task. By dynamically adjusting the PID parameters, the replenishment process can be accurately controlled. The fuzzy adaptive mechanism ensures the optimal performance of the mass flow controller under different operating conditions. Through deviation rate of change signal analysis, overshoot and oscillation are effectively suppressed, guaranteeing that the replenished gas quality accurately meets the standards.

[0063] This application analyzes the gas formulation in real time, dynamically assesses the chemical reactions and adsorption between components, and provides corresponding constraints and correction parameters for filling calculations. Combining a gas interaction library and real-time cylinder environmental data, it dynamically calculates the optimal filling sequence and quantity. By constructing a dilution tree and optimizing the component node sequence, pressure balancing nodes are inserted between nodes with excessive pressure, controlling the pressure and concentration uniformity during the dilution process. Based on the dilution tree, nodes with excessive deviations are identified, and the optimal replenishment quantity and safe replenishment sequence are calculated to obtain the replenishment task. During the replenishment stage, initial PID parameters are matched according to the specific circumstances of the replenishment task, concentration changes are analyzed in real time, and PID control parameters are dynamically adjusted based on deviations and deviation change rates to adaptively control the opening of the mass flow controller. This improves the final gas product's mixing accuracy and reduces waiting and adjustment time during the filling process.

[0064] Furthermore, in response to the user-input gas formula table, the system analyzes the reactions between the components in the gas formula table using a preset gas interaction influence library. Combined with real-time gas cylinder environmental data, it calculates the filling amount and order for each component to achieve the target concentration, generating a filling list, including:

[0065] S201. In response to the gas formula table input by the user, the reaction between each component in the gas formula table is analyzed through a preset gas interaction influence library to obtain a set of reaction parameters for each component.

[0066] S202. Based on the set of reaction parameters and real-time gas cylinder environmental data, calculate the filling amount and filling order of each component to achieve the target concentration, and generate a filling list.

[0067] In this embodiment, in response to the user-input gas formula table, a pre-built gas interaction database is invoked to analyze the reaction between the components. This database is a pre-constructed knowledge database that stores parameters such as the calorific value of common gas combinations, reaction rate constant, types of products, adsorption coefficients, and risk levels. For gas combinations that can be found in publicly available databases, thermodynamic data such as standard enthalpy of formation and standard entropy are directly extracted. For novel gases or special combinations, the calorific value of reaction is obtained by measuring the heat flow change during the mixing process using a laboratory microcalorimeter. The adsorption mass and adsorption rate of the gas on the inner wall material of the gas cylinder are determined using quartz crystal microbalance technology. The types and amounts of products are determined by gas chromatography-mass spectrometry.

[0068] Specifically, a depth-first search algorithm is used to pair all components in the gas formulation table, performing a matching query in a gas interaction database. For component pairs without direct interaction data in the database, interaction parameters are estimated through analogy based on the principle of thermodynamic similarity. For each queried component pair, the interaction data is used as reaction parameters, including the calorific value, reaction rate constant, reaction equilibrium constant, product list, adsorption coefficient, and risk level indicator. After traversing all component pairs, all reaction parameters are integrated to obtain a reaction parameter set. By constructing the reaction parameter set, the interaction characteristics between components can be understood before the actual filling operation, providing constraints for subsequent filling sequence optimization and ensuring applicability to different gas formulations.

[0069] Specifically, based on the set of reaction parameters and real-time gas cylinder environmental data, the filling amount and order for each component to reach the target concentration are calculated, generating a filling list. Analysis of the gas state equation improves the accuracy of the filling amount calculation, and dynamic access to real-time environmental data allows the calculation to adapt to changes in on-site conditions, achieving adaptive control. The generated filling list provides structural support and sequential reference for automated execution.

[0070] Furthermore, based on the set of reaction parameters and real-time gas cylinder environmental data, the filling amount and filling order for each component to reach the target concentration are calculated, generating a filling list, including:

[0071] S301. Based on the set of reaction parameters, calculate the reaction heat values ​​and product types between the components, and construct the reaction matrix between the component reactions;

[0072] S302. Based on the reaction matrix and combined with real-time gas cylinder environmental data, calculate the adsorption rate and desorption rate of each component on the inner wall of the gas cylinder to obtain the corresponding adsorption coefficient.

[0073] S303. Combining the reaction matrix and adsorption coefficient, calculate the filling amount and filling order of each component to achieve the target concentration, and generate a filling list.

[0074] In this embodiment, based on the reaction parameter set, the interaction information between each component pair is structurally reorganized to construct a mathematical matrix that comprehensively reflects the interaction relationships between gas components, thus obtaining the reaction matrix. The number of components in the formula is determined, denoted as n, and an n x n square matrix is ​​constructed, with both row and column indices arranged according to the order of appearance of the components in the formula table. For off-diagonal elements in the matrix, i.e., elements with different row and column numbers, the corresponding reaction calorific value and product type information are extracted from the reaction parameter set. The reaction calorific value is a scalar, with its sign indicating the direction of the thermal effect of the reaction; a positive value indicates an exothermic reaction, and a negative value indicates an endothermic reaction. The magnitude of the value reflects the intensity of the reaction. The product type is a list of strings recording the names of byproducts generated by the component pair. These are converted into calculable indicators using a preset coding rule. Specifically, the byproducts are coded according to their toxicity, corrosiveness, flammability, and explosiveness: non-toxic byproducts are coded as 0, low-toxicity byproducts as 1, moderately toxic byproducts as 2, and highly toxic or explosive byproducts as 3.

[0075] For the diagonal elements of the matrix, representing the interaction between the same component and itself, the heat of reaction and risk level are set to 0. The monomeric property parameters of the component, including but not limited to molecular weight, critical temperature, critical pressure, and eccentricity factor, are then appended to the diagonal elements. All component pairs are iterated through, and the heat of reaction for each pair is stored in the heat of reaction field of the matrix element. The risk level of the encoded product is stored in the risk level field. The completed reaction matrix is ​​a composite data structure, with each matrix element containing two basic fields: heat of reaction and risk level. The diagonal elements also contain the component's physical properties.

[0076] It should be noted that by constructing a reaction matrix, the discrete data originally stored in the reaction parameter set is integrated into a unified mathematical matrix, providing accurate data support for subsequent calculations. By risk coding of the types of products, they are transformed into quantitative numerical indicators, which can automatically identify and compare the degree of danger of different reaction paths. The basic physical property parameters of the components are attached to the diagonal elements to achieve data reuse and improve the efficiency of subsequent calculations.

[0077] Specifically, based on the reaction matrix and combined with real-time acquired gas cylinder environmental data, the adsorption behavior of each component on the inner wall of the gas cylinder is analyzed, and key adsorption kinetic parameters are calculated as adsorption coefficients. The baseline adsorption coefficients of each component are read from the diagonal elements of the reaction matrix. These baseline adsorption coefficients correspond to the adsorption coefficients at standard temperatures and are derived from pre-stored laboratory measurement data in a gas interaction library. The measurement methods employ the dual-chamber pressure decay method or the quartz crystal microbalance method. By measuring the pressure decay curve or mass change curve of the gas on the simulated gas cylinder inner wall material surface, the adsorption rate constant and desorption rate constant are derived, and a functional relationship between adsorption parameters and temperature is established. The current real-time temperature is read. According to adsorption thermodynamics theory, the relationship between the adsorption coefficient and temperature conforms to the Arrhenius form of the correction formula, i.e., the adsorption coefficient decreases when the actual temperature is higher than the standard temperature and increases when the actual temperature is lower than the standard temperature. The Langmuir adsorption model is used to describe the relationship between adsorption amount and pressure. The input parameters of this model include the adsorption rate constant, desorption rate constant, gas phase pressure, and current surface coverage; the outputs are the adsorption rate and desorption rate. By using an iterative calculation method, an initial surface coverage is first assumed, the current adsorption rate is calculated, and then the surface coverage is updated according to the pressure change. This process is repeated until convergence, and finally the adsorption coefficient is output, which reflects the ratio of the adsorption amount to the gas phase pressure under equilibrium conditions.

[0078] It should be noted that the adsorption coefficient reflects the strength of the interaction between gas molecules and the solid surface, directly determining how much gas will be temporarily adsorbed onto the inner wall of the gas cylinder during the filling process, thus affecting the actual gas phase concentration achieved. By calculating the adsorption coefficient, the qualitative understanding of adsorption behavior is transformed into a calculable quantitative parameter. Dynamic correction based on real-time temperature data allows the adsorption coefficient to adapt to changes in on-site operating conditions. In low-temperature winter environments, the system automatically calculates a larger adsorption coefficient and increases the filling amount accordingly; in high-temperature summer environments, it automatically decreases the adsorption coefficient to avoid overfilling. The system calculates the adsorption amount under equilibrium conditions and simulates the dynamic behavior of the adsorption process, providing accurate data support for setting the waiting time during the filling process.

[0079] Specifically, the filling parameters are calculated based on the reaction matrix and adsorption coefficients to obtain a filling list. The reaction matrix stores the heat of reaction and risk level between each component; the adsorption coefficient vector contains the dynamic adsorption coefficients of each component; and real-time environmental data of the gas cylinder, including current pressure and temperature, and the target concentration vector from the user-input gas formula table, is retrieved. The filling quantity is calculated based on the material balance principle, taking into account both real gas non-ideality corrections and adsorption loss corrections. The total number of moles in the current gas cylinder is calculated using the real gas equation of state based on the current pressure and temperature and the nominal volume of the cylinder. The final total number of moles and the theoretical number of moles required for each component are then determined using an iterative method based on the target concentration vector.

[0080] Based on this, adsorption correction is performed. An extended Langmuir competitive adsorption model is used to calculate the number of moles of each component adsorbed in the final state. This model considers the competition for adsorption sites when multiple components are present simultaneously. The adsorption amount of each component is calculated based on its adsorption coefficient and final partial pressure, and then added to the theoretical filling amount to obtain the actual number of moles to be filled from the feed gas cylinder. For component pairs with higher risk levels in the reaction matrix, the degree of reaction is estimated based on the calorific value of the reaction, and then the reaction consumption is calculated. For high-risk component pairs, the system will avoid direct contact between them in subsequent filling sequences, so the reaction loss can be ignored.

[0081] Furthermore, a multi-objective sorting algorithm is employed to group components according to their risk level. High-risk components must be filled first or last and isolated with inert gas. Within the same risk level group, components are sorted from largest to smallest adsorption coefficient, prioritizing those with strong adsorption capacity. If adsorption coefficients are similar, components are sorted from lowest to highest target concentration. The cylinder pressure at the end of each step is predicted based on the filling volume. If the predicted pressure exceeds a preset pressure threshold, a pressure balancing waiting time is automatically inserted after that step. All calculation results are combined to generate a final filling list, including step number, component name, raw material cylinder identifier, set mass flow rate, estimated filling time, estimated waiting time, and estimated end pressure.

[0082] It should be noted that by constructing a filling list, the accuracy of the filling quantity calculation is improved by comprehensively considering three major correction factors: the non-ideal nature of real gas, adsorption loss, and reaction loss. The filling sequence optimization based on risk level can eliminate the safety hazards in the empirical sequence and ensure that the entire filling process is always within the safe pressure range.

[0083] Furthermore, based on the filled list, the concentration changes of each component in the gas formulation table are analyzed, the mass ratio of the corresponding dilution gas is calculated, and a dilution tree is constructed, including:

[0084] S401. Based on the filled list, analyze the concentration changes of each component in the gas formulation table, identify the subordinate relationship between the dilution gas and the corresponding component, and obtain the dilution relationship topology diagram.

[0085] S402. Based on the dilution relationship topology diagram, calculate the mass ratio of the corresponding dilution gas and construct the dilution tree.

[0086] In this embodiment, the dilution process during gas formulation preparation is analyzed based on a filling list. Concentration change trajectory simulation is performed based on the filling list, simulating the dynamic evolution of the concentration of each component within the gas cylinder according to the filling step sequence, obtaining the trajectory curve of each component's concentration changing with each filling step. By analyzing the shape and interrelationships of the trajectory curves, dilution gases that play a role in diluting the concentration of other gases during the filling process are identified. Dilution gases are defined as gaseous components in the formulation used to reduce the concentration of other components, which do not participate in the core function or only serve as a balancing carrier, such as argon and nitrogen.

[0087] Preferably, a lag correlation analysis method is used to calculate the correlation coefficient between the concentration change curve of each component and the filling event sequence of other components. Specifically, for the concentration sequence of component i and the filling event sequence of component j, the correlation coefficient is calculated when the lag difference is not reached. If the concentration of component i decreases significantly in subsequent steps after component j is filled and the correlation coefficient exceeds a preset correlation threshold, then component j is determined to dilute component i. The reaction matrix is ​​called to obtain the reaction risk information between components, and the adsorption coefficient is called to obtain the adsorption characteristics of each component. A Bayesian evidence fusion framework is used to comprehensively reason with the correlation analysis results and information such as reaction risk and adsorption coefficient, and to calculate the posterior probability of a dilution relationship between each pair of components. When the posterior probability exceeds a preset confidence threshold, the dilution relationship is confirmed. The correlation threshold is used to determine whether the decrease in component concentration is significantly temporally associated with the filling of another component. It is usually set to 0.7 and is obtained based on the statistical distribution of correlation coefficients of historical real dilution cases. The confidence threshold is used to determine whether the dilution relationship is finally confirmed. Confirmation is only made when the posterior probability of multiple pieces of evidence exceeds this value. It is usually set to 0.8 to 0.9, based on the tolerance for the risk of misjudgment and omission.

[0088] Specifically, all confirmed dilution relationships are organized into a directed graph structure, where nodes represent gas components and directed edges point from the diluting gas to the diluted component. This directed graph is called the dilution relationship topology graph. By constructing the dilution relationship topology graph, the dilution logic is extracted from the complex sequence of filling operations, allowing the process design ideas originally hidden behind the operations to be explicitly expressed. Through Bayesian evidence fusion mechanism, the interference of dilution effects with adsorption and reaction effects is effectively distinguished, avoiding misjudgments. The construction of the dilution relationship topology graph provides accurate path guidance for subsequent quantitative calculations.

[0089] Specifically, based on the dilution relationship topology, the mass ratio of the corresponding dilution gas is calculated, and a dilution tree is constructed. By constructing the dilution tree, the complex multi-component dilution network is decomposed into corresponding dilution steps, ensuring that the mass ratio that satisfies material balance can be stably solved even in complex multi-level dilution networks, simplifying subsequent optimization operations and improving computational efficiency.

[0090] like Figure 2 As shown, based on the dilution relationship topology graph, the mass ratio of the corresponding dilution gas is calculated, and a dilution tree is constructed, including:

[0091] S501. Match the dilution relationship topology graph with the corresponding filling order, identify component nodes in the dilution relationship topology graph whose order does not conform to the filling order, and optimize their positions to obtain the first dilution tree;

[0092] S502. For each component node in the first dilution tree, calculate the mass ratio of the corresponding dilution gas and construct the corresponding dilution mass ratio matrix.

[0093] S503. Analyze the cylinder pressure based on the dilution mass ratio matrix. During the dilution process, insert pressure balance nodes between component nodes where the cylinder pressure is greater than a preset pressure threshold to obtain a dilution tree.

[0094] In this embodiment, matching analysis is performed based on the dilution relationship topology graph and the filling list to identify and correct logical conflicts, resulting in an optimized first dilution tree. All filling events for each component and their temporal order are extracted from the filling list to establish a component filling timetable, recording the first and last occurrence steps of each component. For each directed edge in the dilution relationship topology graph (i.e., the dilution relationship from the diluting gas to the diluted component), a temporal consistency check is performed. The last filling time of the diluting gas is compared with the earliest filling time of the diluted component. If the last filling time of the diluting gas is later than the earliest filling time of the diluted component, it indicates that there is at least one possible time sequence where the diluting gas is filled after the component, and the dilution relationship is temporally valid. Conversely, if all filling events of the diluting gas are earlier than those of the diluted component, the dilution relationship is temporally invalid and marked as a temporal conflict.

[0095] For identified temporal conflicts, positional optimization is performed, adjusting the filling order. The dilution relationship of the conflict is used as a new constraint to optimize the filling order. Under the premise of satisfying the original constraints such as reaction safety and pressure limits, the filling time of the diluent is adjusted so that it is after the component being diluted. After the conflict is resolved, the updated dilution relationship topology is processed to identify and handle circular structures. For nodes with multiple parent nodes, multiple dilution steps are decomposed into multiple sequentially executed sub-steps according to the temporal order and intermediate nodes are introduced. The original directed graph is transformed into a multi-branch tree, resulting in the first dilution tree. The first dilution tree has a unique root node representing the final product, leaf nodes representing pure component raw materials, and each non-leaf node representing an intermediate mixture. Each node has only one parent node, i.e., a unique dilution source.

[0096] It should be noted that by constructing the first dilution tree, the dilution logic and operation sequence are automatically aligned and verified, eliminating logical contradictions between theoretical design and actual execution, and ensuring that subsequent dilution quality ratio calculations are based on consistency. The multi-parent node decomposition mechanism transforms the complex multi-source dilution relationship into a simple tree structure, providing a clear path for subsequent quantitative calculations.

[0097] Specifically, based on the first dilution tree, the corresponding dilution gas mass ratio is calculated for each node in the tree, and the dilution gas mass ratio values ​​are integrated into a dilution mass ratio matrix. The hierarchical structure of the dilution tree is determined, and a recursive solution is performed from top to bottom starting from the root node. The concentration vector of the root node is known as the target concentration vector input by the user, and its child nodes may be multiple different dilution gases. The dilution order of each child node is determined according to the optimized filling order. For each parent-child node relationship, a material balance equation is established, linking the final target concentration, the concentration of the diluent, and the dilution mass ratio through the principle of mass conservation. A nonlinear least squares method is used for iterative solution, progressively recursively solving from the last dilution step forward. The mass ratio obtained at each step must satisfy the reasonable constraint that all components of the intermediate concentration vector are between 0 and 1 and normalized.

[0098] When a child node is also an intermediate node, the mass ratio and concentration vector of that child node are first recursively calculated, and then the mass ratio of the current node is calculated. After the dilution mass ratios of all nodes are calculated, these ratios are organized into a matrix according to the order of the dilution steps to obtain the dilution mass ratio matrix. The number of rows in the matrix is ​​equal to the total number of dilution steps, and the number of columns is equal to the number of components. Each row of the matrix corresponds to a dilution step, and the elements represent the ratio of the mass of each component in the dilution gas added in that step to the total mass of the current parent gas.

[0099] It should be noted that by constructing a dilution mass ratio matrix, accurate mass ratio parameters are provided for each dilution operation. The recursive backpropagation algorithm solves the coupling problem of concentration transfer in multi-level dilution, ensuring that the final concentration accurately meets the target. The dilution mass ratio matrix can directly use matrix elements without repeatedly traversing the tree structure, improving the flexibility and computational efficiency of the calculation process.

[0100] Specifically, based on the dilution mass ratio matrix and the first dilution tree, the cylinder pressure after each step in the dilution process is predicted and analyzed. Steps exceeding the pressure threshold are identified, and pressure balance nodes are inserted at the corresponding positions to obtain the complete dilution tree. The nominal volume, maximum working pressure, and initial pressure and temperature inside the cylinder are obtained. The dilution sequence of each step is read from the first dilution tree, and the ratio of the mass of diluent added in each step to the current mass of the mother gas is read from the dilution mass ratio matrix.

[0101] Pressure simulation is performed sequentially according to the dilution steps. For each step, based on the total mass and composition of the gas in the cylinder, as well as the mass and composition of the diluent added in that step, the new total mass and composition after that step are calculated. The cylinder pressure under this state is calculated using the real gas law, which considers the non-ideal behavior of gas under high pressure and requires iterative solution of the compressibility factor. The calculated pressure of each step is compared with a preset pressure threshold, which is usually set at 80% of the maximum working pressure of the cylinder. If the predicted pressure of a step is less than or equal to the pressure threshold, the simulation continues to the next step. If the predicted pressure is greater than the pressure threshold, a pressure balance node insertion mechanism is triggered. The type of insertion node is determined according to the degree of overpressure. For slight overpressure and the current pressure is not yet close to the pressure threshold, a waiting node is inserted, pausing the filling process and allowing the gas to mix fully and reach temperature equilibrium. After the pressure stabilizes, a new measurement is taken to determine whether to continue. For severe overpressure or the current pressure is close to the pressure threshold, a venting node is inserted. The amount of gas to be vented is calculated to reduce the pressure to a target safe value, such as 90% of the pressure threshold, and venting is performed through reverse operation of the mass flow controller or a dedicated venting valve.

[0102] After inserting the pressure balance node, the structure of the dilution tree is updated. A new virtual node representing the pressure balance state is inserted between the node corresponding to the overpressure step and the next step node. Simultaneously, the dilution mass ratio matrix is ​​adjusted, either splitting an existing dilution step into multiple sub-steps or keeping the dilution mass ratio matrix unchanged, but pausing when encountering a pressure balance node during control execution. This pressure simulation and node insertion process is repeated until all dilution steps are simulated, resulting in a dilution tree containing the pressure balance node. By constructing the dilution tree, potential overpressure risks can be identified before actual execution, shifting safety management from post-processing to pre-emptive prevention. Inserting the pressure balance node provides flexible adjustment methods for pressure control. Waiting nodes utilize the natural process of temperature equilibrium to reduce pressure without gas loss, while venting nodes provide active depressurization to ensure safe reversal even if the pressure is too high. The dilution tree integrates dilution logic, mass ratio, and pressure control information into a complete and executable model, providing accurate structural support for subsequent calculations.

[0103] Furthermore, components in the dilution tree whose gas concentration deviates from the corresponding target concentration by more than a preset deviation threshold are identified, and the gas replenishment amount for the corresponding components is calculated to obtain the gas replenishment task, including:

[0104] S601. Identify components in the dilution tree whose gas concentrations deviate from their corresponding target concentrations by a preset deviation threshold, and obtain the first node set.

[0105] S602. Based on the first node set, calculate the gas replenishment amount of the corresponding component to obtain the gas replenishment task.

[0106] In this embodiment, based on a dilution tree, deviation detection is performed on the completed dilution process to identify component nodes whose concentrations deviate from the target value by more than the allowable range, and the source of the deviation is traced to obtain a first set of nodes. The actual concentration values ​​of each component in the current gas cylinder are obtained using an online gas analyzer. For each node in the dilution tree, the target concentration stored in that node is read; this target concentration is the theoretical concentration value calculated through backpropagation. The actual concentration of each node is compared with the target concentration to calculate the absolute deviation. The calculated absolute deviation value is then compared with a preset deviation threshold, which is dynamically set according to the accuracy requirements of the gas formulation.

[0107] For components whose absolute deviation values ​​exceed the deviation threshold, they are marked as exceeding the deviation limit, and the node containing the component and its deviation value are recorded, resulting in a preliminary set of deviation events. A root cause tracing method is used to traverse upwards along the dilution tree path. Starting from the node where the deviation was detected, the deviation of the same component in the intermediate dilution stage of its parent node is examined. If a deviation also exists in the parent node, it indicates that the problem originates further upstream, and the focus is shifted to the parent node to continue tracing until the source node where the deviation first appeared is found. After tracing and filtering, all source nodes and their deviation-exceeding component information are integrated to obtain the first node set. Each element in the set contains a node identifier, the name of the deviation-exceeding component and its deviation value, the node's path position in the dilution tree, and the node's current concentration vector and target concentration vector.

[0108] It should be noted that by filtering the first set of nodes, the source node of the deviation can be located, which can avoid blind gas replenishment. The first set of nodes provides a clear target range for subsequent gas replenishment calculation, so that the gas replenishment operation can be carried out on the corresponding dilution stage, thereby improving the accuracy and effectiveness of the gas replenishment process.

[0109] Specifically, based on the first node set, the gas replenishment amount for the corresponding component is calculated, resulting in a gas replenishment task. By constructing a gas replenishment task, the mutual interference problem during multi-component coupled gas replenishment can be resolved, ensuring that the concentrations of all relevant components reach the target simultaneously after a single gas replenishment. By combining all nodes on the dilution tree path, it can be guaranteed that the gas replenishment operation not only corrects the deviation of the source node but also ensures that the concentrations of all subsequent nodes meet expectations, thus guaranteeing the stability and convergence of the optimization process. The generation of the gas replenishment sequence takes into account reaction risks, ensuring the safety of the gas replenishment process and improving production efficiency and consistency.

[0110] Furthermore, based on the first node set, the gas replenishment amount for the corresponding components is calculated to obtain the gas replenishment task, including:

[0111] S701. Based on the first node set, analyze the mass ratio of the corresponding dilution gas for each component and calculate the deviation value of the component.

[0112] S702. Calculate the first gas replenishment amount of the corresponding component based on the deviation value, compare the first deviation between the component concentration after gas replenishment and the corresponding target concentration, optimize the first gas replenishment amount by least squares method until the first deviation is less than the preset deviation threshold, and obtain the second gas replenishment amount.

[0113] S703. Analyze the reaction risk level of each component and construct the corresponding gas replenishment sequence;

[0114] S704. Combining the second replenishment amount and the corresponding replenishment sequence, the replenishment task is obtained.

[0115] In this embodiment, based on the first node set, each source node in the set is analyzed to determine the dilution path of the node and its corresponding dilution gas mass ratio relationship, and the component deviation value is calculated. A source node is selected from the first node set, its identification information is obtained, and its position in the sparse tree is traced back from the dilution tree to determine the complete path from the node to the root node and the dilution gas mass ratio used by the node as the mother gas in subsequent dilution steps. All mass ratio data related to the node are extracted from the dilution mass ratio matrix, including the mass ratio of the dilution steps experienced when the node was formed and the mass ratio of the node as the mother gas participating in subsequent dilution steps. The mass ratio reflects the weight and influence range of the node in the overall dilution process.

[0116] Furthermore, the deviation value of each component exceeding the limit at this node is calculated. The deviation value is calculated using absolute deviation, which is the difference between the current actual concentration and the target concentration. The equivalent deviation mass ratio is calculated, which represents the ratio of the mass of this component that needs to be added to increase its concentration from the current value to the target value, assuming that the concentrations of other components remain unchanged, to the total mass of gas in the current cylinder. The equivalent deviation mass ratio is derived from the mass conservation relationship and reflects the severity of the deviation and the workload of gas replenishment. The calculated deviation values, equivalent deviation mass ratios, node information, target concentration, and current concentration are integrated to obtain a deviation parameter table. By calculating the deviation parameters, the concentration deviation is converted into an equivalent deviation mass ratio, providing accurate data support for subsequent gas replenishment calculations. By analyzing the dilution gas mass ratio, the influence of each source node in the dilution tree is clarified, enabling subsequent gas replenishment calculations to be targeted at the dilution stage of the node, providing complete and accurate data support for deviation correction.

[0117] Specifically, based on the deviation parameter table, the first replenishment amount is calculated based on the equivalent deviation mass ratio. The second replenishment amount is obtained through iterative optimization using the least squares method, ensuring that the concentration deviations of all relevant components are less than a preset deviation threshold. For each source node and its deviation-exceeding component in the deviation parameter table, the first replenishment amount is obtained by multiplying the equivalent deviation mass ratio by the total mass of gas in the current cylinder, i.e., directly replenishing the mass required for a single component. After obtaining the first replenishment amount, the corresponding mass of pure component gas is virtually added to the cylinder according to the first replenishment amount. Based on the dilution mass ratio matrix, the calculation is propagated downwards from the source node along the dilution tree towards the root node. At each step, the state of the next node is calculated based on the current node's state and the dilution mass ratio, until the predicted concentration of the root node is calculated.

[0118] Furthermore, the predicted concentrations of all nodes along the path are compared with their respective target concentrations, and the sum of squared deviations is calculated to obtain the first deviation. If the first deviation is less than a preset deviation threshold, the first gas supply amount is used as the second gas supply amount. If the first deviation is greater than the deviation threshold, the least squares method is used, with the gas supply amount of the component exceeding the deviation limit as the optimization variable, and the sum of squared deviations between the predicted concentrations and target concentrations of all nodes along the path from the source node to the root node as the objective function. An iterative algorithm is used to find the optimal gas supply amount that minimizes the objective function. In each iteration, a virtual gas supply simulation is performed based on the current gas supply amount, the objective function value is calculated, and the partial derivatives of the objective function with respect to each gas supply amount, i.e., the Jacobian matrix, are approximated using the numerical difference method. The search direction and step size are determined based on the partial derivatives to update the gas supply amount. This process is repeated until the change in the objective function value between two adjacent iterations is less than a preset convergence threshold. The obtained gas supply amount is the second gas supply amount. The convergence threshold is used to determine whether the least squares iterative optimization has reached a convergence state, and it is usually set based on the accuracy requirements of numerical calculation and the error range allowed by the actual process.

[0119] Preferably, a virtual gas replenishment simulation is performed again to verify whether the optimized gas replenishment amount makes all deviations less than the deviation threshold. If deviations still exceed the limit, the optimization parameters need to be adjusted or multiple rounds of gas replenishment need to be performed. Using the first gas replenishment amount as the initial condition provides a reasonable starting point for the optimization algorithm, accelerating the convergence process. The iterative optimization of the least squares method ensures that the final second gas replenishment amount can make the concentrations of all relevant components simultaneously approach the target value, solving the mutual interference problem when multiple components are coupled for gas replenishment. By predicting the path concentration, the influence of gas replenishment is propagated downstream along the dilution tree, ensuring that the gas replenishment operation not only corrects the deviation of the source node but also ensures that the concentrations of all subsequent dilution stages meet expectations, thus improving the effectiveness of the gas replenishment process.

[0120] Specifically, the order of gas replenishment operations is determined based on the second gas replenishment volume. A list of components corresponding to all non-zero gas replenishment volumes in the second volume is obtained as the gas replenishment component set. Simultaneously, the composition information of the existing gas in the current cylinder is acquired. The gas interaction influence library is called to analyze the reaction risk between any two components in the gas replenishment component set and the reaction risk between each gas replenishment component and the existing gas in the cylinder. For the risk analysis of a component and the existing gas, the highest reaction risk value between that component and each component in the existing gas is taken as the overall risk level. All analysis results are organized into a risk matrix. The rows and columns of the matrix correspond to all components in the gas replenishment component set plus an entity representing the existing gas. Matrix elements represent the risk level of the mixture of the corresponding two entities. Risk levels are represented by integers from 1 to 5, where 1 represents no risk, 2 represents low risk, 3 represents medium risk, 4 represents high risk, and 5 represents extremely dangerous.

[0121] Preferably, a risk threshold is set, typically 3, indicating that component pairs with medium or higher risk should avoid direct contact during gas replenishment. Based on the risk matrix and the risk threshold, the gas replenishment components are sorted, treating them as nodes in a graph. For component pairs with a risk level greater than or equal to the risk threshold, an edge is connected between them to construct a conflict graph. Then, a greedy coloring algorithm is used to color the conflict graph. After sorting the nodes by the number of connected edges from largest to smallest, each node is assigned the smallest available color number, ensuring that this color is different from the colors of all its colored neighboring nodes. Nodes with the same color number belong to the same group and can be replenished in any order because there is no risk conflict between them.

[0122] Specifically, the gas replenishment order is determined by ascending color numbers. For transitions between different color groups, if there is a risk between the last component of the previous group and the first component of the next group, an inert gas isolation step is inserted between the two groups. This isolation step typically involves adding a small amount of argon or nitrogen as a physical barrier. Within the same color group, components with strong adsorption capacity can be replenished first based on calculated adsorption coefficients to preferentially occupy adsorption sites and reduce interference with subsequent components. Using chemical reaction risk as the primary determinant of the gas replenishment order ensures the safety of the replenishment process. A greedy coloring algorithm transforms complex multi-component risk relationships into color groupings, allowing for rapid calculation of the corresponding gas replenishment order. Inserting an isolation step provides a physical barrier, enabling gas combinations that cannot be replenished simultaneously to be safely completed in the same replenishment task.

[0123] Furthermore, based on the second gas replenishment volume and replenishment sequence, an executable gas replenishment task is established and distributed to the executing agency to guide the actual gas replenishment operation. A unique task identifier is assigned to each replenishment task, formatted as gas cylinder identifier + year / month / day + serial number. The task generation time, corresponding cylinder identifier, and current cylinder status are recorded as initial conditions for task execution. The replenishment sequence list is traversed, with each element representing a replenishment component or isolation step, resulting in a detailed record of the replenishment steps.

[0124] For the gas replenishment step, the replenishment mass of the component is obtained from the second replenishment volume. If the same component appears multiple times in the replenishment sequence list, the total mass is allocated to each occurrence. The allocation strategy typically uses proportional allocation, ensuring that each replenishment volume is not less than the minimum controllable flow rate of the mass flow controller. The estimated replenishment time is calculated based on the replenishment mass and the set mass flow rate. The set mass flow rate can be the same as the initial filling value or selected according to the component characteristics. The gas density and molar volume under standard conditions must be considered during the calculation. For the isolation step, a preset isolation replenishment volume is set, which is 0.5% to 1% of the gas cylinder volume, and the estimated replenishment time is calculated. Each step record also includes the step number, step type, component name, estimated waiting time after the replenishment is completed, and the target node identifier corresponding to the replenishment step for quality traceability.

[0125] After all steps are generated, an integrity check is performed to confirm that all components requiring gas replenishment are included in the steps and that the sum of the gas replenishment mass of each component equals the corresponding value in the second gas replenishment quantity. The total pressure after gas replenishment is estimated using the real gas state equation to ensure that it does not exceed the safe pressure threshold. After the check passes, the gas replenishment task is stored in the database in a standard format, and the task status is marked as pending execution. The gas replenishment task is sent to the actuator controller through the communication interface. The controller parses the task and executes it step by step in sequence. During execution, pressure and temperature data are collected in real time to monitor for anomalies. After the task is completed, the execution result is reported and stored in the database along with the planned data to obtain a complete production process record.

[0126] It should be noted that the calculation optimization results are transformed into an executable instruction sequence, providing accurate data support for automated production and task parsing. The setting of the estimated gas replenishment time and waiting time makes the entire gas replenishment process time predictable, which is convenient for production planning. The real-time monitoring and anomaly handling mechanism during task execution ensures timely response in case of unexpected situations, thereby improving the effectiveness of the gas replenishment task.

[0127] Furthermore, by comparing the real-time compensated gas supply with the target concentration in conjunction with the gas supply task, the opening of the mass flow controller is dynamically controlled via a PID controller, including:

[0128] S801. Combine the real-time compensation of the gas supply with the target concentration in conjunction with the gas supply task, analyze the pressure and temperature data in the gas cylinder, and match the initial parameter set of the PID controller for the corresponding component.

[0129] S802: Based on the initial parameter set, the opening degree of the mass flow controller is dynamically controlled by the PID controller.

[0130] In this embodiment, based on the gas replenishment task, specific gas replenishment operations are executed and control parameters are initialized and configured. Information about the current gas replenishment step is read, including the component name, target gas replenishment mass, and set mass flow rate. Simultaneously, the pressure and temperature values ​​inside the gas cylinder are collected in real-time from pressure and temperature sensors installed on the cylinder. During the replenishment process, the real-time compensation gas volume is calculated. As the pressure inside the cylinder continuously increases during actual replenishment, the mass increment generated by the same standard volume flow rate under high pressure gradually weakens. Therefore, dynamic compensation of the target flow rate is required based on the real-time pressure to ensure that the actual gas mass replenished into the cylinder remains consistent with the target value, while simultaneously ensuring that the concentration steadily approaches the target value.

[0131] Based on this, a pre-built PID parameter initial value knowledge base is invoked to match a suitable set of initial parameters for this gas replenishment. This knowledge base stores recommended PID parameters, including proportional, integral, and derivative coefficients, for different gas types and operating conditions. Using the current gas component, current pressure range, and current temperature range as query conditions, nearest neighbor matching is performed in the PID parameter initial value knowledge base to select the record closest to the operating condition. The corresponding PID parameters are then read as the initial parameter set. After matching, parameter fine-tuning is performed based on the specific characteristics of the gas replenishment task, including but not limited to appropriately reducing the integral coefficient to prevent integral saturation when the gas replenishment volume is large, and appropriately increasing the derivative coefficient to suppress flow disturbances caused by adsorption fluctuations when the gas component has strong adsorption properties. By matching suitable initial parameters for different gases and operating conditions, the control performance degradation problem of traditional fixed-parameter PID is avoided. Real-time compensation of the gas replenishment volume allows the control target to be dynamically adjusted according to changes in cylinder pressure, improving the accuracy of the control process.

[0132] Specifically, based on an initial parameter set, the opening of the mass flow controller is dynamically controlled by a PID controller. This allows the mass flow controller to dynamically adjust its parameters according to the real-time control effect, overcoming the limitations of fixed-parameter PID controllers under nonlinear time-varying process characteristics. The incremental PID algorithm ensures smooth changes in the control output, avoiding impact on the valve actuator, improving the accuracy and effectiveness of the control process, and reducing gas waste and replenishment frequency.

[0133] Furthermore, based on the initial parameter set, the opening degree of the mass flow controller is dynamically controlled by a PID controller, including:

[0134] S901. Based on the initial parameter set, analyze the trajectory curve of concentration change with time during the gas replenishment process, calculate the deviation value and the rate of change of the deviation from the target concentration, and obtain the deviation signal and the rate of change of the deviation signal.

[0135] S902. Based on the deviation signal and the deviation change rate signal, calculate the parameter adjustment amount of the PID controller, and dynamically control the opening degree of the mass flow controller through the adjusted PID controller.

[0136] In this embodiment, based on the initial parameter set and the gas replenishment task, the gas replenishment operation is initiated, and the control effect is monitored in real time. Real-time data on the concentration change of the replenished component over time is acquired using an online gas analyzer, constructing a trajectory curve of concentration change over time. The horizontal axis of this curve represents time, and the vertical axis represents the current concentration value. The system employs a first-order low-pass filter to process the original concentration measurements, with the filter coefficient typically between 0.1 and 0.3, filtering out high-frequency noise from the sensor while preserving the true trend of concentration change.

[0137] Based on the filtered concentration trajectory curve, a deviation signal and a deviation change rate signal are calculated in each control cycle. The deviation signal is the difference between the current actual concentration and the target concentration. The sign of this difference indicates whether the concentration is too high or too low, and the absolute value indicates the degree of deviation. During the gas replenishment process, the deviation signal is usually negative and gradually approaches zero. The deviation change rate signal is the rate of change of the deviation over time, that is, the amount of change in the deviation per unit time. It reflects the speed and direction of the concentration approaching the target value. A negative value indicates that the deviation is decreasing, that is, the concentration is approaching the target value, while a positive value indicates that the deviation is increasing, that is, the concentration is moving away from the target value. The absolute value indicates the rate of change.

[0138] After calculating the deviation signal and the deviation change rate signal, both signals are amplitude-limited. The effective range of the deviation signal is set to ±10% of the target concentration, and the effective range of the deviation change rate is ±5% of the target concentration per second. Values ​​exceeding these ranges are considered outliers and an alarm is recorded to prevent subsequent control algorithm failures due to sensor malfunctions. By quantifying the dynamic changes in concentration during the gas replenishment process, the control effect is converted into specific deviation and deviation change rate signals. Filtering effectively suppresses the interference of sensor noise on the control signal, improving the accuracy and reliability of the calculated deviation change rate signal. The deviation signal reflects the difference between the current concentration and the target value, while the deviation change rate signal provides the concentration change trend, allowing for early prediction of the deviation's trajectory.

[0139] Specifically, based on the deviation signal and the deviation change rate signal, the adjustment of the three parameters of the PID controller is calculated in real time using a fuzzy adaptive PID control algorithm, and the adjusted parameters are used to dynamically control the valve opening of the mass flow controller. The deviation signal and the deviation change rate signal are fuzzified and mapped to a predefined fuzzy set. The fuzzy set is described by seven linguistic values: negative large, negative medium, negative small, zero, positive small, positive medium, and positive large. Each linguistic value corresponds to a Gaussian membership function, defining the degree to which the input value belongs to the set. The universe of discourse for the deviation signal is set to ±10% of the target concentration, and the universe of discourse for the deviation change rate signal is set to ±5% of the target concentration per second.

[0140] Furthermore, a pre-built fuzzy rule base is queried. This base contains forty-nine rules corresponding to input combinations. The fuzzy rule base is established based on classic experience in PID parameter tuning and control knowledge accumulated through long-term practical experience by field engineers. Fuzzy inference is performed using the Mandani inference method. For the current input deviation signal and deviation rate of change signal, the membership degree relative to each fuzzy set is calculated. For each rule, the minimum value of the rule's premise is taken as the rule's activation strength. The activation strength is used to truncate the fuzzy set in the rule's conclusion. All truncated fuzzy sets are then superimposed to obtain the total fuzzy output set. Defuzzification is performed using the centroid method. The abscissa value corresponding to the centroid of the membership function of the fuzzy output set is calculated as the precise PID parameter adjustment amount. This adjustment amount is multiplied by a scaling factor to convert it into a value in actual physical dimensions. The scaling factor is set according to the actual value range of the PID parameters.

[0141] The proportional, integral, and derivative coefficient adjustments obtained from defuzzification are added to the current PID parameters, and the updated parameters are limited to ensure they remain within a preset reasonable range. For example, the proportional coefficient ranges from 0.1 to 2.0, the integral coefficient from 0.01 to 0.5, and the derivative coefficient from 0 to 0.5. An incremental PID algorithm is used to calculate the control output. Using the deviation signals of the current cycle, the previous cycle, and the cycle before that as inputs, and combining them with the updated proportional, integral, and derivative coefficients, the increment of the control quantity is calculated. This increment is accumulated and added to the control output of the previous cycle to obtain the valve opening command for the current cycle. After limiting, this command is sent to the mass flow controller via the analog output module to adjust the valve opening. This process is repeated in each control cycle to achieve online adaptive adjustment of the PID parameters until the cumulative replenished gas mass reaches the target value.

[0142] It should be noted that, through the online adaptive tuning of the PID controller, the controller can automatically adjust its own parameters according to the real-time control effect to overcome the limitations of fixed parameter PID under the nonlinear time-varying characteristics of the process. The use of incremental PID algorithm makes the control output change smoothly and avoids the control quantity jump caused by parameter mutation. The parameter limiting mechanism can ensure that the parameter adjustment is always within the safe range, which enhances the robustness of the system and improves the control accuracy and production efficiency.

[0143] For example, a semiconductor factory needs to prepare a mixed gas for an oxidation diffusion process. The target formula is: silane concentration 2.00%, phosphine concentration 0.10%, and argon as a balance gas concentration of 97.90%. The gas cylinder is a 40-liter aluminum alloy cylinder with a maximum working pressure of 15 MPa. After the operator inputs the gas formula table through the human-machine interface, the system starts the dynamic calculation engine.

[0144] The system invokes the gas interaction library to analyze the reaction between components. The query results show that silane and phosphine do not react violently at room temperature, but both are flammable gases and should be kept away from oxidizers. The current initial pressure of the gas cylinder is 0.1 MPa (argon protection), and the temperature is 22°C. The system calculates the filling quantities based on real-time environmental data: considering adsorption losses, 238.5 g of silane, 9.6 g of phosphine, and 9250 g of argon are required. Based on the reaction risk level and adsorption competition, the filling order is determined to be silane first, then phosphine, and finally argon, generating a filling list.

[0145] After the list filling process is complete, allow it to stand for 20 minutes. The online laser Raman spectrometer detects the actual concentrations as follows: silane 1.97%, phosphine 0.09%, and argon 97.94%. The concentration deviations of silane and phosphine both exceed the preset thresholds (±0.02% and ±0.005%, respectively). The system constructs a dilution tree, identifies argon as the diluent, calculates the dilution mass ratio matrix, and predicts that the pressure after gas replenishment will reach 14.1 MPa, exceeding the safe pressure threshold of 12 MPa. It then automatically inserts a pressure balance node between gas replenishment steps (waiting 3 minutes).

[0146] The gas replenishment analysis module calculates the initial gas replenishment amount: 2.3g of silane and 0.8g of phosphine. Virtual simulation revealed that simultaneous replenishment would cause the argon concentration to drop to 97.85%, so the least squares method was used for optimization. After 5 iterations, the second gas replenishment amount was obtained: 2.2g of silane and 0.7g of phosphine. The replenishment order was determined based on risk analysis: phosphine was replenished first, followed by silane at pressure equilibrium nodes.

[0147] During gas replenishment, the system matches the initial PID parameters (phosphine: proportional coefficient Kp=0.8, integral coefficient Ki=0.15, derivative coefficient Kd=0.05; silane: proportional coefficient Kp=1.2, integral coefficient Ki=0.2, derivative coefficient Kd=0.1) based on the current pressure of 13.8MPa and temperature of 23℃. Flow feedback is collected every 100ms, and the deviation and rate of change of deviation are calculated. The valve opening is adjusted in real time using fuzzy adaptive PID. Ultimately, 0.69g of phosphine and 2.18g of silane were replenished, and the final concentrations all met the standards: silane 2.00%, phosphine 0.10%, and argon 97.90%. The product was then accepted into the warehouse.

[0148] like Figure 3 As shown, a fabrication process control system based on a dynamic computing engine is used to implement a fabrication process control method based on a dynamic computing engine, including:

[0149] The fill list construction module responds to the gas formula table input by the user, analyzes the reaction between the components in the gas formula table through a preset gas interaction influence library, calculates the filling amount and filling order of each component to achieve the target concentration by combining real-time gas cylinder environmental data, and generates a fill list.

[0150] The dilution tree construction module, based on a filled list, analyzes the concentration changes of each component in the gas formulation table, calculates the mass ratio of the corresponding dilution gas, and constructs a dilution tree;

[0151] The gas replenishment analysis module identifies components in the dilution tree whose gas concentration deviates from the corresponding target concentration by more than a preset deviation threshold, calculates the gas replenishment amount for the corresponding components, and obtains the gas replenishment task.

[0152] The preparation process control module compares the real-time compensated gas supply with the target concentration in conjunction with the gas supply task, and dynamically controls the opening of the mass flow controller through a PID controller to control the gas formulation preparation process.

[0153] The above description is merely a preferred embodiment of this application. The scope of protection of this application is not limited to the above embodiments. All technical solutions falling within the scope of this application's concept are within the scope of protection of this application. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of this application should also be considered within the scope of protection of this application.

Claims

1. A preparation process control method based on a dynamic computing engine, characterized in that, include: In response to the gas recipe table input by the user, the system analyzes the reaction between the components in the gas recipe table through a preset gas interaction influence library, and calculates the filling amount and filling order of each component to achieve the target concentration by combining real-time gas cylinder environmental data, and generates a filling list. Based on the filled list, analyze the concentration changes of each component in the gas formulation table, calculate the mass ratio of the corresponding dilution gas, and construct a dilution tree; Identify components in the dilution tree whose gas concentration deviates from the corresponding target concentration by more than a preset deviation threshold, calculate the gas replenishment amount for the corresponding components, and obtain the gas replenishment task. By combining the real-time compensated gas supply with the target concentration in conjunction with the gas supply task, the opening degree of the mass flow controller is dynamically controlled by the PID controller to control the gas formulation preparation process.

2. The preparation process control method based on a dynamic computing engine according to claim 1, characterized in that, The system responds to the user-input gas formula table, analyzes the reactions between the components in the gas formula table using a preset gas interaction influence library, and calculates the filling amount and filling order of each component to achieve the target concentration by combining real-time gas cylinder environmental data, generating a filling list, including: In response to the gas recipe table input by the user, the reaction between the components in the gas recipe table is analyzed through a preset gas interaction influence library to obtain a set of reaction parameters for each component; Based on the set of reaction parameters and real-time gas cylinder environmental data, the filling amount and filling order of each component to achieve the target concentration are calculated, and a filling list is generated.

3. The preparation process control method based on a dynamic computing engine according to claim 2, characterized in that, The process involves calculating the filling amount and filling order of each component to achieve the target concentration based on the set of reaction parameters and real-time gas cylinder environmental data, and generating a filling list, including: Based on the set of reaction parameters, calculate the reaction heat values ​​and product types between components, and construct the reaction matrix between component reactions; Based on the reaction matrix and combined with real-time gas cylinder environmental data, the adsorption rate and desorption rate of each component on the inner wall of the gas cylinder are calculated to obtain the corresponding adsorption coefficient. Based on the reaction matrix and adsorption coefficient, the filling amount and filling order of each component to achieve the target concentration are calculated, and a filling list is generated.

4. The preparation process control method based on a dynamic computing engine according to claim 1, characterized in that, The process of analyzing the concentration changes of each component in the gas formulation table based on a filled list, calculating the mass ratio of the corresponding dilution gas, and constructing a dilution tree includes: Based on the filled list, the concentration changes of each component in the gas formulation table are analyzed, the subordinate relationship between the dilution gas and the corresponding component is identified, and a dilution relationship topology diagram is obtained. Based on the dilution relationship topology, calculate the mass ratio of the corresponding dilution gas and construct a dilution tree.

5. The preparation process control method based on a dynamic computing engine according to claim 4, characterized in that, The step of calculating the mass ratio of the corresponding dilution gas and constructing a dilution tree based on the dilution relationship topology graph includes: The dilution relationship topology is matched with the corresponding filling order. Component nodes whose order in the dilution relationship topology does not conform to the filling order are identified and their positions are optimized to obtain the first dilution tree. For each component node in the first dilution tree, calculate the mass ratio of the corresponding dilution gas and construct the corresponding dilution mass ratio matrix; Based on the dilution mass ratio matrix, the cylinder pressure is analyzed, and pressure balance nodes are inserted between component nodes whose cylinder pressure is greater than a preset pressure threshold during the dilution process to obtain a dilution tree.

6. The preparation process control method based on a dynamic computing engine according to claim 1, characterized in that, The process involves identifying components in the dilution tree whose gas concentration deviates from the corresponding target concentration by a preset deviation threshold, calculating the gas replenishment amount for the corresponding components, and obtaining the gas replenishment task, including: Identify components in the dilution tree whose gas concentration deviates from the corresponding target concentration by a preset deviation threshold, and obtain the first node set; Based on the first node set, calculate the gas replenishment amount of the corresponding component to obtain the gas replenishment task.

7. The preparation process control method based on a dynamic computing engine according to claim 6, characterized in that, The step of calculating the gas replenishment amount of the corresponding component based on the first node set to obtain the gas replenishment task includes: Based on the first node set, analyze the mass ratio of the corresponding dilution gas for each component and calculate the deviation value of the component; Based on the deviation value, the first gas replenishment amount of the corresponding component is calculated. The first deviation between the component concentration after gas replenishment and the corresponding target concentration is compared. The first gas replenishment amount is optimized by the least squares method until the first deviation is less than the preset deviation threshold, and then the second gas replenishment amount is obtained. Analyze the reaction risk of each component to determine the corresponding gas replenishment sequence; By combining the second replenishment amount and the corresponding replenishment sequence, the replenishment task is obtained.

8. The preparation process control method based on a dynamic computing engine according to claim 1, characterized in that, The step of combining the real-time compensated gas supply with the target concentration and dynamically controlling the opening of the mass flow controller via a PID controller includes: By combining the real-time compensated gas supply with the target concentration in conjunction with the gas supply task, the pressure and temperature data inside the gas cylinder are analyzed, and the initial parameter set of the PID controller for the corresponding component is matched. Based on the initial parameter set, the opening degree of the mass flow controller is dynamically controlled by a PID controller.

9. The preparation process control method based on a dynamic computing engine according to claim 8, characterized in that, The method of dynamically controlling the opening of the mass flow controller using a PID controller based on an initial parameter set includes: Based on the initial parameter set, the trajectory curve of concentration change over time during the gas replenishment process is analyzed, and the deviation value and the rate of change of the deviation from the target concentration are calculated to obtain the deviation signal and the rate of change of the deviation signal. Based on the deviation signal and the deviation change rate signal, the parameter adjustment of the PID controller is calculated, and the opening degree of the mass flow controller is dynamically controlled by the adjusted PID controller.

10. A manufacturing process control system based on a dynamic computing engine, characterized in that, The method for controlling the preparation process based on a dynamic computing engine as described in any one of claims 1 to 9 includes: The fill list construction module responds to the gas formula table input by the user, analyzes the reaction between the components in the gas formula table through a preset gas interaction influence library, calculates the filling amount and filling order of each component to achieve the target concentration by combining real-time gas cylinder environmental data, and generates a fill list. The dilution tree construction module, based on a filled list, analyzes the concentration changes of each component in the gas formulation table, calculates the mass ratio of the corresponding dilution gas, and constructs a dilution tree; The gas replenishment analysis module identifies components in the dilution tree whose gas concentration deviates from the corresponding target concentration by more than a preset deviation threshold, calculates the gas replenishment amount for the corresponding components, and obtains the gas replenishment task. The preparation process control module compares the real-time compensated gas supply with the target concentration in conjunction with the gas supply task, and dynamically controls the opening of the mass flow controller through a PID controller to control the gas formulation preparation process.

Images