Online concentration detection method and system for electronic-grade hydrofluoric acid production

Abstract

The invention relates to the technical field of electronic chemical engineering, and discloses an on-line concentration detection method and system for electronic-grade hydrofluoric acid production. The system comprises a real-time concentration data acquisition unit for acquiring real-time concentration data through a sensor array and constructing a concentration kinetic model; the concentration energy function calculation unit is used for calculating a concentration energy function representing concentration deviation energy in combination with real-time data and a kinetic model; the production system equation construction unit is used for constructing a system concentration control equation based on each node model and an energy function; the expected concentration function generating unit is used for generating an expected concentration function describing node concentration errors and relative errors in combination with the model and a preset target; the concentration control algorithm unit is used for calculating a controller equation and a control law equation of each node by utilizing an optimization algorithm; and the control execution unit adjusts the production flow concentration in real time according to the control law equation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electronic chemical engineering, in particular to an online concentration detection method and system for electronic-grade hydrofluoric acid production. BACKGROUND

[0002] In the production process of electronic-grade hydrofluoric acid, the concentration index is directly related to the product quality and the reliability of subsequent electronic device manufacturing. The production process involves multiple consecutive production nodes, and the concentration states between nodes affect each other, forming a complex dynamic correlation system. Traditional concentration detection methods rely mainly on offline sampling analysis, that is, samples are manually taken from each production node at regular intervals and sent to the laboratory for concentration measurement using chemical analysis or precision instruments. This method has significant time lag, and it often takes tens of minutes or even hours from sample collection, transportation to analysis completion. During this period, the concentration in the production process may have changed significantly, resulting in detection results that cannot reflect the real-time production state in a timely manner, making it difficult to provide timely basis for production adjustment.

[0003] Traditional detection methods cannot achieve global concentration monitoring of the entire production system. Since each production node is independently sampled, there is a lack of analysis of the correlation between the nodes, and it is difficult to build a complete production system concentration dynamic model. When an abnormality occurs in the concentration of a node, it is difficult to quickly locate the root cause of the abnormality and predict the impact on other nodes, which can easily trigger a chain reaction, leading to an increase in the unqualified rate of product batches. In addition, the concentration control in traditional production is mainly based on the experience of operators for manual adjustment, and the adjustment strategy lacks the support of scientific mathematical models, making it difficult to achieve precise control according to the dynamic changes of the production system, often resulting in over-adjustment or under-adjustment, which not only affects the stability of production, but also may cause waste of raw materials and increase of energy consumption.

[0004] With the rapid development of the electronic information industry, the purity and concentration stability of electronic-grade hydrofluoric acid are increasingly stringent, and traditional detection and control methods cannot meet the production needs of high precision and high stability. Although some online detection devices have been applied to a single production node in the industry, these devices can only obtain real-time concentration data of a single node, and cannot achieve integrated analysis of multi-node data and system-level concentration modeling. There is also a lack of optimization control algorithm based on global concentration state, making it difficult to fundamentally solve the problem of global concentration dynamic regulation of the production system. Therefore, developing an online concentration detection system that can realize multi-node real-time concentration detection, system-level modeling and precise control has become a problem to be solved in the field of electronic-grade hydrofluoric acid production. SUMMARY

[0005] The purpose of the present application is to provide an online concentration detection method and system for electronic-grade hydrofluoric acid production to solve the problems raised in the background.

[0006] To achieve the above object, the application provides an online concentration detection system for electronic-grade hydrofluoric acid production, which comprises:

[0007] A real-time concentration data acquisition unit is configured to acquire real-time concentration data in the electronic-grade hydrofluoric acid production process through a sensor array and construct a concentration kinetic model according to the real-time concentration data.

[0008] A concentration energy function calculation unit is configured to calculate a concentration energy function of the electronic-grade hydrofluoric acid according to the real-time concentration data and the concentration kinetic model; the concentration energy function is used to represent the deviation energy of the current concentration state from the target concentration.

[0009] A production system equation construction unit is configured to construct a concentration control equation of the electronic-grade hydrofluoric acid production system based on the concentration kinetic model and the concentration energy function of all production nodes.

[0010] A desired concentration function generation unit is configured to generate a desired concentration function of the electronic-grade hydrofluoric acid production system according to the concentration kinetic model of all production nodes and a preset target concentration state; the desired concentration function is used to describe the concentration error of each production node and the relative concentration error between nodes.

[0011] A concentration control algorithm unit is configured to calculate a production system controller equation by using an optimization algorithm according to the concentration control equation and the desired concentration function, and obtain a control law equation of each production node based on the production system controller equation.

[0012] A control execution unit is configured to perform real-time concentration adjustment on the electronic-grade hydrofluoric acid production process by using the control law equation for each production node.

[0013] Preferably, the concentration kinetic model is as follows:

[0014] The real-time concentration data includes the current concentration value, flow value and temperature value of the electronic-grade hydrofluoric acid; the concentration kinetic model is composed of a concentration change rate equation, a mass balance equation and a control input equation; the concentration change rate equation represents the change of concentration with time, the mass balance equation represents the mass relationship of inflow and outflow, and the control input equation represents the input control of additives.

[0015] Preferably, when the concentration energy function calculation unit calculates the concentration energy function, the concentration energy function is calculated based on the square deviation of the current concentration value and the target concentration value and the integral deviation of the flow value; the higher the value of the concentration energy function, the greater the concentration deviation.

[0016] The concentration energy function is constructed based on the square deviation of the current concentration value and the target concentration value and the integral deviation of the flow value; the higher the value of the concentration energy function, the greater the concentration deviation.

[0017] Preferably, the production system equation construction unit constructs the concentration control equation when:

[0018] The concentration control equation is constructed based on a simultaneous form of the concentration kinetic model and a partial derivative form of the concentration energy function; the concentration control equation includes interconnection terms and damping terms for representing concentration coupling relationships between production nodes.

[0019] Preferably, the desired concentration function generation unit generates the desired concentration function when:

[0020] The desired concentration function is constructed based on a difference between a current concentration state of each production node and the target concentration state and a concentration difference between adjacent production nodes; the desired concentration function adjusts error importance of different nodes through a weight coefficient matrix.

[0021] Preferably, the concentration control algorithm unit calculates the production system controller equation when:

[0022] The optimization algorithm is a gradient descent algorithm for minimizing the desired concentration function; the production system controller equation is obtained by solving a matching equation that associates the concentration control equation with the desired concentration function.

[0023] Preferably, the control execution unit applies the control law equation when:

[0024] The control law equation includes a concentration adjustment term, a flow compensation term, and a temperature correction term; the concentration adjustment term is calculated based on a concentration error, the flow compensation term is calculated based on a flow deviation, and the temperature correction term is calculated based on a temperature change.

[0025] Preferably, the system further comprises:

[0026] A concentration risk analysis unit for calculating a concentration deviation risk value of electronic grade hydrofluoric acid production based on the real-time concentration data and historical concentration data; the concentration deviation risk value is used to evaluate the probability of concentration exceeding a safe range;

[0027] A risk response unit for generating a plurality of concentration adjustment schemes and calculating an interference impact value of each concentration adjustment scheme on the production state when the concentration deviation risk value is abnormal;

[0028] An optimal scheme selection unit for selecting a concentration adjustment scheme with the smallest interference impact value as the best concentration adjustment scheme and updating the production control strategy.

[0029] Preferably, the concentration risk analysis unit calculates the concentration deviation risk value when:

[0030] The concentration deviation risk value is based on the deviation of the current concentration from the safety threshold, the real-time flow and temperature fluctuation of the production process; when the concentration deviation risk value is abnormal, it indicates that the concentration may be out of control.

[0031] When the risk response unit generates a concentration adjustment scheme:

[0032] Each concentration adjustment scheme includes an additive injection amount, a flow adjustment value and a temperature setting value; the interference influence value is calculated based on the production interruption time and resource consumption when the scheme is executed; the optimal concentration adjustment scheme is screened through an iterative optimization algorithm.

[0033] Preferably, the application also includes an online concentration detection method for electronic-grade hydrofluoric acid production, which includes all modules and method processes of the above-mentioned online concentration detection system for electronic-grade hydrofluoric acid production.

[0034] Compared with the prior art, the application has the following beneficial effects:

[0035] The online concentration detection system for electronic-grade hydrofluoric acid production realizes real-time acquisition and dynamic modeling of the concentration of each node in the production process through the real-time concentration data acquisition unit, changing the hysteresis problem of traditional offline detection. The sensor array can continuously acquire concentration data of each production node, ensuring the continuity and timeliness of data acquisition, and the concentration kinetic model based on real-time data can accurately depict the law of concentration change of each node with time and production conditions, providing a dynamic perspective for subsequent concentration analysis and control, so that the concentration state in the production process is always under real-time monitoring, avoiding production misjudgment caused by data lag.

[0036] The concentration energy function calculation unit converts the concentration deviation into a quantifiable energy function by combining real-time concentration data and the concentration kinetic model, which can intuitively and accurately represent the deviation degree of the current concentration state and the target concentration. This quantitative method eliminates the subjectivity of traditional empirical judgment, making the evaluation of concentration deviation more scientific. The operator can quickly grasp the overall concentration state of the current production system through the concentration energy function, determine the direction and priority of concentration adjustment, and provide clear basis for the formulation of subsequent control strategies.

[0037] The production system equation construction unit constructs a concentration control equation covering the entire production process based on the kinetic models of all production nodes and the concentration energy function, realizing the integration of the global concentration correlation of the production system. This equation can reflect the influence of concentration changes at each production node on the overall system concentration state, breaking the limitations of traditional single-node detection, allowing operators to understand the concentration dynamic change law from a system perspective. When a node has a concentration anomaly, the potential impact of the anomaly on other nodes can be quickly analyzed through the control equation, making it easier to solve the problem from the root cause and reduce the occurrence of chain reactions.

[0038] The expected concentration function generation unit combines the node kinetics model with the preset target concentration, considers not only the concentration error of a single node but also the relative concentration error between nodes, and can build a concentration control target that better meets the actual production needs. This multi-dimensional error consideration ensures that the concentrations of each node meet their own target requirements while maintaining coordination with other nodes, avoiding production process imbalance due to excessive concentration differences between nodes, and further improving the stability and consistency of the entire production system concentration.

[0039] The concentration control algorithm unit uses an optimization algorithm to generate production system controller equations and node control law equations based on the concentration control equation and the expected concentration function, providing scientific mathematical support for the concentration control strategy. This algorithm can automatically optimize control parameters based on the dynamic changes of the production system, avoiding the randomness and instability of traditional manual experience adjustment, achieving precise regulation of each production node concentration, making concentration adjustment more targeted and effective, reducing over-adjustment or under-adjustment, and ensuring the stability of the production process.

[0040] The control execution unit performs real-time concentration adjustment based on the node control law equation, ensuring the rapid implementation of the control strategy. This unit can directly act on the production process, automatically adjusting production parameters based on the control law equation without human intervention, improving the response speed of concentration adjustment, reducing human error, and enabling the production system to maintain within the target concentration range, which helps to improve product quality stability, reduce substandard products caused by concentration fluctuations, and reduce raw material waste and energy consumption, improving overall production efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 A timing diagram for the online concentration detection system for electronic-grade hydrofluoric acid production described in the present application;

[0042] Figure 2 A working principle diagram of the concentration kinetics model;

[0043] Figure 3 A working principle diagram of the concentration control equation. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the present application.

[0045] Please refer toFigure 1 The present application provides an online concentration detection method and system for electronic-grade hydrofluoric acid production, which comprises: collecting concentration data streams of multiple key nodes in the electronic-grade hydrofluoric acid production process in real time through a high-frequency sensor array, and constructing a concentration kinetic model reflecting dynamic characteristics based on the data; further quantifying the deviation degree between the current production state and the target concentration by calculating the concentration energy function, thereby establishing a numerical representation of the deviation energy. Based on the kinetic model and energy function of all production nodes, the system constructs a system-level concentration control equation that covers the coupling relationship of multiple nodes, which integrates local dynamics and global constraints; combined with the expected concentration function representing the ideal concentration distribution, a controller equation that can achieve multi-objective optimization is generated, and from which the distributed control law of each production node is derived. Finally, the real-time concentration adjustment of the entire production process is realized through the actuator, forming a closed-loop control system from data acquisition to control execution.

[0046] Embodiment 1: refer to Figure 2 The sensor array is deployed at key nodes of the electronic-grade hydrofluoric acid production system, including the inlet and outlet of the reaction kettle, the liquid phase section of the rectifying tower, and the circulating pipeline of the storage tank. These sensors use high-frequency sampling electrochemical probes and electromagnetic flowmeters to synchronously capture instantaneous values of concentration, flow rate, and temperature. The data is transmitted to the preprocessing module through industrial Ethernet, filtered by sliding average and outliers, and then converted to standardized engineering unit dimensions. The concentration kinetic model is composed of three types of core equations: the concentration change rate equation describes the differential relationship between concentration, flow rate, and temperature variables per unit time, and its structure covers convection, diffusion, and reaction source terms; the mass balance equation integrates the conservation relationship between inlet flow rate, outlet flow rate, and container retention based on the principle of fluid continuity, and considers the mass disturbance caused by phase change and additive injection; the control input equation quantifies the adjustment behavior as controllable parameters, such as acid liquid supplement valve opening and cooling water flow rate set value, and is related to the concentration change rate through a transfer function. Model parameters are determined by a combination of offline calibration and online identification. In the offline stage, step response tests are used to obtain system gain and time constant, and in the online stage, recursive least squares method is used to update model coefficients in real time to adapt to the characteristic fluctuations of production materials.

[0047] The calculation of the concentration energy function relies on the real-time data and the output of the kinetic model. The function is designed to represent the deviation energy between the current state and the target state of the system, and its value directly reflects the control quality of the production process. In the calculation process, first, the concentration sampling value at the current time is obtained, and the square deviation is calculated with the preset target concentration value. At the same time, the historical deviation sequence of the current flow and the set flow is integrated. After the weighted combination of the two types of deviations, the energy function value is formed. The weighting coefficient is dynamically adjusted according to the production stage. For example, in the start-up stage, larger transient deviation is allowed, so the weight is reduced, while in the steady-state operation stage, the weight is increased to strengthen the control accuracy. The update frequency of the energy function is synchronized with the sensor sampling, ensuring that it can capture the rapid disturbance in the production process, such as sudden changes in feed concentration or temperature abnormalities caused by heat exchanger fouling. In addition, the function value is normalized to match the output range of the controller input range, which is convenient for subsequent analytical calculation of the control law.

[0048] To enhance the robustness of the function, a sliding time window mechanism is introduced in the calculation process, only using the data sequence within a certain time to evaluate the energy, avoiding the continuous influence of historical abnormal values. At the same time, the energy function output is compared with the safety threshold, when exceeding the limit value, the early warning signal is triggered and sent to the risk analysis unit for subsequent processing. The sensor array provides high-precision and synchronous real-time data stream, the concentration kinetic model converts physical laws into mathematical expressions, and the energy function quantifies the degree of system deviation from the target. This integrated implementation method lays the information foundation for multi-node coordinated control, and its output results are directly used for the construction and optimization of subsequent system-level control equations. In implementation, attention is paid to the calculation efficiency of the algorithm, and incremental calculation and matrix decomposition techniques are used to avoid repeated operations, so that the function update delay control is in the millisecond level, meeting the stringent requirements of real-time control for electronic-grade hydrofluoric acid production.

[0049] With the rectification section of an electronic grade hydrofluoric acid production as an example, the real-time concentration data unit is obtained by installing three groups of sensor arrays on the overhead line, the tower kettle reboiler circulation pipeline and the outlet of the feed preheater to collect real-time production data. Each group of sensor arrays contains an electrochemical concentration sensor, an electromagnetic flow sensor and a platinum resistance temperature sensor to measure the concentration value, flow value and temperature value synchronously at a sampling frequency of 10 times per second. The tower top sensor records the concentration value of 50.15%, the flow value of 352 L / min and the temperature value of 68.3℃, the tower kettle sensor obtains the concentration value of 48.92%, the flow value of 418 L / min and the temperature value of 72.1℃, and the feed sensor detects the concentration value of 49.63%, the flow value of 378 L / min and the temperature value of 70.2℃. These data are transmitted to the data preprocessing module through industrial Ethernet, converted to standard engineering unit dimensions after sliding average filtering and abnormal value elimination. The concentration kinetic model is constructed in a multi-equation combination manner: the concentration change rate equation describes the relationship between the concentration of each node and time, in which the tower top equation focuses on the dynamic of light component enrichment, the tower kettle equation emphasizes the concentration characteristics of heavy components, and the feed equation represents the mixing effect; the mass balance equation is based on the principle of mass conservation, calculates the balance relationship of the inflow and outflow materials, considers the mass transfer caused by phase change and the disturbance caused by additive injection; the control input equation quantifies the operation variables as executable parameters, including the reboiler steam regulating valve opening, the reflux pump frequency and the feed preheating power, etc. The model parameters are determined by combining offline calibration and online identification, in the offline stage, the step response test is used to obtain the system gain and time constant, in the online stage, the model coefficients are updated in real time by recursive least squares method to adapt to the fluctuations of raw material concentration and the changes of equipment state.

[0050] The concentration energy function calculation unit performs operations based on real-time data and kinetic model outputs. The function is constructed using the squared deviation integral of the current concentration value and the target concentration value, with the target concentration at the top of the tower being 50.0%, at the bottom being 49.0%, and at the feed being 49.5%. The flow deviation integral term is included in the calculation process, and the weight coefficient is dynamically adjusted according to the production stage: it is set to 0.7 in the steady-state operation stage and adjusted to 0.4 in the transition stage. The function value is updated every 100 milliseconds, and the output range is controlled between 0-1 through normalization processing. The higher the value, the farther the concentration deviates from the target. When the instantaneous fluctuation of the top concentration is detected to exceed 0.3%, the system automatically increases the flow weight coefficient to 0.9, strengthening the contribution of flow stability in the energy function. A sliding time window mechanism is introduced in the calculation process, only using the data sequence of the last 30 seconds for energy evaluation, avoiding the continuous influence of historical abnormal values. The energy function output is compared with the safety threshold, and when the function value exceeds 0.8, a warning signal is triggered, and the real-time data are sent to the risk analysis unit for subsequent processing. The entire implementation process uses a distributed computing architecture, with each node running local calculations independently, and the central coordinator synchronizes global data every 500 milliseconds to ensure the consistency of the system state. The algorithm implementation uses fixed-point arithmetic and matrix pre-decomposition techniques to control the calculation delay within 5 milliseconds, meeting the real-time requirements of high-precision production.

[0051] Example 2: refer to Figure 3 The concentration control equation is constructed using the multi-body system modeling method, treating each production node as a dynamic subsystem. The mass matrix describes the inertia characteristics of the concentration change in each node, the stiffness matrix reflects the concentration's own recovery ability, and the damping matrix is used to represent the dissipation effect within the system. The interconnection term explicitly expresses the mutual influence of concentration changes between nodes, such as the bidirectional coupling relationship between the top and bottom concentrations of the rectifying tower, and the transmission effect of the outlet concentration of the reaction kettle on the downstream storage tank. These interconnection relationships are embedded in the system equation through partial differential operators, forming a mathematical description of the distributed parameter system. The damping term in the equation is designed using the Rayleigh damping model, whose coefficients are associated with node flow and container volume, which can effectively suppress concentration oscillation caused by additive pulse injection or feed fluctuation, ensuring the smoothness of system response. The derivation of the entire concentration control equation is based on the Lagrangian mechanics framework, with the gradient of the concentration energy function as the generalized force input, thereby establishing a direct link between system dynamics and energy change.

[0052] The generation of the desired concentration function is based on the difference between the preset target concentration state and the real-time state of each node. This function not only includes the absolute deviation of the concentration of a single node from the target value, but also introduces a relative concentration deviation term between adjacent nodes to characterize the coordination of the concentration distribution in the production process. The absolute deviation term is quantified by a quadratic cost function, and the coefficient matrix is configured according to the production process requirements. For example, higher weights are assigned to the final product storage tank nodes to ensure the quality of the finished product, while the restrictions on intermediate reaction nodes are appropriately relaxed. The relative deviation term focuses on node pairs that have direct material exchange, such as the concentration gradient constraints between serial reactors. The calculation uses a norm form to measure the magnitude of the concentration difference, avoiding control blind spots caused by the cancellation of positive and negative deviations. The design of the weight coefficient matrix uses a hierarchical strategy. The basic weights are determined by the process design manual, reflecting the importance level of different nodes in the mass balance. The dynamic adjustment layer is self-adaptively corrected based on real-time production data. For example, when a sensor detects an abnormal temperature in a node, the weight coefficients of its adjacent nodes are automatically increased to prevent chain reactions. The matrix update algorithm is based on Lyapunov stability theory, ensuring that the weight changes do not compromise the controllability of the system. The calculation of the entire desired concentration function uses a distributed optimization structure. Each node controller calculates the local error term in parallel, and the central coordinator aggregates global information and generates a system-level target function, thereby balancing computational efficiency and control consistency. In the implementation process, attention is paid to the numerical realizability of the equations. The concentration control equation is discretized into algebraic equations by finite difference, and implicit integration method is used to ensure numerical stability. The discrete time step is synchronized with the sensor sampling period. The weight matrix in the desired concentration function is stored and operated through sparse matrix technology, reducing the computational complexity to adapt to the processing capacity of industrial controllers. The outputs of the two types of equations are interacted in real time through the data bus, forming a closed-loop optimization framework.

[0053] Taking a specific electronic-grade hydrofluoric acid production as an example, the rectification section as the key production node, its concentration control involves the coordinated regulation of three sub-nodes of tower top production, tower bottom reflux and intermediate feeding. The production system equation construction unit first collects the real-time data of each node: the tower top concentration sensor reads 50.2%, the tower bottom concentration is 48.8%, the feeding concentration is 49.5%, the flow sensor records the tower top production 350L / min, the tower bottom reflux 420L / min, the feeding amount 380L / min, the temperature sensor shows the tower top 68.5℃, the tower bottom 72.3℃, and the feeding temperature 70.1℃. These data are input into the pre-established concentration kinetic model of each node. The tower top model describes the light component concentration dynamic, the tower bottom model describes the heavy component enrichment process, and the feeding model represents the mixing effect. Each model includes a concentration change rate equation, a mass balance equation, and a control input equation, which is associated with the reboiler steam valve opening, the reflux pump frequency, and other actuator parameters.

[0054] When constructing the system-level concentration control equation, the unit adopts the node coupling method to establish the matrix form of the differential equation: the mass matrix is calculated from the liquid holdup of each tower section, reflecting the inertia characteristics of concentration change; the stiffness matrix is derived from the difference in component volatility, representing the self-regulating ability of the system; the damping matrix is determined according to the fluid viscosity and tray efficiency, used to suppress fluctuations. The interconnection term explicitly depicts the material exchange between nodes: changes in tower top production affect the tower bottom liquid level, tower bottom reflux adjustment changes the tower top pressure, and feed fluctuations simultaneously disturb the upper and lower tower sections; these coupling relationships are embedded in the equation through partial differential operators, forming a distributed parameter system covering three nodes. The equation also introduces a damping term based on the energy function, whose coefficient is related to the flow rate and container volume of each node. When the rate of change of the tower top concentration is detected to exceed 0.5% / min, the damping effect is automatically enhanced to prevent oscillation of the tower bottom concentration.

[0055] The desired concentration function generation unit receives the target concentration set value: 50.0% at the tower top, 49.0% at the tower bottom, and 49.5% at the feed, while obtaining real-time concentration state data. The function calculation contains two levels: the absolute deviation term calculates the square of the difference between the current concentration of each node and the target, where the weight coefficient of the tower top node is set to 0.6 (because the product is output from here), and the weight coefficients of the tower bottom and feed nodes are 0.2 and 0.2 respectively; the relative deviation term calculates the concentration difference between adjacent nodes, including the tower top-feed and feed-tower bottom two groups of differences, with a weight coefficient of 0.3. The weight matrix adopts a dynamic adjustment strategy: when the feed flow fluctuation is detected to exceed 10%, the feed node weight is automatically increased to 0.4, and the weights of other nodes are correspondingly reduced; when the tower bottom temperature changes drastically, the weight of the tower top-tower bottom relative deviation is increased to 0.4 to strengthen the temperature compensation effect. The function output is aggregated by the central coordinator to form a multi-objective optimization function that considers both single-point precision and overall balance. Each node controller performs local operation in parallel, with the tower top node focusing on purity control, the tower bottom node focusing on heavy component concentration stability, and the feed node coordinating the material balance of the upper and lower sections; the central coordinator synchronizes global data every 200 milliseconds to correct model mismatch caused by transmission delay. Sparse matrix storage is used in the calculation to reduce the memory occupation of the embedded system; the Gauss-Seidel algorithm is used for iterative solution to ensure a feasible solution under limited computing resources. The final generated concentration control equation and the desired concentration function are transmitted to the downstream control unit through industrial Ethernet.

[0056] Embodiment 3: The concentration control algorithm unit receives the concentration control equation and the desired concentration function from upstream, where the concentration control equation is a multivariable differential equation describing the relationship between the system concentration dynamics and the input and output constraints, and the desired concentration function defines the performance index that the system needs to minimize; the optimization algorithm uses the gradient descent method, the core of which is to iteratively adjust the control variables along the negative gradient direction of the objective function to gradually approach the optimal solution that minimizes the desired function, and the step size parameter is adjusted adaptively according to the Lyapunov exponent of the system during the iteration process, which ensures the convergence speed and avoids oscillation. The solution of the controller equation is achieved by constructing a matching equation, which establishes a mathematical relationship between the system dynamics and the optimization objective, and its form is:

[0057]

[0058] where: is the matching condition residual, is the Hamiltonian function, is the control input vector, is the co-state variable vector, is the system dynamics function; the matching equation is solved using the prediction-correction algorithm, the prediction step estimates the control variable change direction based on the current state, and the correction step updates the optimization trajectory by backtracking through the co-state equation, finally obtaining the analytical form of the controller equation that satisfies the matching condition.

[0059] The calculation of the control law equation is based on the analytically obtained controller equation, and its output includes three components: concentration adjustment term, flow compensation term and temperature correction term; the concentration adjustment term is calculated by proportional-integral-derivative algorithm, the proportional term handles the current concentration error, the integral term accumulates historical deviation to eliminate steady-state error, and the differential term predicts the trend to enhance response speed, and the gain coefficients of each term are adjusted online according to the critical proportionality of the system. The flow compensation term is designed as a feedforward structure, which calculates the required compensation flow value to offset the disturbance by monitoring the deviation between the flow sensor data and the set value, combining the pipeline flow resistance characteristics and the fluid inertia parameters; the temperature correction term establishes a coupling model between temperature and concentration changes, converts temperature fluctuations into equivalent concentration deviation through thermodynamic equations, and adjusts the additive injection strategy accordingly to correct the term coefficients. The control execution unit converts the control law equation into specific actuator instructions, the concentration adjustment term is output to the acid liquid metering pump to adjust the injection rate, the flow compensation term operates the regulating valve to change the opening degree, and the temperature correction term adjusts the heat exchanger power set value; the instruction output adopts an incremental release mode, only adjusting the relative value each time to avoid drastic fluctuations in the production process, and all execution operations are subjected to change rate constraints and amplitude limiting processing to ensure equipment safety and production stability.

[0060] On each production node, the system independently performs local control law calculation, ensuring that each node can autonomously and efficiently handle its own control tasks. The central coordinator focuses on synchronizing global parameters and checking the consistency of each node's calculation results, ensuring the coordination and stability of the entire production system. The calculation period is strictly synchronized with the sensor data collection process, ensuring that the entire process from data input to control instruction output is strictly controlled within milliseconds, meeting the strict real-time requirements of high-precision production. Fixed-point arithmetic techniques are used to significantly improve computational efficiency and reduce resource consumption on embedded platforms. Matrix pre-decomposition techniques are also applied to further optimize the calculation process and improve execution speed. To maintain excellent performance in scenarios requiring high-precision calculations, the system also retains a floating-point backup algorithm as a backup solution for high-precision mode. Through this design, the system forms a hybrid computing strategy that balances speed and accuracy, ensuring fast response in the production process while ensuring calculation accuracy under high-precision requirements, thereby improving the overall performance of the production system.

[0061] Taking an electronic-grade hydrofluoric acid rectification tower concentration control scenario as an example, the concentration control algorithm unit receives system data from upstream: concentration control equations containing the coupling relationship of the tower top, tower bottom, and feed nodes, as well as the expected concentration function composed of absolute and relative deviations. The optimization algorithm uses the adaptive gradient descent method, with a 200-millisecond period for synchronizing sensor data, an initial step size of 0.1, and dynamic adjustment based on the Lyapunov index. When the tower top concentration change rate exceeds 0.3% / s, the step size is automatically reduced to 0.05 to avoid overshooting, and when the system tends to be stable, it is restored to 0.1 to accelerate convergence. The core of the algorithm is to search for the optimal solution along the negative gradient direction of the expected function. Each iteration calculates the partial derivative of the objective function with respect to the control variables, which include the reboiler steam valve opening, reflux pump frequency, and feed preheater power. The solution of the controller equation is achieved by constructing a matching equation that establishes a mathematical relationship between system dynamics and optimization objectives. The solution process uses a two-step prediction-correction method: the prediction step estimates the future state trajectory based on the current control variables, and the correction step retraces the optimized trajectory and corrects the control variables based on the common state equation. The initial value of the common state variable is set to the inverse of the eigenvalue of the system Jacobian matrix, which is updated after each iteration. After 5-6 iterations, the matching condition residual converges below the threshold, and the analytical form of the controller equation is output, which contains 9 gain coefficients (corresponding to the coupling relationship of three control variables and three state variables).

[0062] The control law equation is expanded into three execution components according to the analytical results: the concentration adjustment term adopts an improved PID structure, the proportional term coefficient is determined by 0.6 times the critical gain of the system, the integral term coefficient is related to the integral quantity of the concentration deviation, and the differential term coefficient is dynamically adjusted according to the temperature gradient of the tray. The flow compensation term is designed as a feed-forward-feedback composite structure, the feed-forward part is based on the deviation between the real-time data of the electromagnetic flowmeter and the set value, and is converted into a compensation amount through the pipe flow resistance coefficient; the feedback part adjusts the compensation value through the change rate of the tower kettle liquid level. The temperature correction term establishes a temperature-concentration transfer model to convert the reboiler temperature fluctuation into an equivalent concentration deviation, and the correction coefficient is updated once a week according to the heat exchanger fouling coefficient. The control execution unit converts the control law into specific instructions: the concentration adjustment term outputs a 4-20 mA signal to the hydrofluoric acid metering pump, with a regulation accuracy of 0.1 mL / min; the flow compensation term controls the opening degree of the regulating valve through the PROFIBUS bus, with a step size of 0.5%; the temperature correction term adjusts the heating power of the reboiler through the Modbus protocol, with a regulation granularity of 0.5 kW. The instruction issuing adopts an incremental mode, which only allows the valve opening degree to change by no more than 2% and the power adjustment to change by no more than 5 kW per cycle, and all operations are processed through a rate limiter and an amplitude limiter. After execution, the system enters a monitoring state, and the tower top concentration spectrometer data is collected every 100 milliseconds, and if the deviation expands continuously for three cycles, the re-optimization process is triggered.

[0063] Example 4: Quantitative evaluation of concentration deviation risk and corresponding adjustment scheme generation and selection process, which extends the basic control function to deal with abnormal working conditions in the production process. In a specific electronic-grade hydrofluoric acid production scenario, when the concentration sensor at the outlet of the rectifying tower detects that the concentration value continuously deviates from the set range, the concentration risk analysis unit starts comprehensive evaluation: the unit collects the current concentration data (such as a deviation of 49.8% from the target 50.0%), real-time flow data (fluctuation range ±5%), temperature data (heater outlet temperature change rate 0.5℃ / min), and combines the past 24 hours of historical data (including concentration standard deviation 0.15%, flow range 8L / min, temperature drift record), calculates the concentration deviation risk value through the Bayesian probability model; the risk value calculation sets a safety threshold of 49.5%-50.5%, when the relative deviation of the real-time concentration and the threshold exceeds 30% and the temperature fluctuation intensifies, the system determines that the risk level rises to "abnormal".

[0064] After receiving the high-risk signal, the risk response unit generates three alternative concentration adjustment schemes: scheme A focuses on the rapid correction of the additive injection amount, scheme B emphasizes the smooth transition of flow adjustment, and scheme C adopts the strategy of temperature and flow coordination adjustment. The specific parameters of each scheme are shown in Table 1.

[0065] Table 1: Concentration adjustment scheme parameters and impact evaluation table

[0066] Scheme No. Adjustment amount of additive injection (mL / min) Flow adjustment value (L / min) Temperature set value correction (°C) Production interruption time estimation (min) Resource consumption index Comprehensive interference impact value A +15.0 -2.0 0.0 3.5 0.78 0.67 B +5.0 -4.5 -1.2 8.2 0.42 0.59 C +8.5 -3.0 -0.8 5.0 0.51 0.48

[0067] The calculation of the interference impact value is based on a multi-dimensional evaluation model: the production interruption time is derived from the historical operation log of similar adjustments required for equipment response time and process stabilization time, the resource consumption index comprehensively calculates the additional additive dosage, energy consumption (pump and heater power change) and potential material loss (such as the amount of unqualified products generated during the transition phase), and finally the weighted normalization algorithm is used to obtain the comprehensive evaluation value in the range of 0-1. The optimal solution selection unit adopts the Pareto optimal decision algorithm, compares the impact values of each scheme and identifies non-dominated solutions, and scheme C is selected as the best scheme due to the lowest comprehensive impact value; the production stage constraints (such as currently in the middle of batch production, it is not appropriate to use drastic adjustment) are also considered during the selection process, and finally the parameter combination of scheme C is executed.

[0068] The system implements gradual adjustment during the implementation of scheme C: first, gradually increase the additive injection amount by 8.5 mL / min within 2 minutes, simultaneously reduce the circulation flow by 3 L / min, and decrease the temperature set value by 0.2℃ every 30 seconds until the cumulative decrease is 0.8℃; all operations are issued step by step through the distributed control system, and after each step is executed, the next step is adjusted after waiting for sensor feedback confirmation. At the same time, the risk analysis unit continuously monitors the concentration change trend, updates the risk value every 10 seconds, and when the concentration returns to 49.9% and the risk value decreases to the normal range, the system automatically switches back to the normal control mode, and all operation data of this event are stored in the case library for subsequent risk model optimization. The whole implementation process embodies the closed-loop control logic from risk perception, scheme generation to optimal decision, and balances the demand for adjustment effect and production stability through multi-objective optimization.

[0069] Example 5: The concentration risk analysis unit processes historical concentration data with a sliding time window mechanism, with the window width set to the last 15 batches of operation records according to the production rhythm, each batch containing a complete concentration profile curve from feeding to discharging. In the data preprocessing stage, the original concentration values are aligned and interpolated to eliminate the asynchronous problem caused by differences in sampling frequency. Subsequently, a probability distribution model of concentration values is constructed based on the kernel density estimation method to describe the likelihood of different concentration values. The calculation of risk values integrates real-time sensor readings and statistical model outputs. In the specific calculation, a dynamic adjustment mechanism of the safety threshold is introduced. The threshold not only considers the static range specified in the process card (such as 49.5%-50.5%), but also introduces environmental temperature, raw material purity, and other correction factors, so that the safety boundary floats with the production conditions. Real-time flow fluctuations are incorporated into the risk model by calculating the coefficient of variation, and temperature fluctuations are characterized by the first-order difference value to represent the degree of change. The final risk value is obtained by weighted fusion of the probability deviation, flow stability, and temperature change rate, and its numerical range is normalized to 0-1. The risk judgment section adopts a double-threshold triggering strategy. When the risk value exceeds 0.7 for 3 consecutive samplings, a primary alarm is triggered, and the monitoring is enhanced, but the operation parameters are not adjusted yet. When the risk value breaks through 0.85 with abnormal flow fluctuations, the system determines that it is in an abnormal state and starts the scheme generation process. When generating the concentration adjustment scheme, the risk response unit first establishes a multi-objective optimization framework containing decision variables, constraint conditions, and objective functions. Decision variables include additive injection amount, flow adjustment value, and temperature setting value, and constraint conditions come from equipment capacity limits (such as the maximum injection rate of the pump and the adjustment range of the valve) and process safety boundaries (such as the maximum allowed temperature change rate). The scheme generation adopts the orthogonal experimental design method to select representative level value combinations in the definition domain of each decision variable to form an initial scheme set. For example, the additive adjustment amount selects-10%, 0%, and +10% levels, the flow adjustment selects-5%, 0%, and +5% levels, and the temperature correction selects-1℃, 0℃, and +1℃ levels. Through full combination, 27 candidate schemes are generated.

[0070] The interference impact value of each scheme is calculated by the multi-attribute utility theory, and the evaluation dimensions include three indexes: production interruption time, energy consumption increment and material loss amount. The production interruption time is estimated according to the device response delay and process stabilization time under similar operation conditions in the historical database, the energy consumption increment is calculated by the product of the power change and the duration of the pump, heater and other devices, and the material loss amount is calculated by the empirical model of the unqualified product rate in the transition stage. After normalization, the weight coefficients of each index are assigned, the time weight is 0.5, the energy consumption weight is 0.3, and the material weight is 0.2. The weighted sum is obtained to get the comprehensive impact value in the range of 0-1. The optimal scheme selection uses an iterative optimization algorithm, the first round of screening eliminates schemes that violate hard constraints (such as temperature change exceeding the limit), the second round selects the Pareto optimal solution set based on the impact value ranking, and finally the scheme with the minimum comprehensive impact value is determined as the execution object through the satisfaction decision rule. After the scheme is executed, the system enters the effect tracking mode, the concentration value is collected every 30 seconds and the risk value is recalculated, if the risk value shows a downward trend for 3 consecutive samplings, the scheme re-optimization process is triggered, and the adjustment scheme is generated again using the current production state as the initial point, forming a closed-loop risk control.

[0071] While embodiments of the present application have been shown and described with reference to particular embodiments thereof, it will be understood by those skilled in the art that various changes in form and details can be made therein without departing from the spirit and scope of the application. The scope of the application is defined by the appended claims and their equivalents.

Claims

1. An online concentration detection system for the production of electronic-grade hydrofluoric acid, characterized in that, include: A real-time concentration data acquisition unit is used to collect real-time concentration data in the electronic-grade hydrofluoric acid production process through a sensor array, and to construct a concentration kinetic model based on the real-time concentration data. The concentration energy function calculation unit is used to calculate the concentration energy function of electronic-grade hydrofluoric acid based on the real-time concentration data and the concentration kinetic model; the concentration energy function is used to characterize the energy deviation between the current concentration state and the target concentration. The production system equation construction unit is used to construct the concentration control equations for the electronic-grade hydrofluoric acid production system based on the concentration kinetic model and concentration energy function of all production nodes. The desired concentration function generation unit is used to generate the desired concentration function of the electronic-grade hydrofluoric acid production system based on the concentration kinetic model of all production nodes and the preset target concentration state; the desired concentration function is used to describe the concentration error of each production node and the relative concentration error between nodes. The concentration control algorithm unit is used to calculate the production system controller equation using an optimization algorithm based on the concentration control equation and the desired concentration function, and to obtain the control law equation for each production node based on the production system controller equation. The control execution unit is used to adjust the concentration of electronic-grade hydrofluoric acid in real time for each production node using the control law equation.

2. The online concentration detection system for electronic-grade hydrofluoric acid production according to claim 1, characterized in that, The concentration kinetic model is as follows: The real-time concentration data includes the current concentration, flow rate, and temperature of electronic-grade hydrofluoric acid; the concentration kinetic model consists of a concentration change rate equation, a mass balance equation, and a control input equation; the concentration change rate equation represents the change in concentration over time, the mass balance equation represents the mass relationship between inflow and outflow, and the control input equation represents the input control of the additive.

3. The online concentration detection system for electronic-grade hydrofluoric acid production according to claim 2, characterized in that, When the concentration energy function calculation unit calculates the concentration energy function: The concentration energy function is constructed based on the squared deviation between the current concentration value and the target concentration value and the integral deviation of the flow rate value; the higher the value of the concentration energy function, the greater the concentration deviation.

4. The online concentration detection system for electronic-grade hydrofluoric acid production according to claim 1, characterized in that, When the production system equation building unit constructs the concentration control equation: The concentration control equation is constructed based on the simultaneous form of the concentration kinetic model and the partial derivative form of the concentration energy function; the concentration control equation includes interconnection terms and damping terms, which are used to characterize the concentration coupling relationship between production nodes.

5. The online concentration detection system for electronic-grade hydrofluoric acid production according to claim 4, characterized in that, When the desired concentration function generation unit generates the desired concentration function: The desired concentration function is constructed based on the difference between the current concentration state and the target concentration state of each production node, as well as the concentration difference between adjacent production nodes; the desired concentration function adjusts the error importance of different nodes through a weighting coefficient matrix.

6. The online concentration detection system for electronic-grade hydrofluoric acid production according to claim 5, characterized in that, When the concentration control algorithm unit calculates the controller equation of the production system: The optimization algorithm is a gradient descent algorithm, used to minimize the desired concentration function; the production system controller equation is obtained by solving a matching equation, which associates the concentration control equation with the desired concentration function.

7. The online concentration detection system for electronic-grade hydrofluoric acid production according to claim 6, characterized in that, When the control execution unit applies the control law equation: The control law equation includes a concentration adjustment term, a flow compensation term, and a temperature correction term; the concentration adjustment term is calculated based on the concentration error, the flow compensation term is calculated based on the flow deviation, and the temperature correction term is calculated based on the temperature change.

8. The online concentration detection system for electronic-grade hydrofluoric acid production according to claim 1, characterized in that, Also includes: The concentration risk analysis unit is used to calculate the concentration deviation risk value of electronic-grade hydrofluoric acid production based on the real-time concentration data and historical concentration data. The concentration deviation risk value is used to assess the probability that the concentration exceeds the safe range; The risk response unit is used to generate multiple concentration adjustment schemes when the concentration deviates abnormally from the risk value, and to calculate the interference impact value of each concentration adjustment scheme on the production status. The optimal solution selection unit is used to select the concentration adjustment scheme with the least interference impact as the best concentration adjustment scheme and update the production control strategy.

9. The online concentration detection system for electronic-grade hydrofluoric acid production according to claim 8, characterized in that, When the concentration risk analysis unit calculates the concentration deviation risk value: The concentration deviation risk value is calculated based on the deviation between the current concentration and the safety threshold, as well as the real-time flow rate and temperature fluctuations in the production process; an abnormal concentration deviation risk value indicates that the concentration may be out of control. When the risk response unit generates a concentration adjustment plan: Each concentration adjustment scheme includes the additive injection amount, flow rate adjustment value, and temperature setpoint; the interference impact value is calculated based on the production interruption time and resource consumption during scheme execution; the optimal concentration adjustment scheme is selected through an iterative optimization algorithm.

10. An online concentration detection method for the production of electronic-grade hydrofluoric acid, characterized in that, It includes all modules and method flows of the online concentration detection system for the production of electronic-grade hydrofluoric acid as described in any one of claims 1 to 9.

Images