Intelligent gas purity monitoring system

Abstract

The invention belongs to the technical field of gas monitoring, and discloses an intelligent gas purity monitoring system, which comprises a multi-dimensional sensing unit, a gas regulation execution unit and a man-machine interaction unit, and is characterized in that the multi-dimensional sensing unit is used for collecting gas purity and related process parameters; the gas regulation execution unit executes gas purity regulation and process coordination action; the man-machine interaction unit is used for parameter input and state monitoring; the MPC control unit is used for constructing a'purity-pressure-flow-temperature 'multivariable coupling relationship in the gas purification process into a dynamic state space model capable of being used for online prediction; on this basis, a rolling optimization and feedback correction mechanism of a model prediction control algorithm is introduced, purity change trends at multiple moments in the future are used as optimization targets, an optimal multivariable cooperative adjustment instruction is calculated and executed in advance, and the model precision problem under small sample interference data is solved.

Description

Technical Field

[0001] This invention belongs to the field of gas monitoring and control technology, and in particular relates to an intelligent gas purity monitoring system. Background Technology

[0002] In high-end fields such as semiconductor wafer manufacturing and aseptic pharmaceutical production, gas purity is a core parameter determining product quality and production safety. For example, electronic-grade nitrogen purity below 99.999% can lead to chip oxidation defects, while insufficient purity of medical oxygen directly endangers patient safety. Existing gas purity monitoring and control systems mainly face three major technical bottlenecks:

[0003] Firstly, the monitoring and control links are separated. After the sensor collects data, it requires manual or simple threshold triggering of adjustment commands. The system response lag is generally more than 10 seconds, making it difficult to suppress rapid purity fluctuations and easily exceeding the process threshold of ±0.001%.

[0004] Secondly, the control algorithm is not intelligent enough. It generally adopts traditional methods such as PID or fuzzy control, which can only perform single-variable adjustment (such as only controlling the valve opening). It cannot effectively handle the dynamic coupling relationship between multiple variables such as "purity-pressure-flow-temperature". When the operating conditions such as intake flow fluctuate, the control accuracy can drop by more than 40%.

[0005] Third, it lacks predictive and adaptive capabilities. The system only reacts to the current deviation and cannot predict changes in purity based on historical data and operating trends. Furthermore, it cannot cope with slow time-varying interferences such as adsorbent performance decay and changes in ambient temperature, resulting in a lack of emergency buffer space when sudden purity exceeds the standard. Summary of the Invention

[0006] To address the aforementioned issues and overcome the shortcomings of existing technologies, this invention provides an intelligent gas purity monitoring system. The purpose of this invention is to solve problems such as slow response, weak multivariate coupling processing capability, and lack of interference prediction capability in existing gas purity control systems. It achieves ultra-high precision and rapid response for stable purity control. The system innovatively constructs a dynamic state-space model that can be used for online prediction by coupling the multivariate relationships of "purity-pressure-flow-temperature" in the gas purification process. Based on this, a rolling optimization and feedback correction mechanism of the model predictive control algorithm is introduced. The purity change trend at multiple future moments is used as the optimization target, and the optimal multivariate collaborative adjustment command is calculated and executed in advance. Addressing the challenge of model accuracy under small sample interference data, this invention adopts a modeling method that integrates mechanism and data-driven approaches. Historical interference data is used to identify and adaptively correct the parameters of the mechanism model, significantly improving the model's generalization prediction capability under unknown disturbances. This solution not only achieves precise control of purity within ±0.0005%, but also shortens the response time to less than 2 seconds when interference occurs, providing an intelligent solution for ensuring gas purity in high-end manufacturing.

[0007] The technical solution adopted in this invention is as follows: an intelligent gas purity monitoring system, comprising a multi-dimensional sensing unit, a gas regulation execution unit, an MPC control unit, and a human-machine interaction unit;

[0008] Multi-dimensional sensing unit collects gas purity and related process parameters, including purity C, pressure P, flow rate Q, and temperature T;

[0009] The gas regulation actuator receives control commands and adjusts the opening degree of the electric valve, the air intake flow rate, and the temperature of the adsorption layer.

[0010] The MPC control unit includes a dynamic prediction model module and an MPC optimization module;

[0011] The dynamic prediction model module, based on real-time data collected by multi-dimensional sensing units, employs a mechanism-driven and data-driven fusion method to construct a dynamic prediction model for the gas purification process: C(k+1) = f(C(k), P(k), Q(k), T(k)). Here, C(k+1) represents the model's predicted purity value for the next time step based on the above inputs; k represents the discrete-time index; C(k) represents the real-time measured gas purity value at the current moment; P(k) represents the inlet pressure, Q(k) represents the inlet flow rate, and T(k) both represent the adsorption layer temperature; f(·) represents the prediction model.

[0012] The MPC optimization module, based on the output of the dynamic prediction model, uses a model predictive control algorithm with multivariable constraints and objective function of "minimum purity deviation + minimum control energy consumption" to perform rolling time-domain optimization and generate cooperative control commands.

[0013] The human-machine interaction unit is used for setting target purity, monitoring operating status, alarms, and data traceability.

[0014] Furthermore, the construction process of the dynamic prediction model module specifically includes: first, establishing a state-space equation framework based on the adsorption and purification mechanism; second, using historical operating data containing multiple interference scenarios, correcting the time-varying parameters of the model online through a system identification method; and finally, compensating for the model prediction deviation through an error correction network to form a fusion model.

[0015] Furthermore, the rolling optimization process of the MPC optimization module specifically includes: in each control cycle, starting from the current measurement value, predicting the purity change for the next Np steps using a dynamic prediction model; aiming at minimizing the weighted sum of future purity tracking error and control quantity change, and under multiple constraints such as purity, valve opening, and flow rate, solving for the optimal control sequence for the next Nc steps through quadratic programming; executing only the optimal control quantity at the current moment, and repeating this process in the next cycle.

[0016] By adopting the above solution, the beneficial effects achieved by the present invention are as follows:

[0017] (1) This invention constructs a high-precision dynamic prediction model for gas purification, and for the first time applies the model predictive control algorithm to the field of ultra-high purity gas control. The model accurately depicts the dynamic coupling relationship between "purity-pressure-flow-temperature", fundamentally solving the technical bottleneck of the traditional single-variable control method under multi-variable disturbances, realizing multi-dimensional collaboration from perception to execution, and laying the model foundation for precise control under complex working conditions.

[0018] (2) Furthermore, the rolling optimization and multi-constraint processing mechanism proposed in this invention enables the control system to have the ability to "look ahead". The system can not only react to the current purity deviation, but also predict the purity trend in the next few seconds based on the model, and calculate the optimal adjustment strategy that takes into account both tracking accuracy and actuator smoothness in advance. Compared with the lag response of traditional PID, this invention can start the coordinated adjustment 2 seconds in advance when facing typical disturbances such as sudden changes in intake flow, suppressing purity fluctuations within the threshold, and significantly improving the anti-interference ability and control quality of the system.

[0019] (3) In terms of engineering practicality, the mechanism and data-driven fusion modeling method adopted in this invention effectively overcomes the risk of prediction failure of pure data-driven models in small sample and new interference scenarios. Through continuous learning and correction of mechanism model parameters by historical interference data, the system has a strong adaptability and generalization capability under working conditions. At the same time, the constraint processing function integrated into the MPC framework naturally incorporates engineering constraints such as actuator physical limit and process safety range into the optimization consideration, ensuring the reliability and safety of the system in long-term operation, and finally achieving a leapfrog improvement in purity control accuracy and system comprehensive performance. Attached Figure Description

[0020] The accompanying drawings are provided to further understand the present invention and form part of the specification. They are used together with the embodiments of the present invention to explain the invention and do not constitute a limitation thereof.

[0021] Figure 1 This is a flowchart of the dynamic prediction model parameter adaptation and MPC rolling optimization proposed in this invention. Detailed Implementation

[0022] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0023] Example 1:

[0024] Combination Figure 1 The present invention provides an intelligent gas purity monitoring system, comprising: a multi-dimensional sensing unit, a gas regulation execution unit, an MPC control unit, and a human-machine interaction unit.

[0025] The multi-dimensional sensing unit includes a laser gas analyzer, a pressure sensor, a flow sensor, a temperature sensor, and a data preprocessing module. The laser gas analyzer is fixed to the main gas delivery pipeline and detects the purity C of the target gas in real time. The pressure sensor and flow sensor are installed at the inlet of the purification tower to collect the inlet pressure P and flow rate Q, respectively. The temperature sensor is embedded inside the adsorption layer of the purification tower to collect the temperature T of the adsorption layer. The data preprocessing module uses a Kalman filter algorithm to filter and reduce noise from multiple sources and synchronizes them to a unified timestamp.

[0026] The gas regulation execution unit includes an electric regulating valve, a flow regulating valve, a temperature control module, and a backup purification branch; the electric regulating valve is installed at the outlet of the purification tower to regulate the product gas output ratio; the flow regulating valve is installed in the inlet pipeline to stabilize the inlet gas load; the temperature control module is used to regulate the adsorption layer temperature to optimize adsorption performance; the backup purification branch is connected in parallel with the main pipeline for emergency switching.

[0027] The MPC control unit includes a data receiving module, a dynamic prediction model module, an MPC optimization module, and an instruction output module. The dynamic prediction model module receives synchronous data, constructs and updates the prediction model online using a fusion method of mechanism and data-driven approaches, the MPC optimization module performs rolling optimization based on the model predictions, and the instruction output module converts the optimization results into control signals.

[0028] The human-machine interaction unit includes a touch screen, an alarm module, and a data traceability module, which are used for parameter setting, status monitoring, alarm functions, and historical data query.

[0029] Example 2:

[0030] This embodiment, based on Embodiment 1, details the construction process of the dynamic prediction model module:

[0031] Step M1 (Mechanism Modeling): Based on the physicochemical mechanisms of purification processes such as pressure swing adsorption / membrane separation (e.g., mass conservation, adsorption isotherms, mass transfer rate equations), establish a nonlinear state-space equation framework with key states of the adsorption bed (e.g., adsorption amount) as hidden states and [P(k), Q(k), T(k), C(k)] as inputs and outputs.

[0032] Step M2 (Data-driven correction): Import historical operating data (covering various disturbance conditions), and use system identification techniques such as recursive least squares with forgetting factor to identify and update time-varying parameters related to adsorbent efficiency and environmental disturbance in the mechanism model online, forming parameter adaptive rules.

[0033] Step M3 (fusion and compensation): Design a lightweight neural network as a real-time error corrector. The corrector takes the short-term prediction residual sequence of the mechanism model and the current working condition as input and outputs dynamic compensation amount to finally form a high-precision fusion model: C(k+1)=f(C(k),P(k),Q(k),T(k)), with a prediction error ≤0.0003%.

[0034] Example 3

[0035] This embodiment, based on Embodiment 2, details the data-driven adaptive process in step M2:

[0036] Step M20 (Data Preprocessing and Feature Vector Construction):

[0037] To achieve parallel analysis and clustering of parameters under multiple operating conditions, it is necessary to transform the original time series data with different dimensions into standardized feature vectors.

[0038] The specific process is as follows:

[0039] 1. Data normalization: Extract the pressure sequence {P}, flow rate sequence {Q}, temperature sequence {T}, and purity sequence {C} under the synchronous timestamp from the historical database.

[0040] 2. Sliding window segmentation: Set a fixed-length sliding time window (e.g., 5 minutes) to divide continuous time series data into multiple overlapping or non-overlapping data segments. Each segment contains sampled values ​​of all parameters within a window.

[0041] 3. Feature Extraction and Standardization: For each data segment, calculate the statistical characteristics of each parameter (such as mean, standard deviation, and slope of change over a certain time period). Then, use the max-min normalization method to map each statistical characteristic value to the interval [0,1]. For the i-th data segment, its normalization formula is: x norm =(xx) min ) / (x max -x min );

[0042] Where x is the original feature value, x min and x max These are the minimum and maximum values ​​of this feature across the entire historical dataset.

[0043] 4. Constructing a feature vector: Concatenate all normalized statistical features (e.g., normalized pressure mean, flow rate standard deviation, temperature change slope, etc.) within the same data segment in a fixed order to form a multidimensional feature vector V. i This vector V i This characterizes the overall operating condition of the system within that time period, and the values ​​of each dimension are comparable.

[0044] Step M21 (Feature Vector-Based Condition Clustering): For the feature vector set {V} constructed from all data fragments... i Unsupervised clustering analysis (such as using the K-means algorithm) is performed to automatically divide the system into multiple working condition clusters.

[0045] Step M22 (Parameter Learning): For the data within each working condition cluster, an optimal set of model parameters {θ} is obtained through offline training. cluster}, construct a mapping library for "operating condition = cluster - parameter".

[0046] Step M23 (Online Matching): During system operation, the distance between the current [P(k),Q(k)] and the center of each working condition cluster is calculated in real time, and the best matching model parameter set is dynamically loaded to ensure that the prediction model always adapts to the current operating state and improves the generalization ability.

[0047] Example 4

[0048] This embodiment, based on Embodiment 1, details the working logic of the fault diagnosis and emergency response module in the human-computer interaction unit:

[0049] Step F1 (Multi-dimensional monitoring): Continuously monitor 1) the feasibility of the MPC optimizer (whether it is unsolvable / divergent); 2) whether the dynamic model prediction residuals exceed the confidence interval; 3) whether the physical logic between multi-sensor readings is consistent (e.g., the relationship between flow rate and pressure).

[0050] Step F2 (Graded Emergency Response): Set up a three-level response mechanism. When the purity deviation > 0.0008%, a Level 1 response is triggered, and the MPC optimization module automatically adjusts the objective function weights, prioritizing purity. When the purity deviation > 0.001% or optimization fails, a Level 2 response is triggered, immediately switching to the backup purification branch and controlling the main valve to a safe opening. When a sensor hard fault is detected or the Level 2 response is invalid, a Level 3 response is triggered, executing a system-wide safety interlock shutdown and triggering the highest-level audible and visual alarm, while simultaneously recording a fault snapshot for analysis.

[0051] Example 5 (Typical Application Test)

[0052] This embodiment, based on Embodiment 1, demonstrates the system's performance in supplying electronic-grade nitrogen (target purity 99.9995%) to a semiconductor factory.

[0053] Stable operating conditions: Under stable intake conditions, the system maintains a purity of 99.9995% ± 0.0002% within 30 seconds, with stable valve opening and no overshoot.

[0054] Anti-interference test: Simulate flow surge at 100 seconds (25→38m) 3 Within 0.1 seconds, the system predicted that the purity would fall below the lower limit. The MPC immediately coordinated the adjustment of the outlet valve (increased opening), the inlet valve (slightly closed), and the temperature (increased), and pulled the purity back to 99.9994% in the 102nd second, with no alarms for exceeding the limit throughout the process.

[0055] Long-term performance and energy efficiency: After 72 hours of continuous operation, the purity standard deviation is only 0.00015%. Thanks to the optimized scheduling of MPC, the number of valve actions is reduced by about 40%, and the overall energy consumption is reduced by about 15% compared with the traditional PID system.

Claims

1. An intelligent gas purity monitoring system, characterized in that: It includes a multi-dimensional sensing unit, a gas regulation execution unit, and a human-machine interaction unit. The multi-dimensional sensing unit collects gas purity and related process parameters; the gas regulation execution unit performs gas purity regulation and process coordination actions; the human-machine interaction unit is used for parameter input and status monitoring; and it also includes an MPC control unit. The MPC control unit includes a dynamic prediction model module and an MPC optimization module; The dynamic prediction model module constructs a dynamic prediction model for the gas purification process based on real-time data collected by multi-dimensional sensing units. The MPC optimization module, based on the output of the dynamic prediction model, uses the model predictive control algorithm to perform rolling time-domain optimization and generate cooperative control commands.

2. The intelligent gas purity monitoring system according to claim 1, characterized in that: The dynamic prediction model module is constructed using a fusion method of mechanism and data-driven approaches.

3. The intelligent gas purity monitoring system according to claim 2, characterized in that: The MPC optimization module uses a constrained quadratic programming algorithm to solve for the optimal control sequence.

4. The intelligent gas purity monitoring system according to claim 3, characterized in that: The process of constructing a dynamic prediction model using a fusion approach of mechanism and data-driven methods includes the following steps: Step M1: Based on the mass conservation and adsorption kinetics mechanism of the gas purification process, a preliminary state-space equation is constructed with the inlet pressure P(k), flow rate Q(k), adsorption layer temperature T(k), and current purity C(k) as inputs; Step M2: Collect historical operating data, including normal operating conditions and multiple sets of disturbance scenario data, and identify and correct key parameters in the state-space equations through system identification methods; Step M3: Input the residual between the output of the mechanism model and the real-time monitoring data into an error correction neural network to perform online compensation and correction on the model prediction results.

5. The intelligent gas purity monitoring system according to claim 4, characterized in that: Step M2 specifically includes the following steps: Step M21: Based on the historical running data, divide the data into training and validation sets, and construct input-output sample pairs using the sliding window method; Step M22: Using the mechanistic model as a fixed structure, the recursive least squares method is used to identify the time-varying parameters in the sample pairs online and obtain the parameter trajectories; Step M23: Perform cluster analysis on the parameter trajectory, establish a mapping relationship library between parameters and working conditions, and realize the adaptive adjustment of model parameters according to the current working conditions.

6. The intelligent gas purity monitoring system according to claim 5, characterized in that: The process of performing rolling optimization using a constrained quadratic programming algorithm includes the following steps: Step O1: Set the prediction time domain Np and the control time domain Nc; define the objective function as the weighted sum of the squares of the purity tracking error and the squares of the change amplitude of the control quantity in the future prediction time domain; Step O2: Set optimization constraints, including: output purity constraints, control quantity constraints, and control quantity change rate constraints; Step O3: In each control cycle, with the current measurement value as the initial state, call the dynamic prediction model to predict the system output for the next Np steps; transform the constrained optimization problem into a standard quadratic programming form for online solution to obtain the optimal control sequence; Step O4: Only the first control quantity in the optimal control sequence is sent to the gas regulation execution unit; in the next sampling period, steps O1-O3 are repeated to achieve rolling optimization.

7. The intelligent gas purity monitoring system according to claim 6, characterized in that: The human-computer interaction unit also includes a fault diagnosis and emergency response module, which specifically performs the following steps: Step F1: Monitor the feasibility status of the MPC optimization module solver, the prediction residuals of the dynamic prediction model, and the consistency of sensor data in real time; Step F2: When the purity deviation exceeds the set threshold, the optimization problem is unsolvable, or the key sensor data is abnormal, a multi-level emergency response is triggered.

8. The intelligent gas purity monitoring system according to claim 7, characterized in that: The multi-level emergency response described in step F2 is managed using logic based on finite state machines.

Images