Monitoring system and method for anti-dynamic medium interference based on multi-physical field coupling modeling and adaptive time domain simulation

Abstract

The application discloses a kind of monitoring system and method of anti-dynamic medium interference based on multi-physical field coupling modeling and adaptive time domain simulation, the monitoring system includes: modeling unit, which contains the multi-physical field coupling optical transmission model for describing target medium;Sensor network is used to real-time acquisition several physical parameters of the target medium, the physical parameters are transmitted to the modeling unit;Signal processing engine, an adaptive time domain filtering module and a prediction compensation module are integrated in the signal processing engine;Monitoring and correction unit is used to monitor the performance of the monitoring system and trigger compensation or correction when performance declines;And multi-degree-of-freedom precision calibration platform is connected with the monitoring and correction unit.The system and method of the application significantly improve the detection reliability of weak leakage signal in dynamic environment, its anti-interference performance meets industrial safety standards and has strong environmental adaptability, realizes the high-precision visual monitoring target of chemical equipment operating state.

Description

Technical Field

[0001] This invention relates to a monitoring method for resisting dynamic media interference, specifically to a monitoring system and method for resisting dynamic media interference based on multi-physics field coupling modeling and adaptive time-domain simulation. It is particularly suitable for complex chemical environments where real-time monitoring of toxic gas leaks in tank farms requires simultaneous achievement of leak location accuracy ≤5cm and quantitative detection error rate of leak concentration <3%. Background Technology

[0002] In the field of chemical safety production, real-time monitoring of media leaks under dynamic environments is a core issue for ensuring the stable operation of industrial plants. Steam, as a common energy-carrying medium in chemical processes, presents multiple technical challenges for monitoring leaks: the light transmission distortion effect caused by high-temperature steam leads to a severe deterioration of the signal-to-noise ratio in traditional visual inspection systems, a phenomenon particularly prominent at flange connections in critical equipment such as petrochemical cracking units and polymerization reactors. Precision monitoring standards for high-risk chemical plants typically require a high level of dynamic signal-to-noise ratio improvement (e.g., 8 dB@1kHz), but in actual operating conditions, most existing systems struggle to consistently achieve this target. This technical bottleneck has become a significant obstacle restricting the application of intelligent leak prevention systems.

[0003] Existing technologies primarily employ a combined approach of optical compensation and signal processing. For example, a common industry practice is adaptive filtering systems for turbulent media. These typically establish a quantitative relationship between aerosol particle distribution and light intensity attenuation using a Mie scattering model, combined with a tunable bandpass filter to suppress noise. While such solutions often achieve a certain signal-to-noise ratio improvement in static steam environments, significant model inaccuracies occur under conditions of large temperature gradients. To address this limitation, some existing technologies further incorporate time-domain averaging and use a sliding window Fourier transform to decompose the acquired signal in the frequency domain. These improved versions exhibit better stability under smaller pressure fluctuations. However, these existing methods generally exhibit slow response to transient changes in the optical properties of steam media. The fundamental reason for this is the failure to construct an optical transmission model that couples multiple physical parameters, including temperature, pressure, and flow fields.

[0004] Existing theoretical research and engineering practice indicate that the core deficiency of the current technological system lies in its insufficient adaptability to dynamic environments. Specifically, when the steam medium is under complex dynamic changes in complex refractive index, traditional optical compensation schemes often struggle to achieve accurate matching using a single model, resulting in significant average residuals and directly shortening the effective detection distance. Furthermore, industrial field tests under high Reynolds number turbulent conditions have confirmed that existing time-delay compensation algorithms based on conventional Kalman filters produce non-negligible system delays when processing high-speed fluids. This delay directly causes phase distortion of leakage characteristic signals, leading to a significant increase in image edge blurring and contrast loss. These technological shortcomings have made it difficult to achieve the high signal-to-noise ratio improvement targets required by industry standards, severely hindering the widespread application of intelligent inspection robots in high-risk chemical plants.

[0005] The root causes of the current problems can be attributed to two aspects: First, existing optical distortion compensation models do not consider the coupling effect of transient thermodynamic parameters of the steam medium, treating temperature, pressure, and density as independent variables and linearly superimposing them, resulting in large prediction errors for the attenuation coefficient in the optical transmission equation. Second, the adaptive mechanism of traditional time-domain filtering algorithms is mismatched with the dynamic characteristics of the steam flow field, especially under extreme conditions with high temperature change rates, where fixed-parameter filtering windows struggle to balance signal fidelity and noise suppression efficiency. Industry experience shows that the false alarm rate of existing systems often increases significantly when pressure fluctuations are large, making it difficult to meet the stringent requirements of industrial precision monitoring standards.

[0006] The urgent need to address these technical bottlenecks stems from the dual pressures of economic and social consequences for the safe operation of chemical plants. Statistics show that early detection of flange leaks in refinery pressure reducing units can significantly reduce economic losses caused by accidental shutdowns. Simultaneously, the leakage of sulfur-containing components commonly found in steam can trigger severe air pollution events. This necessitates breakthroughs in key performance indicators such as dynamic signal-to-noise ratio, real-time responsiveness, and environmental adaptability for new leak detection systems.

[0007] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0008] The purpose of this invention is to provide a monitoring system and method for resisting dynamic media interference based on multi-physics coupling modeling and adaptive time-domain simulation. This invention solves the problems of poor adaptability to dynamic environments and insufficient signal-to-noise ratio improvement of existing methods. By establishing a multi-physics coupling optical transmission model, this invention integrates the fluid dynamics, thermodynamics and optical transmission characteristics of dynamic media, providing a precise theoretical basis for real-time monitoring and fundamentally improving the reliability of visual detection in dynamic and harsh environments.

[0009] To achieve the above objectives, this invention provides a monitoring system for resisting dynamic media interference based on multi-physics field coupled modeling and adaptive time-domain simulation. The system includes an image acquisition unit for real-time image capture of the target area and output of raw visual data; a sensor network for real-time acquisition of several physical parameters of the target medium, including temperature and pressure, which transmits these parameters to a modeling unit; a modeling unit containing a multi-physics field coupled optical transmission model describing the target medium, which calculates the turbulent extinction coefficient reflecting light intensity attenuation and the dynamic refractive index distribution reflecting optical path deflection based on the received physical parameters; and a signal processing engine connected to both the image acquisition unit and the modeling unit, which integrates an adaptive time-domain filtering module and a prediction compensation module. The adaptive time-domain filtering module receives data from the modeling unit. The turbulent extinction coefficient and dynamic refractive index distribution transmitted by the module unit are used as filtering references to filter the raw visual data transmitted by the image acquisition unit to extract leakage signals; the prediction compensation module is used to calculate the time shift to compensate for the inherent phase lag generated by the adaptive time-domain filtering module during the time-domain filtering process; the monitoring and correction unit is used to monitor the key performance indicators of the monitoring system and trigger compensation or correction when the key performance indicators decrease; and the multi-degree-of-freedom precision calibration platform is connected to the monitoring and correction unit and the signal processing engine. The signal processing engine provides real-time feedback error, and the monitoring and correction unit transmits compensation or correction information to the multi-degree-of-freedom precision calibration platform. The multi-degree-of-freedom precision calibration platform adjusts the spatial position and angular attitude of the image acquisition unit to perform system self-correction to physically compensate for optical distortion, i.e., optical path difference.

[0010] The multi-physics coupled optical transmission model is as follows: In equations (1) to (3), This represents the real-time density of the steam medium. For time; For Hamiltonian operators; It is a velocity vector; For quality source items; The substance derivative; Real-time fluid pressure; Dynamic viscosity; For the Laplace operator; It is the vector of gravitational acceleration; Specific heat capacity at constant pressure; Thermodynamic temperature; The thermal conductivity coefficient; This is a viscous dissipation term.

[0011] In the multiphysics coupled optical transmission model, the improved Van de Hulst approximation (4) is introduced to calculate the turbulent extinction coefficient caused by steam turbulence. : (4) In equation (4), Sauter mean diameter is a commonly used average particle size based on specific surface area for particles or droplet swarms, obtained through actual measurement using a high-speed microscopic imaging system. Extinction efficiency factor; To detect the wavelength of light; The fluctuation range of the refractive index of the medium; It is the turbulence modulation factor, determined through regression analysis of wind tunnel experimental data.

[0012] In the multi-physics coupled optical transmission model, the parameterized characterization of the light transmission characteristics of the vapor medium is achieved by establishing a dynamic refractive index model with temperature-pressure-density coordinated modulation, as shown in equation (16): (16) In equation (16), The vapor refractive index under the current operating conditions; is the Gladstone-Dale constant, which characterizes the linear relationship between the refractive index and density of a gas; Temperature in Celsius; For reference temperature; For real-time pressure; For reference pressure; Real-time steam density; This is a temperature correction factor; This is the pressure correction factor.

[0013] Preferably, the algorithm of the adaptive time-domain filtering module is to construct a closed-loop control system with predictive compensation capability, whose control law... As shown in formula (5): (5) In equation (5), the control law Precise compensation is achieved through four synergistic effects; This is a proportional term used to provide immediate error correction; This is the integral term, used to eliminate steady-state deviations; The derivative term is used to suppress overshoot; the proportional-integral-derivative (PID) term is used for routine error correction. K p , K i and Kd These are the proportional, integral, and differential gain coefficients, respectively. e ( t The error signal is the difference between the reference signal value or the expected response signal value and the output signal value. For integration variables Historical error signal at any given time; τ For integration time; The prediction term is dynamically calculated using a Kalman predictor to determine the time shift. Actively compensate for phase lag caused by system processing; The gain coefficient for the prediction term; A prediction function to describe changes in the system state.

[0014] Specifically, the prediction function The state transition equation is constructed based on the Extended Kalman Filter (EKF) and is defined as follows: In the formula, This is the system state estimation vector at the current moment. Including temperature deviation Pressure deviation Density deviation and signal-to-noise ratio deviation Four components; This is the system parameter matrix; This is the state transition matrix, used to characterize the evolution of the system state over time; This is to output the observation matrix.

[0015] More preferably, the adaptive time-domain filtering module further includes: a wavelet transform unit, used to perform multi-resolution time-frequency analysis on the acquired signal and achieve wavelet denoising by setting a reasonable threshold; and / or the adaptive time-domain filtering module also dynamically adjusts the processing window length of the filter according to the gradient change of the signal-to-noise ratio and the historical hysteresis error. To adapt to the non-stationary characteristics of the steam medium; and / or, the adaptive time-domain filtering module introduces a prediction compensation module based on an improved PID or Kalman predictor in the filtering stage. This prediction compensation module can predict the system dynamics in the short term based on historical data and the current state, and calculate the compensation amount, thereby actively offsetting the inherent phase lag in the filtering process.

[0016] Preferably, the prediction compensation module adopts an ARIMA model-based prediction compensation module, and its time shift calculation is based on the ARIMA(2,1,2) time series model, as shown in formula (6): (6) In equation (6), For the shift operator; This is for first-order difference calculation; The time series to be predicted; It is a white noise sequence.

[0017] Preferably, the monitoring and calibration unit includes: a distributed parameter acquisition module (e.g., a Schneider Electric TM5 module), which is used to acquire steam flow field parameters, calculate the equivalent wavefront distortion of the optical path, and transmit the equivalent wavefront distortion to the servo driver of the multi-degree-of-freedom precision calibration platform to drive the coordinated motion of each axis of the multi-degree-of-freedom precision calibration platform to compensate for the optical path difference.

[0018] Preferably, the mechanical structure of the multi-degree-of-freedom precision calibration platform includes X / Y / Z linear guides, a pitch / yaw rotary table, and a micro-displacement actuator based on piezoelectric ceramics.

[0019] Preferably, the control algorithm of the multi-degree-of-freedom precision calibration platform adopts an improved incremental PID strategy, and its control output is: (7) In equation (7), Discrete time step Incremental control output; Discrete time step The error; Discrete time step The error (i.e., the error of the previous moment); Discrete time step The error (i.e., the error of the previous moment). , and For incremental PID parameters, feedforward term Pre-compensation is performed based on real-time measured Zernike coefficients. Identified using the least squares method. ; For real-time measurement of the first The Zernike coefficient (used to characterize low-order optical aberrations).

[0020] More preferably, the compensation amount for the optical path difference is calculated in real time based on the Zernike polynomial coefficients: for the defocus term... Through Z-axis displacement compensation, like scattered terms and Higher-order aberrations are corrected at the nanometer level by pitch / yaw stage correction and micro-displacement actuator.

[0021] Preferably, the sensor network adopts a distributed optical fiber sensor network layout based on Reynolds number partitioning, and the spatial spacing of the sensor nodes is dynamically adjusted according to the hydrodynamic parameters of the steam medium. 1) Axial arrangement (8) In equation (8), This refers to the axial spacing. It is the Reynolds number; 2) Radial arrangement Following the principle of one-fifth of the vortex core diameter, as shown in equation (9): (9) In equation (9), Radial spacing; Indicates the diameter of the vortex.

[0022] Preferably, the sensor network adopts a T-shaped topology with high redundancy. ; Preferably, the sensing network uses an OPSENS OPP-W431 temperature sensor, a Keller PA-23Y pressure sensor, and a Shack-Hartmann wavefront sensor.

[0023] Preferably, the sensing network includes: an optical fiber temperature sensor and a pressure sensor for acquiring the thermodynamic parameters of the target medium, and a wavefront sensor for detecting beam wavefront distortion data; the wavefront sensor directly acquires beam wavefront distortion data by capturing the wavefront phase of the laser beam after passing through the steam turbulent medium, and the modeling unit compares the measured wavefront distortion data with the theoretical distortion value calculated by the multiphysics coupled optical transmission model to achieve online calibration of the model.

[0024] A second objective of this invention is to provide a monitoring method for the aforementioned monitoring system, the monitoring method comprising the following steps: (S100) Real-time parameter acquisition Several physical parameters of the target medium are acquired in real time, including temperature and pressure, to provide parameters for the multiphysics coupled optical transmission model. At the same time, the original visual data of the target area is acquired using the image acquisition unit. The multiphysics coupled optical transmission model integrates the fluid dynamics, thermodynamics and optical transmission characteristics of the dynamic medium, and is constructed by coupling the Navier-Stokes momentum equation with the Fourier heat conduction equation and the mass conservation equation. (S200) Adaptive Filtering The raw visual data carrying leakage signals acquired by the image acquisition unit is processed by an adaptive time-domain filtering algorithm that includes prediction compensation. Wavelet decomposition is used to perform multi-resolution time-frequency analysis on the signal to effectively separate noise and signal features at different scales. Furthermore, based on historical data and the current state, the system dynamics in the near future can be predicted and the compensation amount can be calculated to actively offset the inherent phase lag in the filtering process. (S300) Verification and Calibration The signal-to-noise ratio improvement is used as a key performance indicator. The key performance indicator is monitored in real time. When the key performance indicator is lower than the preset threshold, the temperature and pressure compensation algorithm is automatically triggered. The monitoring and correction unit drives the multi-degree-of-freedom precision calibration platform to adjust the spatial position and angle of the image acquisition unit in order to physically compensate for optical distortion.

[0025] Preferably, in step (S200), a multi-layer decomposition tree is constructed using the Symlet wavelet basis function to effectively separate noise of different frequencies from transient characteristic signals of the dynamic medium; the Symlet wavelet basis function is the Symlet2 wavelet basis function, as shown in formula (10): (10) In equation (6), This is the Symlet wavelet function family after scaling and translation; This is the scaling factor (or scaling factor), used to control the bandwidth and resolution of the wavelet function; This is a translation factor (or displacement factor) used to control the position of the wavelet function on the time axis; The Symlet2 mother wavelet function; It is a time variable; Preferably, in step (S200), after the signal is decomposed by wavelet, the effective signal and noise are separated by setting a reasonable threshold. Wavelet coefficients greater than or equal to the threshold are wavelet coefficients of the effective signal, and wavelet coefficients less than the threshold are wavelet coefficients of the noise. The reasonable threshold is set based on the improved SUREShrink threshold function, as shown in formula (11): (11) In equation (11), For threshold; This is an estimate of the noise variance; The signal length; This is the estimated signal-to-noise ratio.

[0026] Preferably, in step (S200), the processing window length of the filter is also dynamically adjusted based on the gradient change of the signal-to-noise ratio and the historical hysteresis error. To adapt to the non-stationary characteristics of dynamic media; the dynamic window length adjustment function is shown in formula (12): (12) In equation (12), Based on the length of the window; For gradient terms, The signal-to-noise ratio gradient; For integration, This is a historical lag error.

[0027] Preferably, in step (S300), the stability verification of dynamic compensation adopts the Lyapunov direct method, which includes: Considering that the perturbation range of the steam medium parameters covers ±35% of the working boundary value, an extended state-space model is established: (13) In equation (13), the state vector Includes temperature, pressure, density, and signal-to-noise ratio deviations; parameter matrix. spectral radius Ensure basic stability; Then, for the extended state-space model, candidate Lyapunov functions are constructed: (14) In equation (14), It is a positive definite matrix; For memory term functions, For time integration variables; For parameter vectors; This is the adaptive rate matrix; The following inequality can be derived: The inequality is proven to remain exponentially convergent under a maximum parameter perturbation of 35%.

[0028] Preferably, in step (S300), a residual feedback loop is established to calibrate the model predictions in real time, and the calibrated extinction coefficient is... The ratio of model predictions to errors, along with integral feedback, determines the outcome. (15) In equation (15), These are the model's predicted values. This represents the residual between the predicted and actual values.

[0029] The monitoring system and method for resisting dynamic media interference based on multi-physics coupling modeling and adaptive time-domain simulation of the present invention solves the problems of poor dynamic environment adaptability and insufficient signal-to-noise ratio improvement of existing methods, and has the following advantages: (1) This invention achieves a leapfrog improvement in anti-interference performance: a breakthrough in signal-to-noise ratio (SNR) is achieved. In a dynamic steam environment, this invention historically improves the SNR of the visual inspection system from the common 5-6 dB level of existing technologies to [a higher value]. dB@1kHz, with a total improvement of up to 64.3%. This indicator not only fully meets, but even exceeds, the 8 dB threshold typically required in the field of conventional industrial non-destructive testing;

[0030] (2) The present invention achieves significantly enhanced detection accuracy and reliability: With the support of a high-precision model, even under extreme conditions such as pressure fluctuation of 0.2 MPa and temperature gradient of 150℃ / m, the detection accuracy of the system can still be maintained above 98.7%, which is 32 percentage points higher than the traditional method. At the same time, the false detection rate of the system has been successfully reduced to 0.23‰, which is far better than the upper limit requirement of 0.5‰ usually recommended in the precision monitoring specifications of the petrochemical industry;

[0031] (3) The present invention achieves significantly improved positioning accuracy: The present invention improves the leakage positioning accuracy by 62%. Simulation tests in a DN400 pipeline show that the positioning error for a 2mm diameter leakage hole can be controlled within ±8.3mm in the axial direction and within ±8.3mm in the radial direction. Within.

[0032] (4) Significant optimization of the real-time response and processing performance of the system of the present invention: Through hardware and software co-optimization, the contradiction between high performance and real-time performance is resolved, achieving true real-time processing, and the overall system processing latency is successfully compressed to Within the millisecond range, it is far below the 10 ms or 33 ms timeliness thresholds typically required by industrial safety interlocks; it boasts excellent computational performance, deployed on a heterogeneous computing architecture of FPGA+GPU, where the GPU can execute predictive control algorithms at a processing speed of up to 12 μs / frame; and it improves compensation accuracy through a predictive compensation mechanism based on the ARIMA model, enhancing the time delay compensation accuracy from that of traditional methods. Upgraded to This effectively avoids phase distortion of the signal; (5) This invention achieves a comprehensive enhancement in high reliability and environmental adaptability: not only is the performance superior, but the stability and adaptability to harsh environments have also been systematically verified. The system stability theory and practical verification, through Lyapunov stability analysis, theoretically proves that the system exists in the presence of parameters. It can maintain exponential convergence even under large perturbations. In a 72-hour continuous test, The fluctuation range is only 0.37 dB, which is better than the 0.5 dB upper limit typically established in the field of precision instrument monitoring; the system boasts excellent reliability, with a mean time between failures (MTBF) of 12,450 hours. A five-fold fault-tolerant system, combined with sensor channel redundancy... Mechanisms such as supercapacitor emergency power supply ensure high availability of the system under extreme conditions; the physical compensation effect is significant, and the integrated five-degree-of-freedom calibration platform can reduce the beam deflection caused by steam turbulence from... Reduce to Wavefront distortion RMS value controlled within Within this range, the quality of signal acquisition is guaranteed from a physical perspective;

[0033] (6) This invention achieves significant economic and social benefits: The application of this invention brings considerable economic returns and social value to chemical enterprises. It creates direct economic benefits by significantly advancing the time of leak detection, effectively avoiding huge economic losses caused by accidental production stoppages, equipment repairs, and environmental fines. In practical application cases, deploying this system significantly reduces annual accident costs and shortens the investment payback period; it improves operation and maintenance efficiency and reduces costs, with high reliability (…). This reduces the system's total lifecycle maintenance costs and generates positive social and environmental benefits. By rapidly responding to leaks of hazardous substances such as sulfur-containing components, it effectively prevents serious air pollution incidents. Simultaneously, the system's high-efficiency design can reduce carbon emissions by hundreds of tons annually, meeting the energy management requirements for green and low-carbon industrial development. Attached Figure Description

[0034] Figure 1 This is the overall logical architecture diagram of the monitoring system for resisting dynamic media interference based on multi-physics coupling modeling and adaptive time-domain filtering according to the present invention.

[0035] Figure 2 This is a functional block diagram of the signal processing engine of the detection method of the present invention.

[0036] Figure 3 This is a schematic diagram of the distributed sensor network layout based on Reynolds number partitioning in this invention.

[0037] Figure 4 This is a diagram of the fault-tolerant logic architecture of a three-level finite state machine based on FPGA of the present invention.

[0038] Explanation of the numbering in the diagram: 101—Sensor Network; 102—Image Acquisition Unit; 103—Modeling Unit; 104—Signal Processing Engine; 105—Monitoring and Correction Unit; 106—Multi-DOF Precision Calibration Platform. Detailed Implementation

[0039] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] It should be noted that: Unless otherwise specified in the examples, conditions should be followed according to standard conditions or the manufacturer's recommendations. Instruments whose manufacturers are not specified are all commercially available products. Raw materials and reagents whose manufacturers are not specified are all commercially available goods or can be prepared using known methods.

[0041] In this invention, all features defined in the form of numerical ranges or percentage ranges, such as numerical values, quantities, contents, and concentrations, are used only for simplicity and convenience. Accordingly, the description of numerical ranges or percentage ranges should be considered as covering and specifically disclosing all possible sub-ranges and individual numerical values ​​(including integers and fractions) within those ranges.

[0042] The features mentioned in this invention can be combined arbitrarily, and all possible combinations should be considered within the scope of this specification, provided that there is no contradiction in the combination of these features. Each feature disclosed in the specification can be replaced by any alternative feature that provides the same, equivalent, or similar purpose. Therefore, unless otherwise specified, the disclosed features are merely general examples of equivalent or similar features.

[0043] In the description of this invention, it should be noted that relational terms such as "first," "second," and "third" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0044] See Figure 1This invention provides a detection system and method based on multiphysics coupling modeling and adaptive temporal filtering to combat dynamic media interference. This addresses the severe interference caused by dynamically changing steam media in chemical safety production, particularly the technical challenge of achieving a signal-to-noise ratio that meets industry standards. The core of this invention lies in constructing a four-dimensional compensation system integrating accurate modeling, real-time sensing, predictive filtering, and high-performance computing, fundamentally improving the reliability of visual detection in dynamic and harsh environments.

[0045] See Figure 2 ( Figure 2 The predictive compensator shown is the predictive compensation module, which is a functional block diagram of the signal processing engine of the detection method of this invention. Its purpose is to extract weak leakage feature signals from the original noisy visual signal full of interference. The original noisy visual signal first enters the "wavelet reset unit" for multi-scale time-frequency analysis, and then the "thresholding and suppression" effectively removes background noise and turbulence interference. Finally, the "reconstructed signal" is generated through the reconstruction stage. In addition, the system uses the "predictive compensation module" and the "monitoring and correction unit" (e.g., monitoring the signal-to-noise ratio improvement value) to further enhance the signal-to-noise ratio. A dynamic "judgment threshold" is generated. Finally, the "reconstructed signal" and the "judgment threshold" work together in the "shape difference information" module to achieve the final output of the leakage characteristic signal after denoising and compensation.

[0046] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be described in detail below with reference to Embodiment 1 and application examples.

[0047] Example 1 A monitoring method for resisting dynamic media interference based on multi-physics coupling modeling and adaptive time-domain simulation specifically includes the following steps: (S100) Optical Distortion Modeling First, a multiphysics coupled optical transmission model is established to describe dynamic media (such as fluid media such as steam, industrial waste gas, or high-temperature oil mist). This model integrates the fluid dynamics, thermodynamics, and optical transmission characteristics of the dynamic media. The multiphysics coupled optical transmission model is constructed by coupling the Navier-Stokes momentum equation with the Fourier heat conduction equation and the mass conservation equation.

[0048] In this model, optical transmission characteristics are specifically manifested as the "dynamic refractive index distribution" and "turbulent extinction coefficient" determined by the temperature, pressure, and density fields. The multiphysics coupled optical transmission model accurately describes the interaction between the temperature, pressure, density, and velocity fields of the dynamic medium, quantifies their synergistic effects on the aforementioned optical parameters, and thus accurately characterizes the complex distortion effects (such as optical path deflection and intensity attenuation) caused by the dynamic medium on light transmission.

[0049] The construction of this model provides the core calculation basis for subsequent steps. The calculated turbulent extinction coefficient affects the signal-to-noise ratio estimate, thereby guiding the setting of the adaptive filtering threshold in the subsequent step (S300). The calculated refractive index distribution is used to solve the equivalent wavefront distortion (optical path difference) of the optical path, and this distortion is used as the compensation basis for the physical correction of the multi-degree-of-freedom platform in (S400), thereby fundamentally solving the compensation failure problem caused by model inaccuracy in traditional methods.

[0050] The Navier-Stokes momentum equation is the fundamental equation describing the motion of Newtonian fluids (such as air and water) with conserved momentum. The Fourier heat conduction equation describes the heat transfer process in the medium, and the mass conservation equation is used to describe the change in steam density. Combining these three equations can more accurately characterize the complex behavior of steam medium in dynamic environments.

[0051] The mass conservation equation is used to describe the change in steam density, as shown in equation (1) below: (1) The momentum conservation equation is used to characterize the steam flow behavior, as shown in equation (2) below: (2) The energy equation is used to characterize the evolution of the steam temperature field, as shown in equation (3) below: (3) In equations (1) to (3), The density of the vapor medium; For time; For Hamiltonian operators; It is a velocity vector; For quality source items; The substance derivative; For fluid pressure; Dynamic viscosity; For the Laplace operator; It is the vector of gravitational acceleration; Specific heat capacity at constant pressure; Thermodynamic temperature; The thermal conductivity coefficient; This is a viscous dissipation term.

[0052] To address the optical distortion effect caused by steam turbulence, this invention innovatively introduces an improved Van de Hulst approximation formula (4) to calculate the turbulent extinction coefficient caused by steam turbulence. This incorporates turbulence, a key interfering factor, into the quantitative analysis system.

[0053] (4) In equation (4), Sauter mean diameter is a commonly used average particle size based on specific surface area for particles or droplet swarms, obtained through actual measurement using a high-speed microscopic imaging system. Extinction efficiency factor; To detect the wavelength of light; The fluctuation range of the refractive index of the medium; It is the turbulence modulation factor, determined through regression analysis of wind tunnel experimental data.

[0054] (S200) Real-time parameter acquisition In order to provide real-time and accurate boundary condition inputs for the aforementioned multiphysics coupled optical transmission model, the system acquires multiple thermodynamic physical parameters of the steam medium in real time through a distributed sensor network, including temperature and pressure; at the same time, it acquires the original visual data of the target area through an image acquisition unit.

[0055] Specifically, the sensor network collects the temperature and pressure of the site in real time at a frequency of no less than 1 kHz. These collected thermodynamic parameters are closely related to the model constructed in step (S100), and they are directly used as input variables of the multiphysics model to solve the current dynamic refractive index distribution and turbulent extinction coefficient in real time; while the raw visual data collected synchronously by the image acquisition unit is the object of subsequent adaptive filtering and distortion correction.

[0056] To acquire model parameters in real time, this embodiment designs a distributed fiber optic sensor network layout scheme based on Reynolds number partitioning characteristics. This scheme pre-optimizes the spatial density of sensor nodes according to the flow field characteristics at different locations in the target area (such as laminar and turbulent regions). Unlike traditional sensors which are independent entities, fiber optic sensing technology can connect multiple fiber grating (FBG) sensing units in series on a single fiber to form a quasi-distributed monitoring network with high spatiotemporal resolution, used for real-time acquisition of on-site temperature and pressure data.

[0057] (S300) Adaptive Filtering To address the lag and mismatch issues of traditional filtering algorithms, this invention designs a novel adaptive time-domain filtering algorithm, namely, an adaptive time-domain filtering algorithm incorporating prediction compensation. This algorithm employs wavelet transform techniques, such as constructing a multi-layer decomposition tree using a Symlet wavelet basis, to perform multi-resolution time-frequency analysis of the signal, effectively separating noise and signal features at different scales. Furthermore, a prediction compensation module based on an improved PID or Kalman predictor is introduced into the filtering stage. This module can predict the system dynamics in the near future based on historical data and the current state, and calculate the compensation amount, thereby actively offsetting the inherent phase lag in the filtering process and ensuring signal fidelity even when steam parameters change drastically.

[0058] The raw visual data carrying the leakage signal, acquired by the image acquisition unit, is input into an adaptive temporal filtering module. This adaptive temporal filtering module filters the raw visual data based on the physical parameters (i.e., temperature and pressure) acquired in real time in step (S200) and in conjunction with a prediction compensation mechanism, in order to suppress interference and extract the leakage signal.

[0059] Specifically, the core of the adaptive time-domain filtering module in this embodiment lies in constructing a closed-loop control system with predictive compensation capabilities, and its control law... As shown in formula (5): (5) In equation (5), the control law Precise compensation is achieved through four synergistic effects; This is a proportional term used to provide immediate error correction; This is the integral term, used to eliminate steady-state deviations; The derivative term is used to suppress overshoot; the proportional-integral-derivative (PID) term is used for routine error correction. K p , K i and K d These are the proportional, integral, and differential gain coefficients, respectively. The error signal is the difference between the reference signal value or the expected response signal value and the output signal value. For integration variables Historical error signal at any given time; This is the integration time variable, and its value ranges from 0 to the current time. ; The prediction term is dynamically calculated using a Kalman predictor to determine the time shift. Actively compensate for phase lag caused by system processing; The gain coefficient for the prediction term; A prediction function to describe changes in the system state.

[0060] Among them, the prediction gain coefficient An adaptive adjustment strategy is adopted. The baseline value of this coefficient is determined through a system initialization calibration process under steady-state steam flow conditions, with the goal of minimizing the system's inherent phase lag. During operation, the real-time monitored rate of change of steam density is used as the reference value. It automatically increases to 1.2 times the baseline value to ensure rapid response capability under sudden operating conditions.

[0061] The adaptive time-domain filtering module in this embodiment also includes a wavelet transform unit for multi-resolution time-frequency analysis of the signal, used to perform multi-resolution time-frequency analysis on the original visual data input by the image acquisition unit. This wavelet transform unit constructs a multi-layer decomposition tree using Symlet wavelet basis functions (preferably Symlet2). The compact support characteristics and approximate symmetry of the Symlet wavelet basis give it excellent localization analysis capabilities in the frequency domain, effectively separating noise and transient vapor characteristic signals at different frequencies. This multi-layer decomposition tree decomposes the original signal into multiple sub-band signals through wavelet transforms at different scales, thereby capturing transient features and periodic components in the signal more precisely. Specifically, the number of decomposition layers is usually dynamically determined based on the complexity of the signal and the sampling frequency to ensure effective noise removal while retaining key information; in this embodiment, a six-layer decomposition tree is preferably constructed. In each decomposition layer, the wavelet transform unit calculates the approximation coefficients and detail coefficients of the signal at different scales. The approximation coefficients represent the low-frequency components of the signal, while the detail coefficients reflect the high-frequency details of the signal. Through this multi-resolution analysis, the adaptive time-domain filtering module can accurately identify and separate the leakage signal from the background noise, providing a reliable basis for subsequent signal reconstruction and leakage detection.

[0062] The Symlet wavelet basis functions mentioned above are specifically the Symlet2 wavelet basis functions, as shown in formula (10): (10) In equation (10), This is the Symlet wavelet function family after scaling and translation; This is the scaling factor (or scaling factor), used to control the bandwidth and resolution of the wavelet function; This is a translation factor (or displacement factor) used to control the position of the wavelet function on the time axis; The Symlet2 mother wavelet function; It is a time variable.

[0063] The compact support properties of Symlet2 wavelet basis functions give it excellent localization analysis capabilities in the frequency domain. Simulation verification and field tests based on typical leakage conditions show that the algorithm achieves accurate multi-resolution analysis of signals in the range of 0.5~5kHz (corresponding to the typical steam jet noise frequency domain) through a 6-layer decomposition tree. Combined with the improved SUREShrink threshold function, it can achieve a high noise suppression ratio while preserving the transient characteristics of steam (retention rate >92%). .

[0064] After wavelet decomposition, the amplitude of the wavelet coefficients of the effective signal is usually greater than that of the noise figure. By setting a reasonable threshold, the two can be separated: wavelet coefficients greater than or equal to the threshold are considered effective signals, while those less than the threshold are considered noise. The key to wavelet denoising lies in the selection of the threshold function, which directly determines the quality of signal denoising. In existing technologies, hard thresholding and soft thresholding are the two most commonly used functions for wavelet denoising.

[0065] However, both of these traditional methods have inherent limitations: the hard threshold function is discontinuous at the threshold, and the reconstructed signal is prone to the pseudo-Gibbs phenomenon, resulting in local oscillations; although the soft threshold function maintains continuity, there is a constant deviation between the processed coefficients and the original coefficients, which can easily lead to amplitude attenuation and edge blurring of the reconstructed signal.

[0066] To address the aforementioned shortcomings, this invention proposes an improved SUREShrink threshold function, aiming to combine the advantages of both and overcome their deficiencies. The improved function is shown in equation (11):

[0067] (11) In equation (11), For threshold; This is an estimate of the noise variance; The signal length; This is the estimated signal-to-noise ratio.

[0068] To verify the superiority of the improved SUREShrink threshold function proposed in this invention, this embodiment constructs a simulated signal containing typical steam leakage characteristics and superimposes Gaussian white noise (input signal-to-noise ratio SNR=5dB) for comparative testing. The experiment compares the denoising effects of traditional "hard threshold function," "soft threshold function," and the "improved SUREShrink threshold function" of this invention.

[0069] The root mean square error (RMSE), output signal-to-noise ratio (SNR), and feature retention rate were used as evaluation indicators, and the comparison results are shown in Table 1.

[0070] Table 1 Comparison of different methods As shown in Table 1 and the experimental results, while the traditional hard thresholding method achieves a reasonable feature retention rate, its oscillation effect leads to a high RMSE. The traditional soft thresholding method, while exhibiting good smoothness, suffers from low feature retention and is prone to losing crucial weak leakage signals. In contrast, the improved method of this invention combines the advantages of both methods. Under the same input conditions, the signal-to-noise ratio is increased to 18.7 dB (meeting the requirement of not less than 18 dB), and the feature retention rate reaches 93.4% (meeting the requirement of exceeding 92%). This fully demonstrates the significant technical advantages of the improved algorithm in processing non-stationary steam leakage signals.

[0071] The adaptive time-domain filtering module in this embodiment also dynamically adjusts the processing window length of the filter based on the gradient change of the signal-to-noise ratio and the historical hysteresis error. To adapt to the non-stationary characteristics of the steam medium, the dynamic window length adjustment function is shown in formula (12):

[0072] (12) In equation (12), Based on the length of the window; For gradient terms, The signal-to-noise ratio gradient; For integration, This is a historical lag error.

[0073] The dynamic window length adjustment function of this invention is designed based on an in-depth analysis of the non-stationary characteristics of the steam medium: the basic window length. To ensure basic time resolution; gradient term The observation window is dynamically expanded based on the rate of change of the signal-to-noise ratio to capture low-frequency disturbances; the integral term... The system then performs compensatory adjustments based on accumulated historical lag errors. This adjustment mechanism can be based on a high-performance heterogeneous computing platform (such as the AMD Zynq UltraScale+ MPSoC, but not limited to it), and can be logic synthesized and implemented using the Verilog hardware description language. Leveraging the platform's hardware-software co-processing capabilities, the system achieves microsecond-level filter-level real-time adjustment, enabling the algorithm to automatically balance signal fidelity and noise suppression efficiency in a wide range of dynamically changing environments.

[0074] The prediction compensation mechanism in this embodiment adopts a dynamic prediction architecture based on the Extended Kalman Filter (EKF). This mechanism is a specific implementation of the aforementioned Kalman predictor in a nonlinear system, aiming to solve the phase lag problem in high-speed fluid monitoring. Specifically, the prediction compensation module does not directly calculate the time shift, but predicts future moments based on the system's state equations. The system status.

[0075] Among them, the predicted step size (instantaneous displacement) It is not a fixed value, but rather based on the fluid flow rate calculated in real time. Computational delay with processing unit Dynamically determined, its calculation relationship satisfies in As a feature scale, This is for the safety factor.

[0076] The core iterative process of the Kalman predictor is shown in equation (21): (twenty one) In equation (21), for Prior state estimate (i.e., predicted value) at time 1; The state transition matrix describes the evolution of the steam leakage signal from the current moment to the next moment. for Posterior state estimation at time t; To control the input matrix; This is the control vector. Through this prediction equation, the system can "predict" the trend of signal changes during the lag time period, thereby achieving real-time compensation with zero phase delay at the output.

[0077] To implement the aforementioned prediction compensation in an FPGA digital system, the calculated continuous time shift needs to be... Discretization to prediction order (Right now ,in (This refers to the system sampling frequency).

[0078] In this example, the prediction order is preferably set to . The basis for its selection lies in: when When the prediction window is too short, it cannot fully cover the inherent phase lag of the system, resulting in insufficient compensation; when While the prediction range expands, the computational complexity of the algorithm increases exponentially, resulting in a measured increase in computational latency of 1.7 ms, exceeding the system's real-time constraint (<1 ms). Experimental data shows that choosing... The configuration significantly improves the system's time delay compensation accuracy from 76% in the traditional method to 98.5%, effectively achieving zero phase delay tracking.

[0079] (S400) Three-dimensional verification and failure early warning To ensure the system's reliability and stability, a three-dimensional verification system encompassing theory, experiments, and industrial field applications was established. Simultaneously, the system incorporates a monitoring and calibration unit, which includes a distributed parameter acquisition module (e.g., a Schneider Electric TM5 module). This unit monitors key performance indicators such as the signal-to-noise ratio improvement in real time. When these indicators fall below preset thresholds (e.g., 6.5 dB), it automatically triggers a temperature and pressure compensation algorithm and performs system self-calibration through a multi-degree-of-freedom precision calibration platform, ensuring the system's stability and accuracy during long-term operation.

[0080] When the key performance indicators of the aforementioned monitoring and correction unit still fail to meet the threshold requirements after triggering the temperature and pressure compensation algorithm, the unit collects steam flow field parameters, calculates the equivalent wavefront distortion of the optical path using a multi-physics coupling model, and converts the equivalent wavefront distortion into multi-axis motion control commands, which are then transmitted to the servo driver of the multi-degree-of-freedom precision calibration platform to drive the coordinated motion of each axis of the multi-degree-of-freedom precision calibration platform to compensate for the optical path difference.

[0081] Set a system performance monitoring metric (e.g., signal-to-noise ratio improvement). When this indicator falls below a preset threshold (e.g., 5dB), the system automatically activates a tiered response mechanism, as follows: First, the software-level 'temperature and pressure compensation algorithm' is triggered: the monitoring and correction unit executes the algorithm to obtain the latest temperature and pressure sensor data, substitutes it into the aforementioned multiphysics coupling model, and recalculates the current dynamic refractive index distribution field; then, the light deflection is calculated based on the principle of light transmission, and pixel-level reverse geometric distortion correction is performed on the original image accordingly to eliminate background distortion caused by the non-uniformity of the medium's refractive index and improve the image signal-to-noise ratio. If the indicators still do not recover after software compensation, the hardware-level "physical correction mechanism" is further triggered: that is, the multi-degree-of-freedom precision calibration platform is driven to fine-tune the spatial attitude (pitch angle and yaw angle) of the image acquisition unit, and the observation optical path is optimized in a physical way to avoid local high turbulence distortion areas.

[0082] Furthermore, in step (S200), in the distributed optical fiber sensor network layout scheme based on Reynolds number partitioning, the spatial spacing of the sensor nodes is dynamically adjusted according to the hydrodynamic parameters of the steam medium, see [link to relevant documentation]. Figure 3 The specific layout is as follows: 1) Axial arrangement (8) Axial spacing Dynamically adjust according to the local flow state: in the laminar flow region ( A 200 mm spacing is used to ensure basic coverage; transition zone ( The diameter was reduced to 120 mm to capture the vortex structure; the strong turbulence region ( Further encryption to 80 mm.

[0083] The aforementioned segmented design enables temperature gradients The capture accuracy is improved by 42%, and the temperature field reconstruction error can be controlled within a certain range in a typical DN300 pipeline, as verified by computational fluid dynamics (CFD) simulation. Within the range.

[0084] 2) Radial arrangement Following the principle of one-fifth of the vortex core diameter, as shown in equation (9): (9) In equation (11), Radial spacing; The vortex diameter is indicated by the vortex core, which is a technical term in fluid dynamics used to describe the core region of a vortex structure. The empirical formula is determined. For pipe diameter, The formula is the Reynolds number, derived from fitting PIV test data of the steam flow field, ensuring that the Mie scattering effect caused by the steam microstructure can be effectively resolved.

[0085] Furthermore, the sensor network topology and redundancy are optimized. The sensor network adopts a T-bus topology, and each monitoring node in the network consists of a pair of temperature and pressure sensors (i.e., two types of sensors are deployed simultaneously at the same monitoring point).

[0086] In terms of specific selection, the OPSENS OPP-W431 fiber optic temperature sensor (range -40 to 350℃) and the Keller PA-23Y pressure sensor (accuracy 0.05%FS, i.e., full scale) were adopted. Deployment between nodes... The spatial sampling theorem must be satisfied: (in To predict the minimum characteristic scale of turbulence and prevent spatial aliasing, the redundancy design ensures that the system's signal-to-noise ratio (SNR) drops by less than 0.5 dB in the event of a single node failure.

[0087] Furthermore, to quantitatively verify the accuracy of the aforementioned multiphysics coupled optical transmission model, this embodiment constructs an active wavefront detection optical path. This optical path employs a transmission-type arrangement: a highly stable collimated laser source (wavelength 532nm, beam diameter 50mm) is placed on one side of the monitoring area as a reference beacon; on the opposite side of the monitoring area, a Shack-Hartmann wavefront sensor (model: HASO4-512) is arranged coaxially with the laser source.

[0088] The sensor directly acquires beam wavefront distortion data (i.e., Zernike coefficients) by capturing the wavefront phase of the laser beam after it passes through the steam turbulence medium. It then compares the measured distortion data with the theoretical distortion value calculated by the multiphysics model to achieve online calibration of the model.

[0089] Calculation of theoretical distortion value by multiphysics model: First, based on the temperature and pressure parameters collected in real time by the sensor network, the dynamic refractive index distribution of the steam medium is calculated using the Gladstone-Dale relationship in formula (16) in the manual; then, the refractive index distribution is integrated along the laser beam transmission path to calculate the theoretical optical path difference (i.e., theoretical wavefront distortion value) that reflects the optical path deflection characteristics.

[0090] The measured wavefront distortion data obtained by the wavefront sensor is transmitted to the modeling unit 103. The modeling unit 103 is the carrier for the operation of the model. The sensor network (including the wavefront sensor) transmits the measured data (T, P and wavefront distortion) to the modeling unit in a unified manner. The modeling unit calculates the theoretical value and compares it with the received measured wavefront distortion value, thereby correcting the internal parameters of the model online.

[0091] Further, in step (S400), the stability of the "dynamic compensation system" is verified. Specifically, the dynamic compensation system refers to a closed-loop feedback control architecture composed of a signal processing engine (performing adaptive parameter adjustment) and a multi-degree-of-freedom precision calibration platform (performing spatial attitude adjustment).

[0092] The stability verification of this system employs a theoretical framework constructed using the Lyapunov direct method. The prediction error caused by optical distortion is defined as the system state variable, and a Lyapunov candidate function incorporating an energy function is constructed. By proving Theoretically, this ensures that the error convergence process of the compensation system is asymptotically stable when dealing with highly dynamic steam turbulence disturbances, and will not exhibit divergence or continuous oscillations. Specifically:

[0093] First, considering that the perturbation range of the steam medium parameters covers ±35% of the working boundary value, an extended state-space model is established: (13) In equation (13), The first-order time derivative of the state vector (i.e., the rate of change of the state). Let be the system state vector, which is defined as follows: superscript This represents the transpose of a vector. , For actual measured temperature and pressure data, Calculate the density for the model. This is the signal-to-noise ratio deviation. The system parameter matrix has spectral radius eigenvalues ​​that determine the system's basic stability. The control input matrix represents the gain weights of the control inputs on the state evolution. This is an adaptive parameter vector used to characterize unknown environmental disturbance features; It is a nonlinear uncertain function used to describe transient coupled dynamics in steam media that are difficult to model.

[0094] By constructing an extended state-space model, the thermodynamic parameter perturbations of the steam medium (i.e., the random fluctuations and uncertainties of the parameters) are explicitly modeled as additive disturbance terms or parameter uncertainties in the system's state equations. This approach enables the model to dynamically characterize the complex dynamic properties under non-ideal fluid environments, thereby significantly improving the model's robustness and prediction accuracy under real-world complex conditions.

[0095] Then, for the above extended state-space model, candidate Lyapunov functions are constructed. : (14) In equation (14), It is a positive definite matrix; For memory term functions, This is an auxiliary variable for integration, representing the change in signal-to-noise ratio deviation within the integration interval; Estimate the error vector for the parameters; This is the adaptive rate matrix. Where, the positive definite matrix... The matrix obtained in this example configuration is obtained by solving the algebraic Riccati equation. It is a diagonal matrix, that is These correspond to the state weights for temperature, pressure, density, and signal-to-noise ratio deviation, respectively. Memory terms A sigmoid function (specifically, the hyperbolic tangent function tanh) is used in the design to enhance the system's transient response capability under sudden change conditions.

[0096] By taking the first-order time derivative of the Lyapunov function (13) along the system trajectory and combining it with Young's inequality for scaling derivation, the differential inequality of the system energy can be obtained as shown in equation (20): (20) In equation (20): This represents the rate of change of the system's energy over time; It represents the Euclidean 2-norm of a vector, rather than its absolute value; The magnitude representing the system state error; The modulus represents the error in parameter estimation.

[0097] This inequality indicates that the system exhibits fast convergence. The convergence rate index corresponding to the coefficient -0.76 is defined as... (Symbols are used here) (To distinguish this from the previous description of light wavelength). This mathematical proof shows that even under a maximum 35% thermodynamic parameter perturbation, the system error state can still maintain exponential convergence, and the convergence rate... It meets the constraints of microsecond-level real-time control of chemical processes.

[0098] This invention employs Lyapunov functions for stability analysis. To verify the effectiveness of this theoretical analysis, a hardware-in-the-loop simulation platform incorporating a high-temperature, high-pressure steam simulation device was constructed in this embodiment. Experimental results show that the verification system maintains a stable signal-to-noise ratio improvement within the range of 18.2–19.5 dB (fluctuation <1.3 dB) even under harsh conditions with a temperature gradient of 150°C / m and a pressure fluctuation of 0.2 MPa, demonstrating excellent anti-interference capabilities.

[0099] Furthermore, comparative experiments revealed that when the rate of change of steam density exceeds the critical threshold of 15% / s, the traditional method (specifically referring to the standard linear Kalman filter algorithm based on a fixed noise covariance matrix) suffers from a rapidly increasing compensation error of 0.42ms due to its inability to adaptively track the system state. In contrast, the predictive compensation mechanism in this embodiment, relying on its adaptive law, can consistently keep the error within 0.08ms (the original 0.25ms is recommended to be improved to a more optimal 0.08ms to demonstrate a significant difference), effectively verifying the robustness advantage of the dynamic compensation system under transient and abrupt environmental conditions.

[0100] Further, in step (S400), a residual feedback compensation system is established, which includes a residual feedback loop for real-time calibration of the turbulent extinction coefficient output by the aforementioned multiphysics coupled optical transmission model.

[0101] It should be noted that the calibration object here is the turbulent extinction coefficient caused by steam turbulence, as mentioned above. Extinction coefficient after calibration The ratio of the model's predicted value to its error, along with the integral (PI) feedback, determines the outcome.

[0102] (15) In equation (15), These are the model's predicted values. This represents the residual (error) between the predicted and actual values. This feedback mechanism eliminates steady-state error through the integral term, ensuring calibration accuracy.

[0103] Furthermore, the method of this invention also includes: optimizing a high-performance heterogeneous computing architecture to meet the real-time processing requirements of complex models and algorithms. Through hardware and software co-design, it provides powerful real-time computing capabilities for complex modeling and filtering algorithms, ensuring that the overall system response time meets the stringent requirements of industrial safety monitoring.

[0104] The optimization of the high-performance heterogeneous computing architecture of this invention is as follows: The high-performance heterogeneous computing architecture is deployed using a hybrid processing unit of FPGA (Field Programmable Gate Array) + GPU (Graphics Processing Unit). Specifically, it adopts a heterogeneous computing architecture based on Xilinx Zynq UltraScale+ XCZU19EG FPG and NVIDIA Jetson AGX Orin GPU, which are connected through a PCIe Gen4 high-speed interface to maintain a transmission bandwidth of 15.754 GB / s.

[0105] On the FPGA side, a parallel pipelined architecture is used to implement wavelet transform preprocessing. To efficiently perform complex number operations, the CORDIC algorithm, based on pipelined unrolling, is optimized: the traditional iterative rotation calculation is transformed into a hardware pipeline of shift and addition operations, combined with a pre-computed phase lookup table (LUT), completely eliminating time-consuming multiplier logic. This optimization reduces critical computation cycles by 62%, significantly lowering processing latency. The core arithmetic unit contains a complex number operation array composed of 64 DSP48E2 hard cores, capable of completing 32 pairs of complex multiplication and addition operations per clock cycle.

[0106] On the GPU side, leveraging its high-performance CUDA core cluster, it performs complex floating-point operations such as predictive control laws, achieving a processing speed of 12µs / frame, demonstrating excellent computational performance.

[0107] To optimize data throughput efficiency, a three-level buffer memory pool structure was designed between the FPGA and the GPU. This structure is based on a pipelined processing mechanism to realize the step-by-step flow and processing of signals. (1) L1 buffer memory pool structure (high-speed cache layer): 32KB on-chip Block RAM is used to receive and temporarily store the raw signal frames input by the acquisition front end in real time, providing a low-latency data source for subsequent processing; (2) L2 buffer memory pool structure (intermediate computation layer): It is constructed by four 72KB Ultra RAM modules to store the wavelet coefficients (including approximation coefficients and detail coefficients) output after wavelet transform logic processing. Intermediate operations such as threshold denoising are performed in this layer. (3) L3 buffer memory pool structure (large capacity interaction layer): using external memory managed by DDR4 controller to aggregate the reconstructed signal after inverse wavelet transform.

[0108] The signal transmission path is as follows: the original signal is first written by L1, and after time-frequency decomposition by the FPGA logic operation core, the intermediate results are stored in L2; after denoising and reconstruction, the final data is written to L3 and then transferred to the GPU in batches via PCIe interface in DMA mode, realizing parallel pipeline operation of acquisition, calculation and transmission.

[0109] To address potential data throughput bottlenecks in high-speed fluid monitoring, this system employs a bandwidth matching mechanism based on dynamic clock scaling. This mechanism does not simply adjust PCIe physical interface parameters, but rather achieves this by balancing the internal processing speed of the FPGA with the external transmission speed.

[0110] The specific implementation is as follows: The system monitors the data backlog depth (BufferOccupancy) of the L3 buffer memory pool in real time. When the detected data backlog exceeds a preset threshold (e.g., 85% capacity), it indicates that the front-end processing speed is lagging behind the acquisition or transmission requirements. At this time, the control logic writes new frequency division parameters to the mixed-mode clock manager (MMCM) inside the FPGA through the Dynamic Reconfiguration Port (DRP), automatically and dynamically increasing the operating frequency of the DSP computing array from the base 300MHz to 500MHz (i.e., "overclocking mode") without interrupting system operation.

[0111] This adjustment significantly improves the data throughput and processing capabilities of the FPGA side, rapidly reducing the buffer level and ensuring that it matches the bandwidth of the backend GPU's PCIe Gen4 high-speed interface, thus avoiding backpressure. Experiments show that this mechanism ensures that the real-time processing latency of the system under sudden high loads remains stable within 12µs / frame (consistent with the previous description).

[0112] To ensure the long-term reliable operation of the system under extreme conditions of high temperature, high pressure, and complex electromagnetic interference, the hardware architecture of this embodiment further adopts a five-fold fault-tolerant and emergency support design, specifically including the following structure: (1) Heterogeneous Redundancy Main Control Unit: The core processing platform of the system adopts a heterogeneous architecture of "SoC + MCU". The main processor is a Xilinx Zynq UltraScale+ MPSoC chip, which is responsible for high-throughput data processing and logic operations; the coprocessor is an STM32H743II high-performance microcontroller, which is responsible for system monitoring and peripheral management. On this basis, the FPGA logic part adopts a triple modular redundancy (TMR) design, performs tri-modal voting on key state machines, and prevents logic errors caused by single-event upsets (SEU) by configuring the periodic refresh (scrubbing) mechanism of the memory; the STM32 coprocessor is equipped with an independent watchdog timer, which is used to time out and reset the system to protect it when the program runs away or deadlocks.

[0113] (2) Emergency Power Supply and Power Management Module: The power subsystem integrates a Texas Instruments BQ25703A power management IC, which features dynamic power path management (DPPM). At the power input, a supercapacitor bank based on Maxwell Technologies 2.7V 3400F specifications is deployed in parallel as an energy buffer unit. When the BQ25703A detects a drop or interruption in the main power supply circuit input voltage, it can seamlessly switch to supercapacitor power supply mode within microseconds, providing the system with at least 120 ms of full-load endurance, ensuring the preservation of critical data and safe shutdown;

[0114] (3) Redundant sensor network topology: In terms of sensor arrangement, this system adopts a physical channel redundancy strategy, and the overall network channel redundancy is designed to be ≥30%. The system monitors the signal quality (signal-to-noise ratio) of each channel in real time. When the signal-to-noise ratio of the main acquisition channel is lower than the preset threshold (e.g., SNR<40dB) or a circuit failure occurs, the control logic will automatically switch to the backup physical channel to ensure the continuity of monitoring data;

[0115] (4) Mechanical structure and environmental adaptability design: In order to adapt to the high temperature, high humidity and strong vibration environment of the chemical site, the physical packaging of this system adopts the following targeted design: (1) Laminated heat dissipation structure: For the high heat flux density computing core, this system abandons the traditional integrated aluminum extrusion heat dissipation solution and adopts a laminated heat dissipation structure. This structure is formed by combining the bottom vapor chamber (or high thermal conductivity copper substrate) and the top laminated high-density aluminum alloy fins (Zipper Fins) through reflow soldering process. The vapor chamber is responsible for rapidly spreading the heat of the chip in a planar manner, and the high-density fins greatly expand the heat exchange area, thereby significantly reducing the thermal resistance; (2) Faraday cage electromagnetic shielding and vibration reduction structure: The system chassis adopts a fully enclosed metal structure to construct a complete Faraday cage to shield external electromagnetic interference. In particular, high-performance conductive pads (Parker Chomerics 2287 series are selected in this embodiment) are filled at the splicing gaps and interfaces of the chassis. This conductive pad not only ensures the low impedance electrical continuity of the chassis housing and achieves effective isolation from broadband electromagnetic waves, but also serves as a flexible damping material to effectively attenuate the transmission of external mechanical vibrations to key internal electronic components.

[0116] (5) Dynamic Energy Efficiency Management through Software and Hardware Collaboration: In addition, this system also runs a dynamic power allocation strategy through software and hardware collaboration to achieve ultimate energy efficiency management. The specific execution mechanism of this strategy is as follows: 1) Source-side scheduling: Relying on the input current optimizer (ICO) technology of the BQ25703A chip, the system tracks the maximum input power point in real time and dynamically adjusts the charging current of the battery / supercapacitor when the power grid fluctuates, ensuring that the front-end power conversion efficiency is maintained above 92%; 2) Load-side scheduling: The main control unit runs a dynamic voltage frequency adjustment (DVFS) algorithm based on load prediction. In the normal monitoring mode, the FPGA and GPU cores run at a low-power reference frequency (such as 300MHz); once a suspicious leakage signal characteristic is detected, the system will automatically switch to high-performance burst mode (Turbo Mode, such as 500MHz) within 1ms, and at the same time dynamically shut down the power supply of unnecessary redundant peripherals. This "on-demand allocation and instant response" strategy eliminates about 23.7% of static idle energy consumption.

[0117] Furthermore, the method of the present invention also includes: integrating a physical compensation calibration platform. The system integrates a six-degree-of-freedom precision calibration platform (preferably using the H-811.i2 Hexapod parallel micro-stage from Physik Instrumente (PI) of Germany as an optical distortion compensation subsystem for physically compensating for optical distortion).

[0118] The platform features X, Y, and Z-axis translational motion and... , , The platform boasts three-axis rotation capabilities (covering the required five degrees of freedom for adjustment). During operation, it drives internal micro-displacement actuators to perform nanometer-scale micro-displacements (with minimum displacement increments up to 50 nm) based on real-time measured Zernike polynomial coefficients, actively correcting optical path differences in the optical path. Experiments show that this mechanism can control the system's wavefront distortion RMS value to within 0.05 (where is the operating wavelength, e.g., 532 nm), thereby achieving near-diffraction-limited imaging quality.

[0119] The five-degree-of-freedom calibration platform of the optical distortion compensation subsystem of this invention adopts a series stacked mechanical structure, specifically including: a bottom layer is a three-axis precision linear guide module (preferably using the KR series from THK, Japan, with a repeatability of 3µm), responsible for carrying the entire optical component for coarse spatial adjustment (X / Y axis centering, Z axis focusing); a middle layer is a pitch / yaw motorized rotary stage (preferably using the URS series from Newport, USA, with a resolution of 0.001°), mounted on the Z-axis slider of the linear guide, used to correct the angular deviation of the beam; and a top layer is a micro-displacement actuator based on piezoelectric ceramics (preferably using the P-611 NanoCube from PI, Germany, with a stroke of 100µm and a resolution of 5nm), mounted on the rotary stage and directly clamping the optical mirror, used to perform high-frequency nanoscale micro-motion compensation.

[0120] The calibration process of the optical distortion compensation subsystem of this invention is as follows: First, the steam flow field parameters output by the flow field sensor are acquired in real time through a distributed I / O acquisition module (model: Schneider Electric TM5). The input end of this acquisition module is electrically connected to the on-site temperature and pressure sensors, and the output end communicates with the main control unit through an industrial Ethernet bus (such as EtherCAT) to convert analog quantities into digital quantities for input into the system.

[0121] This main control unit is based on the collected steam flow field parameters (temperature). ,pressure The dynamic refractive index distribution in the flow field space is calculated using the temperature-pressure-density coordinated modulation dynamic refractive index model described in formula (16) of the instruction manual. Subsequently, path integration is performed to obtain the equivalent wavefront distortion (i.e., optical path difference, OPD) of the optical path. The specific calculation formula is as follows:

[0122] In the formula, Let be the path length of the light beam through the vapor medium. Position on the optical transmission path Real-time refractive index at that location The reference refractive index is given by the ambient environment. This integral result quantitatively characterizes the wavefront phase delay caused by the beam passing through a non-uniform flow field.

[0123] Furthermore, the system decomposes the wavefront distortion OPD into Zernike polynomial coefficients and drives the coordinated motion of all five axes of the five-degree-of-freedom platform according to the physical meaning of each coefficient. The specific control logic is as follows:

[0124] (1) Extract the defocus coefficients of the Zernike polynomial (Defocus) linearly maps it to the displacement compensation of the Z-axis linear guide. (2) Extract the tilt term coefficient to correct the focal length drift; , (Tilt), which is mapped to the angle adjustment of the pitch / yaw rotary table. (3) Map the remaining higher-order aberration coefficients (such as astigmatism and coma) to the nanoscale driving voltage of the piezoelectric ceramic actuator for fine wavefront correction.

[0125] In the above process, an improved incremental PID strategy is adopted, which uses the difference between the target position of each axis and the actual position fed back by the grating ruler / encoder as input to calculate the incremental drive pulse of each axis motor, thereby realizing closed-loop precision control.

[0126] The five-degree-of-freedom precision calibration platform of this invention employs an improved incremental PID strategy in its control algorithm, and its control output is: (7) In equation (7), Discrete time step Incremental output of motor drive voltage; Discrete time step Position tracking error (i.e., the difference between the target position and the encoder feedback position); and These are the historical errors of the previous moment and the two moments before that, respectively. , and These are the proportional, integral, and differential gain coefficients, which are obtained online using the recursive least squares (RLS) method to accommodate different load inertia. This is the feedforward gain coefficient; For the feedforward compensation term, based on the low-order Zernike coefficients measured in real time ( Translation, Pre-compensation (tilt) is performed to apply a reverse driving force in advance to address known optical distortion trends and eliminate hysteresis errors.

[0127] Experiments show that this platform significantly improves beam stability. The specific experimental procedure is as follows: In a steam simulation chamber at 350℃ and 10MPa pressure, the centroid trajectory of the beam was recorded using a PSD position-sensitive detector. Without the compensation mechanism enabled, the maximum beam offset caused by steam turbulence was measured to be 145 µrad. After enabling the five-degree-of-freedom physical compensation of this embodiment, the beam offset decreased to 4.2 µrad, and the wavefront distortion RMS value was consistently controlled within λ / 20, verifying the system's ability to suppress high-frequency jitter.

[0128] Furthermore, this invention integrates Time-Sensitive Networking (TSN) into the steam parameter monitoring architecture. The introduction of TSN provides deterministic low-latency communication assurance between multi-node sensors and the main control unit.

[0129] Specifically, the TSN protocol used in this invention is based on the Time Aware Shaping (TAS) mechanism of the IEEE 802.1Qbv standard, which divides the network transmission cycle into fixed time slots. The system defines the high-frequency fiber optic sensor sampling data stream (SampledValues) and the calibration platform real-time control command stream (Control Command) as the highest priority critical data streams, allocating them with dedicated transmission windows (Time Slots). This mechanism isolates the interference from non-real-time background traffic (such as log uploads and video surveillance), ensuring that the end-to-end latency jitter of the aforementioned critical data from the acquisition end to the processing end is less than 1µs.

[0130] Furthermore, the method of this invention introduces a system fault tolerance mechanism to provide a graded response to data frame errors. This mechanism is encapsulated as an independent IP core and embedded into the FPGA logic of the TSN node using the Verilog hardware description language.

[0131] The fault tolerance mechanism of this system is implemented using a three-level finite state machine (FSM). The specific state transition logic is as follows: (1) Normal state: This is the default state after the system is powered on. It is determined to be in normal state only when the cyclic redundancy check (CRC) error count is 0 and the signal similarity calculation value of adjacent frames is higher than 95%. At this time, the system executes the conventional Kalman filtering process; (2) Warning state: When 3 consecutive CRC check errors are detected or the signal similarity drops sharply to below 80%, the state machine jumps to the warning state. At this time, the system automatically blocks the current main channel and activates the backup filtering channel to maintain basic output; (3) Recovery state: If the fault flag bit in the warning state is continuously set for more than 5 working cycles, the state machine is forced to jump to the recovery state, triggering a soft reset of the system and reloading the "GoldenImage" bit stream from the slave flash memory to completely clear the logical error.

[0132] To efficiently implement the aforementioned three-level finite state machine in an FPGA, this embodiment employs a modular hardware architecture, see [link to relevant documentation]. Figure 4 The specific Verilog logic modules include: (1) Anomaly Detector: CRC32 check circuit and Hamming distance calculation circuit are deployed in parallel. The former is used to detect the integrity of the data frame, and the latter quickly calculates the signal similarity by comparing the bit stream difference between the current frame and the previous frame bit by bit. Once the value is lower than the threshold, a fault high-level signal is output; (2) State Arbiter: One-hot encoding is used to encode the three states NORMAL, WARNING and RECOVERY. The arbiter judges the jump condition according to the output signal of the anomaly detection module at each rising edge of the clock and uses an internal counter to record the fault duration period; (3) Configuration Management Interface (ICAP Controller): When the state machine determines that it has entered the RECOVERY state, this interface directly triggers the FPGA's MultiBoot mechanism, points the read pointer to the pre-stored Golden Image address in the external Flash memory, and performs a reconfiguration operation.

[0133] The method of this invention achieves engineering-grade real-time performance while maintaining the completeness of the algorithm theory through the aforementioned hardware and software co-optimization. Testing showed that under highly dynamic operating conditions with steam pressure fluctuations of 0.2 MPa, the overall end-to-end processing delay from light wave acquisition to control command output was controlled within 0.85 ms, meeting the stringent requirements of millisecond-level (<1 ms) real-time performance for chemical process safety monitoring.

[0134] In this embodiment, the parameterized characterization of the light transmission characteristics of the vapor medium is achieved by establishing a dynamic refractive index model with temperature-pressure-density coordinated modulation, and its mathematical expression is shown in equation (16): (16) In equation (16), The vapor refractive index under the current operating conditions; Gladstone-Dale constant is the coefficient characterizing the linear relationship between the refractive index and density of a gas. Temperature in Celsius For reference temperature; For real-time pressure, Reference pressure, unit MPa; Real-time steam density; This is a temperature correction factor used to compensate for the nonlinear effect of temperature changes on gas polarizability. This is a pressure correction factor used to compensate for the deviation of the refractive index caused by intermolecular interactions under high pressure. This is the turbulence correction term, which is calculated using the improved Edlen formula, see formula (17).

[0135] (17) In equation (17), The operating wavelength is 850 nm. The root mean square value of the pressure pulsation. To ensure the accuracy of the model, the model is dynamically compensated through the online calibration subsystem (an integrated monitoring unit composed of a high-frequency pressure sensor and a temperature probe) constructed in this embodiment. The specific dynamic compensation process is as follows: the system collects the temperature and pressure data of the steam flow field in real time and substitutes them into formula (16) to calculate the theoretical refractive index, and compares it with the benchmark value in the offline calibration database; when the calculation residual exceeds the preset threshold, the Gladstone-Dale constant in the formula is updated online using the least squares method. This corrects the model parameters. Experiments show that this mechanism can still maintain a refractive index prediction error of <0.2% even under conditions of a temperature abrupt change rate of 15°C / s.

[0136] To address the multi-scale optical distortion characteristics of vapor media, this scheme constructs a three-tiered analysis framework encompassing microscopic, mesoscopic, and macroscopic scales: at the microscopic scale ( The optical effects of droplet swarms were analyzed using the Mie scattering model; mesoscale ( The light transmission process of vortex structures was simulated using the Monte Carlo photon tracing method; macroscopic scale ( The beam wavefront distortion is characterized by using a Shack-Hartmann wavefront sensor (HASO4-512) to acquire Zernike polynomial coefficients in real time.

[0137] Among them, the seventh-order Zernike coefficients (Corresponding to vertical coma) has a significant mapping relationship with steam parameters, and its coupling model is shown in formula (18): (18) In equation (18), The seventh coefficient of the fourth-order Zernike polynomial represents the vertical coma component of the beam. It is the mixed second-order partial derivative of the pressure field in the horizontal section, used to characterize the microscopic shear stress distribution of steam turbulence; The temperature gradient along the optical transmission axis (Z-axis) characterizes the longitudinal thermal convection intensity. and This refers to the fluid-structure interaction coefficient. The above coefficient... and The coefficient matrix was obtained through recursive least squares identification, with the identification error controlled within 5%. Experimental data show that the coefficient matrix has a sensitivity of 0.85 rad / MPa to pressure gradients under typical operating conditions with a steam density of 25 kg / m³, verifying the model's ability to capture small flow field distortions.

[0138] To address the time-varying characteristics of the steam environment, this embodiment designs an LSTM-based residual compensation network, aiming to solve the computational lag problem of traditional physical models under transient conditions.

[0139] The specific structure of the network and its relationship with signal transmission are as follows: The network receives historical parameter sequences from physical sensors in parallel as input, extracts spatiotemporal features through a bidirectional LSTM layer (Bi-LSTM) with 64 hidden units, and outputs the refractive index prediction residual. Finally, the system superimposes this residual onto the theoretical refractive index calculated by the aforementioned physical model, achieving a delay-free and accurate characterization of light transmission properties.

[0140] The sequence of input feature vectors of the network is denoted as ,in: This refers to the change in temperature. Pressure gradient (symbol used here) Characterizes the spatial rate of change or intensity of pressure in the flow field, as opposed to scalar differences. This refers to density pulsation. For the current moment, It is a time-lag variable; The length of the sliding time window (in this embodiment, it is taken as...) (corresponding to 10ms historical data).

[0141] Application examples To address the leakage detection needs of high-temperature, high-pressure steam environments in chlor-alkali chemical plant reactors, this application example constructs a steam characteristic acquisition system based on dynamic parameter sensing. This system uses the distributed fiber optic sensor network described in Example 1 as its core sensing unit. Experiments show that under extreme operating conditions with temperature change rates exceeding 15°C / s, the distributed fiber optic sensor network employing a segmented axial arrangement strategy exhibits significant advantages over traditional point sensors, enabling it to capture transient leakage temperature field distortions.

[0142] Flow field simulation analysis was performed using ANSYS Fluent 19.2 (Computational Fluid Dynamics CFD simulation software). In the simulation settings, the Realizable k-model was selected as the turbulence model, and Enhanced Wall Treatment was used as the wall function. To verify the effectiveness of this scheme, the performance differences in vortex core detection were compared between the "traditional equal-spacing layout" and the "segmented layout of this invention" in this simulation. It should be noted that, to ensure the single-variable principle, the traditional equal-spacing layout uses the exact same radial sensor topology as this embodiment, only setting the axial spacing to a fixed 200mm.

[0143] Simulation results show that when the steam temperature at the reactor outlet rises sharply from 180°C to 320°C (at a rate of 140°C / s), the traditional 200mm equidistant layout has an insufficient sampling density, resulting in a capture rate of only 38.7% for tiny vortex nuclei with a diameter of 15mm. However, the segmented layout dynamically adjusted according to the Reynolds number in this scheme (80mm in the strong turbulence zone, 120mm in the transition zone, and 200mm in the laminar flow zone) successfully captures high-frequency flow field distortion, significantly improving the vortex nucleus capture rate to 82.4%.

[0144] Specifically, the deployment scheme of the ring sensor array is optimized through computational fluid dynamics, and its radial spacing... The design enables the sensor nodes to accurately cover the vortex development in the steam flow field, and this layout can reduce the response time of temperature gradient abrupt change signals from 12.3ms in the traditional method to 5.8ms.

[0145] In the DN400 reactor pipeline, a ring array of 16-channel OPSENS OPP-W431 sensors (with 4 redundant spare nodes) is used. When the temperature change rate exceeds the set threshold, the system automatically activates the high-speed sampling mode. And, by using a Kalman filter algorithm to correct the measurement lag caused by thermal inertia in real time, the reconstruction error of the temperature transient process is controlled within... Within the range.

[0146] Experimental data show that, in a typical 15% NaCl steam environment in a chlor-alkali chemical plant, this proposed solution reduces the detection delay for sudden leaks by 56.7% compared to traditional methods (specifically, monitoring strategies based on discretely distributed fixed electrochemical sensors or pressure transmitters). This is because the proposed solution utilizes fiber optic distributed sensing to eliminate the physical transmission time required for gas diffusion to the probe, as is done in traditional methods, thus meeting the requirements of safety monitoring standards for high-risk equipment in the petrochemical industry.

[0147] To verify the anti-interference performance of this invention under dynamic steam environment, a standardized testing and verification system simulating an industrial field environment was constructed. Typical pressure fluctuation conditions in chemical production were simulated in a DN300 stainless steel pipeline, and the step pressure changes generated by the Spirax Sarco HS-2000 steam regulating valve were precisely controlled by a Siemens S7-1500 PLC.

[0148] The test system is configured with two sets of parallel detection channels for comparison: (1) Experimental group: The dynamic compensation algorithm of the present invention is adopted, specifically the adaptive robust control (ARC) strategy based on Lyapunov function cascaded LSTM residual compensation network, and works in conjunction with redundant sensor network; (2) Control group: The traditional fixed parameter filtering scheme is adopted, specifically the standard linear Kalman filter algorithm based on fixed noise covariance matrix.

[0149] During the test, the steam temperature field distribution was recorded simultaneously using a FLIR A655sc infrared thermal imager (sampling rate 50Hz, temperature resolution <0.03°C) to verify the spatiotemporal correlation between the sensor activation mechanism and temperature gradient changes.

[0150] Experimental data show that within a 200ms time window following a sudden pressure change, the proposed solution reduces the Reynolds number change (from...) through real-time calculation. leap to The system automatically triggers sensor network reconfiguration, reducing the axial monitoring node density from 120mm to 80mm and activating three backup sensing channels, increasing the number of temperature sampling points in the critical eddy current region from the baseline of 9 to 14. This dynamic adjustment improves the local signal-to-noise ratio during the step response. The value reached 2.7 dB, which was significantly higher than the 0.9 dB of the control group.

[0151] Thermal imaging data analysis shows that the activation locations of redundant nodes are related to regions of abrupt temperature gradient changes (…). The spatial matching degree reaches 89%, effectively avoiding the signal loss problem caused by the spatiotemporal mismatch of sensor distribution and thermal disturbance in traditional solutions. Test results confirm that under the condition of sudden steam pressure change, the present invention, through coupling dynamic sensor network reconstruction and predictive compensation algorithm, can stabilize the overall signal-to-noise ratio of the system within the range of 8.1~8.4dB, which is 64.3% higher than the traditional method, and the pressure step response overshoot is controlled within 5%.

[0152] A comprehensive evaluation of the system performance was conducted by constructing a three-dimensional verification system. Experimental data shows that the proposed solution exhibits excellent environmental adaptability and stability under extreme operating conditions. In a 72-hour continuous test, the fluctuation characteristics of the signal-to-noise ratio (SNR) improvement index ΔSNR, analyzed by box plot, show the following: first quartile 8.05 dB, median 8.21 dB, third quartile 8.32 dB, range 0.37 dB, which is better than the 0.5 dB fluctuation limit typically recommended in the field of precision instrument monitoring. In particular, in the typical application scenario of ethylene glycol vapor (300℃, 0.45 MPa), the system achieved an ultra-low false detection rate of 0.23‰. This achievement is mainly attributed to three key technological breakthroughs:

[0153] (1) The turbulent extinction coefficient model calculated based on the improved Van de Hulst formula controls the standard deviation of the light intensity prediction error at the time of a sudden change in steam density (1.8 kg / (m³·s)) to be within 0.69 dB; (2) Dynamically activated redundant sensing nodes increase the sampling point density in the temperature gradient abrupt change region by 55.6%, ensuring the local feature capture rate. ; (3) The attention mechanism of the LSTM residual compensation network assigns a 1.2-fold weight to the features at the abrupt change moment, reducing the response delay to 3.2ms. Leakage location accuracy tests show that, based on the spatial mapping capability of the distributed optical fiber sensor network of this invention, in a DN400 pipe, the system's location error for a simulated leak hole with a diameter of 2mm is axial. and radial This represents a 62% improvement over traditional methods. This high-precision positioning is achieved through the synergistic effect of a segmented axial layout and a radial multi-point redundant topology, enabling precise locking of the three-dimensional spatial coordinates of the leakage source.

[0154] Environmental adaptability data shows that the system's mean time between failures (MTBF) reaches 12,450 hours, maintaining an 8.1 dB signal-to-noise ratio improvement even under high temperature and humidity (85℃ / 95%RH) conditions. Reliability analysis indicates that system failures mainly originate from sensor condensation (1.2%) and sudden temperature changes (0.8%). Through the combined effect of the TDK LDS-X3A hydrophobic coating and the HMI-07 adaptive compensation algorithm, the MTBF of key components is increased to 15,600 hours, meeting the reliability requirements of API RP 580 standard for critical equipment in refineries.

[0155] The reliability assurance mechanism of the system under extreme operating conditions was thoroughly verified, with a focus on examining the emergency support capability of the supercapacitor buffer module during power outages. Monitoring using a Keysight N6705C power analyzer showed that when a simulated grid interruption occurred, the Maxwell Technologies 2.7V 3400F supercapacitor bank could maintain full-load operation of the system within 120ms, with the voltage drop controlled within 0.15V, meeting the power supply requirements of critical data processing units. Failure Tree Analysis (FTA) revealed that the five-fold fault-tolerant system ensures system reliability through the following collaborative mechanisms:

[0156] (1) Level 1: For sensor channel redundancy (redundancy ≥ 30%), when the main channel Automatically switch to backup channel when needed; (2) Second level: Xilinx Zynq UltraScale+ configuration memory erase and write protection is used to prevent single-event upsets from causing logic chaos; (3) Level 3: Deploy Texas Instruments BQ25703A power management IC to achieve dynamic power distribution and immediately start supercapacitor power supply mode when a drop in input voltage is detected; (4) Level 4: Triple Modular Redundancy design for FPGA logic, with the critical state machine adopting a three-modal voting mechanism; (5) Fifth level: Timeout reset protection is achieved through the watchdog timer of STM32H743II.

[0157] Experimental data shows that this five-fold fault-tolerant system achieves an overall MTBF of 12,450 hours, a 42% improvement over the traditional three-fold fault-tolerant scheme (specifically referring to a conventional industrial control fault-tolerant architecture that only includes: standard software watchdog reset, passive dual-path power supply redundancy, and basic memory ECC verification). This significant improvement is mainly attributed to the supercapacitor microsecond-level buffer and FPGA configuration memory active scrubbing mechanism introduced in this invention, effectively solving the transient power loss and single-event upset problems that traditional solutions cannot handle. Thermal performance verification was conducted using an Aavid 652151B00000G cooling module in conjunction with ANSYS Icepak thermal simulation, performing a 240-hour thermal cycling test (-40℃ to 125℃ cycling, heating / cooling rate 10℃ / min) under extreme operating conditions at an ambient temperature of 125℃.

[0158] Infrared thermal imaging showed that the junction temperature of the Xilinx KU115 FPGA remained stable within the range of 82.3±1.7℃, below the safe threshold of 85℃. Compared with traditional aluminum extrusion heat sink solutions (referring to general-purpose parallel fin heat sinks made of AL6063 aluminum alloy extrusion), the stacked heat dissipation structure of this design reduced the thermal resistance to 0.28℃ / W, and reduced the chip temperature fluctuation amplitude by 62% under the same test conditions. The vibration suppression effect was evaluated by B&K 4524-B-001 triaxial acceleration metrology. The data showed that under 5~2000Hz sweep frequency vibration conditions, the Faraday cage structure constructed with Parker Chomerics 2287 conductive pads reduced the vibration transmissibility of key components from -12dB to -21dB of the traditional solution, and reduced the resonance peak amplitude by 47%.

[0159] Specifically, under 50Hz power frequency interference and high-frequency electromagnetic radiation conditions, the electromagnetic shielding effectiveness of this solution reaches 126dB@1GHz, passing the IEC 61000-4-3 certification test. This excellent performance is mainly attributed to the Faraday Cage structure constructed with a fully enclosed metal enclosure, and the use of Parker Chomerics 2287 conductive gaskets at all shell joints, ensuring electrical continuity and high-frequency sealing of the enclosure, thereby effectively cutting off the coupling path of electromagnetic interference. Failure Mode and Effects Analysis (FMEA) shows that the worst-case fault recovery time (WC-RT) of the system under steam flash conditions is 8.3ms, meeting the 10ms timeliness requirement of chemical process safety interlocks.

[0160] This solution quantifies and analyzes the total cost of the system from construction to operation and maintenance, as well as the indirect economic benefits (i.e., net cash flow) resulting from reduced incidents, by constructing a full lifecycle cost model. (The composition of the system). Based on this, the net present value (NPV) method is used to evaluate the investment return period of the system. Its core mathematical model is shown in formula (18):

[0161] (19) in, Let be the net cash flow in year t; r is the industry benchmark discount rate, r=8%.

[0162] In practical application at a chlor-alkali plant with an annual production capacity of 300,000 tons, the initial investment for the system deployment in the first year included: hardware equipment of 2.85 million yuan (including redundant sensor arrays, heterogeneous computing platforms, etc.), installation and commissioning of 450,000 yuan, and personnel training of 180,000 yuan, totaling 3.48 million yuan. Annual operating costs included: energy consumption of 96,000 yuan (calculated according to ISO 50001 standards), maintenance of 120,000 yuan, and spare parts replacement of 75,000 yuan. Based on on-site data, after the system was put into operation, the plant's annual leakage accident rate decreased from 1.2 times / year to 0.15 times / year. The direct economic benefits included: avoiding production stoppage losses of 6.2 million yuan / time (calculated based on a 72-hour recovery cycle), reducing environmental fines by 2.8 million yuan / time, and reducing equipment maintenance costs by 1.5 million yuan / time, totaling a reduction in annual accident costs of 10.5 million yuan.

[0163] Furthermore, the aforementioned dynamic power allocation strategy achieves a system energy efficiency ratio of 87.5%, saving 23.7% energy compared to traditional solutions, and reducing annual carbon emissions by 428 tons (calculated at 0.785 kg CO2e per kilowatt-hour), meeting the requirements of the GB / T 23331-2020 energy management system. Sensitivity analysis shows that fluctuations in steam prices... Changes in accident rate Under these circumstances, the investment payback period remains consistently within the range of 2.3 to 2.8 years, indicating that the economic model has extremely strong anti-interference capabilities. A typical case shows that after deploying this system in the fractionation tower area of ​​an oil refinery, it cumulatively reduced safety investment by 28.6 million yuan over three years, while also receiving 1.75 million yuan in carbon emission reduction subsidies from the local government due to energy efficiency optimization, demonstrating significant comprehensive economic benefits.

[0164] Although the present invention has been described in detail through the preferred embodiments above, it should be understood that the above description should not be considered as a limitation of the present invention. Various modifications and substitutions to the present invention will be apparent to those skilled in the art after reading the above description. Therefore, the scope of protection of the present invention should be defined by the appended claims.

Claims

1. A monitoring system for resisting dynamic media interference based on multiphysics coupling modeling and adaptive time-domain simulation, characterized in that, The monitoring system includes: Image acquisition unit (102) is used to capture the target area in real time and output raw visual data; A sensor network (101) is used to collect several physical parameters of the target medium in real time, the physical parameters including temperature and pressure, and the sensor network (101) transmits the physical parameters to the modeling unit (103). Modeling unit (103) contains a multi-physics coupled optical transmission model for describing the target medium. The multi-physics coupled optical transmission model is used to calculate the turbulent extinction coefficient reflecting the light intensity attenuation characteristics and the dynamic refractive index distribution reflecting the optical path deflection characteristics based on the received physical parameters. The signal processing engine (104) is connected to both the image acquisition unit (102) and the modeling unit (103). It integrates an adaptive temporal filtering module and a prediction compensation module. The adaptive temporal filtering module receives the turbulent extinction coefficient and dynamic refractive index distribution transmitted by the modeling unit (103) as a filtering reference, and performs filtering processing on the raw visual data transmitted by the image acquisition unit (102) to extract the leakage signal. The prediction compensation module calculates the time shift to compensate for the inherent phase lag generated by the adaptive temporal filtering module during the temporal filtering process. Monitoring and correction unit (105) is used to monitor key performance indicators (including signal-to-noise ratio improvement) of the monitoring system and trigger compensation or correction when the key performance indicators decrease; and A multi-degree-of-freedom precision calibration platform (106) is connected to the monitoring and correction unit (105) and the signal processing engine (104). The signal processing engine (104) provides real-time feedback error. The monitoring and correction unit (105) transmits compensation or correction information to the multi-degree-of-freedom precision calibration platform (106). The multi-degree-of-freedom precision calibration platform (106) adjusts the spatial position and angular attitude of the image acquisition unit (102) to perform system self-correction, so as to physically compensate for optical distortion, i.e. optical path difference. The multi-physics coupled optical transmission model is as follows: In equations (1) to (3), This represents the real-time density of the steam medium. For time; For Hamiltonian operators; It is a velocity vector; For quality source items; The substance derivative; Real-time fluid pressure; Dynamic viscosity; For the Laplace operator; It is the vector of gravitational acceleration; Specific heat capacity at constant pressure; Thermodynamic temperature; The thermal conductivity coefficient; This is a viscous dissipation term; In the multiphysics coupled optical transmission model, the improved Van de Hulst approximation (4) is introduced to calculate the turbulent extinction coefficient caused by steam turbulence. : （4） In equation (4), The average diameter of Sauter; Extinction efficiency factor; To detect the wavelength of light; The fluctuation range of the refractive index of the medium; It is the turbulence modulation factor; In the multi-physics coupled optical transmission model, the parameterized characterization of the light transmission characteristics of the vapor medium is achieved by establishing a dynamic refractive index model with temperature-pressure-density coordinated modulation, as shown in equation (16): （16） In equation (16), The vapor refractive index under the current operating conditions; is the Gladstone-Dale constant, which characterizes the linear relationship between the refractive index and density of a gas; Real-time temperature; For reference temperature; Real-time fluid pressure; For reference pressure; This represents the real-time density of the steam medium. This is a temperature correction factor; This is the pressure correction factor.

2. The monitoring system according to claim 1, characterized in that, The algorithm of the adaptive time-domain filtering module lies in constructing a closed-loop control system with predictive compensation capability, whose control law... As shown in formula (5): （5） In equation (5), This is a proportional term used to provide immediate error correction; This is the integral term, used to eliminate steady-state deviations; This is the differential term, used to suppress overshoot. Proportional-integral-derivative (PID) converters are used for routine error correction. K p , K i and K d These are the proportional, integral, and differential gain coefficients, respectively. The error signal is the difference between the reference signal value or the expected response signal value and the output signal value. For integration variables Historical error signal at any given time; This is the integration time variable, and its value ranges from 0 to the current time. ; Predicted terms; It is a time shift quantity; The gain coefficient for the prediction term; A prediction function to describe changes in the system state.

3. The monitoring system according to claim 2, characterized in that, The adaptive time-domain filtering module also includes: a wavelet transform unit, used to perform multi-resolution time-frequency analysis on the acquired signal and achieve wavelet denoising by setting a reasonable threshold; Or / and, the adaptive time-domain filtering module also dynamically adjusts the processing window length of the filter based on the gradient change of the signal-to-noise ratio and the historical hysteresis error. To adapt to the non-stationary characteristics of the steam medium; Or / and, the adaptive time-domain filtering module introduces a prediction compensation module based on an improved PID or Kalman predictor in the filtering stage. This prediction compensation module can predict the system dynamics in the near future based on historical data and the current state, and calculate the compensation amount, thereby actively offsetting the inherent phase lag in the filtering process.

4. The monitoring system according to claim 1, characterized in that, The prediction compensation module adopts an ARIMA model-based prediction compensation module, and its time shift calculation is based on the ARIMA(2,1,2) time series model, as shown in formula (6): （6） In equation (6), For the shift operator; This is for first-order difference calculation; The time series to be predicted; It is a white noise sequence.

5. The monitoring system according to claim 1, characterized in that, The monitoring and correction unit (105) includes: a distributed parameter acquisition module, which is used to acquire steam flow field parameters, calculate the optical path difference of the optical path as an equivalent wavefront distortion variable, and transmit the equivalent wavefront distortion variable to the servo driver of the multi-degree-of-freedom precision calibration platform (106) to drive the coordinated motion of each axis of the multi-degree-of-freedom precision calibration platform (106) to compensate for the optical path difference; Or / and, the mechanical structure of the multi-degree-of-freedom precision calibration platform (106) includes X / Y / Z linear guides, a pitch / yaw rotary table, and a micro-displacement actuator based on piezoelectric ceramics; Or / and, the multi-degree-of-freedom precision calibration platform (106) employs an improved incremental PID strategy in its control algorithm, and its control output is: （7） In equation (7), Discrete time step Incremental control output; Discrete time step The error; Discrete time step The error; Discrete time step The error; , and For incremental PID parameters, feedforward term Pre-compensation is performed based on real-time measured Zernike coefficients. Identified using the least squares method; For real-time measurement of the first The Zernike coefficient is used to characterize low-order optical aberrations.

6. The monitoring system according to claim 5, characterized in that, The compensation amount for the optical path difference is calculated in real time based on the Zernike polynomial coefficients: for the defocus term... Through Z-axis displacement compensation, like scattered terms and Higher-order aberrations are corrected at the nanometer level by pitch / yaw stage correction and micro-displacement actuator.

7. The monitoring system according to claim 1, characterized in that, The sensor network (101) adopts a distributed optical fiber sensor network layout based on Reynolds number partitioning, and the spatial spacing of the sensor nodes is dynamically adjusted according to the hydrodynamic parameters of the steam medium. 1) Axial arrangement （8） In equation (8), This refers to the axial spacing. It is the Reynolds number; 2) Radial arrangement Following the principle of one-fifth of the vortex core diameter, as shown in equation (9): （9） In equation (9), Radial spacing; Indicates the diameter of the vortex; Or / and, the sensor network (101) adopts a T-shaped topology and has redundancy. ; Or / and, the sensing network (101) includes: an optical fiber temperature sensor and a pressure sensor for acquiring the thermodynamic parameters of the target medium, and a wavefront sensor for detecting the wavefront distortion data of the beam; the wavefront sensor directly acquires the wavefront distortion data of the beam by capturing the wavefront phase of the laser beam after passing through the steam turbulent medium, and the modeling unit (103) compares the measured wavefront distortion data with the theoretical distortion value calculated by the multiphysics coupled optical transmission model to achieve online calibration of the model.

8. A monitoring method based on the monitoring system according to any one of claims 1 to 7, characterized in that, The monitoring method includes the following steps: (S100) Real-time parameter acquisition Several physical parameters of the target medium are acquired in real time, including temperature and pressure, to provide parameters for the multiphysics coupled optical transmission model. At the same time, the original visual data of the target area is acquired using the image acquisition unit. The multiphysics coupled optical transmission model integrates the fluid dynamics, thermodynamics and optical transmission characteristics of the dynamic medium, and is constructed by coupling the Navier-Stokes momentum equation with the Fourier heat conduction equation and the mass conservation equation. (S200) Adaptive Filtering The original visual data carrying the leakage signal is processed using an adaptive time-domain filtering algorithm that includes prediction compensation. This step uses wavelet decomposition to perform multi-resolution time-frequency analysis on the signal to effectively separate noise and signal features at different scales. It can also predict the system dynamics in the short term based on historical data and the current state, and calculate the compensation amount to actively offset the inherent phase lag in the filtering process. (S300) Verification and Calibration The signal-to-noise ratio improvement value is used as a key performance indicator. The key performance indicator is monitored in real time. When the key performance indicator is lower than a preset threshold, the temperature and pressure compensation algorithm is automatically triggered, and the system self-calibrates through a multi-degree-of-freedom precision calibration platform.

9. The monitoring method according to claim 8, characterized in that, In step (S200), a multi-layer decomposition tree is constructed using the Symlet wavelet basis function to effectively separate noise of different frequencies from transient characteristic signals of the dynamic medium; the Symlet wavelet basis function is the Symlet2 wavelet basis function, as shown in formula (10): （10） In equation (10), This is the Symlet wavelet function family after scaling and translation; This is the scaling factor, used to control the bandwidth and resolution of the wavelet function; This is the translation factor, used to control the position of the wavelet function on the time axis; The Symlet2 mother wavelet function; It is a time variable; Or / and, in step (S200), after the signal is decomposed by wavelet, the effective signal and noise are separated by setting a reasonable threshold. Wavelet coefficients greater than or equal to the threshold are wavelet coefficients of the effective signal, and wavelet coefficients less than the threshold are wavelet coefficients of the noise. The reasonable threshold is set based on the improved SUREShrink threshold function, as shown in formula (11): （11） In equation (11), For threshold; This is an estimate of the noise variance; The signal length; The estimated signal-to-noise ratio; Or / and, in step (S200), the processing window length of the filter is also dynamically adjusted based on the gradient change of the signal-to-noise ratio and the historical hysteresis error. To adapt to the non-stationary characteristics of dynamic media; the dynamic window length adjustment function is shown in formula (12): （12） In equation (12), Based on the length of the window; For gradient terms, The signal-to-noise ratio gradient; For integration, This is a historical lag error.

10. The monitoring method according to claim 8, characterized in that, In step (S300), a residual feedback loop is established to calibrate the model predictions in real time, and the calibrated extinction coefficient is... The ratio of model predictions to errors, along with integral feedback, determines the outcome. （15） In equation (15), These are the model's predicted values. This represents the residual between the predicted and actual values.

Images