Method for realizing real-time monitoring of safety parameters by using deep learning

Abstract

The invention provides a method for realizing real-time monitoring of safety parameters by using deep learning, and the method comprises the steps: carrying out the comparative analysis of the radial and axial gradients of temperature, the gradient direction and the uniformity of temperature distribution, if the radial or axial gradient of the temperature exceeds a safety threshold range, judging that a temperature field has local distortion, and if the radial or axial gradient of the temperature exceeds the safety threshold range, judging that the temperature field has local distortion; triggering an early warning mechanism, and generating a temperature abnormity early warning signal; after the early warning mechanism is triggered, a thermal structure coupling analysis method is adopted, based on material attribute parameters and temperature boundary conditions of the reaction kettle wall, the temperature boundary conditions comprise temperature distribution in the kettle, external environment temperature and the like, the temperature stress distribution state of the kettle wall is calculated, and a stress distribution cloud picture is obtained; if the potential safety hazard level of the equipment is high, an emergency shutdown protection mechanism of the high-temperature cracking device is triggered, meanwhile, equipment safety early warning information is sent to a control center, and the whole dynamic early warning process is completed.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and particularly to a method for realizing real-time monitoring of safety parameters by using deep learning. Background Art

[0002] During the operation of the carbon fiber in the reactor, the distribution gradient characteristics of the temperature field in the radial and axial directions are the key factors affecting the stability of the equipment. The radial temperature gradient usually shows a trend of gradually decreasing from the center of the reactor to the outer wall, while the axial temperature gradient may show complex changes due to factors such as the flow of reactants and uneven distribution of heat sources. When local distortion occurs in the temperature field, for example, when the temperature of a certain area rises abnormally due to local overheating of the reactants or failure of the cooling system, this distortion will quickly affect the temperature distribution of the reactor wall. Therefore, during the high-temperature cracking process, the temperature field distribution in the reactor is extremely complex, with significant temperature gradients in both the radial and axial directions. When the reaction continues, due to the non-uniformity of the materials and the randomness of the flow, the temperature field is prone to local distortion, resulting in a sharp increase in the local temperature of the reactor wall. The wall material will undergo creep and stress relaxation at high temperatures, causing the temperature stress distribution of the wall to continuously evolve. Once the local temperature exceeds the allowable limit of the material, the integrity of the wall will be damaged, triggering a safety accident. Therefore, it is urgent to establish a dynamic temperature field warning mechanism to monitor the temperature distribution characteristics in the reactor in real time, timely capture the early signs of temperature field distortion, analyze the evolution law of the temperature stress of the wall, predict the impact on the safety boundary of the equipment, and thus take necessary control measures to avoid accidents. Summary of the Invention

[0003] The present invention provides a method for realizing real-time monitoring of safety parameters by using deep learning, mainly including:

[0004] Obtain the temperature field distribution data of the reactor during the high-temperature cracking process of carbon fiber, and use the finite element analysis method to establish a three-dimensional model of the temperature field distribution of the reactor in the radial and axial directions, and obtain the gradient magnitude, gradient direction, and temperature distribution uniformity of the temperature in the radial and axial directions in the temperature field;

[0005] Compare and analyze the gradient magnitude, gradient direction, and temperature distribution uniformity of the temperature in the radial and axial directions. If the gradient magnitude of the temperature in the radial or axial direction exceeds the safety threshold range, it is determined that local distortion has occurred in the temperature field, triggering the warning mechanism and generating a temperature anomaly warning signal;

[0006] After triggering the warning mechanism, use the thermal-structural coupling analysis method to calculate the temperature stress distribution state of the reactor wall based on the material property parameters and temperature boundary conditions of the reactor wall. The temperature boundary conditions include the temperature distribution inside the reactor, the external environment temperature, etc., and obtain the stress distribution contour map;

[0007] Extract the temperature gradients in the radial and axial directions, the gradient directions, and the temperature distribution uniformity in the stress concentration region from the stress distribution contour map. Process it through the support vector machine algorithm to obtain the location and stress magnitude of the stress concentration region. Based on the obtained location and stress magnitude of the stress concentration region, determine whether the stress concentration region breaks through the equipment safety boundary;

[0008] If the stress magnitude in the stress concentration region exceeds the yield strength of the equipment material, it is determined that the stress concentration region breaks through the equipment safety boundary. Use the genetic algorithm to process the stress distribution state of the kettle wall to obtain the propagation path of the stress distribution evolution, and obtain the position coordinates and stress peak value of the stress breakthrough point from the propagation path;

[0009] Based on the position coordinates and stress peak value of the stress breakthrough point, analyze the impact of the stress distribution evolution on the equipment safety boundary. The impacts include local plastic deformation, fracture, and fatigue failure caused by stress concentration. Determine the equipment safety hazard level according to the degree to which the stress peak exceeds the yield strength. The equipment safety hazard levels include low level and high level;

[0010] If the equipment safety hazard level is high, trigger the emergency shutdown protection mechanism of the high-temperature cracking device, and at the same time send an equipment safety warning message to the control center to complete the entire process of dynamic warning.

[0011] The technical solution provided by the embodiment of the present invention may include the following beneficial effects:

[0012] The present invention discloses a method for realizing real-time monitoring of safety parameters by using deep learning. The method first uses finite element analysis to establish a three-dimensional model of the temperature field of the reaction kettle and obtains temperature gradient information. When the temperature gradient exceeds the safety threshold, trigger the warning mechanism and perform thermo-structural coupling analysis to generate a stress distribution contour map. Process the stress concentration region data through the support vector machine algorithm to determine whether it breaks through the equipment safety boundary. If the stress exceeds the material yield strength, use the genetic algorithm to analyze the stress propagation path to determine the position and peak value of the stress breakthrough point. Evaluate the equipment safety hazard level according to the analysis results, and trigger the emergency shutdown protection mechanism and send a warning message if necessary. The present invention realizes the real-time monitoring and dynamic warning of the temperature field and stress state of the reaction kettle during the high-temperature cracking process of carbon fiber, effectively improving the equipment operation safety and production efficiency. Description of the Drawings

[0013] Figure 1 It is a flowchart of a method for realizing real-time monitoring of safety parameters by using deep learning according to the present invention.

[0014] Figure 2 It is a schematic diagram of a method for realizing real-time monitoring of safety parameters by using deep learning according to the present invention.

[0015] Figure 3Another schematic diagram of a method for real-time monitoring of safety parameters using deep learning according to the present invention. Detailed implementation manners

[0016] Next, the technical solutions in the embodiments of the present invention will be clearly and detailedly described with reference to the accompanying drawings in the embodiments of the present invention. The described embodiments are only a part of the embodiments of the present invention.

[0017] As Figures 1-3 , a method for real-time monitoring of safety parameters using deep learning in this embodiment may specifically include:

[0018] S101. Obtain the temperature field distribution data of the reaction kettle during the high-temperature pyrolysis process of carbon fiber, and use the finite element analysis method to establish a three-dimensional model of the temperature field distribution of the reaction kettle in the radial and axial directions, so as to obtain the gradient magnitude, gradient direction and temperature distribution uniformity of the temperature in the radial and axial directions in the temperature field.

[0019] Use a thermocouple temperature sensor array to collect real-time data on the surface of the reaction kettle to obtain a reaction kettle temperature monitoring data set; establish a reaction kettle thermodynamic parameter data set according to the reaction kettle temperature monitoring data set and the thermal conductivity of the reaction kettle wall material measured by a thermal conductivity measuring instrument; for the temperature monitoring data set and the thermodynamic parameter data set, use the Laplace heat conduction equation to calculate the internal temperature field distribution of the reaction kettle, and perform Kriging interpolation on the temperature monitoring point data to obtain a three-dimensional temperature field grid data set; for the three-dimensional temperature field grid data set, use the central difference method to calculate the temperature gradient magnitude and direction at the grid points to obtain a temperature gradient vector field data set; use the temperature gradient vector field data as input, and use the least squares method to perform surface fitting on the temperature field to establish a mathematical expression of the temperature field distribution function.

[0020] Specifically, by installing a multi-point temperature sensing array on the surface of the reactor, a total of thirty-two temperature monitoring points are set in the radial and axial directions. Thermocouple temperature sensors are used to collect real-time data at multiple positions of the reactor to obtain a temperature monitoring data set. For the material properties of the reactor, a thermal conductivity measuring instrument is used to obtain the thermal conductivity of the reactor wall material, and the wall thickness value is obtained from the reactor design drawings to establish a data set of the thermodynamic parameters of the reactor including the wall thickness and thermal conductivity. According to the temperature monitoring data set and the reactor thermodynamic parameter data set, the Laplace heat conduction equation q=-k▽T1 is used, where q is the heat flux density vector, k is the thermal conductivity, and T1 is the temperature, to calculate the internal temperature field distribution of the reactor. Kriging interpolation is performed on the data of the temperature monitoring points to establish a three-dimensional temperature field grid data set of the reactor. For the three-dimensional temperature field grid data set, the central difference method is used to calculate the magnitude and direction of the temperature gradient at the grid points to obtain a temperature gradient vector field data set. According to the temperature gradient vector field data set, the least squares method is used to perform surface fitting on the radial and axial temperature distributions to establish a reactor temperature field distribution function equation T(r,z), which can be:

[0021]

[0022] , this formula is the polynomial fitting equation for the temperature field distribution, where T is the temperature field function, r is the radial coordinate, z is the axial coordinate, a are the fitting coefficients, m and n are the highest degrees of the polynomial, and ε is the fitting error term. Based on the temperature field distribution function equation T(r, z), the variance values of the radial and axial temperature fields are calculated to obtain the evaluation index for the uniformity of the reactor temperature field. Among them, based on the temperature field distribution function equation T(r, z), the variance values of the temperature fields of the radial and axial values are calculated. If the variance values of the temperature fields of the radial and axial values are less than the preset threshold, it is determined that the temperature field uniformity meets the requirements. If the variance values of the temperature fields of the radial and axial values are greater than the preset threshold, the gradient descent method is used to optimize the surface fitting parameters. With the optimized surface fitting parameters, the variance values of the temperature fields of the radial and axial values are recalculated. According to the variance values of the temperature fields of the radial and axial values, the evaluation index for the uniformity of the reactor temperature field is determined. During the high-temperature pyrolysis process of carbon fiber, the accurate measurement of the temperature field distribution is crucial for the pyrolysis process. The measurement of the reactor temperature field distribution usually adopts a multi-point temperature sensing array layout scheme. In practical applications, the temperature sensors are arranged radially and axially. There are 8 measuring points arranged radially, with a distance of 60 mm between the measuring points, and 4 measuring points arranged axially, with a distance of 200 mm between the measuring points, thus forming an array of 32 regularly distributed temperature monitoring points. Continuous acquisition is carried out by thermocouple temperature sensors within the temperature range of 800 to 1200 degrees Celsius, with a sampling time interval of 10 seconds and a collection duration of 2 hours to obtain a temperature monitoring data set. The reactor material is selected as special high-temperature resistant stainless steel, and its thermal conductivity changes with temperature. The thermal conductivity is 16.3 watts per meter Kelvin at room temperature of 25 degrees Celsius, and it drops to 12.8 watts per meter Kelvin when the temperature rises to 1000 degrees Celsius. The reactor wall thickness is 25 mm and the inner diameter is 800 mm. The curve of the thermal conductivity changing with temperature is obtained through a thermal conductivity measuring instrument to establish a thermodynamic parameter data set. Based on the obtained temperature monitoring data and thermodynamic parameters, the temperature field distribution is calculated using the Laplace heat conduction equation. Under steady-state conditions, the temperature field satisfies the Laplace equation, and the heat flow propagates along the negative direction of the temperature gradient. The Kriging interpolation method is used to interpolate the discrete measuring point temperature data, and the interpolation grid size is selected as 20 mm to generate high-precision three-dimensional temperature field grid data. Based on the established temperature field grid data, the central difference method is used to calculate the temperature gradient. Taking the grid center point as the reference, the ratio of the temperature difference between adjacent grid points to the distance is calculated to obtain the radial and axial temperature gradient components. Among them, the maximum value of the radial temperature gradient appears near the reactor wall surface, with a value reaching 4.5 Kelvin per centimeter, and the maximum value of the axial temperature gradient appears at the edge of the heating zone, with a value of 2.8 Kelvin per centimeter. The least squares method is used to fit the temperature field distribution to establish the temperature field distribution function T(r, z). The radial temperature distribution is parabolic, with the highest central temperature reaching 1150 degrees Celsius and gradually decreasing to 980 degrees Celsius towards the wall surface.The axial temperature distribution is relatively stable in the middle of the heating zone, and the temperature fluctuation range is controlled within plus or minus 15 degrees Celsius. However, the temperature drops sharply at both ends of the heating zone. The uniformity of the temperature field is evaluated by calculating the variance value of the temperature field. The variance value of the radial temperature field is 156 square degrees Celsius, and the variance value of the axial temperature field is 284 square degrees Celsius, indicating that the uniformity of the axial temperature distribution is lower than that of the radial temperature distribution.

[0023] S102. Compare and analyze the gradient magnitude, gradient direction of the temperature in the radial and axial directions, and the uniformity of the temperature distribution. If the gradient magnitude of the temperature in the radial or axial direction exceeds the safety threshold range, it is determined that a local distortion occurs in the temperature field, the warning mechanism is triggered, and a temperature anomaly warning signal is generated.

[0024] The polynomial fitting method is used to calculate the gradient magnitude of the temperature field in the radial and axial directions, and the radial gradient dataset and axial gradient dataset of the temperature field are obtained; according to the radial gradient dataset and axial gradient dataset of the temperature field, the gradient direction of the temperature field is calculated based on the gradient scalar value and gradient vector of the temperature field, and the radial direction dataset and axial direction dataset of the temperature field are obtained; for the radial direction dataset and axial direction dataset of the temperature field, judgments are made according to the preset safety threshold interval of the temperature gradient, and the position points where the temperature gradient exceeds the limit are marked; the random forest algorithm is used to fuse the radial direction dataset, axial direction dataset and temperature gradient over-limit position points of the temperature field to obtain the position and range of the temperature anomaly area, and the distortion amplitude of the temperature field in the temperature anomaly area is calculated by Gaussian kernel density estimation.

[0025] Specifically, according to the three-dimensional distribution data of the temperature field, the polynomial fitting method is used to calculate the magnitude of the temperature gradient in the radial and axial directions of the temperature field, obtaining the radial gradient dataset and axial gradient dataset of the temperature field, and extracting the maximum and minimum values of the temperature gradient magnitude from the temperature field gradient datasets. For the radial gradient dataset and axial gradient dataset of the temperature field, based on the temperature field gradient scalar value and gradient vector, calculate the temperature gradient directions in the radial and axial directions of the temperature field, obtaining the radial direction dataset and axial direction dataset of the temperature field. According to the preset radial temperature gradient safety threshold interval and axial temperature gradient safety threshold interval, perform threshold judgment on the radial gradient dataset and axial gradient dataset of the temperature field, and mark the position points where the temperature gradient exceeds the limit. The random forest algorithm is used to fuse the radial direction dataset, axial direction dataset of the temperature field and the position points where the temperature gradient exceeds the limit to identify the location and range of the temperature anomaly area. For the temperature anomaly area, the Gaussian kernel density estimation is used to calculate the distortion amplitude of the temperature field, where the kernel function bandwidth parameter is set to the standard deviation of the temperature gradient, obtaining the temperature field distortion degree dataset. According to the temperature field distortion degree dataset, generate temperature anomaly warning signal data including the location, anomaly range and distortion amplitude of the temperature anomaly area. The temperature field monitoring during the high-temperature pyrolysis process of carbon fiber involves the analysis of temperature gradients in two key directions, radial and axial. By performing polynomial fitting on the three-dimensional distribution data of the temperature field, the magnitudes of the radial and axial temperature gradients are calculated respectively. The radial temperature gradient reflects the heat transfer characteristics on the cross-section of the reactor, while the axial temperature gradient characterizes the heat transfer along the length direction of the reactor. Based on the actual measurement data, the numerical range of the radial temperature gradient is between 2 and 8 Kelvin per centimeter, and the numerical range of the axial temperature gradient is between 1 and 5 Kelvin per centimeter. The calculation of the temperature field gradient direction is based on the gradient vector. The radial gradient direction is represented as the angle with the central axis of the reactor, and the axial gradient direction is represented as the angle with the axial direction of the reactor. In actual production, under normal operating conditions, the radial gradient direction is perpendicular to the central axis, and the angle fluctuates within the range of 85 to 95 degrees, and the axial gradient direction is parallel to the axial direction, and the angle fluctuates within the range of -5 to 5 degrees. The setting of the temperature gradient safety threshold interval is based on the requirements of the carbon fiber pyrolysis process. The radial temperature gradient safety threshold interval is set between 0 and 6 Kelvin per centimeter, and the axial temperature gradient safety threshold interval is set between 0 and 4 Kelvin per centimeter. When it is detected that the temperature gradient exceeds the safety threshold interval, mark this position point as a potential temperature anomaly point, and then use the random forest algorithm to classify these anomaly points and identify the temperature anomaly area in combination with the gradient direction data. For the identified temperature anomaly area, the Gaussian kernel density estimation method is used to evaluate the distortion degree of the temperature field. The kernel function bandwidth parameter is selected as the standard deviation of the temperature gradient, and this value is about 1.2 Kelvin per centimeter in practical applications.The amplitude of temperature field distortion obtained through kernel density estimation reflects the severity of temperature anomalies. When the distortion amplitude exceeds 4 Kelvin per centimeter, it indicates that the temperature distribution in this area has deviated significantly from the normal state. The temperature anomaly warning signal data contains multiple key information: the spatial coordinates of the abnormal area, the size of the abnormal range, and the amplitude of temperature distortion. During actual operation, when a temperature anomaly is detected at a radial position of 350 mm and an axial position of 1200 mm, the abnormal range is elliptical, with a major axis of 100 mm and a minor axis of 60 mm. The maximum temperature distortion amplitude within the area reaches 5.2 Kelvin per centimeter. This data directly reflects the inhomogeneity of the temperature field distribution and provides an important basis for adjusting process parameters.

[0026] S103. After triggering the warning mechanism, adopt the thermo-structural coupling analysis method. Based on the material property parameters and temperature boundary conditions of the reactor wall, where the temperature boundary conditions include the temperature distribution inside the reactor, the external environmental temperature, etc., calculate the temperature stress distribution state of the reactor wall and obtain the stress distribution nephogram.

[0027] Call the material database according to the temperature anomaly warning signal data of the reactor to obtain a dataset of material properties including the elastic modulus, Poisson's ratio, and thermal expansion coefficient of the reactor wall; use the temperature field distribution data and the data of the external environmental temperature sensor for the dataset of material properties to establish a dataset of temperature boundary conditions including the temperature distribution inside the reactor and the environmental temperature; use the dataset of temperature boundary conditions to discretize the reactor wall through the hexahedron mesh generation method to obtain the reactor wall mesh dataset; according to the dataset of material properties and the mesh dataset, use the finite difference method to solve the thermal stress control equation to obtain the reactor wall stress distribution dataset.

[0028] Specifically, according to the data of the abnormal warning signal of the reactor temperature, the material database is called to obtain the values of the elastic modulus, Poisson's ratio, and coefficient of thermal expansion corresponding to the reactor wall at different temperatures, and a dataset of the material properties of the reactor is established. Using the temperature field distribution data and the data of the external environment temperature sensor, a dataset of temperature boundary conditions including the temperature distribution inside the reactor and the ambient temperature is established. For the dataset of the material properties of the reactor and the dataset of temperature boundary conditions, the support vector regression algorithm is used to fit the correlation between the thermophysical parameters of the material and the temperature, and a dataset of material property curves related to the reactor wall temperature is obtained. Based on the hexahedron mesh generation method, the reactor wall is discretized, and the mesh size is set according to one-tenth of the reactor wall thickness, and a dataset of the reactor wall mesh is obtained. According to the dataset of the material property curves and the mesh dataset, the finite difference method is used to solve the thermal stress control equation σij = 2μεij + λεkkδij, where σij is the stress tensor, εij is the strain tensor, μ and λ are the Lame constants, which are the inherent elastic modulus of the material and are used to describe the elastic properties of the material, and εkk is the sum of the diagonal elements of the strain tensor, and a dataset of the stress distribution of the reactor wall is obtained. For the dataset of the stress distribution of the reactor wall, the maximum principal stress σ1, the intermediate principal stress σ2, and the minimum principal stress σ3 are calculated, and the von Mises criterion is used to calculate the equivalent stress σe. Based on the principal stress data and the equivalent stress data, the Kriging interpolation method is used to reconstruct the stress field of the reactor wall, and stress distribution contour data including the magnitude of the principal stress, the stress direction, and the value of the equivalent stress are generated. After the abnormal warning of the carbon fiber reactor temperature is triggered, it is necessary to deeply analyze the stress state of the reactor wall, and the material property parameters play a key role in this process. The reactor is made of special stainless steel, and its elastic modulus changes significantly with temperature, which is 210 GPa at room temperature of 25 °C and drops to 160 GPa when the temperature rises to 800 °C, while Poisson's ratio is relatively stable and varies in the range of 0.28 to 0.32, and the coefficient of thermal expansion increases with the increase of temperature, increasing from 12.5 μm / (m·K) at room temperature to 15.8 μm / (m·K) at 800 °C. The accurate acquisition of temperature boundary conditions is crucial for stress analysis. The temperature distribution inside the reactor is obtained through temperature field monitoring. When the temperature of the inner surface of the reactor wall reaches 950 °C, the temperature of the outer surface is about 120 °C, and the ambient temperature remains at about 25 °C. The support vector regression algorithm fits the thermophysical parameters of the material and establishes the corresponding relationship between temperature and material properties, and the fitting accuracy reaches more than 98%. The reactor wall is meshed with hexahedrons, the wall thickness is 250 mm, the mesh size is set to 25 mm, and 10 layers of mesh elements are formed radially. The number of meshes in the axial and circumferential directions is determined according to the actual size. Usually, 80 layers are set axially and 36 layers are set circumferentially, and a total of 28,800 mesh elements are generated. The mesh quality control indicators include that the mesh distortion degree is less than 0.3 and the mesh orthogonality is greater than 0.85.By solving the thermal stress control equation, the stress distribution data is calculated. The maximum principal stress usually appears on the inner surface of the kettle wall, with a value reaching 320 MPa. The intermediate principal stress is about 260 MPa, and the minimum principal stress is around 140 MPa. The maximum value of the equivalent stress calculated using the von Mises criterion appears in the region with the largest temperature gradient, reaching 280 MPa. These stress values are compared with the yield strength of the material. When the equivalent stress exceeds 80% of the yield strength of the material at this temperature, it indicates the existence of potential structural safety hazards. The stress distribution nephogram uses a multi-color annotation method, with different stress levels represented by different colors. Usually, the stress range is divided into 10 intervals, with the maximum stress area represented by red and the minimum stress area represented by blue. The stress vector direction is also marked on the nephogram. The principal stress direction is indicated by an arrow, and the arrow length is proportional to the stress magnitude. In the local stress concentration area, the nephogram resolution can be increased to 5 MPa per color level to more finely display the details of the stress distribution. The Kriging interpolation algorithm ensures the smooth transition of the stress nephogram and eliminates the sudden change phenomenon caused by grid meshing.

[0029] S104. Extract the temperature gradient magnitude, gradient direction, and temperature distribution uniformity in the radial and axial directions of the stress concentration area from the stress distribution nephogram, and process them through the support vector machine algorithm to obtain the position and stress magnitude of the stress concentration area. Based on the obtained position and stress magnitude of the stress concentration area, determine whether the stress concentration area breaks through the equipment safety boundary.

[0030] Perform regional segmentation on the stress values according to the stress distribution nephogram data using the region growing algorithm, and obtain the spatial contour data of the stress concentration area through the region growing algorithm based on the multiple threshold of the arithmetic mean of the local area stress; calculate the temperature field gradient numerical value and direction for the spatial contour data of the stress concentration area. The temperature field gradient includes radial and axial components to obtain the temperature gradient vector data set; establish a stress feature vector according to the temperature gradient vector data set using the support vector machine algorithm. The stress feature vector includes dimensions such as stress numerical value, temperature gradient magnitude, gradient direction angle, and temperature uniformity index to obtain the stress concentration area feature vector data set; perform regional clustering on the stress concentration area feature vector data set using the Gaussian mixture density algorithm. The Gaussian mixture density algorithm classifies the feature vectors based on the preset number of Gaussian components to obtain the center coordinate point and the maximum stress numerical value of the stress concentration area. Determine whether the stress numerical value of the stress concentration area exceeds the equipment stress safety limit range according to the stress safety threshold.

[0031] Specifically, according to the stress distribution contour dataset, the region growth algorithm is used to perform regional segmentation on the stress values. The growth threshold is set to 1.5 times the arithmetic mean of the local regional stress, and the spatial contour data of the stress concentration region is obtained. For the spatial contour data of the stress concentration region, the gradient values and directions of the temperature field in the radial and axial directions are calculated to obtain the temperature gradient vector dataset. According to the temperature gradient vector dataset, the temperature field variance value is calculated as the temperature uniformity index to obtain the temperature distribution characteristic dataset of the stress concentration region. For the temperature distribution characteristic dataset of the stress concentration region, the support vector machine algorithm is used to establish the stress feature vector. The feature dimensions include stress value, temperature gradient magnitude, gradient direction angle, and temperature uniformity index, and the stress concentration region feature vector dataset is obtained. Based on the stress concentration region feature vector dataset, the Gaussian mixture density algorithm is used for regional clustering, and the number of Gaussian components is set to 3 to obtain the spatial distribution dataset of the stress concentration region. According to the spatial distribution dataset of the stress concentration region, the center coordinate point and the maximum stress value of the stress concentration region are extracted to generate the key feature dataset of the stress concentration region. For the key feature dataset of the stress concentration region, the stress value is evaluated based on the stress safety threshold criterion of the reactor equipment. The stress safety threshold is set to 80% of the material yield strength, and it is judged whether the stress value in the stress concentration region exceeds the equipment stress safety limit range. The analysis of the stress distribution contour starts from identifying the stress concentration region. The region growth algorithm determines the boundary of the high-stress region by setting the growth threshold. In practical applications, when the average stress value in a certain region of the reactor wall is 180 MPa, the growth threshold is set to 270 MPa. Any region with a stress value exceeding this threshold is marked as a potential stress concentration region. The stress concentration regions identified in this way usually have an irregular shape. Within the identified stress concentration region, the calculation of the temperature field gradient reflects the characteristics of heat transfer. The typical value of the radial temperature gradient is 4.2 Kelvin per centimeter, and the axial temperature gradient is 2.8 Kelvin per centimeter. The gradient direction is represented by an angle. The angle between the radial gradient and the central axis is between 87 degrees and 93 degrees, and the angle between the axial gradient and the axis is between -3 degrees and 3 degrees. The temperature field uniformity is characterized by the variance value. Under normal operating conditions, the variance value is within 200 square Kelvin. When the variance value exceeds 300 square Kelvin, it indicates that the temperature distribution is significantly non-uniform. The construction of the stress feature vector integrates multiple physical quantities, including stress value, temperature gradient, and uniformity index. The support vector machine algorithm classifies these features, and each dimension of the feature vector is normalized. The stress value is divided by the material yield strength, the temperature gradient is divided by the designed maximum gradient value, the direction angle is divided by 180 degrees, and the uniformity index is divided by the reference variance value.The Gaussian mixture density algorithm performs a fine classification on the stress concentration region through three Gaussian components. For the first type of region, the stress value is between 240 and 280 MPa, and the temperature gradient is relatively small; for the second type of region, the stress value is between 280 and 320 MPa, and the temperature gradient is medium; for the third type of region, the stress value exceeds 320 MPa, and the temperature gradient is large. The center point coordinates and the maximum stress value of each type of region are recorded to form a spatial distribution map of the stress concentration region. The stress safety threshold of the reactor equipment is determined based on the high-temperature performance of the material. At a working temperature of 800 degrees Celsius, the yield strength of the special stainless steel is 420 MPa, and the safety threshold is set at 336 MPa. When the maximum stress value of the stress concentration region approaches or exceeds this threshold, it indicates that there is a safety hazard in this region. During specific judgment, the spatial range of the stress concentration region also needs to be considered. When the area of the stress-overlimit region exceeds 5% of the total area of the kettle wall, even if the maximum stress value does not reach the safety threshold, it is also determined to exceed the safety boundary. The stress concentration region usually appears in positions with a large temperature gradient, such as the transition zone between the heating area and the non-heating area, or where the kettle body structure is discontinuous.

[0032] Identify the range of the stress concentration region from the stress distribution contour map, extract the radial and axial temperature gradient magnitude data, temperature gradient direction data, and temperature distribution uniformity data of the stress concentration region, and perform vectorization processing. Construct a support vector machine classification model to classify the stress concentration region. After obtaining the classification boundary, determine the stress concentration position and obtain the stress value at the stress concentration position to obtain the stress magnitude of the stress concentration region.

[0033] Determine the growth threshold according to the local mean of the stress. Use the Sobel operator to calculate the stress value gradient. When the stress gradient value exceeds the growth threshold, it is marked as the boundary point of the stress concentration region, and the spatial range data of the stress concentration region is obtained; calculate the radial and axial temperature gradient values and the gradient direction angle for the spatial range data of the stress concentration region, and calculate the temperature uniformity index of the stress concentration region using the mean square deviation of the temperature field; use the min-max normalization method to standardize the temperature gradient value, the gradient direction angle, and the temperature uniformity index, and construct a feature vector matrix containing three components: temperature gradient value, gradient direction angle, and temperature uniformity; use a support vector machine classifier to classify the feature vector matrix, extract the discrete point coordinates of the boundary curve, and obtain the position coordinates and stress value data of the stress concentration region.

[0034] Specifically, the region growing segmentation method is used to process the stress distribution nephogram. The growth threshold is set to 1.5 times the local mean of the stress. The stress value gradient is calculated based on the Sobel operator. When the stress gradient value exceeds the growth threshold, it is marked as the boundary point of the stress concentration region, and the spatial range data of the stress concentration region is obtained. According to the temperature monitoring point data within the spatial range of the stress concentration region, the central difference method is used to calculate the radial and axial temperature gradient values. Based on the gradient vector, the temperature gradient direction angle is calculated to obtain the temperature gradient data of the stress concentration region. For the temperature gradient data of the stress concentration region, the average variance of the temperature field in this region is calculated as the temperature uniformity index to obtain the temperature distribution characteristic data. Based on the temperature distribution characteristic data, the minimum-maximum normalization method is used to standardize the temperature gradient value, the gradient direction angle, and the temperature uniformity index. The normalization function is x'=(x - xmin) / (xmax - xmin), and the standardized characteristic data is generated. For the standardized characteristic data, a feature vector matrix is constructed. Each row of the matrix represents the feature vector of a spatial position point, including three components: the temperature gradient value, the gradient direction angle, and the temperature uniformity. The support vector machine classifier is used to classify the feature vector matrix. The radial basis function is selected as the kernel function, and the penalty parameter is set to 1 to obtain the classification boundary data of the stress concentration region. According to the classification boundary data of the stress concentration region, the discrete point coordinates of the boundary curve are extracted, and the stress values within the boundary curve are extracted to obtain the position coordinates and stress value data of the stress concentration region. The processing of the stress distribution nephogram first uses the region growing segmentation method to identify the stress concentration region, and a reasonable growth threshold is set to ensure accurate capture of the high-stress region. When the local mean of the stress in a certain region of the reactor wall is 200 MPa, the growth threshold is set to 300 MPa. The points where the stress gradient value calculated by the Sobel operator exceeds this threshold are marked as boundary points. This method shows good boundary recognition ability in the region where the stress changes sharply, especially at the edge of the stress concentration region. The calculation of the temperature gradient uses the central difference method, and the difference operation is performed by taking adjacent measurement points in the radial and axial directions respectively. In practical applications, the radial measurement point spacing is 60 mm, and the axial measurement point spacing is 200 mm. The typical value of the calculated radial temperature gradient is 4.5 Kelvin per centimeter, and the axial temperature gradient is 2.8 Kelvin per centimeter. The temperature gradient direction is obtained by calculating the gradient vector. The angle between the radial temperature gradient and the central axis is between 88 degrees and 92 degrees, and the angle between the axial temperature gradient and the axis is between -2 degrees and 2 degrees. The temperature field uniformity index reflects the concentration degree of the temperature distribution and is obtained by calculating the average variance of the temperature field. When the temperature distribution is uniform, the average variance value is within 150 square Kelvin. When the temperature distribution is non-uniform, the average variance value can reach 400 square Kelvin. The size of the average variance value directly reflects the disturbance degree of the temperature field and is an important index for evaluating the stability of the temperature field. The standardization processing of the characteristic data uses the minimum-maximum normalization method to convert the characteristics with different dimensions to the same scale space.The typical range of the temperature gradient value is from 0 to 8 Kelvin per centimeter, which is mapped to the range of 0 to 1 after normalization; the original range of the gradient direction angle is from 0 to 180 degrees, and it is also mapped to the range of 0 to 1 after normalization; the temperature uniformity index is converted from 0 to 500 square Kelvin into a normalized value. The normalized eigenvectors form an eigenmatrix, and each row of the matrix corresponds to a position point in space, including three eigencomponents: the temperature gradient value, the direction angle, and the uniformity of this point. The support vector machine uses a radial basis kernel function for classification. The kernel function parameter γ is taken as 0.1, and the classification penalty parameter C is taken as 1.0. This set of parameter settings shows good classification effects in practical applications. The boundary curve of the classification result is obtained by extracting discrete points on the classification boundary, forming a closed boundary curve on the reactor wall surface. The stress concentration area usually presents an irregular shape, and the area enclosed by the boundary curve ranges from several hundred square millimeters to several thousand square millimeters. The maximum stress value in the area usually appears at the center position of the area, and the value can reach 320 MPa, exceeding 80% of the material design stress.

[0035] S105. If the stress magnitude in the stress concentration area exceeds the yield strength of the equipment material, it is determined that the stress concentration area breaks through the equipment safety boundary. The genetic algorithm is used to process the stress distribution state of the reactor wall to obtain the propagation path of the stress distribution evolution, and the position coordinates and stress peak value of the stress breakthrough point are obtained from the propagation path.

[0036] Regarding the yield strength of the reactor wall material, it is judged whether the maximum stress value in the stress concentration area exceeds the yield strength. If it exceeds the yield strength, the stress breakthrough point coordinates and stress value are obtained from the stress concentration area; according to the stress breakthrough point coordinates and stress value, the genetic algorithm is used to construct the stress propagation path. The real number coding of the stress distribution coordinate points is adopted by the genetic algorithm to obtain the initial stress distribution evolution path; for the initial stress distribution evolution path, a fitness function is set to calculate the stress gradient value between adjacent points, and the maximum stress gradient propagation path data is obtained through iterative optimization; according to the maximum stress gradient propagation path data, key nodes with a stress gradient difference exceeding a preset threshold are selected, and the stress propagation direction between nodes is calculated by the gradient tracking method, and the stress distribution evolution data set is reconstructed through cubic spline interpolation.

[0037] Specifically, according to the yield strength values of the reactor wall materials in the material database, the stress distribution data in the stress concentration area is judged. If the maximum stress value in the stress concentration area exceeds the material yield strength, the stress breakthrough point coordinates and stress values are extracted from the stress concentration area spatial range data. For the stress distribution data of the reactor wall, a genetic algorithm is used to construct a stress propagation path calculation scheme. The gene coding is represented by real numbers of stress distribution coordinate points. The population size is set to 100, the crossover rate is set to 0.8, and the mutation rate is set to 0.1 to obtain the initial stress distribution evolution path. According to the initial stress distribution evolution path, the fitness function f = ∑(σi+1 - σi) / l i is set, where σi represents the stress value at the i-th point on the path, and l i represents the distance between adjacent points. The propagation path data with the maximum stress gradient is obtained through iterative optimization. For the stress propagation path data, key nodes are selected on the path according to the magnitude of the stress gradient value. When the stress gradient difference between nodes is greater than the preset threshold, it is marked as a stress propagation node to obtain the stress propagation key node data. According to the stress propagation key node data, the gradient tracking method is used to calculate the stress propagation direction between nodes, and the direction vector is determined by the stress gradient vector at the nodes to obtain the stress propagation direction data. For the stress propagation direction data, the cubic spline interpolation method is used to continuously reconstruct the stress distribution between nodes, and the natural boundary condition is used as the boundary condition to obtain the complete stress distribution evolution data set. According to the stress distribution evolution data set, the stress breakthrough point position coordinates and stress peak data are extracted to generate the stress breakthrough characteristic data. The stress evolution analysis when the reactor stress exceeds the material yield strength involves multiple key links. First, the yield strength values of the reactor wall materials at different temperatures are obtained through the material database. At a working temperature of 800 degrees Celsius, the yield strength of special stainless steel is 420 MPa. When the stress in the stress concentration area exceeds this value, it indicates that the material has entered the plastic deformation stage. In actual monitoring, it is found that the stress breakthrough point position usually appears in the area with the largest temperature gradient, and the stress value can reach 440 MPa. The application of the genetic algorithm in stress propagation path calculation uses a real number coding method. Each chromosome represents a possible stress propagation path, and the chromosome length is related to the number of spatial discrete points. The population size is set to 100 to ensure sufficient coverage of the search space. The settings of the crossover rate of 0.8 and the mutation rate of 0.1 show good convergence characteristics in practical applications. After 300 generations of iterative optimization, the optimal individual in the population represents the most likely stress propagation path. The design of the fitness function directly affects the recognition accuracy of the stress propagation path. The function value calculation takes into account the stress gradient and distance factors between adjacent points. In practical applications, the distance value between adjacent points ranges from 20 to 50 mm, and the stress gradient is usually between 0.5 and 2.0 MPa per mm. When the stress gradient of a certain section on the path exceeds 1.5 MPa per mm, this section is marked as a key area of stress propagation.The selection of key nodes for stress propagation is based on the change of stress gradient value. The gradient difference threshold is set to 0.3 MPa / mm. When the stress gradient difference between adjacent regions exceeds this threshold, this position is marked as a propagation node. In an actual reactor, usually 15 to 20 key propagation nodes are identified, and these nodes form the skeleton path of stress propagation. The gradient tracking method determines the stress propagation direction by calculating the stress gradient vector, and calculates the magnitude and direction of the stress gradient at each key node. The stress gradient vector at the node points to the direction where the stress value increases fastest. The typical gradient magnitude is between 1.2 and 1.8 MPa / mm, and the included angle between the direction angle and the structure surface is in the range of 75 to 85 degrees. Cubic spline interpolation realizes the continuous reconstruction of the stress propagation path, and natural boundary conditions are adopted to ensure the smooth transition of the curve at the endpoints. The division of the interpolation interval adopts uneven segmentation. The segment length in the region with a large stress gradient is 20 mm, and the segment length in the region with a small gradient can be relaxed to 40 mm. The reconstructed stress distribution curve clearly shows the trend of stress spreading from the breakthrough point to the periphery, and the stress peak shows an approximately exponential decay characteristic with the increase of distance.

[0038] S106. Analyze the influence of the stress distribution evolution on the equipment safety boundary according to the position coordinates and stress peak of the stress breakthrough point. The influences include local plastic deformation, fracture, and fatigue failure caused by stress concentration. Determine the equipment safety hazard level according to the degree to which the stress peak exceeds the yield strength. The equipment safety hazard levels include low level and high level.

[0039] According to the ratio of the yield strength of the reactor material to the stress peak at the stress breakthrough point, establish a stress-strain curve using the plastic strain theory to obtain a plastic deformation data set at the stress breakthrough point; for the plastic deformation data set at the stress breakthrough point, use the support vector regression algorithm to fit the stress-strain curve. The support vector regression algorithm selects the radial basis function as the kernel function to obtain the stress-strain fitting curve data; according to the stress-strain fitting curve data, calculate the critical stress intensity factor K I C at the stress breakthrough point using the fracture mechanics criterion, where σ represents stress, a represents crack length, to obtain fracture parameter characteristic data; for the fracture parameter characteristic data, use the Miner linear cumulative damage criterion to calculate the fatigue damage degree D at the stress breakthrough point, where D = ∑(n i / N i), n i represents the actual number of cycles, N i represents the number of fracture cycles, to obtain fatigue characteristic data; combine the plastic deformation data set at the stress breakthrough point and the fatigue characteristic data to divide the safety hazard level of the stress breakthrough region.

[0040] Specifically, according to the yield strength values of the reactor materials in the material database, calculate the ratio of the stress peak at the stress breakthrough point to the yield strength of the material. Use the plastic strain theory to establish the stress-strain curve σ = Kεn, where K is the strength coefficient and n is the strain hardening index, to obtain the plastic deformation data set at the stress breakthrough point. For the plastic deformation data set at the stress breakthrough point, use the support vector regression algorithm to fit the stress-strain curve, select the radial basis function as the kernel function, and set the penalty parameter to 1.0 to obtain the stress-strain fitting curve data. Based on the stress-strain fitting curve data, calculate the critical stress intensity factor at the stress breakthrough point according to the fracture mechanics criterion Among them, σ is the stress, a is the crack length, and characteristic data of fracture parameters are obtained. For the characteristic data of fracture parameters, the rain-flow counting method is used to statistically analyze the stress cycle amplitude at the stress breakthrough point. The cycle counting unit is set to half a cycle, and characteristic data of stress cycles are obtained. According to the characteristic data of stress cycles, the Miner linear cumulative damage criterion D = ∑(n i / N i) is adopted, where n i is the actual number of cycles and N i is the fracture number of cycles, to calculate the fatigue damage degree at the stress breakthrough point and obtain fatigue characteristic data. For the plastic deformation data set and fatigue characteristic data, a fuzzy membership function μ(x) is constructed, and the membership grading threshold is set to [0.3, 0.7] to obtain membership data of potential hazard levels. According to the membership data of potential hazard levels, the potential hazard levels of the stress breakthrough area are divided. A membership less than 0.3 is judged as a low-level hazard, and a membership greater than 0.7 is judged as a high-level hazard to obtain the determination result of potential hazard levels. The safety assessment of the reactor wall material in a high-temperature environment involves the calculation and analysis of multiple key physical quantities. When the stress peak at the stress breakthrough point reaches 440 MPa, exceeding the material yield strength of 420 MPa, the material enters the plastic deformation stage. The stress-strain relationship curve established using the plastic strain theory shows that the strength coefficient K of the material is 850 MPa and the strain hardening index n is 0.15. This set of parameters reflects the work hardening characteristics of the material in the plastic deformation stage. The radial basis kernel function is used for the fitting of the stress-strain curve by the support vector regression algorithm. When the strain value is in the range of 0.02 to 0.05, the root mean square error between the fitting curve and the experimental data remains within 3%. The fitting result shows that the equivalent plastic strain at the stress breakthrough point reaches 0.038, which has exceeded the uniform deformation limit of the material of 0.035. In fracture mechanics analysis, the calculation of the critical stress intensity factor takes into account the stress intensity at the crack tip. When the crack length is 2 mm, the calculated stress intensity factor is 68 MPa·m square root, close to the fracture toughness of the material of 75 MPa·m square root. This value indicates that the crack is in the subcritical growth stage and has a tendency of slow growth. The rain-flow counting method uses a half-cycle counting unit for the statistical analysis of stress cycle characteristics to identify the stress cycle amplitude distribution experienced by the stress breakthrough point. The main stress cycle amplitudes are concentrated in the range of 280 to 320 MPa, the cycle frequency is about 4 times per hour, and the peak-valley ratio reaches 0.85. This cycle characteristic has a significant impact on the fatigue life of the material. When using the Miner linear cumulative damage criterion to evaluate the fatigue damage degree, the ratio of the actual number of cycles at each stress level to the fracture number of cycles at that stress level is calculated. The cumulative damage degree reaches 0.72, which has exceeded the safety limit of 0.6, indicating that the material has entered the rapid development stage of fatigue damage. The construction of the fuzzy membership function comprehensively considers three indicators: the degree of plastic deformation, the stress intensity factor, and the fatigue damage degree.The membership degree grading threshold is set to [0.3, 0.7]. When the membership degree is less than 0.3, it is determined as a low-level safety hazard, indicating that the structure is still in a controllable state; when the membership degree is greater than 0.7, it is determined as a high-level safety hazard, indicating that the structure has reached the critical state of failure. In practical applications, the membership degree value of the stress breakthrough point reaches 0.82, which belongs to a high-level safety hazard, reflecting that the equipment is already in a dangerous working condition.

[0041] S107. If the equipment safety hazard level is high, trigger the emergency shutdown protection mechanism of the high-temperature cracking device, and at the same time send an equipment safety warning message to the control center to complete the entire process of dynamic warning.

[0042] Use a random forest classifier to evaluate the equipment status, and obtain the determination data of the high-level safety hazard status of the equipment according to the stress peak value, the position of the stress breakthrough point, and the hazard level index; establish a state transition diagram for the determination data of the high-level safety hazard status of the equipment. If it is detected that the warning state transfers to the shutdown state, generate a shutdown protection instruction data set; encode the control instructions according to the shutdown protection instruction data set and set the priorities, where the priority of closing the feed valve is the first priority, the priority of reducing the heating power is the second priority, and the priority of opening the gas vent valve is the third priority, to obtain the control instruction execution sequence data; use a real-time sequence scheduling method for the control instruction execution sequence data to generate an instruction execution time sequence table, set the adjacent instruction execution interval duration, construct a warning information data packet including the time stamp, instruction code, and execution status, and send it to the control center through the message queue mechanism.

[0043] Specifically, according to the judgment result of the safety hazard level, a random forest classifier is used to conduct real-time evaluation of the equipment status. The depth of the tree is set to 8, the number of trees is 100, and the feature dimension includes the stress peak value, the position of the stress breakthrough point, and the hazard level index, generating the determination data of the advanced safety hazard status of the equipment. For the determination data of the advanced safety hazard status of the equipment, a state transition diagram including three states of operation, warning, and shutdown is established. When transferring from the warning state to the shutdown state, a shutdown protection instruction dataset is generated. Based on the shutdown protection instruction dataset, the closing instruction of the feed valve, the reducing instruction of the heating power, and the opening instruction of the gas vent valve are encoded to obtain the control instruction identification data. For the control instruction identification data, the priority of the closing instruction of the feed valve is set to 1, the priority of the reducing instruction of the heating power is set to 2, and the priority of the opening instruction of the gas vent valve is set to 3, generating the control instruction execution sequence data. According to the control instruction execution sequence data, a real-time sequence scheduling method is used to generate an instruction execution timing table, and the adjacent instruction execution interval is set to 100 milliseconds, obtaining the equipment protection execution timing data. For the equipment protection execution timing data, a warning information data structure including three fields of timestamp, instruction code, and execution status is constructed, obtaining a warning information data packet. Based on the warning information data packet, a publish-subscribe channel is established using the message queue mechanism, the message queue length is set to 128, and the warning information data is sent to the control center. In the safety hazard warning and protection mechanism of the high-temperature pyrolysis device, the real-time evaluation of the equipment status is realized by using a random forest classifier. By setting a classifier with 8 layers of depth and 100 decision trees, features such as the stress peak value, the position of the stress breakthrough point, and the hazard level are evaluated. When the stress peak value exceeds 420 MPa, the stress breakthrough point is located at the edge of the heating zone, and the hazard level is in the advanced state, the classifier outputs the determination result of the advanced safety hazard status. The equipment state transition diagram defines the state transition rules during the operation of the device. Under the normal operation state, parameters such as the equipment temperature and pressure are within the set range; when entering the warning state, at least one monitoring parameter exceeds the threshold; when it is determined as an advanced safety hazard, the state machine triggers the transfer to the shutdown state, and at the same time, a shutdown protection instruction set is generated. The shutdown instructions include three types of instructions: closing the feed valve, reducing the heating power, and opening the gas vent valve. The execution order of the control instructions directly affects the safe shutdown process of the equipment. The closing instruction of the feed valve has the highest priority and is used to block the continuous entry of materials into the reaction kettle; followed by the reducing instruction of the heating power, which slows down the reaction process by reducing the heat input; finally, the opening instruction of the gas vent valve is executed to ensure the safe release of the pressure in the reaction kettle. The timing arrangement of the instruction execution uses an interval of 100 milliseconds to ensure that each actuator has sufficient response time.The data structure design of the warning information includes three key fields. The timestamp records the instruction generation time, accurate to the millisecond level. The instruction code is represented by a 4-byte integer. The high byte represents the instruction type, and the low byte represents the specific parameter. The execution status field is used to track the execution progress of the instruction, including three states: to be executed, executing, and executed. The message queue mechanism uses the publish-subscribe mode to transmit warning information, and the queue length is set to 128 to ensure the buffer space for information transmission. The publisher, i.e., the device control unit, packs the warning information and sends it to the message queue. The subscriber is the control center, which receives and processes the warning information in real time. When the warning information data packet enters the queue, the system automatically assigns a priority identifier to it. The warning information related to high-level security hazards obtains the highest priority to ensure that this type of information is processed by the control center first. In practical applications, the end-to-end delay from the sending to the receiving of the warning information is controlled within 50 milliseconds, ensuring the real-time nature of the warning mechanism.

[0044] The above is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A method for real-time monitoring of safety parameters using deep learning, characterized in that: The method comprises: The temperature field distribution data of the reactor during the high-temperature pyrolysis of carbon fiber is obtained, and the finite element analysis method is used to establish a three-dimensional model of the radial and axial temperature field distribution of the reactor, and the radial and axial temperature gradient size, gradient direction and uniformity of temperature distribution in the temperature field are obtained; Compare and analyze the radial and axial temperature gradients, gradient directions, and uniformity of temperature distribution. If the radial or axial temperature gradients exceed the safety threshold, it is determined that the temperature field is locally distorted, triggering the early warning mechanism and generating an abnormal temperature warning signal. After the early warning mechanism is triggered, the thermal-structural coupling analysis method is used to calculate the temperature stress distribution state of the reactor wall based on the material property parameters and temperature boundary conditions of the reactor wall. The temperature boundary conditions include the temperature distribution inside the reactor and the external environment temperature, and obtain the stress distribution cloud map. The radial and axial gradients, gradient directions, and uniformity of temperature distribution of the stress concentration area are extracted from the stress distribution cloud map, and the position and stress magnitude of the stress concentration area are obtained through the support vector machine algorithm. Based on the obtained position and stress magnitude of the stress concentration area, it is determined whether the stress concentration area exceeds the safety boundary of the equipment. If the stress in the stress concentration area exceeds the yield strength of the equipment material, it is determined that the stress concentration area has broken through the equipment safety boundary. The genetic algorithm is used to process the stress distribution state of the kettle wall to obtain the propagation path of the stress distribution evolution, and the position coordinates and stress peak value of the stress breakthrough point are obtained from the propagation path. According to the position coordinates of the stress breakthrough point and the stress peak, analyze the impact of stress distribution evolution on the equipment safety boundary. The impact includes local plastic deformation, fracture, and fatigue failure caused by stress concentration. The equipment safety hazard level is determined according to the degree to which the stress peak exceeds the yield strength. The equipment safety hazard level includes low and high levels. If the equipment safety hazard level is high, the emergency shutdown protection mechanism of the high-temperature cracking unit will be triggered, and the equipment safety warning information will be sent to the control center to complete the entire dynamic warning process.

2. The method according to claim 1, characterized in that The method of obtaining the temperature field distribution data of the reactor during the high temperature cracking of the carbon fiber, using the finite element analysis method, establishing a three-dimensional model of the radial and axial temperature field distribution of the reactor, and obtaining the radial and axial temperature gradient size, gradient direction and uniformity of the temperature distribution in the temperature field, includes: A thermocouple temperature sensor array is used to collect real-time data on the reactor surface to obtain a reactor temperature monitoring data set; Establishing a reactor thermodynamic parameter data set according to the reactor temperature monitoring data set and the thermal conductivity of the reactor wall material measured by a thermal conductivity measuring instrument; For the temperature monitoring data set and the thermodynamic parameter data set, the Laplace heat conduction equation is used to calculate the temperature field distribution inside the reactor, and the temperature monitoring point data is subjected to Kriging interpolation to obtain a three-dimensional temperature field grid data set; For the three-dimensional temperature field grid data set, the central difference method is used on the grid points to calculate the size and direction of the temperature gradient to obtain a temperature gradient vector field data set; Taking the temperature gradient vector field data as input, the least square method is used to perform surface fitting on the temperature field and the mathematical expression of the temperature field distribution function is established.

3. The method according to claim 1, characterized in that The temperature gradient size in radial and axial directions, the gradient direction and the uniformity of temperature distribution are compared and analyzed. If the temperature gradient size in radial or axial directions exceeds the safety threshold range, it is determined that the temperature field is locally distorted, the early warning mechanism is triggered, and a temperature abnormality early warning signal is generated, including: The temperature gradient of the temperature field in radial and axial directions is calculated by using the polynomial fitting method to obtain the radial gradient data set and the axial gradient data set of the temperature field. According to the temperature field radial gradient data set and the axial gradient data set, the temperature field gradient direction is calculated based on the temperature field gradient scalar value and the gradient vector to obtain the temperature field radial direction data set and the axial direction data set; For the radial direction data set and the axial direction data set of the temperature field, a judgment is made according to a preset temperature gradient safety threshold interval, and a temperature gradient exceeding limit position point is marked; The random forest algorithm is used to fuse the radial and axial temperature field datasets with the temperature gradient exceeding limit position points to obtain the position and range of the temperature anomaly area, and the temperature field distortion amplitude of the temperature anomaly area is calculated by Gaussian kernel density estimation.

4. The method according to claim 1, characterized in that: After the early warning mechanism is triggered, a thermal structure coupling analysis method is used to calculate the temperature stress distribution state of the reactor wall based on the material property parameters and temperature boundary conditions of the reactor wall. The temperature boundary conditions include the temperature distribution in the reactor and the external environment temperature, and obtain the stress distribution cloud map, including: The material database is called according to the abnormal temperature warning signal data of the reactor to obtain the material property data set including the elastic modulus, Poisson's ratio and thermal expansion coefficient of the reactor wall; Using the temperature field distribution data and the external environment temperature sensor data for the material property data set, a temperature boundary condition data set including the temperature distribution in the kettle and the environment temperature is established; Using the temperature boundary condition data set, the reactor wall is discretized by a hexahedral mesh generation method to obtain a reactor wall mesh data set; According to the material property data set and the grid data set, a finite difference method is used to solve the thermal stress control equation to obtain a reactor wall stress distribution data set.

5. The method according to claim 1, characterized in that The method extracts the radial and axial gradients, gradient directions, and uniformity of temperature distribution of the stress concentration area according to the stress distribution cloud map, processes the gradients through a support vector machine algorithm, obtains the position and stress magnitude of the stress concentration area, and determines whether the stress concentration area exceeds the safety boundary of the equipment according to the obtained position and stress magnitude of the stress concentration area, including: According to the stress distribution cloud map data, the stress value is segmented into regions using the regional growing algorithm, and the spatial contour data of the stress concentration area is obtained based on the multiple threshold of the arithmetic mean value of the local regional stress through the regional growing algorithm; Calculating the value and direction of the temperature field gradient based on the spatial contour data of the stress concentration area, wherein the temperature field gradient includes radial and axial components, and obtaining a temperature gradient vector data set; A stress characteristic vector is established according to the temperature gradient vector data set using a support vector machine algorithm, wherein the stress characteristic vector includes a stress value, a temperature gradient magnitude and a gradient direction angle, and a temperature uniformity index dimension, to obtain a stress concentration area characteristic vector data set; A Gaussian mixture density algorithm is used to perform regional clustering on the characteristic vector data set of the stress concentration area. The Gaussian mixture density algorithm classifies the characteristic vectors based on a preset number of Gaussian components, obtains the central coordinate point and the maximum stress value of the stress concentration area, and determines whether the stress value of the stress concentration area exceeds the equipment stress safety limit range according to the stress safety threshold; The method also includes: identifying the range of the stress concentration area from the stress distribution cloud map, extracting the radial and axial temperature gradient size data, temperature gradient direction data and temperature distribution uniformity data of the stress concentration area, and vectorizing the data, constructing a support vector machine classification model, classifying the stress concentration area, and after obtaining the classification boundary, determining the stress concentration position, obtaining the stress value of the stress concentration position, and obtaining the stress size of the stress concentration area.

6. The method according to claim 5, characterized in that The method includes identifying the range of the stress concentration area from the stress distribution cloud map, extracting radial and axial temperature gradient size data, temperature gradient direction data, and temperature distribution uniformity data of the stress concentration area, vectorizing the data, constructing a support vector machine classification model, classifying the stress concentration area, and determining the stress concentration position after obtaining the classification boundary, and obtaining the stress value of the stress concentration position to obtain the stress size of the stress concentration area, including: Determine a growth threshold according to the local mean value of stress, calculate the stress value gradient using the Sobel operator, mark it as a boundary point of the stress concentration area when the stress gradient value exceeds the growth threshold, and obtain spatial range data of the stress concentration area; Calculate radial and axial temperature gradient values ​​and gradient direction angles based on the spatial range data of the stress concentration area, and calculate the temperature uniformity index of the stress concentration area using the temperature field mean square error; The temperature gradient value, the gradient direction angle, and the temperature uniformity index are standardized by using a minimum-maximum normalization method to construct a characteristic vector matrix including three components: the temperature gradient value, the gradient direction angle, and the temperature uniformity; The support vector machine classifier is used to classify the characteristic vector matrix, and the coordinates of discrete points of the boundary curve are extracted to obtain the position coordinates of the stress concentration area and the stress numerical data.

7. The method according to claim 1, characterized in that If the stress in the stress concentration area exceeds the yield strength of the equipment material, it is determined that the stress concentration area breaks through the equipment safety boundary, and a genetic algorithm is used to process the stress distribution state of the kettle wall to obtain a propagation path of the stress distribution evolution, and the position coordinates and stress peak value of the stress breakthrough point are obtained from the propagation path, including: According to the yield strength of the reactor wall material, it is judged whether the maximum stress value in the stress concentration area exceeds the yield strength. If it exceeds the yield strength, the stress breakthrough point coordinates and stress value are obtained from the stress concentration area; A genetic algorithm is used to construct a stress propagation path according to the stress breakthrough point coordinates and stress values, and an initial stress distribution evolution path is obtained by using the genetic algorithm to encode the stress distribution coordinate points in real numbers; For the initial evolution path of the stress distribution, a fitness function is set to calculate the stress gradient values ​​of adjacent points, and the data of the maximum propagation path of the stress gradient is obtained through iterative optimization; According to the stress gradient maximum propagation path data, key nodes whose stress gradient difference exceeds a preset threshold are selected, the gradient tracking method is used to calculate the stress propagation direction between nodes, and the stress distribution evolution data set is obtained through cubic spline interpolation reconstruction.

8. The method according to claim 1, characterized in that According to the position coordinates of the stress breakthrough point and the stress peak value, the influence of the stress distribution evolution on the equipment safety boundary is analyzed, and the influence includes local plastic deformation, fracture, and fatigue failure caused by stress concentration. The equipment safety hazard level is determined according to the degree to which the stress peak value exceeds the yield strength. The equipment safety hazard level includes low and high levels, including: According to the ratio of the yield strength of the reactor material to the peak stress at the stress breakthrough point, the stress-strain curve is established using the plastic strain theory to obtain the plastic deformation data set at the stress breakthrough point; For the stress breakthrough point plastic deformation data set, a support vector regression algorithm is used to fit the stress-strain curve, and the support vector regression algorithm selects radial basis function as the kernel function to obtain stress-strain fitting curve data; According to the stress-strain fitting curve data, the critical stress intensity factor KIC at the stress breakthrough point is calculated using the fracture mechanics criterion, where σ represents stress, a represents crack length, and characteristic data of fracture parameters are obtained; According to the fracture parameter characteristic data, the fatigue damage degree D of the stress breakthrough point is calculated by using the Miner linear cumulative damage criterion, where D=∑(ni / Ni), ni represents the actual number of cycles, and Ni represents the number of fracture cycles, to obtain fatigue characteristic data; Combining the plastic deformation data set and fatigue characteristic data of the stress breakthrough point, the safety hazard level of the stress breakthrough area is divided.

9. The method according to claim 1, characterized in that: If the equipment safety hazard level is high, the emergency shutdown protection mechanism of the high-temperature cracking device is triggered, and the equipment safety warning information is sent to the control center to complete the dynamic warning process, including: The random forest classifier is used to evaluate the equipment status, and the equipment high-level safety hazard status judgment data is obtained based on the stress peak, stress breakthrough point location and hazard level indicators; A state transition diagram is established for the high-level safety hazard state determination data of the equipment, and if it is detected that the warning state is transferred to the shutdown state, a shutdown protection instruction data set is generated; The control instructions are encoded and prioritized according to the shutdown protection instruction data set, wherein the closing priority of the feed valve is the first priority, the heating power reduction priority is the second priority, and the opening priority of the gas venting valve is the third priority, and control instruction execution sequence data is obtained; A real-time sequence scheduling method is used to generate an instruction execution timing table for the control instruction execution sequence data, set the interval length of adjacent instruction executions, construct an early warning information data packet containing a timestamp, instruction code and execution status, and send it to the control center through a message queue mechanism.

Images