Testing system and method for simulating fragment casting characteristics during leakage of storage tank

Abstract

The invention belongs to the technical field of storage tank safety testing, and particularly relates to a testing system and method for simulating fragment casting characteristics during leakage of a storage tank. The storage tank simulation device comprises a cylindrical stainless steel storage tank, a storage tank leakage opening is sequentially provided with an electromagnetic valve and a compression ring pore plate, the compression ring pore plate adjusts the size of the leakage opening, fragments abut against the compression ring pore plate, and after the electromagnetic valve is opened, jet flow acts on the fragments; the analysis processing unit processes the image data through an image recognition algorithm and reconstructs a three-dimensional motion trail of the fragment to obtain a state variable parameter of the fragment, combines a physical differential equation with a neural network when constructing a motion trail model, and parameterizes unknown items in the physical differential equation by using the neural network to obtain a motion trail model; and optimizing the trajectory in real time through a self-adaptive solver. Through fusion of a physical model and a machine learning algorithm, the physical interpretability of the motion track model can be ensured, complex nonlinear dynamics during debris casting can be captured, and the method is suitable for casting motion analysis in a real complex environment.

Description

Technical Field

[0001] The invention belongs to the technical field of storage tank safety testing, and in particular relates to a testing system and method for simulating the debris ejection characteristics when a storage tank leaks. Background Art

[0002] When a pressure vessel storing liquefied gas leaks, it can rupture and produce fragments. These fragments possess high kinetic energy, posing a serious threat to surrounding personnel and equipment. Currently, research on the motion patterns of fragments focuses primarily on theoretical analysis and numerical simulation. Experimental data is obtained by analyzing past accident cases and using large, near-end-of-life tanks. This is not only costly and dangerous, but also results in poor reproducibility. Therefore, a low-cost, parameter-controllable, and reusable testing device is needed.

[0003] In addition, there are currently three main types of debris trajectory models: empirical models, physical models, and machine learning models. The empirical model fits empirical formulas based on experimental data, such as establishing a relationship between debris velocity and leakage pressure through regression analysis, or fitting a trajectory curve with a polynomial. This method has poor universality, relies on specific experimental conditions, lacks physical meaning, and cannot explain causal relationships. The physical model combines fluid mechanics (leakage jet thrust) and Newtonian mechanics (fragment force analysis) to establish differential equations, and solves the debris trajectory through numerical solutions. The modeling process of this method needs to be simplified, and it is difficult to solve equations in complex scenarios. It relies on accurate initial conditions and parameter measurements. The machine learning model uses a neural network algorithm to predict trajectory parameters with leakage parameters and fragment initial conditions as input. This method relies on a large amount of experimental data or engineering data and is labor-intensive. Summary of the Invention

[0004] To solve the above technical problems, the present invention provides a testing system and method for simulating the debris ejection characteristics during tank leakage. Using a small experimental device, the speed, angle and other parameters of the debris ejection are measured, revealing the motion mechanism of the debris ejection caused by BLEVE in small-sized containers. The effectiveness of the specific motion trajectory model is verified through experimental data, providing strong support for predicting and evaluating the hazards of debris in tank leakage accidents.

[0005] To achieve the above-mentioned purpose, the technical solution adopted by the present invention is: a test system for simulating the debris ejection characteristics when a storage tank leaks, comprising an environmental simulation chamber, a storage tank simulation device, an information acquisition unit, and an analysis and processing unit.

[0006] The environmental simulation chamber realizes coupled control of temperature and wind speed, simulating the influence of various complex weather conditions on debris ejection, and the storage tank simulation device is installed inside the environmental simulation chamber;

[0007] The tank simulation device includes a cylindrical stainless steel tank, a solenoid valve and a pressure ring orifice plate are sequentially provided at the tank leakage port, the pressure ring orifice plate adjusts the size of the leakage port, the debris is close to the pressure ring orifice plate, and the jet flow acts on the debris after the solenoid valve is opened;

[0008] The information acquisition unit includes at least two high-speed cameras and a pressure sensor, a thermocouple, and a liquid level gauge installed on the storage tank, which transmits the measured experimental data to the analysis and processing unit;

[0009] The analysis and processing unit processes the image data through an image recognition algorithm and reconstructs the three-dimensional motion trajectory of the fragment to obtain the state variable parameters of the fragment. When constructing the motion trajectory model, the physical differential equation is combined with the neural network, the unknown terms in the physical differential equation are parameterized by the neural network, and the trajectory is optimized in real time through an adaptive solver.

[0010] Preferably, the analysis and processing unit includes a motion trajectory model module, which integrates a physical differential equation component, a neural network component, and an adaptive solver component. The physical differential equation component establishes a physical differential equation in the debris ejection process based on classical mechanics theory; the neural network component represents the unknown force in the physical differential equation as the output of the neural network component; the adaptive solver component embeds the unknown force output by the neural network component into the physical differential equation solving process, and calculates the debris motion trajectory in real time through an adaptive integrator; the initial state variables and environmental parameters of the fragments are used as inputs of the motion trajectory model, and the motion trajectory of the fragments is used as the model output.

[0011] Preferably, the physical differential equation component expresses the physical differential equation as:

[0012]

[0013] Where Z = [X, V] is the state variable of the fragment, X is the position of the fragment (x, y, z), V is the speed of the fragment (vx, vy, vz), and t is the time; h 已知 (Z, t) is the known force in the form of fragment state variables; f θ (Z, t) is the unknown force f that the fragments are subjected to during the ejection process. 未知力 (X, V, t) is parameterized as a neural network, and θ is the neural network parameter.

[0014] Preferably, the neural network component takes the fragment state variable Z and the environmental parameter p as input, and the unknown term f in the physical differential equation θ (Z, t) is used as the output, and a fully connected network or a convolutional neural network is used as the network structure. The environmental parameters include the pressure in the tank, the air density, the debris shape coefficient, the wind speed and temperature in the environmental simulation cabin.

[0015] Preferably, the adaptive solver component embeds the unknown terms in the differential equation output by the neural network component into the physical differential equation solving process, calculates the motion trajectory in real time through the Runge-Kutta method, and predicts the motion trajectory at any time t from the initial condition Z(t0) of the fragment. i The state Z(t i ),Right now

[0016] Preferably, the total loss function L of the motion trajectory model is 总 as follows:

[0017] L 总 =L1+λ1L2+λ2L3;

[0018] Among them, L1 is the data fitting loss, Z 预测 The model predicts the trajectory of the debris movement, Z 真实 The actual trajectory captured by the high-speed camera;

[0019] L2 is the residual loss of the differential equation, ensuring that the predicted derivatives satisfy the known physical terms. λ1 is the balance weight corresponding to the residual loss of the differential equation;

[0020] L3 is the energy conservation and momentum conservation loss, L3 = |E(Z(t))-E(Z(0))|, E(Z(t)) and E(Z(0)) are the system energy functions at time t and time 0 respectively, and λ2 is the balance weight corresponding to the conservation loss.

[0021] Preferably, the analysis and processing unit includes an image data processing module, which uses a threshold segmentation method to separate fragment targets and a target tracking algorithm to obtain the movement trajectory of the fragments; for the i-th moment t i The debris track position X i , calculate the velocity V by numerical differentiation i , acceleration a i , construct the data set R = {(X i ,V i ,a i ,t i ,p i )},p i represents the i-th moment t i Corresponding environmental parameters.

[0022] Accordingly, the test method for simulating the debris projection characteristics when the tank leaks includes the following steps:

[0023] S1. Build a simulation test system and conduct experiments. Change the heating coil temperature, tank liquid level, debris type, wind speed and temperature in the environmental simulation chamber to simulate the debris ejection characteristics of a tank leak under different working conditions. Record the complete video sequence of the debris moving from the ejection point to the end point, and save the image data of the debris under different working conditions.

[0024] S2. Perform stereo calibration on multiple high-speed cameras, calculate camera intrinsic and extrinsic parameters, and establish a three-dimensional world coordinate system;

[0025] S3, use the threshold segmentation method to separate the fragment targets, use the target tracking algorithm to obtain the movement trajectory of the fragments, calculate the velocity and acceleration of the fragments by numerical differentiation, and construct the data set R = {(X i ,V i ,a i ,t i ,p i )},X i Indicates that the fragment is at time t i The trajectory position, V i Indicates that the fragment is at time t i The speed of a i Indicates that the fragment is at time t i The acceleration, p i represents the i-th time point t i Corresponding environmental parameters;

[0026] S4. Establishing physical differential equations for debris ejection characteristics based on classical mechanics theory The unknown force f in the physical differential equation 未知力 (X, V, t) is expressed as the output of the neural network component. The adaptive solver component embeds the unknown force output by the neural network component into the physical differential equation solution process and calculates the fragment motion trajectory in real time through the adaptive integrator.

[0027] Preferably, step S3 includes the following process:

[0028] S31. Analyze the original video frame by frame, extract single-frame images, optimize image quality through noise reduction, brightness / contrast adjustment, distortion correction, etc., use threshold segmentation to separate fragment targets from the background, and identify fragment feature points;

[0029] S32. Based on the multi-frame image sequence, cross-frame matching is performed on the feature points of the same fragment using a target tracking algorithm to obtain its spatial coordinates at different times;

[0030] S33, combining camera calibration parameters and multi-view parallax principle, calculating the actual 3D spatial position of the fragments through 3D reconstruction technology;

[0031] S34. Calculate the instantaneous velocity, acceleration and other motion parameters of the fragments based on the time interval and displacement between adjacent frames, and analyze the direction and change trend of the velocity vector.

[0032] The present invention has the following beneficial effects:

[0033] 1. The motion trajectory model of the present invention integrates a physical differential equation component, a neural network component, and an adaptive solver component. The neural network component represents the unknown forces in the physical differential equation as the output of the neural network component. The adaptive solver component embeds the unknown forces output by the neural network component into the physical differential equation solution process, and an adaptive integrator is used to calculate the debris motion trajectory in real time. By integrating the causal logic of the physical model with the data fitting capabilities of the machine learning algorithm, the motion trajectory model of the debris ejection characteristics is not only physically interpretable, but also breaks through the reliance of traditional modeling on simplifying assumptions. The neural network is used to capture the complex nonlinear dynamics of debris ejection. This is particularly suitable for analyzing and predicting the motion of debris ejection during tank leaks in complex real-world environments, providing a data basis for accident and disaster prevention. Furthermore, by leveraging the continuity of differential equations, the model can handle debris trajectory predictions at any time resolution.

[0034] 2. The test system of the present invention scales down the storage tanks used in actual production, and at the same time sets a heating module at the bottom of the storage tank and fragments on the top of the storage tank. This can simulate the situation in actual production where the storage tank is heated, the pressure inside the tank increases, and the weak environment ruptures, inducing a BLEVE accident, resulting in container rupture and fragment impact. The tank simulation device is placed in an environmental simulation cabin to simulate the debris ejection under different temperature and wind speed environments, eliminating the influence of environmental factors on the experimental results, and improving the repeatability of the experiment and the reliability of the data. The speed and angle of the debris ejection are explored through fragment ejection experiments under different liquid levels, pressures and leakage areas, and the range of the fragment impact is limited to a controllable range by a glass cover.

[0035] 3. The testing system of this invention innovatively evenly applies a luminescent coating to the surface of the debris and is equipped with professional lighting equipment for supplemental lighting to locate the debris device, greatly enhancing the visibility of the debris and providing clear image data for subsequent observation and analysis. Highly reliable experimental data was obtained through a small-sized container, revealing the range of debris ejection and the safe distance from the storage tank, helping to prevent damage to the surrounding area caused by debris ejection and block the spread of the domino effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a test flow chart of the simulation test system of the present invention. DETAILED DESCRIPTION

[0037] The following will be combined with the accompanying drawings to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Unless otherwise specified, the technical means used in the embodiments are conventional means well known to those skilled in the art.

[0038] It should be noted that when an element is referred to as being “disposed on” or “moved to” another element, it may be directly on the other element or indirectly on the other element. When an element is referred to as being “connected to” another element, it may be directly connected to the other element or indirectly connected to the other element.

[0039] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the quantity of the indicated technical features.

[0040] The present invention discloses a test system for simulating the debris ejection characteristics when a storage tank leaks, comprising an environmental simulation cabin, a storage tank simulation device, an information acquisition unit, and an analysis and processing unit. The storage tank simulation device is placed in the environmental simulation cabin.

[0041] The environmental simulation chamber can achieve coupled control of temperature and wind speed, simulating the impact of various complex weather conditions on debris ejection, while effectively eliminating the interference of random variables in the natural environment, improving the repeatability and reliability of experimental data. Specifically, the environmental simulation chamber can adjust the wind speed within the range of 0-15m / s, and can simulate a variety of wind speed scenarios including no wind, light breeze, and strong wind. The temperature can be adjusted within the range of -30℃-50℃, and can simulate complex weather conditions from low temperature to high temperature, providing a comprehensive temperature environment for ejection experiments. The environmental simulation chamber is also equipped with professional lighting equipment for fill light, such as high-intensity LED or laser light sources, to ensure uniform illumination of the shooting area. At the same time, a calibration plate of known size can be placed on the floor inside the environmental simulation chamber to facilitate the subsequent conversion between image pixels and actual physical size.

[0042] The storage tank simulation device includes a cylindrical stainless steel storage tank that is proportionally reduced in size to the actual container. The tank leakage port is provided with a solenoid valve and a pressure ring orifice plate according to the gas flow direction, that is, the pressure ring orifice plate is fixedly installed at the outlet end of the solenoid valve, and the pressure ring orifice plate can adjust the size of the leakage port. The fragments are close to the pressure ring orifice plate. After the solenoid valve is opened, the jet flow acts on the fragments, thereby ejecting the fragments; a support base is fixedly installed on the ground, and a groove is provided on the support base that is adapted to the bottom of the cylindrical stainless steel storage tank, so that the cylindrical stainless steel storage tank is fixedly installed on the support base through the groove, and a cavity is provided inside the support base and below the cylindrical stainless steel storage tank, and a heating coil is provided in the cavity. The heating coil is used to heat the liquid gas in the cylindrical stainless steel storage tank to gasify it, and the heating coil is adjusted by adjusting the pressure ring orifice plate. The heating temperature simulates different leakage intensities, and the heating temperature and the liquid level height in the tank jointly affect the pressure in the tank; the cylindrical stainless steel tank is also equipped with pressure sensors, thermocouples, liquid level gauges and other measuring elements. The pressure sensor measures the pressure in the tank, the thermocouple measures the temperature in the tank, and the liquid level gauge measures the liquid level in the tank. These measuring elements are connected to the information acquisition unit. When the internal pressure of the cylindrical stainless steel tank reaches the target value, the solenoid valve is remotely opened to simulate the tank leakage; in particular, due to the influence of the mist jet formed during the leakage and the dark environment, direct observation is difficult, so a luminescent coating is evenly applied to the surface of the fragments to facilitate monitoring their movement trajectory; fragments of different shapes and sizes are made to study the movement trajectories of fragments of different forms;

[0043] The information collection unit includes multiple cameras, which shoot simultaneously from multiple positions (at least from two orthogonal perspectives). For example, a high-speed camera is deployed on the front and side of a cylindrical stainless steel storage tank. Time codes or optical synchronization signals are added to the video to ensure time alignment of multi-camera data. The high-speed cameras shoot synchronously from two or more perspectives at a sampling frequency of ≥1000 frames per second, which can accurately record the movement trajectory of debris. The shutter speed of the high-speed camera is ≤1 / 10000s to avoid motion blur. The aperture and ISO are adjusted according to the intensity of the light source to ensure clear images. Multiple cameras are started synchronously by a trigger signal. The trigger signal can be the moment when the pressure sensor on the cylindrical stainless steel storage tank senses a leak or the moment after the solenoid valve is opened.

[0044] The analysis and processing unit processes the fragment images captured by the camera through an image recognition algorithm, tracks the fragment motion trajectory and reconstructs the three-dimensional trajectory of the fragments to obtain the state variable parameters of the fragments; when constructing the motion trajectory model, the physical differential equations are combined with neural networks, and the unknown terms in the physical differential equations are parameterized by the neural network. The trajectory is optimized in real time through an adaptive solver, and the constructed motion trajectory model is used to reveal the projection characteristics of the fragments when the tank leaks.

[0045] In a further embodiment, the analysis and processing unit is a very important part of the test system, including an image data processing module and a motion trajectory model module. The plurality of high-speed cameras are communicatively connected to the image data processing module, and the image data processing module is communicatively connected to the motion trajectory model module.

[0046] The image data processing module uses Zhang's calibration method to perform stereo calibration on dual cameras or multiple cameras, calculates the camera's internal parameters (focal length, distortion coefficient) and external parameters (rotation matrix, translation vector), establishes a three-dimensional world coordinate system, and sets the origin at the center of the leak. The pixel coordinates are converted into physical coordinates using the Matlab / Simulink computer vision toolbox. The threshold segmentation method is used to separate the debris targets, and the target tracking algorithm is used to obtain the motion trajectory of the debris. i The debris track position X i , calculate the velocity V by numerical differentiation i , acceleration a i , construct the data set R = {(X i ,V i ,a i ,t i ,p i )},p i represents the i-th moment t i Corresponding environmental parameters.

[0047] The motion trajectory model module includes a physical differential equation component, a neural network component, and an adaptive solver component. The physical differential equation component establishes the physical differential equations in the debris ejection process based on the theory of classical mechanics. The neural network component transforms the unknown force f in the physical differential equation 未知力 (X, V, t) is represented as the output of the neural network component; the adaptive solver component embeds the unknown force output by the neural network component into the physical differential equation solution process, and calculates the fragment motion trajectory in real time through the adaptive integrator.

[0048] In a further implementation, the physical differential equation component uses the relationship between the acceleration and the force of the fragment as the core physical constraint for motion trajectory modeling. The physical differential equation is expressed as follows:

[0049]

[0050] Among them, X represents the position of the fragment (x, y, z), V represents the speed of the fragment (v x ,v y ,v z ), t is time; f 已知力 is the known net force (f x ,f y ,fz ), including air resistance and self-weight, f 已知力 =f 重力 +f 空气阻力 =-mge z -kV, m is the mass of the debris, g is the acceleration due to gravity, e z is the vertical unit vector, k is the linear air resistance coefficient, which is related to air density and debris size; f 未知力 is the unknown net force acting on the fragment at time t;

[0051] Then the unknown force f that the fragments are subjected to during the ejection process is 未知力 (X,V,t) is parameterized as a neural network f θ (Z, t), θ is the neural network parameter, and the physical differential equation is expressed as:

[0052]

[0053] Among them, Z = [X, V] is the state variable of the fragment; h 已知 are the known forces expressed in terms of fragment state variables.

[0054] In a further embodiment, the neural network component is used to parameterize the unknown terms in the physical differential equations that are difficult to model analytically, and to learn the complex physical relationships in the debris ejection process through data-driven learning. The state variables Z of the debris and the environmental parameters p are used as inputs, and the unknown terms f in the physical differential equations are used as θ (Z, t) (time derivative of the state variable) is used as the output, and a fully connected network, convolutional neural network, etc. are used as the network structure.

[0055] Specifically, the input layer of the neural network architecture is the concatenated state vector [X,V,t] (7-dimensional vector, (x,y,zy,v x ,v y ,v z ,t)), environmental parameters p, including the pressure inside the tank at the time of leakage, air density, fragment shape coefficient, fragment mass, wind speed and temperature in the environmental simulation chamber, etc. The fragment shape coefficient can be customized. The hidden layer is set as a 2-layer fully connected network with 64 neurons in each layer. The activation function uses the Swish function. The output layer is the unknown force f θ (Z,t)(three-dimensional vector, (f` x ,f` y ,f` z )).

[0056] Embed physical constraints in the neural network, for example, add a projection layer after the output layer so that the angle between the unknown force and the velocity direction is less than 90°, but the direction of the resistance is opposite to the direction of motion; constrain the energy and momentum conservation of the system through the loss function L3, L3 can be the sum of energy conservation and momentum conservation losses, energy conservation loss, or momentum conservation loss, L3 = |E(Z(t))-E(Z(0))|, E(Z(t)) and E(Z(0)) are the system energy functions at time t and time 0 respectively. For example, if only the energy conservation loss is considered, then z(t) is the height of the fragment at time t, V0 is the velocity of the fragment at time 0, and z0 is the height of the fragment at time 0.

[0057] In a further embodiment, the adaptive solver component embeds the unknown term in the differential equation output by the neural network component into the physical differential equation In the solution process, the motion trajectory is calculated in real time by the Runge-Kutta method, and the initial condition Z(t0) of the fragment is used to predict the motion trajectory at any time t i The state Z(t i ),Right now At each time step, the neural network component predicts unknown forces based on the current state variables, while the physical differential equation component updates the position and velocity of the fragment based on the total force acting on it. The neural differential equations transform the discrete computations between network layers into a continuous differential process through a numerical integrator with an adaptive step size, reducing the discretization error of traditional neural networks.

[0058] In a further implementation method, the principle of the motion trajectory model is to parameterize other forces (unknown terms) that are difficult to express with mathematical formulas into neural networks, and combine them with known physical terms to construct physical differential equations, integrate physical differential equation components, neural network components, and adaptive solver components, use the initial state variables and environmental parameters of the fragments as model inputs, and use the motion trajectory of the fragments as output. During the motion estimation model training process, 80% of the experimental data of different pressures and different fragment types are used as training sets to train the motion estimation model, and the remaining 20% ​​of the experimental data are used as validation sets to verify the model accuracy. The AdamW optimizer is used to iteratively update the parameters, and the initial learning rate e is used as the training set. -3 , decays by 0.9 every 50 epochs, and the weight decays by 0.01. The total loss function L of the motion trajectory model 总 as follows:

[0059] L 总 =L1+λ1L2+λ2L3;

[0060] Among them, L1 is the data fitting loss, Z 预测 The model predicts the trajectory of the debris movement, Z 真实is the actual trajectory captured by the high-speed camera; L2 is the residual loss of the differential equation, which ensures that the predicted derivative satisfies the known physical terms. λ1 is the balance weight corresponding to the residual loss of the differential equation; L3 is the energy conservation and momentum conservation loss, L3 = |E(Z(t))-E(Z(0))|, E(Z(t)) and E(Z(0)) are the system energy functions at time t and time 0, respectively, and λ2 is the balance weight corresponding to the conservation loss.

[0061] The training process of the motion trajectory model is to input the initial state Z(t0), environmental parameters and time series t1, t2, ..., t T , calculate Z(t i )’s predicted value, calculate the total loss function L 总 And update the network parameters θ. The network parameters θ are the weights and biases of the neural network components. Through training, θ is adjusted so that the calculated f θ (Z, t) is close to the unknown forces in real physical scenarios, making the entire physical differential equation more consistent with actual observation data.

[0062] It should be noted that the linear air resistance coefficient k in the physical differential equation was obtained by fitting the initial experimental data and used as a known term input to the motion trajectory model to reduce the burden on the neural network. The neural network weights were initialized using Xavier, and the bias was set to 0.

[0063] like Figure 1 As shown, the present invention also discloses a test method for simulating the debris ejection characteristics when a storage tank leaks, comprising the following steps:

[0064] S1. Build a simulation test system and conduct experiments. Change the heating coil temperature, tank liquid level, debris type, wind speed and temperature in the environmental simulation chamber to simulate the debris ejection characteristics of a tank leak under different working conditions. Record the complete video sequence of the debris moving from the ejection point to the end point, and save the image data of the debris under different working conditions.

[0065] S2. Perform stereo calibration on multiple high-speed cameras, calculate camera intrinsic and extrinsic parameters, and establish a three-dimensional world coordinate system;

[0066] S3, use the threshold segmentation method to separate the fragment targets, use the target tracking algorithm to obtain the movement trajectory of the fragments, calculate the velocity and acceleration of the fragments by numerical differentiation, and construct the data set R = {(X i ,V i ,a i ,t i ,p i )},X i Indicates that the fragment is at time t i The trajectory position, Vi Indicates that the fragment is at time t i The speed of a i Indicates that the fragment is at time t i The acceleration, p i represents the i-th time point t i Corresponding environmental parameters;

[0067] S4. Establishing physical differential equations for debris ejection characteristics based on classical mechanics theory The unknown force f in the physical differential equation 未知力 (X, V, t) is expressed as the output of the neural network component. The adaptive solver component embeds the unknown force output by the neural network component into the physical differential equation solution process and calculates the fragment motion trajectory in real time through the adaptive integrator.

[0068] In a further embodiment, step S3 includes the following process:

[0069] S31. Analyze the original video frame by frame, extract single-frame images, optimize image quality through noise reduction, brightness / contrast adjustment, distortion correction, etc., use threshold segmentation to separate fragment targets from the background, and identify fragment feature points;

[0070] S32. Based on the multi-frame image sequence, cross-frame matching is performed on the feature points of the same fragment using a target tracking algorithm to obtain its spatial coordinates at different times;

[0071] S33, combining camera calibration parameters and multi-view parallax principle, calculating the actual 3D spatial position of the fragments through 3D reconstruction technology;

[0072] S34. Calculate the instantaneous velocity, acceleration and other motion parameters of the fragments based on the time interval between adjacent frames (determined by the camera frame rate) and the displacement, and analyze the direction and change trend of the velocity vector.

[0073] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various deformations, modifications, and substitutions made to the technical solutions of the present invention by ordinary technicians in this field should fall within the scope of protection determined by the claims of the present invention.

Claims

1. A test system for simulating the debris projection characteristics of a tank leak, characterized by: Including environmental simulation cabin, tank simulation device, information collection unit, analysis and processing unit, The environmental simulation chamber realizes coupled control of temperature and wind speed, simulating the influence of various complex weather conditions on debris ejection, and the storage tank simulation device is installed inside the environmental simulation chamber; The tank simulation device includes a cylindrical stainless steel tank, a solenoid valve and a pressure ring orifice plate are sequentially provided at the tank leakage port, the pressure ring orifice plate adjusts the size of the leakage port, the debris is close to the pressure ring orifice plate, and the jet flow acts on the debris after the solenoid valve is opened; The information acquisition unit includes at least two high-speed cameras and a pressure sensor, a thermocouple, and a liquid level gauge installed on the storage tank, which transmits the measured experimental data to the analysis and processing unit; The analysis and processing unit processes the image data through an image recognition algorithm and reconstructs the three-dimensional motion trajectory of the fragment to obtain the state variable parameters of the fragment. When constructing the motion trajectory model, the physical differential equation is combined with the neural network, the unknown terms in the physical differential equation are parameterized by the neural network, and the trajectory is optimized in real time through an adaptive solver.

2. The test system for simulating the debris projection characteristics of a storage tank leak according to claim 1, characterized in that: The analysis and processing unit includes a motion trajectory model module, which integrates a physical differential equation component, a neural network component, and an adaptive solver component. The physical differential equation component establishes the physical differential equations in the debris ejection process based on classical mechanics theory; the neural network component represents the unknown forces in the physical differential equations as outputs of the neural network component. The adaptive solver component embeds the unknown forces output by the neural network component into the physical differential equation solution process and calculates the debris motion trajectory in real time through the adaptive integrator; The initial state variables and environmental parameters of the fragments are used as the input of the motion trajectory model, and the motion trajectory of the fragments is used as the model output.

3. The test system for simulating the debris projection characteristics of a storage tank leak according to claim 2, characterized in that: The physical differential equation component expresses the physical differential equation as: Among them, Z = [X, V] is the state variable of the fragment, X is the position of the fragment (x, y, z), V is the speed of the fragment (v x ,v y ,v z ), t is time; h 已知 (Z, t) is the known force in the form of fragment state variables; f θ (Z, t) is the unknown force f that the fragments are subjected to during the ejection process. 未知力 (X, V, t) is parameterized as a neural network, the output of the neural network, and θ is the neural network parameter.

4. The test system for simulating the debris projection characteristics of a storage tank leak according to claim 3, characterized in that: The neural network component takes the fragment state variable Z and the environmental parameter p as input, and the unknown term f in the physical differential equation θ (Z, t) is used as the output, and a fully connected network or a convolutional neural network is used as the network structure. The environmental parameters include the pressure in the tank, the air density, the debris shape coefficient, the wind speed and temperature in the environmental simulation cabin.

5. The testing system for simulating the debris ejection characteristics of a storage tank leak according to claim 4, characterized in that: The adaptive solver component embeds the unknown terms in the differential equation output by the neural network component into the physical differential equation solving process, calculates the motion trajectory in real time through the Runge-Kutta method, and predicts the motion trajectory at any time t from the initial condition Z(t0) of the fragment. i The state Z(t i ),Right now 6. The testing system for simulating the debris projection characteristics of a storage tank leak according to claim 2, characterized in that: The total loss function L of the motion trajectory model 总 as follows: L 总 =L1+λ1L2+λ2L3; Among them, L1 is the data fitting loss, Z 预测 The model predicts the trajectory of the debris movement, Z 真实 The actual trajectory captured by the high-speed camera; L2 is the residual loss of the differential equation, ensuring that the predicted derivatives satisfy the known physical terms. λ1 is the balance weight corresponding to the residual loss of the differential equation; L3 is the energy conservation and momentum conservation loss, L3 = |E(Z(t))-E(Z(0))|, E(Z(t)) and E(Z(0)) are the system energy functions at time t and time 0 respectively, and λ2 is the balance weight corresponding to the conservation loss.

7. The testing system for simulating the debris projection characteristics of a storage tank leak according to claim 2, characterized in that: The analysis and processing unit includes an image data processing module, which uses a threshold segmentation method to separate fragment targets and a target tracking algorithm to obtain the movement trajectory of the fragments; i The debris track position X i , calculate the velocity V by numerical differentiation i , acceleration a i , construct the data set R = {(X i ,V i ,a i ,t i ,p i )},p i represents the i-th moment t i Corresponding environmental parameters.

8. A test method for simulating the projection characteristics of debris during a tank leak, characterized by: The steps include: S1. Build a simulation test system and conduct experiments. Change the heating coil temperature, tank liquid level, debris type, wind speed and temperature in the environmental simulation chamber to simulate the debris ejection characteristics of a tank leak under different working conditions. Record the complete video sequence of the debris moving from the ejection point to the end point, and save the image data of the debris under different working conditions. S2. Perform stereo calibration on multiple high-speed cameras, calculate camera intrinsic and extrinsic parameters, and establish a three-dimensional world coordinate system; S3, use the threshold segmentation method to separate the fragment targets, use the target tracking algorithm to obtain the movement trajectory of the fragments, calculate the velocity and acceleration of the fragments by numerical differentiation, and construct the data set R = {(X i ,V i ,a i ,t i ,p i )},X i Indicates that the fragment is at time t i The trajectory position, V i Indicates that the fragment is at time t i The speed of a i Indicates that the fragment is at time t i The acceleration, p i represents the i-th time point t i Corresponding environmental parameters; S4. Establishing physical differential equations for debris ejection characteristics based on classical mechanics theory The unknown force f in the physical differential equation 未知力 (X, V, t) is expressed as the output of the neural network component. The adaptive solver component embeds the unknown force output by the neural network component into the physical differential equation solution process and calculates the fragment motion trajectory in real time through the adaptive integrator.

9. The method for testing the debris projection characteristics when simulating a tank leak according to claim 8, characterized in that: The step S3 includes the following process: S31. Analyze the original video frame by frame, extract single-frame images, optimize image quality through noise reduction, brightness / contrast adjustment, distortion correction, etc., use threshold segmentation to separate fragment targets from the background, and identify fragment feature points; S32. Based on the multi-frame image sequence, cross-frame matching is performed on the feature points of the same fragment using a target tracking algorithm to obtain its spatial coordinates at different times; S33, combining camera calibration parameters and multi-view parallax principle, calculating the actual 3D spatial position of the fragments through 3D reconstruction technology; S34. Calculate the instantaneous velocity, acceleration and other motion parameters of the fragments based on the time interval and displacement between adjacent frames, and analyze the direction and change trend of the velocity vector.

Images