Explosion venting design method based on explosion theory and extreme gradient boosting algorithm

Abstract

This invention, based on explosion theory and the extreme gradient boosting algorithm, relates to the field of explosion venting safety protection technology. The method is grounded in the combustion-explosion reaction pressure rise mechanism, venting pressure drop mechanism, and choke flow theory during explosion venting. It collects data such as vent parameters, rupture disc parameters, and initial venting conditions, and constructs a dimensionless feature parameter matrix with physical meaning using explosion theory. This matrix is ​​then used as model input and trained using an extreme gradient boosting tree model to predict the maximum venting pressure. Constrained by the container's allowable pressure or the target venting pressure, the method iteratively searches for the venting area of ​​candidate rupture discs, outputting the effective venting area of ​​the rupture disc that satisfies the safety constraints. This invention improves the stability, interpretability, and engineering applicability of explosion venting pressure prediction under limited data conditions and can be used for rupture disc venting area design in gas / dust conveying, storage, and processing scenarios.

Description

Technical Field

[0001] This invention relates to a safety protection design method for explosion venting, specifically a design method for explosion rupture disc venting based on explosion theory and extreme gradient boosting algorithm, belonging to the field of explosion venting safety protection technology. Background Technology

[0002] Dust, as an important metallic powder material, is widely used in industries such as metal processing, additive manufacturing, aerospace, chemicals, coatings, pyrotechnics, and powder metallurgy. Due to its small particle size, large specific surface area, high chemical activity, and ease of suspension and dispersion, dust can easily form combustible dust clouds during production, transportation, screening, dust removal, storage, and recycling. When combustible gas or dust clouds are within a certain concentration range and encounter ignition sources such as open flames, electric sparks, or mechanical friction sparks, violent dust explosions may occur, posing a serious threat to equipment integrity, plant structure, and personnel safety.

[0003] Explosion venting is a commonly used method of explosion protection. Its basic principle is that when the explosion pressure inside a container or pipeline rises to the opening pressure of a rupture disc, the disc ruptures, forming a vent that allows explosion products, unburned materials, and high-temperature gases to escape, thereby reducing the maximum explosion pressure inside the container. The rationality of the rupture disc venting area design directly affects the explosion venting effect and equipment safety. If the venting area is too small, the explosion pressure cannot be released in time, and the internal pressure of the container may exceed the design pressure limit; if the venting area is too large, it will increase equipment manufacturing costs, make layout difficult, expand the secondary flame jet range, and complicate the design of subsequent protection systems. Therefore, accurately determining the rupture disc venting area is of significant engineering importance.

[0004] Currently, the design of explosion relief areas largely relies on empirical or semi-empirical formulas, and its applicability is significantly limited by the physicochemical properties of the fuel gas or dust. When applied to high-explosive-index metal dust, it may produce substantial deviations. With the development of data-driven technologies, machine learning methods for predicting maximum explosion relief pressure have become more feasible. However, while existing pure machine learning methods can fit nonlinear relationships, directly using raw parameters as input can lead to problems such as inconsistent dimensions, unclear physical meanings, and poor generalization ability under insufficient sample size, making it difficult to meet the safety and interpretability requirements of rupture disc relief area design. Furthermore, after the rupture disc opens, when the pressure ratio inside and outside the container reaches a critical condition, congestion may occur at the relief port, limiting the relief mass flow rate and reducing the pressure release capacity inside the container. Traditional design methods often struggle to effectively couple explosion theories such as explosion pressure rise, relief pressure drop, and congestion flow effects with machine learning prediction models, affecting the accuracy of relief pressure prediction. Therefore, it is necessary to propose a design method for the rupture disc venting area that combines explosion venting theory with machine learning methods. By constructing physically meaningful input features through explosion theory, and then using machine learning models to learn the nonlinear relationship between explosion venting pressure and venting area, the accurate design of the rupture disc venting area can be achieved. Summary of the Invention

[0005] To address the problems existing in rupture disc venting design methods, such as the limited applicability of empirical / semi-empirical formulas, difficulty in describing nonlinear coupling relationships, poor physical interpretability of machine learning models, and insufficient prediction stability under small sample conditions, this invention provides a rupture disc venting area design method based on explosion theory and extreme gradient boosting algorithm.

[0006] The purpose of this invention is to: construct a dimensionless characteristic parameter matrix based on the physical mechanism of the competition between the explosion reaction pressure rise and the release pressure drop during the explosion release process and the theory of release congestion flow; predict the maximum explosion release pressure using an extreme gradient boosting tree model; and further, back-calculate the rupture disc release area based on allowable pressure constraints and safety margins, thereby achieving rapid and accurate design of the rupture disc release area.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a method for designing the rupture disc venting area based on explosion theory and extreme gradient boosting algorithm. Based on the combustion-explosion reaction pressure rise mechanism, venting pressure drop mechanism, and choke flow theory during the explosion venting process, a machine learning model is constructed to predict the maximum venting pressure. The rupture disc venting area that meets safety constraints is then determined based on the prediction results. Specifically, the method includes the following steps:

[0008] S1. Dataset Construction;

[0009] Data from explosion-sealing tests and explosion-venting tests were collected to form a raw sample database. The data included at least the maximum venting explosion pressure P of the combustible material. red (bar), maximum explosion pressure in a sealed container P max (bar), dust explosion index K st (bar·m / s), discharge area A v (m²), static opening pressure of the rupture disc P stat (bar), Combustible material concentration C (g / m³) 3 ), median particle size d of dust 50 (μm), container volume V (m³), container characteristic cross-sectional area A c (m²), congestion discrimination pressure P ref (bar);

[0010] S2. Analysis of control parameters for explosion theory:

[0011] The explosion venting process can be understood as a competition between combustion pressurization and venting depressurization. After ignition, the combustible material undergoes rapid combustion and oxidation reactions. The expansion of combustion products and high-temperature gases causes the internal pressure of the container to rise. When the internal pressure of the container reaches the opening pressure of the rupture disc, the rupture disc breaks and forms a vent. Unburned mixture, combustion products, and high-temperature gases are discharged from the vent, causing the internal pressure of the container to decrease.

[0012] Therefore, the maximum pressure relief is mainly related to the following factors:

[0013] 1) The explosive intensity of the combustible material itself;

[0014] 2) The matching relationship between the rupture disc venting area and the container volume;

[0015] 3) Static opening pressure of the rupture disc;

[0016] 4) The concentration of combustible materials and the degree to which it deviates from the optimum explosive concentration;

[0017] 5) Dust particle size and its effect on reactivity;

[0018] 6) Whether there is any blockage in the flow at the vent and the degree of blockage.

[0019] Based on the above analysis, the original parameters are transformed into a dimensionless feature matrix that can be recognized by the machine learning model:

[0020] ;

[0021] The parameters are defined as follows:

[0022] The dimensionless parameter for the sealed explosion pressure is:

[0023] ;

[0024] The dimensionless parameter of the discharge area is:

[0025] ;

[0026] The dimensionless parameter for explosive intensity is:

[0027]

[0028] Among them, K st,max It is the maximum explosive intensity of the same type of flammable material.

[0029] The dimensionless parameter for the pressure at which the explosion is released is:

[0030] ;

[0031] The characteristic parameters of combustible material concentration are:

[0032] ;

[0033] Among them, C ref This is the optimal explosive concentration.

[0034] The dimensionless parameter for dust particle size is:

[0035] ;

[0036] Parameters for determining congested flow and congestion margin parameters Determined based on the critical pressure ratio of the vent.

[0037] S3. Construction of congestion effect characteristics:

[0038] After the rupture disc opens, the flow state at the vent has a significant impact on the venting capacity. When the ratio of the internal pressure to the external pressure reaches a critical pressure ratio, the venting flow enters a choking state, and the venting mass flow rate is limited. The critical pressure ratio for choking flow at the vent is:

[0039] ;

[0040] When R≥R c When it is determined that the vent is blocked, the following is set:

[0041] ;

[0042] When R <R c If it is determined that there is no obstruction in the flow at the vent, then:

[0043] ;

[0044] Further construct continuous congestion margin parameters:

[0045] ;

[0046] By inputting simultaneously and This enables machine learning models to identify both whether blockage occurs during the venting process and the impact of the degree of blockage on the venting pressure.

[0047] S4. Training objective processing:

[0048] The maximum explosion pressure P red Dimensionlessization:

[0049] ;

[0050] To reduce the order-of-magnitude difference between high-pressure and low-pressure samples and improve the model's stability under small sample conditions, a logarithmic transformation is performed on the target value:

[0051] ;

[0052] Use y as the training label for the extreme gradient boosting tree model.

[0053] S5. Training of extreme gradient boosting tree models:

[0054] Input the characteristic parameter matrix of the explosion theory The training labels y are input into an extreme gradient boosting tree model for training. The model is represented as:

[0055] ;

[0056] Where K is the total number of regression trees, k Number the regression tree, f k For the k-th regression tree, Predict labels for the model.

[0057] The objective function for model training is:

[0058] ;

[0059] in, The loss function; This is a penalty term for the complexity of the regression tree.

[0060] To enhance the physical consistency of the model, physical monotonic constraints are applied to some input features:

[0061] , , , , , ;

[0062] Among them, the dimensionless parameter P of the sealed explosion pressure max,i The dimensionless parameter K of the explosion index st,i , 1. Dimensionless parameter P of the opening pressure stat,i and the congestion margin parameter η c,i Predicting labels for the model Applying a positive constraint, the dimensionless parameter A of the discharge area. v,i Predicting labels for the model Apply a negative constraint.

[0063] S6. Explosion relief pressure prediction:

[0064] For the design condition, input the maximum pressure of the sealed explosion, the concentration of combustibles, the dust particle size, the explosion relief opening pressure, the container geometric parameters, and the release area of ​​the candidate rupture discs, and construct the corresponding dimensionless feature matrix. .Will Input the trained extreme gradient boosting tree model to obtain the predicted labels:

[0065] ;

[0066] The predicted maximum discharge pressure is obtained through inverse transformation:

[0067] ;

[0068] in, The maximum explosion relief pressure is predicted for the explosion relief area corresponding to the candidate rupture disc venting area.

[0069] S7. Reverse design of rupture disc discharge area:

[0070] In the reverse design step of the rupture disc venting area, within the candidate venting area range [A] min A max The search function within the search area determines the effective discharge area of ​​the rupture disc that satisfies the following formula:

[0071] ;

[0072] in, Candidate discharge area A v The corresponding model predicts the maximum release pressure, where δ is the safety margin coefficient, and P al The allowable pressure for the container;

[0073] Final effective rupture fragment release area Represented as:

[0074] ;

[0075] Compared with the prior art, the present invention has the following beneficial effects:

[0076] Traditional explosion venting design methods typically rely on empirical or semi-empirical formulas. Their applicability is limited by factors such as experimental conditions, combustible material type, explosion intensity, static operating pressure, and container structure. A single calculation method is insufficient to comprehensively apply to explosion venting area design under different operating conditions. Furthermore, using machine learning methods solely for predicting venting pressure or area, if raw experimental parameters are directly used as input, problems arise such as inconsistent dimensions, unclear physical meaning, poor generalization ability under insufficient training samples, and prediction trends that do not conform to explosion laws. These issues fail to meet the safety, reliability, and interpretability requirements of rupture disc venting area design. This invention combines explosion venting theory with a machine learning model. Based on the explosion pressure formation mechanism and venting flow process, relevant parameters are dimensionless, constructing input features with clear physical meaning. This allows the model to not only learn complex nonlinear relationships but also reflect the comprehensive influence of factors such as venting area, explosion opening pressure, combustible material concentration, dust particle size, and congestion effects on dust explosion venting pressure, thus providing a more comprehensive description of the coupling relationship between pressure increase and venting pressure drop during dust explosion venting.

[0077] This invention employs an extreme gradient boosting tree model, which is adaptable to the characteristics of limited sample size, strong nonlinearity of variable relationships, and significant differences in sample sources in explosion venting test data. Compared to deep neural networks, it exhibits better training stability and engineering applicability under small sample conditions. Furthermore, this invention introduces physical monotonic constraints during model training, ensuring that the prediction results better conform to the basic laws of explosion venting when key variables such as venting opening pressure, venting area, and explosion intensity change, reducing the possibility of unreasonable prediction trends. In addition, this invention characterizes the flow state at the vent outlet through choking flow discrimination parameters and choking margin parameters, which helps improve the prediction accuracy and safety conservatism under high-pressure venting conditions, providing a technical solution for explosion venting pressure prediction and rupture disc venting area design that combines prediction accuracy, physical interpretability, and engineering reliability. Attached Figure Description

[0078] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0079] Figure 1 This is a schematic diagram of the explosion venting test device.

[0080] Figure 2The maximum sealed explosion pressure diagram is shown in the example.

[0081] Figure 3 The diagram shows the maximum pressure relief under different operating conditions in the example.

[0082] Figure 4 This is a graph showing the model's prediction results. Detailed Implementation

[0083] Certain exemplary embodiments are described below. As will be appreciated by those skilled in the art, the described embodiments can be modified in various ways without departing from the spirit or scope of the invention. Therefore, the drawings and description are to be considered exemplary in nature and not restrictive.

[0084] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0085] This invention constructs physically meaningful input features based on the fundamental theory of explosion venting, then uses an extreme gradient boosting tree model to learn the nonlinear mapping relationship between the maximum venting pressure and the venting area of ​​the rupture disc, and finally determines the effective venting area of ​​the rupture disc that meets safety requirements based on the container's allowable pressure.

[0086] An explosion venting test apparatus includes a standard 20L explosion container, a rupture disc venting device, a pressure sensor, a dust supply system, a gas supply system, an ignition control system, and a data acquisition system. The dust supply system regulates the powder injection pressure via an air cylinder, a pressure reducing valve, and a solenoid valve, controlling the dust's entry from the storage chamber into the explosion container. During the experiment, a certain mass of dust is placed in a 0.6L pressure vessel, and compressed air is introduced. After the solenoid valve opens, the compressed air carries the dust into the 20L spherical explosion container, causing the dust to form a uniform suspended dust cloud within the container. The ignition control system activates the ignition source after a preset delay time, preferably 60ms. The pressure sensor is located inside the explosion container and is used to collect real-time data on the pressure changes within the container over time. The pressure-time curve is recorded using the data acquisition system or a Yokogawa oscilloscope, thereby obtaining the maximum venting pressure of the dust explosion under different operating conditions.

[0087] The model usage method of this invention is as follows: First, the maximum explosion pressure and dust explosion index of a certain concentration of combustible material in a sealed environment are obtained through a sealed explosion test. Then, the maximum release pressure under different rupture disc release areas, different opening pressures, and different operating conditions is obtained through a venting test. Next, the original test parameters are converted into dimensionless explosion theory features. These features are then input into an extreme gradient boosting tree model based on explosion theory for training. After training, for the operating condition to be designed, under the given container allowable pressure and safety margin, the input parameters of the candidate rupture disc release areas are gradually changed, and the trained model is used to predict the maximum venting pressure corresponding to each candidate release area. Finally, the smallest release area that satisfies the safety constraints is selected as the effective release area of ​​the rupture disc. The specific steps are as follows:

[0088] S1. Dataset Construction;

[0089] Data from explosion-sealing tests and explosion-venting tests were collected to form a raw sample database. The data included at least the maximum venting explosion pressure P of the combustible material. red (bar), maximum explosion pressure in a sealed container P max (bar), Explosiveness Index of Combustible Materials (K) st (bar·m / s), discharge area A v (m²), static opening pressure of the rupture disc P stat (bar), dust concentration C (g / m³) 3 ), median particle size d of dust 50 (μm), container volume V (m³), container characteristic cross-sectional area A c (m²), congestion discrimination pressure P ref (bar);

[0090] S2. Analysis of control parameters for explosion theory:

[0091] The explosion venting process can be understood as a competition between combustion pressurization and venting depressurization. After ignition, the combustible material undergoes rapid combustion and oxidation, and the expansion of combustion products and high-temperature gases causes the internal pressure of the container to rise. When the internal pressure reaches the detonation pressure of the rupture disc, the rupture disc breaks and forms a vent, allowing unburned mixture, combustion products, and high-temperature gases to escape, thus reducing the internal pressure of the container. The original parameters are then transformed into a dimensionless feature matrix recognizable by a machine learning model.

[0092] ;

[0093] The parameters are defined as follows:

[0094] The dimensionless parameter for the sealed explosion pressure is:

[0095] ;

[0096] Where P0 is environmental pressure;

[0097] The dimensionless parameter of the discharge area is:

[0098] ;

[0099] The dimensionless parameter for the explosive intensity of combustible materials is:

[0100] ;

[0101] Among them, K st,max It is the maximum explosive intensity of a certain combustible material.

[0102] The dimensionless parameter for the pressure at which the explosion is released is:

[0103] ;

[0104] The characteristic parameters of combustible material concentration are:

[0105] ;

[0106] Among them, C ref This is the optimal explosive concentration.

[0107] The dimensionless parameter for dust particle size is:

[0108] ;

[0109] L c It is the diameter of the spherical dust particle used as a reference.

[0110] Parameters for determining congested flow and congestion margin parameters Determined based on the critical pressure ratio of the vent.

[0111] S3. Construction of congestion effect characteristics:

[0112] After the rupture disc is activated, the flow state at the vent has a significant impact on the venting capacity. When the ratio of the internal pressure to the external pressure reaches the critical pressure ratio, the venting flow enters a choking state. At this point, the venting mass flow rate reaches its upper limit, and even if the internal pressure continues to rise, the venting capacity will hardly increase proportionally, which may lead to an increase in the maximum venting pressure inside the container. The critical pressure ratio for choking flow at the vent is:

[0113] ;

[0114] in, The equivalent adiabatic index of the vented gas-powder mixture.

[0115] The reference pressure P is determined based on the congestion. refCalculate the pressure ratio inside and outside the vent:

[0116] ;

[0117] When R i ≥R c When it is determined that the vent is blocked, the following is set:

[0118] ;

[0119] When R i <R c If it is determined that there is no obstruction in the flow at the vent, then:

[0120] ;

[0121] Further construct continuous congestion margin parameters:

[0122] ;

[0123] By inputting simultaneously and This enables machine learning models to identify both whether blockage occurs during the venting process and the impact of the degree of blockage on the venting pressure.

[0124] S4. Training objective processing:

[0125] The maximum explosion pressure P red Dimensionlessization:

[0126] ;

[0127] To reduce the order-of-magnitude difference between high-pressure and low-pressure samples and improve the model's stability under small sample conditions, a logarithmic transformation is performed on the target value:

[0128] ;

[0129] Use y as the training label for the extreme gradient boosting tree model.

[0130] S5. Extreme gradient boosting tree model training:

[0131] Input the characteristic parameter matrix of the explosion theory The training labels y are input into an extreme gradient boosting tree model for training. The model is represented as:

[0132] ;

[0133] Where K is the total number of regression trees, k Number the regression tree, f k For the k-th regression tree, Predict labels for the model.

[0134] The objective function for model training is:

[0135] ;

[0136] in, The loss function; This is a penalty term for the complexity of the regression tree.

[0137] To enhance the physical consistency of the model, physical monotonic constraints are applied to some input features:

[0138] , , , , , ;

[0139] Among them, the dimensionless parameter P of the sealed explosion pressure max,i The dimensionless parameter K of the combustible material explosion index st,i , 1. Dimensionless parameter P of the opening pressure stat,i and the congestion margin parameter η c,i Predicting labels for the model Applying a positive constraint, the dimensionless parameter A of the discharge area. v,i Predicting labels for the model Apply a negative constraint.

[0140] S6. Explosion relief pressure prediction:

[0141] For the design condition, input the maximum pressure of the sealed explosion, the concentration of combustibles, the dust particle size, the explosion relief opening pressure, the container geometric parameters, and the release area of ​​the candidate rupture discs, and construct the corresponding dimensionless feature matrix. .Will Input the trained extreme gradient boosting tree model to obtain the predicted labels:

[0142] ;

[0143] The predicted maximum discharge pressure is obtained through inverse transformation:

[0144] ;

[0145] in, The maximum explosion relief pressure is predicted for the explosion relief area corresponding to the candidate rupture disc venting area.

[0146] S7. Reverse design of rupture disc discharge area:

[0147] In the reverse design step of the rupture disc venting area, within the candidate venting area range [A]min A max The search function within the search area determines the effective discharge area of ​​the rupture disc that satisfies the following formula:

[0148] ;

[0149] in, Candidate discharge area A v The corresponding model predicts the maximum release pressure. P is the safety margin factor. al The allowable pressure for the container;

[0150] Final effective rupture fragment release area Represented as:

[0151] ;

[0152] The trained model is deployed on a computer, controller, or engineering design software. When designing the rupture disc venting area, engineers only need to input parameters such as the container volume, characteristic cross-sectional area, allowable pressure, rupture disc opening pressure, combustible material concentration, dust particle size, ignition location, maximum sealed explosion pressure, and explosion index of the equipment to be designed. The method of this invention can then automatically construct a dimensionless feature matrix and search for the minimum rupture disc venting area that meets safety constraints within a preset candidate venting area range.

[0153] Example 1:

[0154] Using 30μm aluminum powder as the research object, through methods such as Figure 1 The dust explosion release test apparatus shown measured 400 g / m³. 3 600g / m 3 800g / m 3 1200g / m 3 Maximum sealed explosion pressure such as Figure 2 As shown, by changing parameters such as static operating pressure, venting area, and dust concentration, the maximum explosion relief pressure under different operating conditions was measured. Figure 3 As shown. Based on the above experimental data, the model is constructed in the following steps:

[0155] 1. Dataset construction;

[0156] Experimental data on aluminum powder explosion venting under different operating conditions were compiled to form an original sample database. The data includes the maximum venting explosion pressure P of the dust cloud. red (bar), maximum explosion pressure in a sealed container P max (bar), dust explosion index K st (bar·m / s), discharge area A v (m) 2), static opening pressure P of the rupture disc stat (bar), dust concentration C (g / m³) 3 ), median particle size d of dust 50 (μm), container volume V (m) 3 ), container characteristic cross-sectional area A c (m) 2 ), congestion discrimination pressure P ref (bar);

[0157] 2. Analysis of control parameters in explosion theory:

[0158] The dust explosion venting process can be understood as a competition between combustion pressurization and venting depressurization. After ignition, the dust cloud undergoes rapid combustion and oxidation, and the expansion of combustion products and high-temperature gases causes the internal pressure of the container to rise. When the internal pressure reaches the opening pressure of the rupture disc, the rupture disc breaks, forming a vent. Unburned mixture, combustion products, and high-temperature gases are discharged from the vent, reducing the internal pressure of the container. The original parameters are then transformed into a dimensionless feature matrix recognizable by a machine learning model.

[0159] ;

[0160] The parameters are defined as follows:

[0161] The dimensionless parameter for the sealed explosion pressure is:

[0162] ;

[0163] The dimensionless parameter of the discharge area is:

[0164] ;

[0165] The dimensionless parameter for dust explosion intensity is:

[0166] ;

[0167] Among them, K st,max It is the maximum explosive intensity of a certain dust at the same particle size.

[0168] The dimensionless parameter for the pressure at which the explosion is released is:

[0169] ;

[0170] The characteristic parameters of dust concentration are:

[0171] ;

[0172] Among them, C ref This is the optimal explosive concentration.

[0173] The dimensionless parameter for dust particle size is:

[0174] ;

[0175] Parameters for determining congested flow and congestion margin parameters Determined based on the critical pressure ratio of the vent.

[0176] 3. Construction of congestion effect characteristics:

[0177] After the rupture disc is activated, the flow state at the vent has a significant impact on the venting capacity. When the ratio of the internal pressure to the external pressure reaches the critical pressure ratio, the venting flow enters a choking state. At this point, the venting mass flow rate reaches its upper limit, and even if the internal pressure continues to rise, the venting capacity will hardly increase proportionally, which may lead to an increase in the maximum venting pressure inside the container. The critical pressure ratio for choking flow at the vent is:

[0178] ;

[0179] in, The equivalent adiabatic index of the vented gas-powder mixture.

[0180] The reference pressure P is determined based on the congestion. ref Calculate the pressure ratio inside and outside the vent:

[0181] ;

[0182] When R i ≥R c When it is determined that the vent is blocked, the following is set:

[0183] ;

[0184] When R i <R c If it is determined that there is no obstruction in the flow at the vent, then:

[0185] ;

[0186] Further construct continuous congestion margin parameters:

[0187] ;

[0188] By inputting simultaneously and This enables machine learning models to identify both whether blockage occurs during the venting process and the impact of the degree of blockage on the venting pressure.

[0189] 4. Training objective processing:

[0190] The maximum discharge pressure P of the dust explosion red Dimensionlessization:

[0191] ;

[0192] To reduce the order-of-magnitude difference between high-pressure and low-pressure samples and improve the model's stability under small sample conditions, a logarithmic transformation is performed on the target value:

[0193] ;

[0194] Use y as the training label for the extreme gradient boosting tree model.

[0195] 5. Training of extreme gradient boosting tree models:

[0196] Input the characteristic parameter matrix of the explosion theory The training labels y are input into an extreme gradient boosting tree model for training. The model is represented as:

[0197] ;

[0198] Where K is the total number of regression trees, k Number the regression tree, f k For the k-th regression tree, Predict labels for the model.

[0199] The objective function for model training is:

[0200] ;

[0201] in, The loss function; This is a penalty term for the complexity of the regression tree.

[0202] To enhance the physical consistency of the model, physical monotonic constraints are applied to some input features:

[0203] , , , , , ;

[0204] Among them, the dimensionless parameter P of the sealed explosion pressure max,i The dimensionless parameter K of the dust explosion index st,i , 1. Dimensionless parameter P of the opening pressure stat,i and the congestion margin parameter η c,i Predicting labels for the model Applying a positive constraint, the dimensionless parameter A of the discharge area. v,i Predicting labels for the model Apply a negative constraint.

[0205] 6. Dust explosion release pressure prediction:

[0206] For the design condition, input the maximum pressure of the dust-sealed explosion, dust concentration, dust particle size, explosion relief opening pressure, container geometric parameters, and candidate rupture disc venting area to construct the corresponding dimensionless feature matrix. .Will Input the trained extreme gradient boosting tree model to obtain the predicted labels:

[0207] ;

[0208] The predicted maximum discharge pressure is obtained through inverse transformation:

[0209] ;

[0210] in, The maximum predicted discharge pressure for dust explosion is the area corresponding to the discharge area of ​​the candidate rupture disc.

[0211] 7. Reverse design of rupture disc venting area:

[0212] In the reverse design step of the rupture disc venting area, within the candidate venting area range [A] min A max The search function within the search area determines the effective discharge area of ​​the rupture disc that satisfies the following formula:

[0213] ;

[0214] in, Candidate discharge area A v The corresponding model predicts the maximum release pressure. P is the safety margin factor. al The allowable pressure for the container;

[0215] Final effective rupture fragment release area Represented as:

[0216] ;

[0217] 8. Model accuracy verification:

[0218] 80% of the obtained experimental data was randomly selected as the training set for model parameter learning and nonlinear mapping construction; the remaining 20% ​​was used as the test set for independent verification of the model's prediction accuracy. The explosion theory characteristic parameters corresponding to each working condition in the test set were input into the trained model to obtain the predicted maximum deflation pressure, which was then compared with the experimentally measured maximum deflation pressure. The results are as follows: Figure 4As shown in the figure, the model's predicted values ​​are generally distributed around the experimental values, indicating good consistency between the two. The calculated mean relative error (MRE) of the model is 4.93%, and the coefficient of determination (R²) is... 2 The value of 0.92 indicates that the model has high prediction accuracy and good generalization ability. Furthermore, the explosion theory features constructed in this invention can effectively characterize the dust explosion and release process, and the trained model can be used for the design of rupture disc release areas.

[0219] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A design method for explosive discharge based on explosion theory and extreme gradient boosting algorithm, characterized in that: Based on the combustion-explosion reaction pressure rise mechanism, the release pressure drop mechanism, and the choke flow theory during the explosion release process, a machine learning model is constructed to predict the maximum explosion release pressure. The model is then used to determine the rupture disc release area that meets safety constraints based on the prediction results. Specifically, the following steps are included: S1. Dataset Construction; Collect data from explosion sealing tests and explosion relief tests to form a raw sample database; Data includes at least a certain concentration dust cloud maximum release explosion pressure P red , closed maximum explosion pressure P max , dust explosion index K st , release area A v , static opening pressure P of rupture disc stat , combustible concentration C, median particle size d of dust 50 , container volume V, characteristic cross-sectional area A of container c , choke discrimination pressure P ref ; S2. Analysis of control parameters for explosion theory: Transform the original parameters into a dimensionless feature matrix that can be recognized by the machine learning model: ； The parameters are defined as follows: The dimensionless parameter for the sealed explosion pressure is: ； Where P0 is environmental pressure; The dimensionless parameter of the discharge area is: ； The dimensionless parameter for the explosive intensity of combustible materials is: ； where K st, max is the maximum explosion intensity of the combustible; The dimensionless parameter for the pressure at which the explosion is released is: ； The characteristic parameters of combustible material concentration are: ； where C ref is the optimum explosion concentration; The dimensionless parameter for dust particle size is: ； L c It is the diameter of the spherical dust particle used as a reference. Parameters for determining congested flow and congestion margin parameters Determined based on the critical pressure ratio of the vent outlet; S3. Construction of congestion effect characteristics: When the ratio of the internal pressure to the external pressure of the container reaches the critical pressure ratio, the venting flow enters a choking state, and the venting mass flow rate is limited; the critical pressure ratio for choking flow at the vent outlet is: ； It is the equivalent adiabatic index of the combustion mixture; When R≥R c When it is determined that the vent is blocked, the following is set: ； When R <R c If it is determined that there is no obstruction in the flow at the vent, then: ； Constructing continuous congestion margin parameters: ； By inputting simultaneously and This enables machine learning models to identify whether blockage occurs during the release process and to identify the impact of the degree of blockage on the release pressure. S4. Training objective processing: The maximum release pressure P of the combustible material explosion red Dimensionlessization: ； Perform a logarithmic transformation on the target value to reduce the order-of-magnitude difference between high-pressure and low-pressure samples: ； Use y as the training label for the extreme gradient boosting tree model; S5. Extreme gradient boosting tree model training; Input the characteristic parameter matrix of the explosion theory The training label y is input into the extreme gradient boosting tree model for training; the model is represented as: ； Where K is the total number of regression trees, k Number the regression tree, f k For the k-th regression tree, Predict labels for the model; The objective function for model training is: ； in, The loss function; This is a penalty term for the complexity of the regression tree; Imposing physical monotonic constraints on some input features enhances the physical consistency of the model: ， ， ， ， ， ； Among them, the dimensionless parameter P of the sealed explosion pressure max,i The dimensionless parameter K of the combustible material explosion index st,i , 1. Dimensionless parameter P of the opening pressure stat,i and the congestion margin parameter η c,i Predicting labels for the model Applying a positive constraint, the dimensionless parameter A of the discharge area. v,i Predicting labels for the model Apply negative constraints; S6. Dust explosion release pressure prediction: For the design condition, input the maximum pressure of a sealed explosion of combustible material, combustible material concentration, dust particle size, explosion relief opening pressure, container geometric parameters, and candidate rupture disc venting area to construct the corresponding dimensionless feature matrix. ;Will Input the trained extreme gradient boosting tree model to obtain the predicted labels: ； The predicted maximum discharge pressure is obtained through inverse transformation: ； in, The maximum predicted discharge pressure for dust explosion corresponding to the discharge area of ​​the candidate rupture disc; S7. Reverse design of rupture disc discharge area: In the reverse design step of the rupture disc venting area, within the candidate venting area range [A] min A max The search function within the search area determines the effective discharge area of ​​the rupture disc that satisfies the following formula: ； in, Candidate discharge area A v The corresponding model predicts the maximum release pressure, where δ is the safety margin coefficient, and P al The allowable pressure for the container; Final effective rupture fragment release area Represented as: 。

Images