Dust explosion dynamic blocking method and system based on multi-modal data fusion

Abstract

The invention discloses a dust explosion dynamic blocking method based on multi-modal data fusion. The method comprises the following steps: S1, constructing five core accident trees; s2, collecting multi-source data of dust explosion in real time and preprocessing the multi-source data, and extracting a multi-dimensional feature vector containing temporal and spatial variation features; s3, converting the multi-dimensional feature vector into a real-time state vector of five core accident trees through an event activation rule, and completing multi-source data fusion; s4, according to the real-time state vector, calculating a top event probability of each accident tree by using a Monte Carlo simulation method, and then obtaining an accident tree dynamic coupling risk value through multi-tree coupling analysis; s5, according to the dynamic coupling risk value, using a spatial interpolation method and voxelization processing to obtain a normalized risk value, and generating a dynamic risk thermodynamic diagram; and S6, according to the risk value of each area in the dynamic risk thermodynamic diagram, determining the risk level of dust explosion, then carrying out multi-level interlocking control, and triggering a corresponding blocking strategy to realize dynamic blocking.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial safety monitoring, and in particular to a dust explosion dynamic blocking method based on multi-modal data fusion. BACKGROUND

[0002] As a major risk source in the field of industrial safety production, dust working environment (such as petrochemical storage tank area, coal powder preparation workshop) is a high-risk scene of industrial safety production. The suspended dust (PM10~PM100) in a certain concentration will cause a violent explosion when encountering an ignition source. Especially for airtight space with dust cloud concentration exceeding 30g / m³, once the five elements of explosion (combustible dust, oxidant, ignition source, diffusion condition, and confined space) are coupled, the accident consequences will be serious. Although the current national standard (such as GB 15577 Dust Explosion Prevention Safety Regulations) proposes basic requirements for dust concentration threshold, there are serious defects in the dynamic risk prevention and control level: the existing risk classification relies on static parameters such as storage tank volume or dust category, and does not quantify dynamic causes such as static electricity accumulation, temperature gradient mutation, and air flow disturbance; the traditional method relies on manual inspection and offline sampling, and cannot capture the millisecond-level coupling process of dust cloud formation, ignition source triggering, and protection failure in real time; the existing method only triggers global shutdown, and cannot implement precise blocking according to the explosion evolution path, resulting in more than 90% of the loss of false shutdown. The current dust explosion prevention and control technology mainly has the following bottlenecks:

[0003] Single monitoring dimension: excessive reliance on dust concentration or a single environmental parameter (such as temperature) for static threshold judgment, making it difficult to comprehensively capture the coupling risk dynamics of the five elements of explosion (combustible dust, oxidant, ignition source, diffusion, and confined space) such as dust cloud formation, ignition source occurrence (such as static spark, high-temperature surface), and confined space;

[0004] Lagging and static: the early warning model based on fixed threshold cannot adapt to complex and variable field conditions (such as material disturbance, environmental temperature and humidity fluctuation), is prone to false positives or false negatives, and usually only alerts when the risk is close to the critical state, leaving a very short time window for emergency disposal;

[0005] Lack of active intervention capability: existing systems focus on "warning", lack of "active prevention and control" mechanism that can automatically or assistively trigger precise blocking measures deeply bound to risk evolution logic.

[0006] Therefore, there is an urgent need for an intelligent active prevention and control system and method that can integrate multi-dimensional real-time sensing, have dynamic risk assessment capability, and implement precise blocking according to the risk evolution path, to solve the above problems and significantly improve the early identification and active blocking level of dust explosion risk. SUMMARY

[0007] The application aims to provide a dust explosion dynamic blocking method based on multi-modal data fusion, which realizes real-time coupling analysis of five elements of dust explosion through dynamic risk assessment of multi-modal data fusion, locates high-risk areas based on three-dimensional space mapping technology, performs multi-cascade lock control, triggers accurate blocking strategy, and forms a "perception-analysis-blocking" full-chain closed-loop prevention and control.

[0008] To solve the above technical problems, the technical scheme of the application is:

[0009] The first aspect provides a dust explosion dynamic blocking method based on multi-modal data fusion, comprising the following steps:

[0010] S1: constructing five core accident trees based on historical accident data;

[0011] S2: collecting multi-source data of dust explosion in real time and preprocessing, and extracting a multi-dimensional feature vector containing space-time change characteristics from the multi-source data;

[0012] S3: converting the multi-dimensional feature vector into a real-time state vector of the five core accident trees through a pre-defined event activation rule, and completing multi-source data fusion;

[0013] S4: according to the real-time state vector, first calculating the top event probability of each accident tree by using the Monte Carlo simulation method, and then obtaining the dynamic coupling risk value of the accident tree through multi-tree coupling analysis;

[0014] S5: according to the dynamic coupling risk value, using the spatial interpolation method and voxelization processing to obtain the normalized risk value, and generating a dynamic risk heat map;

[0015] S6: determining the risk level of dust explosion according to the risk value of each region in the dynamic risk heat map, and then performing multi-cascade lock control to trigger the corresponding blocking strategy, and realizing the dynamic blocking of dust explosion.

[0016] Preferably, in step 1, the five core accident trees include a combustible environment formation accident tree, an ignition source triggering accident tree, an oxygen concentration anomaly accident tree, a dust diffusion out-of-control accident tree, and a protection failure accident tree.

[0017] Preferably, in step 2, the multi-source data includes dust concentration data, temperature distribution data, electrostatic potential and discharge energy data, air flow speed and direction data, oxygen concentration data, and environmental air pressure data.

[0018] Preferably, step 2 comprises:

[0019] Step 2.1: performing dynamic noise reduction processing on the multi-source data collected in real time by Kalman filtering;

[0020] Step 2.2: millisecond-level spatio-temporal alignment of the denoised multi-source data based on a sliding window of timestamps, and extracting the multi-dimensional feature vector from the aligned data; the multi-dimensional feature vector includes a statistical feature vector, a spectral feature vector, a spatial relationship vector, and a spatio-temporal change feature.

[0021] Preferably, in step 4, the Monte Carlo simulation method is used to calculate the top event probability of each fault tree, including:

[0022] Step 4A.1: extracting the real-time occurrence probability of all basic events in the fault tree according to the real-time state vector, to obtain a basic event probability vector;

[0023] Step 4A.2: setting the number of random simulations to, and determining the state of each basic event by random sampling according to the real-time probability of each basic event in the basic event probability vector;

[0024] Step 4A.3: according to the state of each basic event, following the logic gate structure of the fault tree, calculating from bottom to top layer by layer, judging the state of the top event in the simulation calculation, and estimating the top event probability through the frequency of the occurrence of the top event in the simulation result;

[0025] The top event probability , and the estimation formula is as follows:

[0026]

[0027] , wherein, is an indicator function, taking the value of 1 when the top event occurs in the th simulation, and otherwise taking the value of 0, is the number of random simulations.

[0028] Preferably, in step 4, the dynamic coupling risk value of the fault tree is obtained by multi-tree coupling analysis, including:

[0029] Step 4B.1: determining a five-dimensional coupling matrix based on historical accident data;

[0030] Step 4B.2: obtaining a single-tree top event probability vector based on the top event probability of the fault tree calculated in step 4A.3, and then combining the predicted dust concentration to calculate a comprehensive risk value;

[0031] Step 4B.3: superimposing the spatio-temporal correlation factor and the conditional probability, and normalizing the comprehensive risk value to obtain the dynamic coupling risk value.

[0032] Preferably, the predicted dust concentration in step 4B.2 is the future 60-second dust concentration prediction value obtained based on a long short-term memory network prediction model, and the prediction process is as follows: based on the real-time collected dust concentration time series as the input of the prediction model, the time series modeling is performed through two layers of memory network, and the future 60-second dust concentration prediction value is output.

[0033] Preferably, step 5 comprises:

[0034] Step 5.1: constructing a global coordinate system according to the factory BIM model;

[0035] Step 5.2: generating a continuous risk field by spatial interpolation method by combining the dynamic coupling risk value and the spatial coordinates of the multi-source sensor;

[0036] Step 5.3: voxelizing the risk field to calculate the normalized risk value of each voxel unit in the risk field;

[0037] Step 5.4: generating a dynamic risk heat map by RGB- mixed rendering based on the normalized risk value.

[0038] Preferably, step 6 comprises:

[0039] (1) When the risk value approaches 0 and the transparency approaches 0, the region is displayed in green, representing a risk level of I level, and triggering global inactivation and device strategy;

[0040] (2) When the risk value and the transparency are both between 0 and 1, the region is proportionally mapped to a transition color between green and red, representing a risk level of II level, and starting local explosion suppression and ventilation enhancement strategy;

[0041] (3) When the risk value approaches 1 and the transparency approaches 1, the region is displayed in red, representing a risk level of III level, and executing the target device shutdown strategy.

[0042] The second aspect also provides a dust explosion dynamic blocking system based on multi-modal data fusion, which comprises a perception layer, an analysis layer and an execution layer;

[0043] The perception layer comprises a multi-source sensor array for real-time acquisition of multi-source data of dust explosion;

[0044] The analysis layer comprises an accident tree fusion evaluator for fusion and dynamic risk assessment of multi-source data to determine the risk level of dust explosion;

[0045] The execution layer comprises an intelligent blocking terminal for multi-level interlock control according to the risk level obtained by the analysis layer, triggering the corresponding blocking strategy to realize the active blocking of dust explosion.

[0046] Compared with the prior art, the present application has the following beneficial effects:

[0047] The present application breaks through the limitation of single parameter monitoring, captures the coupling risk dynamics of the five elements of explosion in all directions through the deep fusion of dust, temperature, static electricity, airflow and oxygen concentration data of multi-source sensing array, realizes the leap from static threshold judgment to dynamic probability evaluation based on the fault tree model and Monte Carlo simulation, the risk assessment update frequency reaches 200 milliseconds, greatly improving the real-time and accuracy of early warning, converts the risk value into a visual heat map using three-dimensional space mapping technology, intuitively locates the risk hotspots in the plant area, provides spatial decision support for precise blocking, automatically triggers the hierarchical interlocking blocking strategy according to the dynamic risk level, changes passive alarm to active intervention, realizes precise cutting at the early stage of the accident chain, and effectively avoids false shutdown; integrates the LSTM prediction model, and the predicted value of dust concentration in the next 60 seconds is included in the risk assessment, which has the risk trend prediction ability and strengthens the pre-control ability. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the description of the embodiments of the present application. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0049] Figure 1 The present application is a process flow diagram of a dust explosion dynamic blocking method based on multi-modal data fusion.

[0050] Figure 2 The present application is a running schematic diagram of a dust explosion dynamic blocking method based on multi-modal data fusion.

[0051] Figure 3 The present application is a deployment architecture schematic diagram of a dynamic blocking system.

[0052] Figure 4 The present application is a fault tree schematic diagram of a combustible environment formation.

[0053] Figure 5 The present application is a fault tree schematic diagram of an ignition source.

[0054] Figure 6 The present application is a fault tree schematic diagram of abnormal oxygen concentration.

[0055] Figure 7 The present application is a fault tree schematic diagram of dust diffusion out of control.

[0056] Figure 8 The present application is a fault tree schematic diagram of protection failure. DETAILED DESCRIPTION

[0057] The present invention will now be described in further detail with reference to the accompanying drawings. The terminal technical solutions of the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0058] like Figures 1-2 As shown, this invention provides a dynamic dust explosion prevention method based on multimodal data fusion, breaking through the limitations of static thresholds. Based on the dynamically coupled risk values, a three-level interlocking control mechanism is established to achieve precise prevention and control from fine-tuning of target equipment to emergency inertia. Coupled with time series prediction, it forecasts the risk trend for the next 60 seconds, transforming "passive alarm" into proactive prevention. This advances dust explosion prevention from the era of "threshold alarm" to the era of "precisely cutting off the accident chain," constructing a new paradigm of intrinsic safety that is calculable, predictable, and preventable for high-risk industrial scenarios.

[0059] A dynamic dust explosion prevention method based on multimodal data fusion includes the following steps:

[0060] S1: Five core accident trees are constructed based on historical accident data, including the flammable environment formation accident tree, the ignition source triggering accident tree, the oxygen concentration anomaly accident tree, the dust diffusion runaway accident tree, and the protection failure accident tree. Figures 4-8 As shown.

[0061] As shown in Figure 4, the fault tree for the formation of the combustible environment strictly follows the IEC 61025 standard. Its basic structure uses "formation of combustible dust environment" as the top event, connecting two key intermediate events via AND gates: "dust concentration ≥ lower explosive limit (LEL)" and "dust cloud persistence". The intermediate event "dust concentration ≥ LEL" is associated with three basic events via OR gates: excessive dust generation rate (FB001), ventilation system failure (FB002), and dust removal system failure (FB003). Similarly, the intermediate event "dust cloud persistence" is associated with three basic events via OR gates: disturbed deposited dust (FB004), stagnant airflow in enclosed space (FB005), and uncontrolled secondary dust (FB006). All basic events are activated through a dynamic threshold determination mechanism. For example, the activation condition for excessive dust generation rate (FB001) is a real-time dust concentration > 28 g / m³ and a dust particle size < 200 μm, with LEL = 0.05 × K × e. (-0.06RH) K is the dust reactivity coefficient, and RH is the relative humidity. The final result is a two-level logic tree structure containing one top event, two intermediate events, and six basic events. Each path in the tree represents a combination of failure modes formed by a flammable environment.

[0062] As Figure 5 shown, the ignition source trigger accident tree is drawn according to ISO 17776 risk analysis standard, and the basic structure takes the occurrence of an effective ignition source as the top event, and connects four intermediate events through a logic or gate: mechanical spark, electrostatic discharge, hot surface, and open flame; among them, the intermediate event "mechanical spark" is associated with two basic events of metal collision (IB001) and missing of spark tool (IB002) through a logic and gate; the intermediate event "electrostatic discharge" is associated with two basic events of static electricity accumulation energy > 300 mJ (IB003) and grounding resistance > 10 Ω (IB004) through a logic and gate; the intermediate event "hot surface" is connected by a logic or gate to two basic events of bearing temperature > 150°C (IB005) and motor overload (IB006); the intermediate event "open flame" is directly associated with two basic events of illegal hot work (IB007) and exposure of spontaneous combustion material (IB008). All basic events are dynamically determined through real-time monitoring data, for example: the activation condition of bearing temperature > 150°C (IB005) is that the infrared thermal imager detects a ≥ 150°C hot spot for 2 seconds, and the trigger threshold of static electricity accumulation energy > 300 mJ (IB003) is that the electrostatic sensor captures a single discharge energy E = 0.5CV²> 300 mJ. Finally, a three-layer tree structure containing 1 top event, 4 intermediate events, and 8 basic events is formed, each path in the tree represents a failure logic chain formed by a specific ignition source, and the necessary environmental prerequisite of "dust cloud exists" (Cond) is marked through a conditional event symbol.

[0063] As Figure 6As shown, the drawing of the oxygen concentration abnormality fault tree follows the ISO 14224 equipment reliability standard, and its basic structure takes "combustion-supporting environment abnormality" as the top event, connects two intermediate events "oxygen concentration > 18%" and "inerting system failure" through a logic OR gate; wherein, the intermediate event "oxygen concentration > 18%" is associated with two basic events of fresh air excess (OB001) and oxygen leakage (OB002) through a logic AND gate, and is marked with an environmental pressure correction formula (actual threshold value = 18% x P / 101.3 kPa, in the oxygen concentration abnormality fault tree, the real-time air pressure data P used to correct the oxygen concentration threshold value is collected by a digital air pressure sensor deployed in the monitoring area in real time, the sampling frequency is 10 Hz, and the data is transmitted to the analysis layer through the data bus of the multi-source sensing array for dynamic calculation of the actual risk threshold value of the oxygen concentration); the intermediate event "inerting system failure" is connected through a logic OR gate with three basic events: nitrogen pressure deficiency (OB003, trigger condition < 0.6 MPa), control system failure (OB004, safety circuit open circuit), and monitoring probe failure (OB005, signal drift > ± 0.5%vol), and is marked with a conditional event symbol as "inerting system has started" as a prerequisite. Finally, a tree structure containing 1 top event, 2 intermediate events and 5 basic events is formed, and all basic events are bound to real-time sensors, for example: oxygen leakage (OB002) detects oxygen concentration sudden increase gradient > 0.2% / s through an electrochemical oxygen concentration sensor, and ensures that each logic path in the tree matches the actual process state through dynamic calculation.

[0064] As shown in Figure 7 As shown, the drawing of the dust diffusion out-of-control fault tree strictly follows the IEC 60079 dust explosion-proof standard, and its basic structure takes "dust diffusion out-of-control" as the top event, connects two core intermediate events "equipment sealing failure" and "dust suppression system failure" through a logic OR gate; wherein, the intermediate event "equipment sealing failure" is associated with three basic events of flange sealing damage (DB001, differential pressure sensor detects > 500 Pa and lasts > 5 s), maintenance door not closed (DB002, position sensor continuous open state > 30 s) and conveying pipeline rupture (DB003, acoustic emission sensor captures > 100 dB anomaly) through a logic OR gate; the intermediate event "dust suppression system failure" is connected through a logic OR gate with three basic events of insufficient spray water pressure (DB004, pressure transmitter < 0.5 MPa), dust suppressant depletion (DB005, liquid level sensor < 10%) and airflow speed exceeding the standard (DB006, ultrasonic speed measurement > 2.5 m / s), and is marked with a transfer symbol as "ventilation system state" as an environmental coupling factor. Finally, a three-layer tree network containing 1 top event, 2 intermediate events and 6 basic events is formed, and all basic events are bound to a fluid mechanics model, for example: the trigger threshold value of airflow speed exceeding the standard (DB006) is corrected by Reynolds number: when Re > 2300, the flow rate threshold value m / s), wherein, is the dust density; 1.2 is the standard density of air, by the diffusion kinetic equation the physical plausibility of each logical path in the tree is verified, wherein, is the dust concentration, is the time, is the diffusion coefficient, is the air flow velocity vector, is the dust source term, which enables dynamic quantitative assessment of the dust cloud diffusion conditions.

[0065] As Figure 8 shown, the drawing of the protection failure fault tree strictly follows the IEC 61511 safety instrumented system standard, and the basic structure takes "protection system failure" as the top event, connects two core intermediate events "explosion suppression failure" and "isolation failure" through a logic or gate; wherein, the intermediate event "explosion suppression failure" is associated with two basic events of explosion suppression agent not released (PB001) and detector response delay (PB002) through a logic or gate, and the explosion suppression agent not released (PB001) is further decomposed into two bottom events of air path blockage (PB003, pressure sensor < 0.2 MPa) and control signal loss (PB004, 24VDC safety loop circuit break) through a logic and gate; the intermediate event "isolation failure" is connected in series through a logic or gate of explosion-proof valve jamming (PB005, position feedback timeout > 100 ms) and interlocking device failure (PB006, safety relay contact welding), and at the same time, "explosion detector has triggered" is marked as a conditional event symbol as an execution prerequisite. Finally, a three-level tree structure containing 1 top event, 2 intermediate events and 6 basic events is formed, and all basic events are bound to real-time monitoring mechanism, for example: the delay judgment of detector response delay (PB002) adopts time stamp comparison: more than 80 ms from detector alarm to execution signal output activates; explosion-proof valve jamming (PB005) verifies the jamming state through valve position sensor and cylinder pressure double variables. The core role of this protection failure fault tree is: through real-time state monitoring and logical diagnosis of the explosion suppression system, isolation device and its associated control circuit, the failure probability of the entire active protection system at the current time is dynamically quantitatively evaluated; it decomposes the complex instrument safety function (SIF) into basic events that can be independently monitored, realizes the dynamic reliability evaluation of the key link from "explosion detector has triggered" to "protection action successfully executed", provides the failure risk input of "the last line of defense" for comprehensive risk assessment, and ensures that when the explosion risk escalates, the trigger mechanism of the blocking strategy itself is highly reliable.

[0066] S2: Real-time acquisition of multi-source data of dust explosion and preprocessing, extraction of multi-dimensional feature vector. The multi-source data includes dust concentration data, temperature distribution data, static potential and discharge energy data, airflow velocity and direction data, oxygen concentration data, and environmental air pressure data. Including:

[0067] Step 2.1: Dynamic noise reduction processing of real-time collected multi-source data by Kalman filtering, filtering parameter process noise covariance Q = 0.1, measurement noise covariance R = 0.01.

[0068] Step 2.2: Millisecond-level spatio-temporal alignment of multi-source data after noise reduction based on time stamp sliding window, extraction of multi-dimensional feature vector.

[0069] The multi-dimensional feature vector includes statistical feature vector, spectral feature vector and spatial relationship vector, specifically: the statistical feature vector is extracted in the time domain dimension, including dust concentration gradient change rate, temperature field change slope, mean, variance and peak value; the spectral feature vector is extracted in the frequency domain dimension, including the main frequency component, energy distribution and frequency domain entropy obtained by fast Fourier transform (FFT) analysis of static electric signal; the spatial relationship vector is extracted in the spatial dimension, including the relative position of the sensor, the spatial gradient field of the thermal imaging temperature distribution matrix, the risk gradient distribution and the spatial clustering characteristics.

[0070] S3: Convert the statistical feature vector, spectral feature vector, and spatial relationship vector into real-time state vectors of the five core accident trees through pre-defined event activation rules, and complete multi-source data fusion.

[0071] S4: According to the real-time state vector, first calculate the top event probability of each accident tree using the Monte Carlo simulation method, and then obtain the dynamic coupling risk value of the accident tree through multi-tree coupling analysis.

[0072] First, the Monte Carlo simulation method is used to calculate the top event probability of each accident tree, with 5000 iterations each time and the calculation result updated every 200 milliseconds to ensure that the error is controlled within , providing real-time input parameters for multi-tree coupling analysis, including:

[0073] Step 4A.1: According to the real-time state vector, extract the real-time occurrence probability of all basic events in a single accident tree to obtain the basic event probability vector of a single accident tree ; Wherein, is the total number of basic events in the accident tree, is the real-time occurrence probability of the th basic event.

[0074] Step 4A.2: In each calculation period, set the random simulation times Based on the real-time probability of each basic event in the basic event probability vector, the state of each basic event is determined by random sampling. The basic event state includes 1 for occurrence and 0 for non-occurrence.

[0075] Step 4A.3: Based on the states of each basic event, following the logic gate structure of the fault tree, calculate layer by layer from bottom to top, determine the state of the top event in each simulation calculation, and then... The frequency of occurrence of the top event in the simulation is used to estimate the probability of the top event; the state of the top event also includes 1 for occurrence and 0 for non-occurrence.

[0076] The probability of the top event The estimation formula is as follows:

[0077]

[0078] in, For indicator functions, when the first The value is 1 when the top event occurs in this simulation, and 0 otherwise. The number of random simulations.

[0079] In this embodiment, the top event probability of the formation of a fault tree in a flammable environment is calculated in parallel according to the above steps. The probability of an ignition source triggering the top event of the fault tree. Oxygen concentration anomaly fault tree top event probability The probability of the top event in the accident tree of dust diffusion runaway and the probability of the top event in the fault tree of protection failure It updates every 200 milliseconds.

[0080] (1) Probability of the top event in the fault tree formation of a flammable environment

[0081] The top event "formation of a combustible dust environment" is connected by an AND gate to two intermediate events: dust concentration ≥ lower explosive limit LEL (M1) and dust cloud persistence (M2); the probability of the top event is calculated as follows:

[0082]

[0083]

[0084]

[0085] in, These represent the real-time occurrence probabilities of the intermediate events "dust concentration ≥ lower explosive limit (LEL)" and "dust cloud persists," respectively. This is the index number of the basic event, used to traverse all basic events belonging to the same intermediate event. The real-time occurrence probability of the 1th basic event, The real-time occurrence probability of the 1th basic event, The real-time occurrence probability of the 1th basic event,

[0086] The top event probability of ignition source triggered accident tree

[0087] The top event "effective ignition source" is connected by logical or gate with 4 intermediate events: mechanical spark (M3), electrostatic discharge (M4), hot surface (M5), open flame (M6); the top event probability is calculated as follows:

[0088]

[0089]

[0090]

[0091]

[0092]

[0093] The real-time occurrence probability of the 1th basic event, The real-time occurrence probability of the 1th basic event, The real-time occurrence probability of the 1th basic event,

[0094] The top event probability of oxygen concentration abnormality accident tree

[0095] The top event "combustion-supporting environment abnormality" is connected by logical or gate with 2 intermediate events: "oxygen concentration > 18%" (M7), inerting system failure (M8); the top event probability is calculated as follows:

[0096]

[0097]

[0098] The real-time occurrence probability of the 1th basic event, The real-time occurrence probability of the 1th basic event, The index number of the basic event, used to traverse all basic events under the same intermediate event; The real-time occurrence probability of the 1th basic event, The real-time occurrence probability of the 1th basic event, The real-time occurrence probability of the 1th basic event,

[0099] (4) Probability of the top event in the accident tree of dust diffusion runaway

[0100] The top event "Dust diffusion runaway" is connected to two intermediate events by a logic OR gate: equipment seal failure (M9) and dust suppression system failure (M10); the probability of the top event is calculated as follows:

[0101]

[0102]

[0103]

[0104] in, These represent the real-time occurrence probabilities of the intermediate events "equipment seal failure" and "dust suppression system failure," respectively. This is the index number of the basic event, used to traverse all basic events belonging to the same intermediate event; For the first The real-time probability of occurrence of a basic event; These represent the real-time occurrence probability of the corresponding basic events in the dust diffusion runaway accident tree.

[0105] (5) Probability of the top event in the fault tree of protection failure

[0106] The top event "Protection System Failure" is connected to two core intermediate events via a logic OR gate: Explosion Suppression Failure (M11) and Isolation Failure (M12); the probability of the top event is calculated as follows:

[0107]

[0108]

[0109]

[0110]

[0111] in, These represent the real-time occurrence probabilities of the intermediate events "explosion suppression failure" and "isolation failure," respectively. These represent the real-time occurrence probability of the corresponding basic events in the protection failure incident tree.

[0112] Then, the dynamic coupling risk value of the fault tree is obtained through multi-tree coupling analysis, which is implemented by a dynamic weight matrix algorithm, including:

[0113] Step 4B.1: Determine the five-dimensional coupling matrix based on historical accident data. The expression is as follows:

[0114]

[0115] wherein, is the coupling matrix element, representing the coupling enhancement weight of the top event of the i-th accident tree and the top event of the j-th accident tree on the comprehensive risk, is the coupling enhancement weight of the top event of the i-th accident tree and the top event of the j-th accident tree on the comprehensive risk, F, I, O, D, P represent the five types of accident trees of combustible environment formation, ignition source triggering, oxygen concentration anomaly, dust diffusion out of control and protection failure, respectively, and the weight value of each accident tree is determined by statistical regression analysis of historical explosion cases;

[0116] Step 4B.2: Based on the top event probability of each accident tree calculated in step 4A.3, the single-tree top event probability vector P is obtained , and the predicted dust concentration is combined to calculate the comprehensive risk value R, and the calculation formula is as follows;

[0117]

[0118] wherein, is the coupling matrix element, representing the coupling enhancement weight of the top event of the i-th accident tree and the top event of the j-th accident tree on the comprehensive risk, is the coupling enhancement weight of the top event of the i-th accident tree and the top event of the j-th accident tree on the comprehensive risk, and are the i-th and j-th elements in the single-tree top event probability vector P, is the concentration time-varying factor and , wherein, is the time interval between the current time and the last risk assessment time point, is a natural constant. The predicted dust concentration is the future 60-second dust concentration prediction value obtained based on a long short-term memory network (LSTM) prediction model, and the prediction process is as follows: based on the real-time collected dust concentration time series (sampling frequency 10 Hz) as the input of the prediction model, the time series modeling is performed through two layers of LSTM network (the number of hidden layer units is 32), and the future 60-second dust concentration prediction value

[0119] is output. Step 4B.3: Superimpose the space-time correlation factor and the conditional probability

[0120] , normalize the comprehensive risk value R to obtain the dynamic coupling risk value R.

[0121] ​​​​​​​​

[0122]

[0123]

[0124] in, Let be the element of the coupling matrix, representing the . Fault tree and the first The coupling weight of the combined risk is enhanced when similar tree-top events occur simultaneously. and The probability vector of the top event of a single tree The first in The and the first One element, The concentration time-varying factor and ,in, The time interval between the current moment and the last risk assessment time point. It is a natural constant. The first Fault tree and the first Fault tree type, representing the fault tree type index. In the fault tree and The conditional probability of an explosion occurring given that both top events occur simultaneously. The first Class and the The main risk contribution moment of the probability of a treetop event similar to an accident. For the first Class and the The spatial association function for fault trees is calculated by spatial distance.

[0125] In this embodiment, the dynamic coupling risk value is output at a refresh rate of 100Hz, and the calculation error is controlled within a certain range. Within the range, and the error is within the 95% confidence interval.

[0126] Multi-source data collected by the multi-source sensor array, such as temperature, static electricity, airflow, and oxygen concentration, are not directly used for the final risk value calculation. Instead, they serve as front-end inputs to construct the real-time state vectors of the five core fault trees. Monte Carlo simulation is then used to calculate the probability of the top event in each fault tree. These top event probabilities constitute the probability vector of the top event in each tree. The vector Essentially, it is the quantized output of multi-source data after the first layer of fusion. Subsequently, in multi-tree coupling analysis, this vector... With predicted dust concentration Together, they serve as the core input for calculating the comprehensive risk value. Therefore, all the information of the multi-source data is embodied in the final risk assessment result indirectly and sufficiently.

[0127] S5: According to the dynamic coupling risk value, the normalized risk value is obtained by using the spatial interpolation method and voxelization processing, and a dynamic risk heat map is generated, including:

[0128] Step 5.1: According to the factory BIM model, a global coordinate system is constructed.

[0129] Step 5.2: The dynamic coupling risk value and the multi-source sensor spatial coordinates are generated into a continuous risk field by using the spatial interpolation method; the spatial interpolation method is Kriging, and the semi-variogram function is: wherein, is the lag distance, that is, the Euclidean distance between any two sensor sampling points in space. The semi-variogram function model indicates that the spatial correlation range of the system is about 8.2 meters, the nugget value is 0.02, and the sill value is 0.15.

[0130] Step 5.3: The risk field is voxelized, and the normalized risk value of each voxel unit in the risk field is calculated, and the calculation formula is as follows:

[0131]

[0132] wherein, is the weight of the i-th sensor, is the distance from the voxel to the i-th sensor, is a smoothing coefficient and , is the Euclidean distance between the center point coordinate of the current voxel unit and the spatial coordinate of the i-th sensor , is the normalized risk value at the center point coordinate of the current voxel unit after spatial interpolation and Gaussian smoothing, and the value range is mapped to [0, 1], is the total number of sensors participating in the fusion calculation. Step 5.4: Based on the normalized risk value , a dynamic risk heat map is generated by RGB- mixed rendering.

[0133] Step 5.4: Based on the normalized risk value , a dynamic risk heat map is generated by RGB- mixed rendering.

[0134] S6: According to the risk value of each region in the dynamic risk heat map, the risk level of the dust explosion is determined, and the multi-level interlock control is performed to trigger the corresponding blocking strategy, so as to realize the dynamic blocking of the dust explosion. ​​

[0135] (1) When the risk value approaches 0 and the transparency approaches 0, i.e. and , the region is displayed as green (this color is very transparent), representing that the risk level is level I, very safe, and then triggering the global inerting and device strategy, achieving ≤200 ms response through the 8 bar high-pressure nitrogen injection system.

[0136] (2) When the risk value and the transparency are both between 0 and 1, the region is proportionally mapped to a transition color between green and red (this color is between transparent and opaque), for example, yellow, orange, representing that the risk level is level II, moderate danger, and then starting the local explosion suppression and ventilation enhancement strategy, deploying the ultra-fine water mist system to achieve ≤150 ms activation.

[0137] (3) When the risk value approaches 1 and the transparency approaches 1, i.e. and , the region is displayed as red (this color is opaque), representing that the risk level is level III, extremely dangerous, and then immediately executing the target device shutdown strategy, using a SIL3 certified safety relay to achieve ≤50 ms action cut-off.

[0138] As shown in Figure 3 , the application also provides a dust explosion dynamic blocking system based on multi-modal data fusion, which comprises a perception layer, an analysis layer and an execution layer.

[0139] The perception layer comprises a multi-source sensing array for real-time collection of multi-source data of dust explosion; the multi-source sensing array comprises a laser dust sensor, an infrared thermal imager, an electrostatic detection array, an ultrasonic airflow sensor, an electrochemical oxygen concentration sensor and a digital air pressure sensor.

[0140] The analysis layer comprises an accident tree fusion evaluator for fusion and dynamic risk assessment of the multi-source data to determine the risk level of dust explosion.

[0141] The execution layer comprises an intelligent blocking terminal for multi-level interlock control according to the risk level obtained by the analysis layer, triggering the corresponding blocking strategy to achieve active blocking of dust explosion.

[0142] The parts not involved in the application are the same as or realized by using the prior art.

[0143] The above content is a further detailed description of the application in combination with specific embodiments, and the specific implementation of the application should not be limited to these descriptions. For ordinary skilled persons in the technical field of the application, without departing from the concept of the application, a number of simple deductions or substitutions can be made, which should be considered as falling within the protection scope of the application.

Claims

1. A dust explosion dynamic blocking method based on multi-modal data fusion, characterized in that: The method comprises the following steps: S1: constructing five core accident trees based on historical accident data; S2: collecting multi-source data of dust explosion in real time and preprocessing, and extracting a multi-dimensional feature vector containing space-time change characteristics from the multi-source data; S3: converting the multi-dimensional feature vector into a real-time state vector of the five core accident trees through a predefined event activation rule, and completing multi-source data fusion; S4: according to the real-time state vector, first calculating the top event probability of each accident tree by using the Monte Carlo simulation method, and then obtaining the dynamic coupling risk value of the accident tree through multi-tree coupling analysis; S5: according to the dynamic coupling risk value, using spatial interpolation method and voxel processing to obtain normalized risk value, and generating dynamic risk heat map; S6: determining the risk level of dust explosion according to the risk value of each region in the dynamic risk heat map, and then performing multi-level interlock control to trigger the corresponding blocking strategy, thereby realizing the dynamic blocking of dust explosion.

2. The dust explosion dynamic blocking method based on multi-modal data fusion according to claim 1, characterized in that: In step 1, the five core accident trees include combustible environment formation accident tree, ignition source triggering accident tree, oxygen concentration abnormality accident tree, dust diffusion out-of-control accident tree and protection failure accident tree.

3. The dust explosion dynamic blocking method based on multi-modal data fusion according to claim 1, characterized in that: In step 2, the multi-source data includes dust concentration data, temperature distribution data, electrostatic potential and discharge energy data, airflow velocity and direction data, oxygen concentration data, and environmental air pressure data.

4. The dust explosion dynamic blocking method based on multi-modal data fusion according to claim 1, characterized in that: Step 2 comprises: Step 2.1: dynamically denoising the multi-source data collected in real time by Kalman filtering; Step 2.2: performing millisecond-level space-time alignment on the denoised multi-source data based on a time-stamped sliding window, and extracting the multi-dimensional feature vector from the aligned data; the multi-dimensional feature vector includes a statistical feature vector, a frequency spectrum feature vector, a spatial relationship vector, and a space-time change characteristic.

5. The dust explosion dynamic blocking method based on multi-modal data fusion according to claim 1, characterized in that: In step 4, the Monte Carlo simulation method is used to calculate the top event probability of each accident tree, which comprises: Step 4A.1: according to the real-time state vector, extracting the real-time occurrence probability of all basic events in the accident tree to obtain a basic event probability vector; Step 4A.2: setting the random simulation number as, determining the state of each basic event by random sampling according to the real-time probability of each basic event in the basic event probability vector; Step 4A.3: according to the state of each basic event, following the logic gate structure of the accident tree, calculating layer by layer from bottom to top, judging the state of the top event in the simulation calculation, and estimating the top event probability through the frequency of the top event occurrence in the simulation result; The top event probability The estimation formula is as follows: wherein, is an indicator function that takes the value 1 when the top event occurs in the jth simulation, and 0 otherwise, is an indicator function that takes the value 1 when the top event occurs in the jth simulation, and 0 otherwise, is the number of stochastic simulations.

6. The dust explosion dynamic blocking method based on multi-modal data fusion according to claim 1, characterized in that: In step 4, the dynamic coupling risk value of the accident tree is obtained through multi-tree coupling analysis, which comprises: Step 4B.1: determining a five-dimensional coupling matrix based on historical accident data; Step 4B.2: obtaining a single-tree top event probability vector based on the top event probability of the accident tree calculated in step 4A.3, and then combining the predicted dust concentration to calculate a comprehensive risk value; Step 4B.3: superimposing the space-time correlation factor and the conditional probability, and normalizing the comprehensive risk value to obtain the dynamic coupling risk value.

7. The dust explosion dynamic blocking method based on multi-modal data fusion according to claim 6, characterized in that: The predicted dust concentration in step 4B.2 is a 60-second future dust concentration prediction value obtained based on a long short-term memory network prediction model. The prediction process is as follows: based on the real-time collected dust concentration time series as the input of the prediction model, the two-layer memory network is used for time series modeling, and the 60-second future dust concentration prediction value is output.

8. The dust explosion dynamic blocking method based on multi-modal data fusion according to claim 1, characterized in that: Step 5 includes: Step 5.1: Construct a global coordinate system according to the factory BIM model; Step 5.2: Generate a continuous risk field by spatial interpolation of the dynamic coupling risk value and the multi-source sensor spatial coordinates; Step 5.3: Perform voxelization processing on the risk field to calculate the normalized risk value of each voxel unit in the risk field; Step 5.4: Based on the normalized risk values, a dynamic risk heat map is generated by RGB- hybrid rendering.

9. The dust explosion dynamic blocking method based on multi-modal data fusion according to claim 1, characterized in that: Step 6 includes: (1) When the risk value approaches 0 and the transparency approaches 0, the area is displayed in green, representing a risk level of I level, and triggering global inerting and equipment strategy; (2) When the risk value and transparency are both between 0 and 1, the area is proportionally mapped to a transition color between green and red, representing a risk level of II level, and starting local explosion suppression and ventilation enhancement strategy; (3) When the risk value approaches 1 and the transparency approaches 1, the area is displayed in red, representing a risk level of III level, and executing the target equipment shutdown strategy.

10. A dust explosion dynamic blocking system based on multi-modal data fusion, characterized in that: The system includes a perception layer, an analysis layer, and an execution layer; The perception layer includes a multi-source sensor array for real-time acquisition of multi-source data of dust explosion; The analysis layer includes an accident tree fusion evaluator for fusion and dynamic risk assessment of multi-source data to determine the risk level of dust explosion; The execution layer includes an intelligent blocking terminal for multi-level interlock control according to the risk level obtained by the analysis layer to trigger the corresponding blocking strategy and realize the active blocking of dust explosion.

Images