Dust explosion-proof early warning method and system

Abstract

The invention provides a dust explosion-proof early warning method and system, and relates to the technical field of safety monitoring and early warning, and the method comprises the steps: collecting and standardizing the multi-source data of a dust movement track, an airflow field and electrostatic potential distribution, calculating a vortex intensity index and a dust gathering trend coefficient, and constructing a multilayer neural network electrostatic coupling effect model, and nonlinear coupling mapping of dust concentration, electrostatic field intensity and vortex intensity is realized. And when the danger increment output by the model and the growth rate of the dust gathering trend coefficient exceed a preset threshold value, triggering dust explosion-proof early warning. Therefore, the risk change under the action of the electrostatic field can be detected, the recognition capability of abnormal working conditions and a coupling mechanism is remarkably improved, and the safety control level is greatly improved through early-stage accurate early warning and intelligent disposal of dust explosion hidden dangers.

Description

Technical Field

[0001] The present invention relates to the technical field of safety monitoring and early warning, and more specifically, to a dust explosion prevention early warning method and system. Background Art

[0002] With the increasing automation and intelligence of industrial production, dust explosion prevention and control has become a key research topic in the field of industrial safety. Traditionally, dust explosion early warning relies primarily on fixed-point detection of ambient dust concentration and alarms for exceeding limits, or static risk assessments based on the three explosion elements: combustible dust, oxidants, and ignition sources. In recent years, with the advancement of sensor technology, image processing, and intelligent algorithms, dynamic monitoring and risk assessment methods for dust explosions have gradually emerged. For example, some studies have used laser particle counters, industrial cameras, airflow sensors, and electrostatic field sensors to conduct real-time monitoring of dust concentration, flow field distribution, and electrostatic parameters in workshop environments, providing data support for early warning and prevention of dust explosion risks. Furthermore, the introduction of artificial intelligence algorithms and deep learning models has enabled the fusion of multi-source data and risk prediction under complex working conditions, driving the development of dust explosion early warning technology towards intelligent and systematic approaches.

[0003] However, existing dust explosion warning technologies generally suffer from the following deficiencies: First, existing methods often use a single parameter, such as dust concentration, as a risk criterion. These methods lack dynamic monitoring and analysis of multi-physics interactions, such as airflow disturbances, vortex structures, and electrostatic coupling, making it difficult to fully reflect the complexity of explosion triggers. Second, the localized enrichment and aggregation of dust under strong airflow disturbances is difficult to accurately capture with traditional concentration sensors, which can easily lead to missed detections in high-risk areas. Third, the nonlinear relationship between the spatial distribution of electrostatic field strength and parameters such as dust concentration and airflow disturbances has not been fully modeled, making it difficult to promptly identify explosion precursors caused by the coupling of multiple factors. Furthermore, existing explosion warning systems mostly employ rule-based thresholding methods, lacking intelligent identification and dynamic response to evolving hazard trends. Consequently, the timeliness and accuracy of warnings remain to be improved. Therefore, developing a dust explosion warning method based on the collaborative fusion of multi-source data, coupled physical mechanism modeling, and intelligent trend prediction has become a pressing challenge for current technology. Summary of the Invention

[0004] To address the aforementioned technical issues, the present invention provides a dust explosion prevention early warning method and system that, to a certain extent, addresses the problems of imperfect modeling of multi-physics field interactions, which makes it difficult to fully reflect the complex coupling relationships between parameters such as dust concentration, airflow disturbance, and electrostatic field strength, as well as the limited dynamic perception of dust accumulation, which makes it difficult to promptly detect the formation of localized high-risk dust clusters.

[0005] According to one aspect of the present invention, a dust explosion prevention early warning method is provided, which includes:

[0006] Collect dust movement trajectory data, airflow field data and electrostatic potential distribution data, and perform standardization processing;

[0007] Calculating a vortex intensity index based on the airflow field data, and obtaining a dust aggregation tendency coefficient in combination with the dust movement trajectory data;

[0008] An electrostatic coupling effect model is constructed based on the electrostatic potential distribution data and the dust aggregation tendency coefficient, and a nonlinear mapping relationship between dust concentration, electrostatic field strength and vortex strength is established using a multi-layer neural network;

[0009] When the hazard increment and the growth rate of the dust aggregation tendency coefficient output by the electrostatic coupling effect model exceed a preset threshold, a dust explosion prevention warning signal is triggered.

[0010] Furthermore, the dust movement trajectory data is acquired through a high-speed camera system, the airflow field data is acquired through an airflow field sensor array, and the electrostatic potential distribution data is acquired through an electrostatic field strength sensor.

[0011] Furthermore, three-dimensional velocity vector decomposition is performed based on the airflow field data, and the velocity of each measuring point is decomposed into radial and tangential components in the cylindrical coordinate system. The velocity gradients in each direction are calculated using coordinate transformation and central difference method to obtain the three components of the vorticity vector, which reflect the local rotation characteristics and spatial vortex structure of the airflow field, and the vorticity scalar field characterizing the rotation intensity is obtained through the vorticity vector modulus.

[0012] Furthermore, tensorization processing is performed based on the three components of the vorticity vector, a velocity gradient tensor is constructed in a cylindrical coordinate system, and its antisymmetric part is extracted to obtain the vorticity tensor. Combined with the calculated strain rate tensor, the Q criterion is used to quantitatively characterize the vortex intensity index of each region in the airflow field to identify the vortex core area.

[0013] Furthermore, based on the vortex intensity index and the dust motion trajectory data, the collected dust motion trajectory data is divided into time windows and image processed, the center of mass position of the dust group in each window is extracted, and combined with the center of mass changes between adjacent windows, the movement direction and speed of the dust are calculated, the angle between the movement direction of the dust group and the vortex center direction is judged, and the dust aggregation trend coefficient is calculated.

[0014] Furthermore, based on the dust aggregation tendency coefficient, combined with multi-source data of dust concentration, electric field intensity and airflow field, a structured feature matrix is ​​constructed. Using this as input, a multi-layer neural network is used to establish an electrostatic coupling effect model to monitor the interaction between dust, airflow and static electricity in the space.

[0015] Furthermore, the multi-layer neural network adopts an hourglass-shaped structure of "narrow layer-wide layer-narrow layer". The input layer contains normalized dust concentration, electric field strength and vortex characteristics. Missing data is supplemented by interpolation or substitution. The first hidden layer performs feature compression and extracts physical quantity coupling characteristics through principal component analysis and weighted merging. Low-contribution features will be eliminated. The number of nodes in the middle hidden layer is increased for reorganization and coupling feature mining, and the coupling term weights are dynamically adjusted. ReLU activation is used, and the feature extraction layer further refines the dominant coupling mechanism. Tanh activation is used to ensure physical rationality. The output layer corresponds to each key physical quantity, and normalization and different activation functions are used to ensure the physical meaning and effective range of the output.

[0016] Furthermore, training is performed based on the electrostatic coupling effect model, and training, validation, and test sets are divided. For typical working conditions with insufficient samples, data is supplemented through numerical simulation of physical equations, and physical constraints are introduced in training to prevent the predicted values ​​from exceeding the bounds. After each round of training, the prediction accuracy of the electrostatic field and dust movement is evaluated separately. The training for this stage is completed when the accuracy meets the standards and converges. The prediction ability and effectiveness of the model are verified through test sets and actual engineering data.

[0017] Furthermore, based on the output results of the electrostatic coupling effect model, the hazard increment is obtained by calculating the product of the normalized electrostatic field intensity and the spatial gradient of the dust concentration distribution. The dust aggregation trend coefficient under the action of the electrostatic field is determined by the dust movement speed and concentration change rate. Thresholds are set for the hazard increment and the aggregation trend coefficient respectively to achieve graded warning:

[0018] When a strong electrostatic field is generated during equipment operation and key indicators rapidly exceed thresholds, a Level 1 warning is triggered and the equipment is immediately slowed down, dusted, or shut down.

[0019] When chain-like danger spreads across multiple areas, triggering a system-level warning, all lines must be decelerated and zoned for processing;

[0020] After all emergency responses, a comprehensive assessment is conducted on operating parameters, anti-static measures and dust treatment efficiency, and the warning thresholds and response plans are optimized accordingly to ensure safe operation.

[0021] According to another aspect of the present invention, there is provided a dust explosion prevention warning system, which includes: a data acquisition unit, a calculation processing unit, and a warning execution unit;

[0022] The data acquisition unit is used to collect dust movement trajectory data, airflow field data and electrostatic potential distribution data, and perform standardization processing;

[0023] The computing processing unit calculates the vortex intensity index using the airflow field data and obtains the dust aggregation tendency coefficient in combination with the dust motion trajectory data; then, based on the electrostatic potential distribution data and the dust aggregation tendency coefficient, a multi-layer neural network is used to construct an electrostatic coupling effect model to achieve a nonlinear mapping relationship between dust concentration, electrostatic field strength, and vortex intensity;

[0024] The warning execution unit triggers a dust explosion prevention warning signal when the hazard increment and the growth rate of the dust aggregation tendency coefficient output by the electrostatic coupling effect model exceed a preset threshold.

[0025] Compared with existing technologies, this invention achieves dynamic identification and quantitative analysis of the coupling mechanism between electrostatic fields, airflow, and dust movement by integrating high-speed video, airflow field measurement, multi-source data processing, and multi-layer neural network modeling. It can not only accurately capture the aggregation behavior of dust under complex working conditions, but also reflect the dangerous changes under the action of the electrostatic field in real time, significantly improving the ability to identify abnormal working conditions and coupling mechanisms. At the same time, based on the hierarchical warning and emergency response mechanism, it can achieve early and accurate warning and intelligent disposal of dust explosion hazards, greatly improving the safety management and control level of the system. It not only improves the real-time monitoring, warning, and decision-making capabilities of the electrostatic-dust system, but also provides a scalable intelligent method for multi-physical field coupling analysis in complex industrial environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0027] Figure 1 Flowchart of a dust explosion prevention early warning method according to an embodiment of the present invention;

[0028] Figure 2 1 is a warning flow chart of the electrostatic coupling effect model in the embodiment of the present invention. DETAILED DESCRIPTION

[0029] Below, the exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments of the present invention, and it should be understood that the present invention is not limited to the exemplary embodiments described herein.

[0030] As mentioned in the above background technology, there are two major outstanding problems in the existing technology: First, the monitoring system generally relies on only a single physical quantity for risk assessment, which makes it difficult to reveal the complex coupling effects and dynamic evolution processes among dust, airflow and electrostatic fields, resulting in limited ability to identify dust accumulation and explosion hazards; Second, the existing data modeling methods mostly use linear or static analysis, which cannot effectively handle the nonlinear correlation and spatiotemporal coupling characteristics between multi-source data, especially under special or extreme working conditions. The model generalization ability is insufficient and the warning accuracy is low.

[0031] Figure 1 FIG. 1 is a flow chart of a dust explosion prevention early warning method according to an embodiment of the present invention. Figure 1 As shown, the dust explosion prevention early warning method includes:

[0032] S1: Collect dust movement trajectory data, airflow field data and electrostatic potential distribution data, and perform standardization processing.

[0033] Multiple high-speed camera systems are installed in key work areas within the workshop. These systems utilize industrial-grade high-speed camera systems, including a multi-channel, directionally mounted high-speed camera array. These cameras are mounted using fixed brackets in key monitoring areas on the workshop's roof and sidewalls. Each high-speed camera captures data at a rate of at least 1000 frames per second and is equipped with an independent image processing module. This module uses real-time image sequence analysis to extract characteristic parameters of dust cluster motion, including their spatial position coordinates, velocity vector, and rate of acceleration. This image processing module utilizes a deep learning target recognition algorithm to accurately track and locate dust clusters of varying sizes, effectively eliminating interference from ambient light fluctuations.

[0034] Furthermore, the image processing module preprocesses the image sequences captured by the high-speed camera, including contrast enhancement and noise removal. The image is divided into M×N subregions, and the mean grayscale value and standard deviation of each subregion are calculated. Based on the grayscale difference between adjacent subregions, the gain coefficient of the grayscale mapping function is determined to achieve image brightness balance. Image noise is removed using a 3×3 pixel window median filter and a three-layer db4 wavelet decomposition denoising method.

[0035] When the preprocessed image is input to the processing unit, the characteristic grayscale value of the dust cluster is first determined as the seed point. The growth criterion is set as follows: the grayscale value difference between adjacent pixels is less than the set regional growth threshold T1, and the standard deviation of pixels within the region is less than the regional growth threshold T2. Adjacent pixels that meet the growth criterion are iteratively updated and grouped into the same dust cluster region. If the growth area of ​​a region exceeds the preset threshold, the region is marked as a potential high-concentration dust cluster.

[0036] When the dust cluster position information is obtained in consecutive frames, the position offset of the dust cluster center of mass in two adjacent frames is calculated to obtain the velocity vector, and the acceleration is calculated based on the velocity change in three consecutive frames. When the distance between multiple dust clusters in the same area is detected to be less than the preset spacing D, the ratio of the cumulative sum of the pixel grayscale values ​​in the area to the unit area is calculated to obtain the dust concentration change rate;

[0037] When the average brightness of the detection area exceeds the preset range [L1, L2], the image enhancement gain coefficient is adjusted according to a linear mapping relationship. When the dust cluster area obtained during region growing is less than the minimum threshold S1 or greater than the maximum threshold S2, the region growing thresholds T1 and T2 are increased or decreased by 10%, respectively. The processing results are transmitted to the data acquisition host via the industrial communication network.

[0038] Airflow field data collection utilizes a multi-level sensor array layout consisting of hot-wire and ultrasonic wind speed sensor units. The hot-wire wind speed sensor unit collects instantaneous local airflow velocity, with a measurement accuracy better than 0.1 m / s. The sensor utilizes a constant-temperature thermistor and features automatic temperature compensation. The ultrasonic wind speed sensor unit utilizes a dual-probe structure to acquire three-dimensional airflow velocity components, with a sampling frequency of no less than 10 Hz. These two types of sensors are deployed within the workshop monitoring area according to spatial grid division rules, forming a multi-point, three-dimensional monitoring network structure that comprehensively captures the distribution characteristics of the workshop's airflow field.

[0039] The electrostatic potential distribution data acquisition system consists of a network of electrostatic field strength sensors. These sensors utilize a capacitive design and cover a wide measurement range of 0-100 kV / m. These sensors are housed in explosion-proof enclosures and feature a built-in electrostatic shielding layer to ensure signal stability. The sensors are connected to the data acquisition system in an array via an industrial bus, forming a regularly distributed array of monitoring points within the workshop. Each sensor operates at a sampling frequency of at least 100 Hz and is equipped with independent signal conditioning circuitry and digital filtering technology to mitigate electromagnetic interference in the industrial environment.

[0040] The three types of data acquisition systems are centrally managed by an industrial-grade data acquisition host. This host utilizes a redundant architecture and is equipped with a high-performance industrial control computing unit, enabling data acquisition and processing speeds of at least 1000MB / s. Using a distributed data acquisition structure, each sensor node is synchronized with the system's master clock, achieving clock synchronization accuracy better than 1ms to ensure data timestamp consistency. The host also incorporates a built-in sensor status monitoring module, enabling real-time monitoring of the operating status of all sensors, ensuring the reliability of the data acquisition process.

[0041] After standardized processing, collected data enters the data management system, which uses a tiered storage architecture to enable rapid access to hot data and efficient management of historical data. The data management system has data backup and recovery capabilities to ensure data security.

[0042] S2: Calculating a vortex intensity index based on the airflow field data, and obtaining a dust aggregation tendency coefficient in combination with the dust movement trajectory data.

[0043] The collected airflow field data is subjected to three-dimensional velocity vector decomposition. The velocity vector V at each measuring point in the airflow field is decomposed into a radial component Vr and a tangential component Vθ. The radial component represents the radial diffusion characteristics of the airflow, and the tangential component represents the rotational characteristics of the airflow. By establishing a cylindrical coordinate system and using a coordinate transformation matrix to transform the velocity components from the rectangular coordinate system to the cylindrical coordinate system, the radial and tangential velocity values ​​at each measuring point are obtained.

[0044] Specifically, in the cylindrical coordinate system (r, θ, z), the conversion relationship of the velocity components is:

[0045]

[0046]

[0047] Where Vx and Vy are the velocity components in the rectangular coordinate system. This conversion process is performed synchronously for all measuring points in the monitoring space to form a complete velocity component field.

[0048] The vorticity field distribution is calculated based on the velocity components, including: using the central difference format to calculate the spatial derivative of the velocity field to obtain the vorticity vector ω. The vorticity calculation takes into account the gradient changes of the velocity field in three directions, including the radial velocity gradient ∂Vr / ∂r, the tangential velocity gradient ∂Vθ / ∂θ, and the axial velocity gradient ∂Vz / ∂z. The spatial derivative of the velocity field is calculated using the central difference format with fourth-order accuracy. The three components of the vorticity vector ω are expressed as:

[0049]

[0050]

[0051]

[0052] Where ∂Vz / ∂θ represents the rate of change of the axial velocity Vz in the tangential θ direction, r represents the radial distance from the rotational axis, ∂Vθ / ∂z represents the rate of change of the tangential velocity Vθ in the axial z direction, ∂Vr / ∂z represents the rate of change of the radial velocity Vr in the axial z direction, and ∂Vz / ∂r represents the rate of change of the axial velocity Vz in the radial r direction. ∂Vθ / ∂r represents the rate of change of the tangential velocity Vθ in the radial r direction, Vθ / r represents the ratio of the tangential velocity to the radial distance, reflecting the angular velocity, and ∂Vr / ∂θ represents the rate of change of the radial velocity Vr in the tangential θ direction. ωr represents the intensity of rotation about the radial axis, ωθ represents the intensity of rotation about the tangential axis, and ωz represents the intensity of rotation about the axial axis. By calculating the modulus of the vorticity vector, the vorticity scalar field representing the local rotation intensity is obtained.

[0053] The vorticity tensor Ω is constructed based on the obtained vorticity components ωr, ωθ and ωz, including:

[0054] In the cylindrical coordinate system, the spatial derivatives of the velocity field V(r, θ, z) are combined to form the velocity gradient tensor ∇V. Specifically, the derivatives of the velocity field in the radial, tangential, and axial directions are calculated, including the partial derivatives of the radial velocity Vr with respect to r, θ, and z, the partial derivatives of the tangential velocity Vθ with respect to r, θ, and z, and the partial derivatives of the axial velocity Vz with respect to r, θ, and z, forming complete velocity gradient information.

[0055] Extract the antisymmetric part of the velocity gradient tensor and construct the vorticity tensor Ω. Specifically, the components of the vorticity tensor Ω correspond to the obtained vorticity components as follows:

[0056] The vorticity component ωr corresponds to the component of the vorticity tensor that represents the rotation around the r axis, reflecting the difference between the change of the tangential velocity in the z direction and the change of the axial velocity in the θ direction;

[0057] The vorticity component ωθ corresponds to the component of the vorticity tensor that represents the rotation around the θ axis, reflecting the difference between the change of the axial velocity in the r direction and the change of the radial velocity in the z direction;

[0058] The vorticity component ωz corresponds to the component of the vorticity tensor that represents the rotation around the z-axis, reflecting the difference between the change of radial velocity in the θ direction, the change of tangential velocity in the r direction, and the angular velocity term.

[0059] Based on the above correspondence, the three vorticity components are converted into the matrix form of the vorticity tensor. Specifically, the vorticity tensor Ω is a 3×3 antisymmetric matrix, whose upper triangular elements are negative values ​​of the vorticity components, the lower triangular elements are positive values ​​of the vorticity components, and the main diagonal elements are zero.

[0060] Based on the velocity component field in the established cylindrical coordinate system, the velocity change rate in each direction is calculated, and the strain rate tensor S is constructed. Its main diagonal elements correspond to the radial, tangential and axial strain rates, respectively, and the non-main diagonal elements correspond to the shear strain rates on each surface. The elements in the upper and lower symmetrical positions are equal.

[0061] The vortex intensity index Q is calculated based on the strain rate tensor S and the vorticity tensor Ω, which is expressed as:

[0062] Q = 1 / 2(|Ω|² - |S|²)

[0063] When the Q value is positive, it means that the vortex motion is dominant, and the larger the Q value, the stronger the vortex intensity. By analyzing the Q value distribution of the entire monitoring space, the vortex core area is identified.

[0064] The dust accumulation tendency coefficient is derived based on the vortex intensity index combined with dust trajectory data. Specifically, the dust trajectory data collected by the high-speed camera system is divided into time windows, each lasting 0.5 seconds. For each image sequence within a time window, image processing techniques are used to extract the position of dust particle clusters, and the center of mass of the dust cluster within that time window is calculated.

[0065] By calculating the change in the center of mass position between adjacent time windows, the movement direction and speed of the dust cluster are determined. The vortex center is also located, and the direction from the center of mass to the vortex center is calculated. If the angle between the dust cluster's movement direction and the direction toward the vortex center is less than 30 degrees, and the movement speed exceeds a preset speed threshold of 0.5 m / s, the dust cluster is considered to be converging toward the vortex center.

[0066] Based on the movement speed of the dust group, the angle between the movement direction and the vortex center direction, and the preset speed threshold, a calculation model for the dust aggregation tendency coefficient is established.

[0067] Specifically, the construction of the calculation model of dust aggregation tendency coefficient includes three components, namely, speed influence factor, direction influence factor and vortex intensity influence factor;

[0068] Regarding the speed impact factor Kv, specifically:

[0069] When the dust movement speed V is less than the preset minimum speed threshold Vmin, the speed impact factor Kv takes a value of 0;

[0070] When the dust movement speed V is greater than the preset maximum speed threshold Vmax, the speed impact factor Kv takes a value of 1;

[0071] When the dust movement velocity V is between Vmin and Vmax, the velocity influence factor Kv is calculated using a piecewise linear function so that it changes smoothly between 0 and 1.

[0072] Directional influence factor Kd, specifically:

[0073] When the angle α between the dust movement direction and the direction pointing to the vortex center is 0 degrees, the direction influence factor Kd takes the value of 1, indicating that it is completely pointing to the vortex center;

[0074] When the angle α is 180 degrees, the direction influence factor Kd takes a value of 0, indicating that it is completely away from the vortex center;

[0075] When the angle α is between 0 and 180 degrees, the cosine function is used to calculate the directional influence factor Kd so that it decays smoothly as the angle increases.

[0076] The vortex intensity influencing factor Kq is specifically:

[0077] When the local Q value is less than the preset minimum threshold Qmin, the vortex intensity influence factor Kq takes the value of 0;

[0078] When the local Q value is greater than the preset maximum threshold Qmax, the vortex intensity influence factor Kq takes the value of 1;

[0079] When the local Q value is between Qmin and Qmax, a piecewise linear function is used to calculate the vortex intensity influence factor Kq so that it varies smoothly between 0 and 1.

[0080] The three influencing factors are integrated to establish a basic calculation model, specifically:

[0081] When the velocity influence factor is 0, the final aggregation tendency coefficient is 0 regardless of the direction and vortex intensity;

[0082] When the directional influence factor is 0, it means that the dust is far away from the vortex center, and the final aggregation tendency coefficient approaches 0;

[0083] When the vortex intensity influence factor is 0, it means that there is no obvious vortex structure locally and the final aggregation tendency coefficient is small;

[0084] When all three influencing factors are not 0, the final clustering trend coefficient is calculated using the weighted product method.

[0085] The weight coefficient for the weighted product is initially 1.0. If an aggregation trend is detected in three consecutive time windows, the weight coefficient is increased by 20%; if no aggregation trend is detected, the weight coefficient is decreased by 10%. The dust aggregation trend coefficient ranges from 0 to 1, with larger values ​​indicating a higher degree of localized dust enrichment under the influence of the vortex airflow.

[0086] S3: An electrostatic coupling effect model is constructed based on the electrostatic potential distribution data and the dust aggregation tendency coefficient, and a multi-layer neural network is used to establish a nonlinear mapping relationship between dust concentration, electrostatic field strength and vortex intensity.

[0087] When dust particles move through the air, friction and collisions generate static charges. These charged particles create a charge distribution in space, generating a localized electrostatic field. Once established, the electrostatic field exerts a Coulomb force on the charged particles, altering their trajectory. As the dust's motion changes, the spatial charge distribution shifts accordingly, causing the localized electrostatic field strength to adjust. This interaction between the electrostatic field and the movement of the dust particles constitutes a continuous dynamic coupling process.

[0088] To address this dynamic coupling process, developing an analytical model requires simultaneous consideration of the electrostatic field equations, the dust motion equations, and the charge transport equations. These equations exhibit strong nonlinear coupling, and the boundary conditions dynamically change over time. Simplified approaches fail to reflect the true behavior of the system. Maintaining the full form of the equations would be computationally intensive and difficult to ensure convergence. Therefore, a neural network algorithm capable of handling nonlinearities and possessing adaptive learning capabilities is required for modeling.

[0089] Based on the obtained electrostatic potential distribution data and dust aggregation tendency coefficient, a multi-layer neural network is used to construct an electrostatic coupling effect model, including:

[0090] Collect and organize multi-source data from the monitoring space, including the spatiotemporal distribution of dust concentration, electrostatic potential distribution, airflow characteristics, and the calculated dust accumulation tendency coefficient. Dust concentration data requires normalization to eliminate dimensionality. For electrostatic potential data, the magnitude and direction of the electric field strength are determined by calculating the potential gradient. For airflow data, vectors representing vortex characteristics are constructed based on the previously calculated Q-criterion and the dust accumulation tendency coefficient. This preprocessed data is combined row by row to construct a structured feature matrix, which serves as the input to the neural network.

[0091] When constructing a multi-layer neural network, an hourglass-shaped neural network structure of "narrow layer-wide layer-narrow layer" is adopted, including the input layer, the first hidden layer, the middle hidden layer, the feature extraction layer and the output layer.

[0092] The number of nodes in the input layer corresponds to the dimensions of the feature matrix, which includes three key parameters: dust concentration distribution, electric field intensity distribution, and vortex characteristics. If dust concentration data is missing, it is supplemented based on the average value of nearby sampling points. When electric field intensity data is missing, interpolation estimation is performed using the electric field distribution law. If vortex characteristic data is missing, data from adjacent time points is used as a replacement. Once the feature matrix is ​​complete, the maximum and minimum value normalization method is used to map all eigenvalues ​​to a unified interval to eliminate dimensionality effects.

[0093] The first hidden layer uses a smaller number of nodes for feature compression, with the number of nodes set to one-third of the total number of input physical quantities. If the Pearson correlation coefficient between the local electric field strength and the dust concentration distribution is detected to be greater than 0.8, indicating that there is significant electrostatic coupling in the area, the coupling characteristics of these two physical quantities are extracted using principal component analysis. When the electrostatic field distribution characteristics of adjacent monitoring points are similar, that is, the Euclidean distance is less than 10% of the minimum characteristic scale of the monitoring space, the field strength characteristics of these points are merged using a weighted average method based on the inverse distance. If the contribution of a physical quantity to the coupling effect is always less than 5% of the average contribution after 50 consecutive iterations, the physical quantity is eliminated from the coupling analysis. The Sigmoid function is selected to normalize the physical quantities, so that physical quantities of different dimensions can be coupled.

[0094] The middle hidden layer uses a larger number of nodes for feature reorganization, setting the number of nodes to twice the number of nodes in the input layer. To characterize the coupling effect of dust concentration and vortex intensity, a polynomial combination of these two features is constructed, including a linear term, a square term, and a cross term. The initial weight values ​​corresponding to these combinations are set to random numbers from a standard normal distribution. When the partial correlation coefficient of a coupling term is found to be greater than 0.7 on the validation set, it indicates that the coupling mechanism is significant. The learning rate of the weight corresponding to the coupling term is increased to 1.5 times the baseline learning rate. The ReLU function is selected as the activation function. When the input is positive, a linear mapping relationship is maintained, and when the input is negative, the output is zero.

[0095] The feature extraction layer uses a larger number of nodes based on the reorganized features to extract the dominant factors, setting the number of nodes to twice that of the output layer. If the mutual information between the dust accumulation trend and the local electric field strength in a region exceeds 80% of the system information entropy, a polynomial for the electrostatic coupling effect in that region is established, consisting of an electrostatic force drive term, a spatial charge distribution term, and a field intensity gradient term. When a coupling mechanism's predicted contribution on the test set exceeds 30%, it is identified as the dominant coupling mechanism, and its interaction with other physical quantities is strengthened. The tanh function is selected as the activation function, and the eigenvalues ​​are mapped to the interval [-1, 1] to keep the coupling effect strength within a physically reasonable range.

[0096] The output layer sets the number of nodes to the number of physical quantities to be predicted, and the output nodes correspond to the key physical quantities of the electrostatic coupling effect. The first type of node outputs the local electrostatic field intensity distribution, which characterizes the electrostatic interaction strength at each point in space. The second type of node outputs the dust velocity vector, describing the motion of dust under the influence of electrostatic forces. The third type of node outputs the dust concentration distribution, reflecting the spatial distribution characteristics of dust under the influence of electrostatic forces. When processing these physical quantity outputs, a dimensional normalization method is used: the electric field intensity is divided by the standard atmospheric breakdown field strength, the velocity is divided by the characteristic dust settling velocity, and the concentration is divided by the safe limit concentration. The tanh function is used for the electric field intensity output node to ensure that the predicted value is within a reasonable range. The Softplus function is used for the concentration output node to ensure that the predicted value is always positive. The velocity output node uses linear activation to maintain vector characteristics.

[0097] Based on the constructed electrostatic coupling effect model, when the training process starts, data including the electrostatic field intensity distribution, dust concentration distribution and its evolution under different working conditions are collected and divided into training set, validation set and test set in a ratio of 7:2:1.

[0098] If there are insufficient samples for certain typical coupling conditions (such as the dust aggregation process under a strong electrostatic field), numerical simulations are performed based on the physical equations of electrostatic force and dust motion to generate supplementary samples. If the predicted values ​​of physical quantities exceed the reasonable range during training, penalty terms based on physical laws are introduced to constrain them. After each round of training, the electrostatic field prediction accuracy and dust motion prediction accuracy are evaluated separately. If the relative error of both is less than 5% and the rate of change is less than 0.1% for 10 consecutive rounds, the training for that stage is completed. If the prediction results under certain coupling conditions deviate significantly from the actual observed data, the coupling parameters under that condition are optimized until the error is reduced to within the allowable range.

[0099] After the initial training is completed, the test set data is used to verify the model's coupling effect prediction ability.

[0100] If the relative error between the model's prediction of the electrostatic field distribution or dust movement trend and the actual observation value exceeds 10%, it is necessary to recheck whether the construction of the coupling equation is complete.

[0101] When it is found that the prediction accuracy is insufficient under certain special working conditions (such as the strong coupling effect caused by high concentration of dust), additional monitoring data under this working condition should be added, and the number of additional samples should be no less than 5% of the original training set.

[0102] When the model is applied to actual projects, the effectiveness of the coupling model is evaluated by comparing the time series deviation between the predicted and measured values. If the predicted deviation is found to be increasing, it is necessary to adjust the key parameters in the coupling equations in accordance with the latest physical knowledge.

[0103] S4: When the hazard increment and the growth rate of the dust aggregation trend coefficient output by the electrostatic coupling effect model exceed the preset threshold, the dust explosion prevention warning signal is triggered.

[0104] The output layer based on the electrostatic coupling effect model outputs the local electrostatic field intensity distribution, dust movement velocity vector and dust concentration distribution;

[0105] The hazard increment is calculated by integrating the normalized electrostatic field intensity distribution and the spatial gradient product of the dust concentration distribution, reflecting the coupling strength between the electrostatic field and the dust distribution.

[0106] The dust aggregation tendency coefficient is determined based on the normalized dust motion velocity vector and the spatiotemporal change rate of concentration distribution to characterize the aggregation dynamics of dust under the action of electrostatic force.

[0107] It should be noted that the dust aggregation tendency coefficient here is the dust movement characteristic under the action of the electrostatic field, and is obtained based on the dust aggregation tendency coefficient obtained based on the airflow field data and dust movement trajectory data.

[0108] Thresholds are set based on the hazard increment and the dust accumulation trend coefficient. The threshold for the hazard increment is set at 30% of the standard atmospheric breakdown field strength, and the growth rate of the dust accumulation trend coefficient is twice the growth rate of the safety concentration limit.

[0109] When the equipment runs at high speed and generates a strong electrostatic field, the electrostatic field strength node in the output layer outputs a value close to the breakdown field strength, and the concentration distribution node shows a significant change. At this time, the calculated hazard increment quickly exceeds the preset threshold, and the growth rate of the dust accumulation tendency coefficient also exceeds the preset threshold, triggering a level 1 warning signal.

[0110] Immediately reduce the equipment's operating speed to below 50% of the safe speed limit, and simultaneously activate the emergency dust removal system to increase dust removal frequency. If the increase in risk persists above the preset threshold after the speed reduction, an emergency shutdown and temporary grounding are required to eliminate excess static charge. Only after confirming that the static field strength has returned to a safe level and the dust concentration has dropped below the standard value can the system be resumed and monitoring frequency increased.

[0111] When the conveying device continues to operate, the output layer shows that the electric field intensity remains at a medium level, and the velocity vector node points to the local accumulation area. If the increase in danger is observed to be continuously above 80% of the preset threshold, and the dust accumulation trend coefficient continues to grow, when its growth rate reaches 90% of the preset threshold, a secondary warning signal is triggered;

[0112] Adjust conveyor operating parameters, including reducing conveying speeds and adjusting material loading. Increase local dust removal and ventilation measures, paying particular attention to areas with significant accumulation. If the dust accumulation coefficient continues to increase despite these measures, extend equipment downtime to allow for comprehensive static elimination. Upon resuming operation, implement a progressive speed increase to ensure stable performance.

[0113] When a sudden increase in electrostatic field intensity occurs in a certain local area, the electric field intensity output node in the area quickly approaches the saturation value, and the velocity vector output shows a strong convergence trend toward the area. If the increase in the danger level of the area quickly exceeds the preset threshold, and the growth rate of the local aggregation trend coefficient exceeds the preset threshold, a special warning signal for the local area is triggered;

[0114] The area experiencing the sudden change in field strength must be immediately isolated, and the directional dust removal system activated. If the increased risk in the area does not effectively decrease after isolation measures, emergency spraying and localized static elimination treatment will be required. Once local indicators return to normal, the area should be inspected to identify the cause of static accumulation and, if necessary, replace or install additional anti-static devices.

[0115] When electrostatic field enhancement occurs simultaneously in multiple regions, the physical quantity nodes in the output layer display continuous spatial variation characteristics. If the risk increment shows a propagation characteristic between adjacent regions, and the aggregation trend coefficient shows a chain development trend, when the indicator of any region exceeds the preset threshold, a system-level warning signal is triggered;

[0116] The entire production line must be slowed down, with all dust removal and ventilation systems activated. If the chain reaction is not effectively controlled, production will need to be gradually shut down in sections, prioritizing areas with the highest risk of increased risk. Before the system is restored to normal, a comprehensive ESD safety test should be conducted on the entire production line, and anti-static measures should be optimized based on the test results.

[0117] After all system indicators return to normal, a comprehensive safety assessment is required. This assessment includes: analyzing the rationality of equipment operating parameters, verifying the effectiveness of anti-static measures, and evaluating the efficiency of the dust treatment system. Based on the assessment results, the early warning threshold settings are updated and improved, and the emergency response process is optimized.

[0118] In summary, the dust explosion prevention warning method based on the embodiment of the present invention is explained. By integrating high-speed video, airflow field measurement, multi-source data processing and multi-layer neural network modeling, it realizes the dynamic identification and quantitative analysis of the coupling mechanism between the electrostatic field, airflow and dust movement. It can not only accurately capture the aggregation behavior of dust under complex working conditions, but also reflect the dangerous changes under the action of the electrostatic field in real time, significantly improving the ability to identify abnormal working conditions and coupling mechanisms. At the same time, based on the hierarchical warning and emergency response mechanism, it can achieve early and accurate warning and intelligent disposal of dust explosion hazards, greatly improving the safety management and control level of the system. It not only improves the real-time monitoring, warning and decision-making capabilities of the electrostatic-dust system, but also provides a scalable intelligent method for multi-physical field coupling analysis in complex industrial environments.

Claims

1. A dust explosion prevention early warning method, characterized in that: include: Collect dust movement trajectory data, airflow field data and electrostatic potential distribution data, and perform standardization processing; Calculating a vortex intensity index based on the airflow field data, and obtaining a dust aggregation tendency coefficient in combination with the dust movement trajectory data; An electrostatic coupling effect model is constructed based on the electrostatic potential distribution data and the dust aggregation tendency coefficient, and a nonlinear mapping relationship between dust concentration, electrostatic field strength and vortex strength is established using a multi-layer neural network; When the hazard increment and the growth rate of the dust aggregation tendency coefficient output by the electrostatic coupling effect model exceed a preset threshold, a dust explosion prevention warning signal is triggered.

2. The dust explosion prevention early warning method according to claim 1, characterized in that: The dust movement trajectory data is acquired through a high-speed camera system, the airflow field data is acquired through an airflow field sensor array, and the electrostatic potential distribution data is acquired through an electrostatic field strength sensor.

3. The dust explosion prevention early warning method according to claim 2, characterized in that: Based on the airflow field data, three-dimensional velocity vector decomposition is performed, and the velocity of each measuring point is decomposed into radial and tangential components in the cylindrical coordinate system. The velocity gradients in each direction are calculated using coordinate transformation and central difference method to obtain the three components of the vorticity vector, which reflect the local rotation characteristics and spatial vortex structure of the airflow field. The vorticity scalar field characterizing the rotation intensity is obtained through the vorticity vector modulus.

4. The dust explosion prevention early warning method according to claim 3, characterized in that: Based on the three components of the vorticity vector, tensorization processing is performed, a velocity gradient tensor is constructed in a cylindrical coordinate system, and its antisymmetric part is extracted to obtain the vorticity tensor. Combined with the calculated strain rate tensor, the Q criterion is used to quantitatively characterize the vortex intensity index of each region in the airflow field and identify the vortex core area.

5. The dust explosion prevention early warning method according to claim 4, characterized in that: Based on the vortex intensity index and the dust motion trajectory data, the collected dust motion trajectory data is divided into time windows and image processed, the center of mass position of the dust group in each window is extracted, and combined with the center of mass changes between adjacent windows, the movement direction and speed of the dust are calculated, the angle between the movement direction of the dust group and the direction of the vortex center is judged, and the dust aggregation trend coefficient is calculated.

6. The dust explosion prevention early warning method according to claim 5, characterized in that: Based on the dust aggregation tendency coefficient, combined with multi-source data of dust concentration, electric field intensity and airflow field, a structured feature matrix is ​​constructed. With this as input, a multi-layer neural network is used to establish an electrostatic coupling effect model to monitor the interaction between dust, airflow and static electricity in the space.

7. The dust explosion prevention early warning method according to claim 6, characterized in that: The multi-layer neural network adopts an hourglass structure of "narrow layer-wide layer-narrow layer". The input layer contains normalized dust concentration, electric field intensity and vortex characteristics. Missing data is supplemented by interpolation or substitution. The first hidden layer performs feature compression and extracts physical quantity coupling characteristics through principal component analysis and weighted merging. Low-contribution features will be eliminated. The number of nodes in the middle hidden layer is increased for reorganization and coupling feature mining, and the coupling term weights are dynamically adjusted. ReLU activation is used. The feature extraction layer further refines the dominant coupling mechanism, and tanh activation is used to ensure physical rationality. The output layer corresponds to each key physical quantity, and normalization and different activation functions are used to ensure the physical meaning and effective range of the output.

8. The dust explosion prevention early warning method according to claim 7, characterized in that: Training is performed based on the electrostatic coupling effect model, and training, validation, and test sets are divided. For typical working conditions with insufficient samples, data is supplemented through numerical simulation of physical equations. Physical constraints are introduced during training to prevent the predicted values ​​from exceeding the bounds. After each round of training, the prediction accuracy of the electrostatic field and dust movement is evaluated separately. The training for this stage is completed when the accuracy meets the standards and converges. The prediction ability and effectiveness of the model are verified through test sets and actual engineering data.

9. The dust explosion prevention early warning method according to claim 8, characterized in that: Based on the output results of the electrostatic coupling effect model, the hazard increment is obtained by calculating the product of the normalized electrostatic field intensity and the spatial gradient of the dust concentration distribution. The dust aggregation trend coefficient under the action of the electrostatic field is determined by the dust movement speed and concentration change rate. Thresholds are set for the hazard increment and the aggregation trend coefficient respectively to achieve graded warning: When a strong electrostatic field is generated during equipment operation and key indicators rapidly exceed thresholds, a Level 1 warning is triggered and the equipment is immediately slowed down, dusted, or shut down. When chain-like danger spreads across multiple areas, triggering a system-level warning, all lines must be decelerated and zoned for processing; After all emergency responses, a comprehensive assessment is conducted on operating parameters, anti-static measures and dust treatment efficiency, and the warning thresholds and response plans are optimized accordingly to ensure safe operation.

10. A dust explosion warning system, which uses the method according to any one of claims 1 to 9 to provide early warning for dust explosion, characterized in that: It includes a data acquisition unit, a calculation and processing unit, and an early warning execution unit; The data acquisition unit is used to collect dust movement trajectory data, airflow field data and electrostatic potential distribution data, and perform standardization processing; The computing processing unit calculates the vortex intensity index using the airflow field data and obtains the dust aggregation tendency coefficient in combination with the dust motion trajectory data; then, based on the electrostatic potential distribution data and the dust aggregation tendency coefficient, a multi-layer neural network is used to construct an electrostatic coupling effect model to achieve a nonlinear mapping relationship between dust concentration, electrostatic field strength, and vortex intensity; The warning execution unit triggers a dust explosion prevention warning signal when the hazard increment and the growth rate of the dust aggregation tendency coefficient output by the electrostatic coupling effect model exceed a preset threshold.

Images