An internal temperature rise real-time monitoring system for an explosion-proof electrical device

Abstract

The application discloses an internal temperature rise real-time monitoring system of an explosion-proof electrical equipment, relates to the technical field of intelligent sensors, and collects temperature change data of multiple thermal nodes on the surface layer of the explosion-proof electrical equipment; carries out time-frequency joint deconstruction on the temperature data, extracts nonlinear temperature rise characteristics; constructs a multi-node thermal anomaly coupling model based on a graph neural network, generates a heat conduction path weight matrix; fuses the weight matrix and historical temperature gradients, inputs a gated recurrent unit neural network to perform temperature rise evolution prediction, and calculates a heat accumulation rate; generates a temperature rise risk probability atlas by comparing probability density functions, dynamically adjusts a sampling strategy; when the temperature rise rate exceeds a threshold value, a dynamic thermal field atlas is constructed, compared with a failure mechanism database, and potential fault types and early warning levels are output; the application realizes real-time identification, positioning and early warning of internal temperature rise anomalies of the equipment, and improves the active safety management capability of the explosion-proof equipment.

Description

Technical Field

[0001] This invention relates to the field of intelligent sensor technology, and more specifically to a real-time monitoring system for internal temperature rise of explosion-proof electrical equipment. Background Technology

[0002] Explosion-proof electrical equipment is widely used in flammable and explosive environments such as coal mines, petrochemical plants, and natural gas plants. Its operational safety is directly related to the safety of personnel and property. However, such equipment is usually complex in structure, highly sealed, and accumulates heat quickly. During long-term operation, problems such as loose internal contacts, abnormal loads, and aging insulation often lead to localized heating or even abnormal temperature rises, which can easily ignite explosive gases and cause serious accidents.

[0003] Traditional temperature rise monitoring technologies often rely on fixed-point sampling by external temperature sensors, such as thermocouples, platinum resistance temperature sensors (e.g., Pt100), or thermistors attached to the outer surface of explosion-proof enclosures. Because explosion-proof enclosures are typically thick metal structures with long heat conduction paths and high thermal inertia, internal heat transfer to the outer surface undergoes significant attenuation and delay. This results in a severe discrepancy between the temperature measured by the sensor and the actual internal hot spot temperature, making it difficult to penetrate the explosion-proof enclosure and perceive the dynamic evolution of the internal heat distribution in real time. This type of fixed-point temperature sensor arrangement suffers from low spatial resolution, long response delays, and numerous monitoring blind spots, making it particularly difficult to detect early abnormal changes in small, localized heat sources.

[0004] Existing solutions also attempt to use infrared thermal imagers for non-contact temperature rise monitoring. However, due to the strong shielding effect of explosion-proof enclosures on thermal radiation, and the presence of dirt, oxide layers, or non-uniform emissivity on the equipment surface, infrared imaging results are easily affected by environmental interference and cannot accurately reflect the temperature field of internal components. Furthermore, the installation of contact sensors such as thermocouples inside explosion-proof equipment is limited and difficult to modify. In complex electrical equipment with multiple circuits and contacts, nonlinear temperature rise changes caused by a slight loosening of a single contact are often masked by overall thermal inertia, making timely warnings difficult with conventional monitoring methods. Summary of the Invention

[0005] The purpose of this invention is to provide a real-time monitoring system for internal temperature rise of explosion-proof electrical equipment to address the shortcomings of the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a real-time temperature rise monitoring system for the internal environment of explosion-proof electrical equipment, comprising: The temperature data acquisition module acquires the raw temperature signals of multiple thermal nodes during the operation of the explosion-proof electrical equipment, and collects temperature change data in key areas on the surface of the explosion-proof electrical equipment. The feature extraction module performs nonlinear temperature perturbation decoupling processing on the collected temperature change data based on time-frequency joint deconstruction, and extracts the temperature rise characteristic change trend induced by potential internal heat sources. The thermal anomaly coupling modeling module constructs a multi-node thermal anomaly coupling model based on the temperature rise characteristic change trend, and uses a graph neural network to establish the heat conduction path weight matrix between nodes. The micro-temperature rise evolution prediction module inputs the heat conduction path weight matrix into the micro-temperature rise evolution prediction model, calculates the heat accumulation rate of potential hidden danger points, and generates an internal temperature rise risk probability map. The dynamic sampling and control module divides high-risk monitoring zones based on the internal temperature rise risk probability map and adjusts the sampling density and frequency of the surface temperature of the explosion-proof electrical equipment. The dynamic thermal field construction module locates potential heat sources and forms a dynamic thermal field map when the temperature rise rate of any node in the internal temperature rise risk probability map exceeds a preset safety threshold. The warning level output module compares the dynamic thermal field spectrum with the preset failure mechanism database to output the potential thermal fault type and warning level.

[0007] Preferably, the feature extraction module includes: Segmented sampling of temperature change data in the time domain is performed to extract transient temperature rise fluctuation segments from continuous temperature change curves; Based on the transient temperature rise fluctuation segment, the frequency components at different scales are decomposed to obtain the multi-scale frequency distribution matrix of temperature disturbance; The nonlinear amplitude response curve of temperature disturbance is calculated based on the multi-scale frequency distribution matrix, and the characteristic temperature rise component with the highest correlation to the internal potential heat source is extracted.

[0008] Preferably, the thermal anomaly coupling modeling module includes: The temperature rise characteristic change trend of each thermal node is aligned with the data in a time synchronization manner, and an initial node set with node temperature gradient as the attribute vector is constructed. Based on the spatial position and thermal conductivity of the thermal nodes, the initial thermal conduction connection relationship is generated using the distance decay function, forming a multi-node thermally coupled topology graph containing node and edge attributes. The multi-node thermally coupled topology graph is input into the graph neural network model. The node attribute vectors are updated through graph convolution operations, and the heat conduction contribution value of each edge after multiple iterations is calculated. Construct a heat conduction path weight matrix based on the updated node attribute vectors.

[0009] Preferably, the microscopic temperature rise evolution prediction module includes: The heat conduction path weight matrix is ​​fused with the historical temperature gradient data of each node to construct a dynamic sequence of heat migration, which serves as the input feature set for the prediction model. A micro-temperature rise evolution prediction model is constructed based on a gated recurrent unit neural network. After inputting the feature set, the temperature rise trend of each node in the future period is obtained through time series learning. The heat accumulation rate of each node within a specified time window is calculated based on the prediction results. This is the integral of the product of the continuous temperature rise value and the time interval within the window. Based on the deviation measurement of the heat accumulation rate from the statistical distribution under historical normal conditions, a probability density function model is used to generate an internal temperature rise risk probability map.

[0010] Preferably, the generation of the internal temperature rise risk probability map includes: Historical data on the heat accumulation rate of each thermal node is backtracked to construct a rate distribution dataset under normal operating conditions of the equipment, and indexed and classified by node number. Based on the historical rate distribution of each node, the corresponding probability density function model is constructed using the kernel density estimation method. Substitute the currently predicted heat accumulation rate into the probability density function corresponding to each node, calculate its probability density value in the normal distribution, and evaluate its deviation accordingly. The deviation of all nodes is mapped onto a two-dimensional spatial location map using color gradients, forming an internal temperature rise risk probability map that characterizes the thermal anomaly risk level of each node.

[0011] Preferably, the dynamic sampling and control module includes: Based on the risk level of each thermally sensitive node in the internal temperature rise risk probability map, the internal space of the equipment is divided into high-risk monitoring zone, medium-risk monitoring zone and low-risk monitoring zone; Enhanced sampling strategies were implemented for nodes within high-risk monitoring areas; The default sampling strategy is maintained for medium-risk areas, while the frequency reduction strategy is implemented for low-risk areas.

[0012] Preferably, the dynamic thermal field construction module includes: When the rate of temperature rise of any node in the internal temperature rise risk probability map exceeds the preset safety threshold, the center location of the potential heat source is determined by performing directional analysis on the temperature rise gradient changes of the surrounding adjacent nodes based on the node's position in the three-dimensional structural coordinates of the equipment. The center of the heat source is taken as the initial core point for thermal field reconstruction. The spatial distance from the core point to all nodes is obtained, and the instantaneous heat propagation intensity of each node is calculated by combining the heat conduction path weight matrix. The heat transfer intensity of each node is continuously processed to generate temperature field distribution data covering the internal space of the equipment; A dynamic thermodynamic field map is constructed based on continuous temperature field data, and the thermodynamic field intensity is updated in a time series manner to form a dynamic thermodynamic field map that can reflect the evolution trend of potential heat sources.

[0013] Preferably, the warning level output module includes: The dynamic thermal field map is split into multiple temperature field snapshots according to the time series, and the temperature distribution feature vector in each snapshot is extracted, including the maximum temperature value, average temperature gradient, thermal center offset and propagation rate. Retrieve fault sample maps that match the current equipment type from the preset failure mechanism database, and establish a multi-dimensional feature template set containing typical thermal failure modes; The minimum Euclidean distance method is used to compare the similarity between the current feature vector and the template set. If the distance is lower than the set recognition threshold, it is determined to be a known thermal fault type. Based on the historical severity level of the fault type, the fault evolution time, and the spatial expansion range, the corresponding warning level is output in combination with the set classification rules.

[0014] The technical effects and advantages provided by the present invention in the above technical solution are as follows: 1. This invention constructs a microscopic temperature rise evolution monitoring method that integrates multi-node heat conduction path weighting, nonlinear temperature rise feature extraction, and dynamic thermodynamic field reconstruction. This method achieves high spatiotemporal resolution perception and risk prediction of internal temperature rise anomalies in explosion-proof electrical equipment. Compared to traditional monitoring methods based on single-point temperature sensors, this invention can dynamically identify heat source locations and predict fault evolution trends, effectively overcoming technical bottlenecks such as heat conduction hysteresis in explosion-proof enclosures and delayed monitoring response, significantly improving the accuracy of thermal hazard identification.

[0015] 2. This invention uses graph neural networks and gated recurrent neural networks to collaboratively model the correlation mapping and time series prediction between thermistor nodes of multi-source temperature sensors, and outputs the risk level by combining a fault template comparison mechanism. It has the technical advantages of high intelligence, adaptability and scalability. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0017] Figure 1 This is a flowchart of a real-time monitoring system module for internal temperature rise of explosion-proof electrical equipment according to the present invention.

[0018] Figure 2 This is a flowchart of the method for generating the internal temperature rise risk probability map of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] For examples, please refer to Figure 1 As shown in this embodiment, a real-time temperature rise monitoring system for the internal temperature of explosion-proof electrical equipment includes: The temperature data acquisition module is used to acquire the raw temperature signals of multiple thermal nodes during the operation of explosion-proof electrical equipment, and to collect temperature change data in key areas on the surface of the explosion-proof electrical equipment.

[0021] This embodiment uses the ZKSC series intelligent control data acquisition explosion-proof box as the object, selecting a typical three-phase circuit power distribution wiring structure as the experimental platform. Without compromising the integrity of the explosion-proof enclosure, a high-sensitivity NTC thermistor (response time ≤1s, accuracy ±0.2°C) is used as the temperature acquisition unit, combined with a fiber optic temperature sensor for high electromagnetic interference areas at key points. The sensor placement locations are as follows: Circuit input terminal blocks ×3 (L1, L2, L3); Intermediate overcurrent relay contacts ×2; Output terminal block × 3; One heat source point (near the cable inlet) on the inner wall of the outer casing; One weak heat dissipation area at the top of the casing; A total of 10 temperature acquisition points were deployed and connected to the local edge computing terminal via RS485 bus. The sampling frequency was 10 Hz per second, and data was collected continuously for 24 hours.

[0022] The equipment was powered on and operated for 8 hours under simulated high load conditions, during which the load current gradually increased from 40% to 90% of the rated load. The temperature changes at each sampling point were recorded in real time, and the temperature rise trend and time characteristics were monitored.

[0023] Table 1 shows the temperature variation data (unit: °C) at some of the data collection points during operation:

[0024] As shown in Table 1, although the temperature change on the surface of the device casing is small (e.g., the temperature rise at the top of the casing is only 7.3°C), the temperature rise of the internal terminals and contacts has exceeded 30°C, showing a clear trend of local overheating. This verifies the necessity and effectiveness of forming a thermal node network through multi-point data acquisition.

[0025] All temperature signals are normalized in real time by local edge nodes and bound to timestamps to form a temperature time-series data stream.

[0026] The feature extraction module performs nonlinear temperature perturbation decoupling processing on the collected temperature change data based on time-frequency joint deconstruction, and extracts the temperature rise characteristic change trend induced by potential internal heat sources.

[0027] First, the acquired multi-channel temperature change data is segmented using a sliding time window approach. Each sampling period is set to 120 seconds, with an overlap rate of 50%, meaning adjacent windows overlap by 60 seconds. The first derivative (i.e., the rate of temperature change) of each temperature curve is calculated. If, within any given time period, the rate of temperature change exceeds 2°C / min across five consecutive sampling points, that segment is considered a transient temperature rise fluctuation segment.

[0028] The start and end points of each transient fluctuation segment are recorded as feature candidate intervals, which serve as the input data source for subsequent frequency analysis.

[0029] The extracted transient temperature rise fluctuation segments are input into a multi-scale analysis model based on continuous wavelet transform for frequency deconstruction. Using Morlet mother wavelet as the basis function, frequency response analysis is performed on each fluctuation segment at 16 scales (corresponding to frequency components in the range of 0.01 Hz to 1 Hz).

[0030] For each time period and scale combination, wavelet coefficients are calculated, and the energy of that frequency component is represented by the square of the coefficient amplitude. Finally, a two-dimensional matrix is ​​constructed, where rows represent time and columns represent scale (or frequency). Each element in the matrix represents the temperature perturbation energy at a specific time and frequency, which is called the multi-scale frequency distribution matrix of the temperature perturbation.

[0031] Based on the frequency distribution matrix, a nonlinear amplitude response modeling method is used to evaluate the trajectory of perturbation energy variation in the time domain at various scales. First, the wavelet energy-time curve at each scale is normalized, and then its gradient is calculated. The nonlinear amplitude response curve is defined as the time integral curve of the product of the normalized energy and its gradient; this response value can effectively characterize the contribution intensity of the frequency component to the transient temperature rise.

[0032] Based on a preset heat source correlation threshold Th, all frequency components with nonlinear amplitude response values ​​exceeding Th are selected as characteristic temperature rise components. In this embodiment, Th is set to the mean of the response of each frequency channel plus 1.5 times the standard deviation to ensure that the screening results are statistically significant.

[0033] The thermal anomaly coupling modeling module constructs a multi-node thermal anomaly coupling model based on the temperature rise characteristic change trend, and uses a graph neural network to establish the heat conduction path weight matrix between nodes.

[0034] The temperature rise characteristic trend curves of each thermistor node are aligned along a unified time axis to ensure the comparability of data from all nodes at the same time scale. For the temperature value sequence of each thermistor node within a continuous time slice, its first derivative is calculated to obtain the temperature gradient of the node at each time point. Let the temperature of the i-th thermistor node at time t be Ti(t), then its temperature gradient is expressed as: Temperature Gradient Where Δt represents the sampling time interval, which is 1 second. The attribute vector of each node is constructed using this temperature gradient, serving as the initial features of the nodes in the graph model, forming the initial node set.

[0035] Based on the spatial relationship between each pair of heat-sensitive nodes and the thermal conductivity of the heat conduction path material, connection edges between nodes are constructed, and connection strength is calculated. Spatial distances are determined through 3D modeling of the equipment structure, and thermal conductivity is provided in the equipment material handbook.

[0036] The connection weights are defined using the distance decay function: Initial Connection Weights Where Wij is the connection weight between node i and node j, dij is the spatial distance between the two nodes, σ is the distance sensitivity coefficient with a value of 100 mm, and k is the thermal conductivity normalization coefficient, which is linearly mapped to the [0,1] interval based on the thermal conductivity of the material. This results in a multi-node thermally coupled topology graph containing node attributes and edge weights, which is used as input for graph neural network modeling.

[0037] A graph convolutional neural network model is used to iteratively update the thermally coupled topology graph. In each graph convolution operation, the attribute vector of the current node is updated by a weighted average of the attribute vectors of its neighboring nodes. The graph convolution formula is defined as follows: Where H(l) represents the node feature matrix of the l-th layer, A is the adjacency matrix, D is the degree matrix, W(l) is the trainable weight matrix of the l-th layer, and σ is the non-linear activation function, using a modified linear unit function. The graph convolution process iterates through 3 layers, ultimately obtaining the context-aware attribute vectors of all nodes in the graph structure.

[0038] Meanwhile, for each edge eij, its weight contribution in each convolutional update is calculated, and the cumulative heat conduction contribution value between node i and node j is obtained as the energy propagation intensity index in the edge attributes.

[0039] After training the graph neural network, the final heat conduction contribution value of each edge is extracted and a two-dimensional matrix M is constructed according to the node number, where Mij represents the heat conduction path weight from node i to node j. This heat conduction path weight matrix not only contains spatial conduction intensity information, but also the correlation pattern in the temperature rise evolution process. It serves as input data for the subsequent temperature rise evolution prediction process to infer the propagation path of potential heat sources and the risk level distribution.

[0040] Please see Figure 2 As shown, the micro-temperature rise evolution prediction module is used to input the heat conduction path weight matrix into the micro-temperature rise evolution prediction model, calculate the heat accumulation rate of potential hidden danger points, and generate an internal temperature rise risk probability map.

[0041] The heat conduction path weight matrix obtained in the previous processing step is fused with the historical temperature gradient data of each node. The temperature gradient is defined as the rate of temperature change between two adjacent time points, in degrees Celsius per second. Let Ti(t) be the temperature of node i at time t, then the temperature gradient... For each time step, the temperature gradient value of each node at that time is weighted and combined with its heat conduction path weight (obtained from the weight matrix) to form a dynamic sequence of heat migration. This sequence captures the spatial propagation trend of heat flow and the temporal characteristics of temperature changes at each node, constituting the input feature set of the subsequent prediction model.

[0042] In this embodiment, a gated recurrent unit neural network model is used to predict the microscopic temperature rise evolution. This model has strong time series modeling capabilities and can learn the complex dependency between nodal temperature rise changes and heat conduction.

[0043] The input data is the aforementioned dynamic sequence of heat migration, with a sequence length of 60 time steps, each step corresponding to 1 second. The model structure includes an input layer, a gated recurrent neural network layer with 128 units, and a fully connected output layer, the output of which is the predicted temperature rise sequence for each node in the next 30 seconds.

[0044] The loss function is the mean squared error function, the optimization algorithm uses the adaptive moment estimation method, the training rounds are 200, and the training data comes from temperature rise data samples under actual operating conditions.

[0045] Based on the predicted temperature rise sequence, the rate of heat accumulation at each node within a specified time window is calculated. This rate is defined as the definite integral of the temperature rise curve within that window, as shown in the following formula: The heat accumulation rate Qi = tt + ΔtTi(τ)·dτ; where Ti(τ) is the predicted temperature curve, and Δt is the window length, set to 30 seconds in this embodiment. Numerically, the trapezoidal integral method is used for approximate calculation. This accumulation rate is used to characterize the heat growth trend of the node in the short term and is a key indicator for assessing potential thermal risks.

[0046] For each thermal node, the heat accumulation rate under normal operating conditions (i.e., no faults and no alarms) in historical operating data is traced back to form a dataset with node number index. Each node contains at least 1,000 rate samples for modeling reference baseline.

[0047] A probability density function model of the velocity distribution is constructed using the kernel density estimation method. The Gaussian kernel function is chosen as the kernel function, and its expression is as follows: Where n is the number of samples, h is the bandwidth parameter, and K is the Gaussian function. The bandwidth parameter h is set according to the Silverman rule, that is: , where σ is the sample standard deviation.

[0048] Substitute the predicted current heat accumulation rate value Qi into the probability density function fi(x) corresponding to the node to calculate its probability density value fi(Qi). If this value is significantly lower than the principal density interval of the normal rate of the node (e.g., fi(Qi) < μ-2σ), it is judged as an abnormal deviation.

[0049] The deviation indices of all nodes (i.e., the logarithmic reciprocal of the probability density values) are normalized and mapped to color levels. A two-dimensional coordinate graph is used to display the position and risk level of each node in the equipment's spatial structure, ultimately forming an internal temperature rise risk probability map, which is used to visualize the distribution characteristics of internal thermal risks in the equipment.

[0050] This map can not only be used to quickly locate abnormal heat sources, but also provide a basis for subsequent early warning mechanisms and maintenance strategies.

[0051] The dynamic sampling and control module divides high-risk monitoring zones based on the internal temperature rise risk probability map and adjusts the sampling density and frequency of the surface temperature of the explosion-proof electrical equipment.

[0052] First, obtain the internal temperature rise risk probability map generated in the previous module. The risk level of each thermosensitive node in the map is represented numerically, ranging from 0 to 1. Risk level intervals are then determined based on this numerical value. Nodes with a risk value greater than 0.7 are classified as high-risk monitoring nodes; Nodes with risk values ​​between 0.3 and 0.7 are classified as medium-risk monitoring nodes; Nodes with a risk value of 0.3 or less are classified as low-risk monitoring nodes.

[0053] Based on the spatial coordinates of nodes, the internal space of the equipment is divided into three monitoring areas: high, medium, and low risk, according to node clustering. Density-based spatial clustering is used to spatially aggregate neighboring nodes with similar risk levels, ultimately resulting in a monitoring area division map with risk labels.

[0054] In high-risk monitoring areas, to promptly capture localized temperature rise anomalies, the system automatically executes an enhanced sampling strategy, which includes the following two aspects: The sampling frequency has been increased from once per second in the basic configuration to 10 times per second in order to improve the temporal resolution of heat changes. The sampling density is enhanced by redeploying supplementary thermistors in the area, so that the distribution density of sensing points reaches one temperature measurement point per 100 square millimeters, ensuring that the spatial resolution meets the requirements for local overheat detection.

[0055] The above strategy is transmitted to the edge acquisition terminal via instructions, and the sampling strategy is updated in real time by the local scheduler.

[0056] The medium-risk monitoring area will continue to use the default sampling strategy, which is to keep the sampling frequency at once per second and the sensor deployment density at one sampling point per 200 square millimeters to ensure the integrity of the monitoring coverage.

[0057] In low-risk monitoring areas, to conserve processing resources and reduce power consumption, a frequency reduction and sparse sampling strategy is implemented, as follows: Reduce sampling frequency: Set to sample once every 10 seconds; The sampling density was reduced, and the deployment density was adjusted to one sampling point per 400 square millimeters, maintaining only the basic thermal monitoring coverage capability.

[0058] This tiered control strategy automatically generates a scheduling table using a dynamic sampling and scheduling algorithm, which is then written into the edge acquisition control program for execution in real time. If the risk level changes subsequently, the system will trigger a reclassification process and update the regional strategy accordingly.

[0059] The dynamic thermal field construction module locates potential heat sources and forms a dynamic thermal field map when the temperature rise rate of any node in the internal temperature rise risk probability map exceeds a preset safety threshold.

[0060] First, based on the internal temperature rise risk probability map, it is determined whether the temperature rise rate of any node exceeds a preset safety threshold. This safety threshold is defined as the mean temperature rise rate of the node under normal operating conditions plus three standard deviations to ensure statistical significance. When the temperature rise rate of a node exceeds this threshold, the heat source location process is triggered.

[0061] By reading the position of the node in the device's three-dimensional structural coordinates, the temperature rise gradient of its surrounding neighboring nodes is obtained. The temperature rise gradient is calculated by dividing the temperature difference between adjacent time points by the time interval. Using a directional analysis method, the main heat propagation direction is calculated based on the direction and magnitude of the gradient vector, and the point with the maximum gradient along this direction is selected as the potential heat source center location. Directional analysis is achieved through vector angle calculation, that is, the direction with the smallest angle to the gradient vector is selected as the main heat flow direction.

[0062] The determined center location of the heat source is set as the initial core point for thermal field reconstruction. The spatial distances from the core point to all nodes are calculated using the Euclidean distance formula through the three-dimensional structural model of the equipment.

[0063] Then, the heat propagation intensity at each node is calculated using the heat conduction path weight matrix. The propagation intensity is defined as the heat conduction weight multiplied by the distance attenuation factor. The distance attenuation factor is represented by an exponential attenuation function, which takes the form: propagation intensity equals heat conduction weight multiplied by an exponential function (negative distance divided by the attenuation coefficient). The attenuation coefficient is set based on the thermal conductivity of the equipment material; in this embodiment, it is set to 200 mm.

[0064] The propagation intensity is used to describe the degree of influence of the heat source on different nodes at the current moment.

[0065] To obtain continuous temperature field data covering the internal space of the equipment, the heat propagation intensity of discrete nodes is processed to be continuous. This embodiment employs a radial basis function interpolation method, using a quadratic radial basis function, expressed as the square root of the sum of the squared distances between nodes and the squared smoothing factor. The smoothing factor is set to 10% of the average distance between nodes.

[0066] Based on this radial basis function, the heat propagation intensity at each node is interpolated to generate a temperature field distribution covering the entire three-dimensional space inside the device. The interpolation results can reflect the spatial attenuation trend of heat propagation from the core point outwards.

[0067] Based on the continuous temperature field data described above, a dynamic thermal field map is constructed. The intensity values ​​of the map are displayed using a color gradient according to the degree of temperature rise in space, with higher temperature rise areas represented in red and lower temperature rise areas represented in blue.

[0068] To present the evolution trend of the heat source, the thermal field data is updated at fixed time intervals. This embodiment uses a time resolution of updating once per second. The new temperature field data is generated from the latest nodal temperature rise values ​​and propagation intensity calculation results, realizing the dynamic display of changes in the thermal field.

[0069] Warning level output module: Based on the dynamic thermal field spectrum and the preset failure mechanism database, the module compares the spectrum and outputs the potential thermal fault type and warning level.

[0070] First, the dynamic thermal field map is divided into multiple temperature field snapshots at set time intervals (every 10 seconds in this embodiment). Each snapshot corresponds to the spatial temperature distribution of the device at a certain moment.

[0071] For each snapshot, the following four features are extracted from its three-dimensional temperature data to form a temperature distribution feature vector: Maximum temperature: The highest temperature in the thermal field, expressed in degrees Celsius; Average temperature gradient: the average rate of temperature rise across all nodes, expressed in degrees Celsius per second; Thermal center offset: The spatial distance between the current thermal center point and the thermal center point in the previous snapshot, calculated as three-dimensional Euclidean distance; Heat transfer rate: the speed at which a thermal center moves per unit time, calculated by dividing the offset by the snapshot time interval, and measured in millimeters per second.

[0072] The above four types of feature values ​​are combined into a feature vector of length 4, which is used for subsequent comparison with the fault sample map.

[0073] The pre-defined failure mechanism database stores a large amount of graphical data on historical thermal failure cases. Each record in the database includes: Fault type label (e.g., overheating of terminal blocks, loose busbars, aging cables, etc.); A sequence of snapshots of the thermal field during the corresponding time period; Fault confirmation information and processing results.

[0074] Select the sample data that matches the current monitoring equipment type, and extract the corresponding feature vector for each fault sample map in the manner described above.

[0075] All sample feature vectors constitute a multidimensional feature template set. Each template contains a feature vector and a corresponding fault type label, which are used for subsequent similarity matching.

[0076] The similarity between the current thermal field snapshot feature vector and all template vectors in the fault feature template set is calculated. Euclidean distance is used as the similarity metric, and the specific calculation method is: D=[(A1-B1)²+(A2-B2)²+(A3-B3)²+(A4-B4)²], where A is the current feature vector and B is the template feature vector.

[0077] The template with the smallest distance is selected. If this smallest distance is less than the identification threshold Tr, the current thermal anomaly behavior is determined to belong to the fault type identified by that template. The identification threshold Tr is an empirically set value, determined by combining the sample mean and standard deviation. In this embodiment, it is taken as the historical sample distance mean minus one standard deviation.

[0078] If no template matching result meets the threshold condition, it is marked as an unknown fault type and enters the manual review queue or triggers a deep learning supplementary model for secondary identification.

[0079] After identifying the fault type, the warning level is determined based on the following three indicators: Historical Severity Level: The severity level of this fault type in the historical database, categorized as mild, moderate, and severe; Fault evolution time: The cumulative time that the current thermal anomaly has lasted, in seconds; Thermal expansion range: The spatial volume of the area affected by the current temperature rise, in cubic millimeters.

[0080] The above three parameters are input into the preset hierarchical rule judgment model. This embodiment uses a segmented judgment method: If the severity level is severe, the evolution time exceeds 300 seconds, and the thermal expansion range exceeds 20% of the total equipment volume, then a Level 1 warning will be issued. If the severity is moderate, and the duration exceeds 180 seconds and the affected area exceeds 10%, a Level 2 warning will be issued. All other situations are considered as Level 3 or low-level warnings.

[0081] The final warning level, along with the fault type, will be uploaded to the monitoring terminal for control response triggering and maintenance task scheduling.

[0082] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A real-time monitoring system for internal temperature rise of explosion-proof electrical equipment, characterized in that, include: The temperature data acquisition module acquires the raw temperature signals of multiple thermal nodes during the operation of the explosion-proof electrical equipment, and collects temperature change data in key areas on the surface of the explosion-proof electrical equipment. The feature extraction module performs nonlinear temperature perturbation decoupling processing on the collected temperature change data based on time-frequency joint deconstruction, and extracts the temperature rise characteristic change trend induced by potential internal heat sources. The thermal anomaly coupling modeling module constructs a multi-node thermal anomaly coupling model based on the temperature rise characteristic change trend, and uses a graph neural network to establish the heat conduction path weight matrix between nodes. The micro-temperature rise evolution prediction module inputs the heat conduction path weight matrix into the micro-temperature rise evolution prediction model, calculates the heat accumulation rate of potential hidden danger points, and generates an internal temperature rise risk probability map. The dynamic sampling and control module divides high-risk monitoring zones based on the internal temperature rise risk probability map and adjusts the sampling density and frequency of the surface temperature of the explosion-proof electrical equipment. The dynamic thermal field construction module locates potential heat sources and forms a dynamic thermal field map when the temperature rise rate of any node in the internal temperature rise risk probability map exceeds a preset safety threshold. The warning level output module compares the dynamic thermal field spectrum with the preset failure mechanism database to output the potential thermal fault type and warning level.

2. The real-time temperature rise monitoring system for the internal environment of explosion-proof electrical equipment according to claim 1, characterized in that, The feature extraction module includes: Segmented sampling of temperature change data in the time domain is performed to extract transient temperature rise fluctuation segments from continuous temperature change curves; Based on the transient temperature rise fluctuation segment, the frequency components at different scales are decomposed to obtain the multi-scale frequency distribution matrix of temperature disturbance; The nonlinear amplitude response curve of temperature disturbance is calculated based on the multi-scale frequency distribution matrix, and the characteristic temperature rise component with the highest correlation to the internal potential heat source is extracted.

3. The real-time temperature rise monitoring system for the internal temperature of explosion-proof electrical equipment according to claim 1, characterized in that, The thermal anomaly coupling modeling module includes: The temperature rise characteristic change trend of each thermal node is aligned with the data in a time synchronization manner, and an initial node set with node temperature gradient as the attribute vector is constructed. Based on the spatial position and thermal conductivity of the thermal nodes, the initial thermal conduction connection relationship is generated using the distance decay function, forming a multi-node thermally coupled topology graph containing node and edge attributes. The multi-node thermally coupled topology graph is input into the graph neural network model. The node attribute vectors are updated through graph convolution operations, and the heat conduction contribution value of each edge after multiple iterations is calculated. Construct a heat conduction path weight matrix based on the updated node attribute vectors.

4. The real-time temperature rise monitoring system for the internal temperature of explosion-proof electrical equipment according to claim 1, characterized in that, The microscopic temperature rise evolution prediction module includes: The heat conduction path weight matrix is ​​fused with the historical temperature gradient data of each node to construct a dynamic sequence of heat migration, which serves as the input feature set for the prediction model. A micro-temperature rise evolution prediction model is constructed based on a gated recurrent unit neural network. After inputting the feature set, the temperature rise trend of each node in the future period is obtained through time series learning. The heat accumulation rate of each node within a specified time window is calculated based on the prediction results. This is the integral of the product of the continuous temperature rise value and the time interval within the window. Based on the deviation measurement of the heat accumulation rate from the statistical distribution under historical normal conditions, a probability density function model is used to generate an internal temperature rise risk probability map.

5. A real-time monitoring system for internal temperature rise of explosion-proof electrical equipment according to claim 4, characterized in that, The generation of the internal temperature rise risk probability map includes: Historical data on the heat accumulation rate of each thermal node is backtracked to construct a rate distribution dataset under normal operating conditions of the equipment, and indexed and classified by node number. Based on the historical rate distribution of each node, the corresponding probability density function model is constructed using the kernel density estimation method. Substitute the currently predicted heat accumulation rate into the probability density function corresponding to each node, calculate its probability density value in the normal distribution, and evaluate its deviation accordingly. The deviation of all nodes is mapped onto a two-dimensional spatial location map using color gradients, forming an internal temperature rise risk probability map that characterizes the thermal anomaly risk level of each node.

6. The real-time temperature rise monitoring system inside explosion-proof electrical equipment according to claim 1, characterized in that, The dynamic sampling and control module includes: Based on the risk level of each thermally sensitive node in the internal temperature rise risk probability map, the internal space of the equipment is divided into high-risk monitoring zone, medium-risk monitoring zone and low-risk monitoring zone; Enhanced sampling strategies were implemented for nodes within high-risk monitoring areas; The default sampling strategy is maintained for medium-risk areas, while the frequency reduction strategy is implemented for low-risk areas.

7. The real-time temperature rise monitoring system for the internal temperature of explosion-proof electrical equipment according to claim 1, characterized in that, The dynamic thermal field construction module includes: When the rate of temperature rise of any node in the internal temperature rise risk probability map exceeds the preset safety threshold, the center location of the potential heat source is determined by performing directional analysis on the temperature rise gradient changes of the surrounding adjacent nodes based on the node's position in the three-dimensional structural coordinates of the equipment. The center of the heat source is taken as the initial core point for thermal field reconstruction. The spatial distance from the core point to all nodes is obtained, and the instantaneous heat propagation intensity of each node is calculated by combining the heat conduction path weight matrix. The heat transfer intensity of each node is continuously processed to generate temperature field distribution data covering the internal space of the equipment; A dynamic thermodynamic field map is constructed based on continuous temperature field data, and the thermodynamic field intensity is updated in a time series manner to form a dynamic thermodynamic field map that can reflect the evolution trend of potential heat sources.

8. The real-time temperature rise monitoring system for the internal environment of explosion-proof electrical equipment according to claim 1, characterized in that, The warning level output module includes: The dynamic thermal field map is split into multiple temperature field snapshots according to the time series, and the temperature distribution feature vector in each snapshot is extracted, including the maximum temperature value, average temperature gradient, thermal center offset and propagation rate. Retrieve fault sample maps that match the current equipment type from the preset failure mechanism database, and establish a multi-dimensional feature template set containing typical thermal failure modes; The minimum Euclidean distance method is used to compare the similarity between the current feature vector and the template set. If the distance is lower than the set recognition threshold, it is determined to be a known thermal fault type. Based on the historical severity level of the fault type, the fault evolution time, and the spatial expansion range, the corresponding warning level is output in combination with the set classification rules.

Images