Explosive safety prediction method and system based on digital twin model

Abstract

The invention relates to the technical field of safety prediction, in particular to an explosive safety prediction method and system based on a digital twin model. The method comprises the following steps: acquiring original explosive data and extracting particle size distribution data, constructing an equal-proportion simulation explosive model based on the data, recording an explosive molding process, extracting and marking working condition characteristics to obtain explosive working condition data of each stage, and performing real-time stability analysis to obtain a simulation explosive model. The method comprises the steps of generating explosive real-time stability data, building a stability environment field, conducting stacking storage simulation on a simulation explosive model, obtaining simulation stacking data, deducing stacking bearing pressure data, conducting real-time stacking safety evaluation according to the stacking bearing pressure and working condition data, reckoning stacking cumulative bearing pressure, and conducting stacking safety evaluation according to the stacking bearing pressure and the working condition data. And continuous pressure deformation prediction is carried out by using stacked accumulated bearing pressure, and comprehensive safety evaluation is carried out in combination with real-time safety. The explosive safety prediction method is more efficient and safer.

Description

Technical Field

[0001] This invention relates to the field of safety prediction technology, and in particular to a method and system for predicting the safety of explosives based on a digital twin model. Background Technology

[0002] Traditional methods for predicting explosive safety based on digital twin models often rely on experience and static models for safety assessment, lacking dynamic monitoring of explosive performance and environmental changes. This leads to the inability to identify potential safety hazards in a timely manner in practical applications. Especially in high-risk blasting operations, the stability and stacking safety of explosives are often overlooked, significantly increasing the risk of accidents. Existing technologies often fail to conduct comprehensive data analysis when dealing with explosive particle size distribution and molding process characteristics, resulting in insufficient accuracy and reliability of the models, which in turn affects the effectiveness of safety assessments. Traditional explosive stacking simulation methods typically lack the integration of real-time operating data and fail to establish a dynamic stability environment field, making the assessment of the pressure borne by the explosive stack unable to reflect the actual situation. Especially under complex storage and transportation conditions, the deformation and stability changes of explosives are difficult to predict. The lack of scientific assessment tools makes it difficult to implement safety management measures. Research on stacking safety assessment often relies on single static data and fails to comprehensively consider the impact of multiple operating conditions on explosive performance. Summary of the Invention

[0003] Therefore, it is necessary to provide a method and system for predicting the safety of explosives based on a digital twin model to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, the explosive safety prediction method based on a digital twin model includes the following steps:

[0005] Step S1: Obtain the raw explosive data; extract the particle size distribution data from the raw explosive data; construct a scaled-down simulation explosive model based on the particle size distribution data.

[0006] Step S2: Record the explosive forming process; extract the working condition characteristics in the explosive forming process, and label the explosive working condition data of each stage of the explosive forming process based on the working condition characteristics;

[0007] Step S3: Perform real-time stability analysis of the explosive using explosive working condition data to obtain real-time stability data of the explosive; construct the explosive stability environment field based on the real-time stability data of the explosive;

[0008] Step S4: Simulate the storage of explosives in a simulated explosive stacking environment based on the explosive stability environment to obtain simulated explosive stacking data; infer the pressure data that the explosive stack will withstand based on the simulated explosive stacking data;

[0009] Step S5: Perform a real-time stacking safety assessment based on the pressure borne by the explosive stack and the explosive working condition data to generate real-time explosive stacking safety; calculate the cumulative pressure borne by the stack based on the explosive stacking pressure data;

[0010] Step S6: Based on the cumulative pressure of the stack, predict the continuous pressure deformation of the simulated explosive model to obtain the predicted explosive deformation data; predict the continuous stacking safety based on the predicted explosive deformation data, and conduct a safety assessment in combination with the real-time explosive stacking safety.

[0011] This invention ensures the accuracy and comprehensiveness of raw explosive data by acquiring it. The extracted particle size distribution data provides the necessary foundation for constructing the simulation model. The scaled-down simulated explosive model generated based on this data achieves accurate simulation of explosive properties, significantly improving the model's reliability and practicality. Recording the explosive forming process ensures comprehensive analysis of the process characteristics at each stage. The extracted operating condition characteristics provide crucial support for the explosive's stability assessment. Real-time stability analysis, based on dynamic monitoring of operating condition data, makes the assessment of explosive stability more scientific and accurate. The constructed stability environment field provides environmental basis for subsequent stacking and storage simulation. The combination of explosive stacking and storage simulation and the stability environment field effectively reflects the pressure borne by the stack under different operating conditions. The inferred pressure data from the stack provides important parameters for safety assessment. Real-time stack safety assessment, combined with operating condition data, enhances the scientific nature and effectiveness of management decisions. The calculation of cumulative pressure on the stack provides data support for long-term safety. The prediction of continuous pressure deformation, combined with real-time data, provides an early warning mechanism for the long-term storage safety of explosives. The resulting comprehensive safety assessment system promotes the intelligentization and modernization of explosives management, significantly improves the industry's safety level, ensures the stability and safety of explosives under various usage conditions, provides strong technical support for the safety management of future explosive applications, promotes technological progress and standardized development in the explosives industry, and ensures the efficiency and safety of explosives in production, storage, and use.

[0012] The present invention also provides an explosive safety prediction system based on a digital twin model, used to execute the explosive safety prediction method based on a digital twin model as described above. The explosive safety prediction system based on a digital twin model includes:

[0013] The model simulation module is used to acquire raw explosive data; extract particle size distribution data from the raw explosive data; and construct a scaled-down simulated explosive model based on the particle size distribution data.

[0014] The working condition recording module is used to record the explosive forming process; extract the working condition characteristics in the explosive forming process, and annotate the explosive working condition data of each stage of the explosive forming process based on the working condition characteristics.

[0015] The stability analysis module is used to perform real-time stability analysis of explosives using explosive operating condition data, thereby obtaining real-time stability data of the explosives; and to construct the stability environment field of the explosives based on the real-time stability data of the explosives.

[0016] The stacking simulation module is used to simulate the stacking and storage of explosives based on the explosive stability environment field to obtain simulated explosive stacking data; and to infer the pressure data that the explosive stacks bear based on the simulated explosive stacking data.

[0017] The real-time safety assessment module is used to perform real-time stack safety assessment based on the pressure borne by the explosive stack and the explosive working condition data, so as to generate real-time explosive stack safety; and to calculate the cumulative pressure borne by the stack through the explosive stack pressure data.

[0018] The safety joint assessment module is used to predict the continuous pressure deformation of the simulated explosive model based on the cumulative pressure of the stack, so as to obtain the predicted explosive deformation data; based on the predicted explosive deformation data, the continuous stack safety is predicted, and the safety assessment is carried out in combination with the real-time explosive stack safety.

[0019] This invention achieves comprehensive acquisition of raw explosive data through a model simulation module, ensuring data integrity and accuracy. The extracted particle size distribution data provides a scientific basis for constructing the simulated explosive model. The generated scaled-down simulated explosive model realistically reproduces the explosive's characteristics, significantly improving the reliability and practicality of the simulation. The working condition recording module comprehensively records the explosive forming process, and the extracted working condition characteristics provide crucial support for subsequent stability analysis. The real-time stability analysis module enhances the dynamic assessment of explosive safety through real-time monitoring of explosive working condition data. The constructed stability environment field provides a scientific basis for simulating stacked storage. The stacking simulation module, combined with the environmental field, effectively reflects the pressure under different stacking conditions, and infers... Stack pressure data provides key parameters for safety assessment. The real-time safety assessment module combines operating data and pressure to conduct a comprehensive stack safety assessment and generate a real-time safety report. The calculation of cumulative stack pressure provides data support for long-term safety. The joint safety assessment module can predict continuous pressure deformation, providing early warning for the long-term storage and use of explosives. The resulting comprehensive safety assessment system promotes the intelligentization and modernization of explosives management, significantly improves the industry's safety standards and technical level, ensures the stability and safety of explosives in various environments, provides strong technical support for the safety management of future explosive applications, promotes the technological progress and standardized development of the explosives industry, and enhances the overall safety management capabilities and efficiency. Attached Figure Description

[0020] Figure 1This is a flowchart illustrating the steps of a method for predicting the safety of explosives based on a digital twin model.

[0021] Figure 2 This is a detailed flowchart illustrating the implementation steps of step S2;

[0022] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0023] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0024] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0025] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0026] To achieve the above objectives, please refer to Figures 1 to 2 A method for predicting the safety of explosives based on a digital twin model includes the following steps:

[0027] Step S1: Obtain the raw explosive data; extract the particle size distribution data from the raw explosive data; construct a scaled-down simulation explosive model based on the particle size distribution data.

[0028] In this embodiment, a multi-channel laser particle size analyzer is installed in the production line. The Analyzer continuously collects particle size data of explosive raw materials passing over the conveyor belt at a scanning frequency of 10Hz. The single sampling results undergo three multi-segment screening and calibration processes. First, the particle size measurement range is set between 0.5 μm and 1500 μm, and a total of 300 particle size channels are constructed with each 5 μm particle size interval as a separate interval. Then, the collected raw data is fitted with a multi-interval distribution, and a back-projection calibration strategy is used to control the error range within 2%. Based on the acquired particle size distribution data, a 3D modeling module is used with the OpenCASCADE modeling kernel driven by Python scripts to construct a simulated explosive model. The model is scaled at a volume ratio of 1:100. Particles in each size segment are stacked in the simulation space according to the Poisson distribution rule. Contact detectors based on surface normal coincidence constraints are used to correct particle contact boundaries during the modeling process. Finally, the 3D simulated explosive model data is obtained. The data is stored in structured STL (Stereolithography) format with no fewer than 100,000 facets to ensure a simulation accuracy of 0.1 mm.

[0029] Step S2: Record the explosive forming process; extract the working condition characteristics in the explosive forming process, and label the explosive working condition data of each stage of the explosive forming process based on the working condition characteristics;

[0030] In this embodiment, a full-process data acquisition platform configured in the pressurization and filling stage is used. The acquisition cycle is 15 frames per second, recording five core data parameters, including the pressure head application rate, the rate of change of the filling material mass, the internal pressure of the mold cavity, the mold temperature, and the vibration amplitude. The pressure head application rate is limited to between 0.3 and 1.2 mm / s, the rate of change of the filling material mass is limited to no more than 1.8 kg / min, and the mold cavity pressure is maintained between 0.6 and 2.5 MPa. Data acquisition lasts for more than 180 minutes. Based on this, an explosive process timeline is established, with each process node... Using 15 seconds as the minimum granularity, the time axis is segmented into stages using a labeling algorithm. The threshold difference segmentation method is used to divide the entire process time into four stages: pre-loading, steady-state pressurization, final compaction, and stabilization. Then, feature extraction is performed on the working condition data of each stage. The extracted features include three indicators: stability threshold change rate, pressure fluctuation frequency, and loading rate gradient. By setting the sensitivity range of each indicator and constructing an indicator space vector labeling template, the stage process data is labeled with explosive working condition data, and finally, an explosive working condition dataset with a unique working condition identifier for each stage is generated.

[0031] Step S3: Perform real-time stability analysis of the explosive using explosive working condition data to obtain real-time stability data of the explosive; construct the explosive stability environment field based on the real-time stability data of the explosive;

[0032] In this embodiment, the explosive working condition data is reorganized into a structured time series matrix according to stages. The temporal states are extracted using gating units in an LSTM (Long Short-Term Memory) neural network architecture. The model input is a 20-dimensional working condition vector sequence with a step size of 10. The model is trained for 200 epochs using the Adam optimizer with a learning rate of 0.001. The model output is then mapped to a real-time stability score for the explosive. The stability score ranges from 0 to 1, and data with scores less than 0.35 are marked as unstable. All stability scores are input into a three-dimensional stability environment field construction module in the form of a structure. This module generates an environmental field tension domain based on the spatial gradient and temporal variation amplitude of the stability scores. The tension domain is divided using a three-dimensional mesh, with each cell being 1 cm in size. 3 And insert local stability vectors into each cell to form a complete explosive stability environment field.

[0033] Step S4: Simulate the storage of explosives in a simulated explosive stacking environment based on the explosive stability environment to obtain simulated explosive stacking data; infer the pressure data that the explosive stack will withstand based on the simulated explosive stacking data;

[0034] In this embodiment, the environmental field is used as the simulation background. The previously constructed simulated explosive model data is imported, and a physics engine (such as Bullet Physics) is used to perform a stacking simulation. The stacking rule is based on the decreasing particle size hierarchy. The simulation area is set to a three-dimensional space of 30cm×30cm×50cm. The total number of models is controlled between 1200 and 1600. The stacking order is arranged from low to high according to the stability field tension gradient. The simulation time is 30 minutes, and a snapshot of the stacking structure is output every 2 seconds. All simulation snapshots are combined to form simulated explosive stacking data. The local contact force on each explosive model is calculated by the contact point data recorded in the simulation. The pressure superposition structure formed by each stacking layer is deduced by combining the stacking sequence order. The pressure distribution curve is constructed using the Force Chain Propagation Model and archived as explosive stack pressure data.

[0035] Step S5: Perform a real-time stacking safety assessment based on the pressure borne by the explosive stack and the explosive working condition data to generate real-time explosive stacking safety; calculate the cumulative pressure borne by the stack based on the explosive stacking pressure data;

[0036] In this embodiment, based on the aforementioned explosive stack pressure data and explosive operating condition data, a cross-pressure mapping structure is used to assess the safety of the real-time explosive stack state. A pressure peak-operating condition sensitive threshold mapping model is used for coupling processing, and stack stress threshold surface and operating condition disturbance response surface are established respectively. The intersection area is defined as the safety critical surface. A real-time explosive stack safety judgment standard is constructed, and the real-time safety is scored every 3 seconds to generate real-time explosive stack safety score data. At the same time, the stack pressure at all time points is longitudinally superimposed and accumulated. A stress integral algorithm is used to accumulate each unit thickness in the stack height direction, with an integration step size of 2mm and a maximum accumulation layer of 25 layers, thereby generating stack cumulative pressure data.

[0037] Step S6: Based on the cumulative pressure of the stack, predict the continuous pressure deformation of the simulated explosive model to obtain the predicted explosive deformation data; predict the continuous stacking safety based on the predicted explosive deformation data, and conduct a safety assessment in combination with the real-time explosive stacking safety.

[0038] In this embodiment, based on the accumulated pressure data of the stack, the static finite element structural analysis module is used to simulate the continuous pressure loading of the simulated explosive model. The simulation continuous loading time is 3600 seconds, the pressure loading is in 60-second cycles, and the pressure value increases by 0.1 MPa. SolidWorks is used for this simulation. The Simulation module performs stress-strain coupling modeling, dividing the simulated explosive model into a hexahedral element mesh, with each element having a side length no greater than 0.5 mm. The loading direction is adjusted in real time by coupling changes in the pressure loading path and the center of gravity trajectory, thereby obtaining continuous pressure simulation data. Based on this, the aforementioned explosive pressure response data is imported into the deformation response prediction network. A deformation transmission prediction algorithm based on graph convolutional neural networks (GCN) is used to input the continuous pressure simulation data. The input dimensions of each node are the applied pressure, direction angle, and initial material density. The output is the rate of change of deformation per unit time. Predicted explosive deformation data is generated through 10 rounds of prediction iterations. Finally, the predicted explosive deformation data is fused and compared with real-time explosive stacking safety data on both axes. The final evaluation value is corrected based on the hyperbolic stability mapping formula to determine the stacking safety level of the simulated explosive model under the current working conditions, forming complete continuous stacking safety prediction output data.

[0039] Preferably, step S1 includes the following steps:

[0040] Step S11: Obtain raw explosive data; extract the basic particle properties from the raw explosive data;

[0041] Step S12: Perform particle size classification measurement based on the basic physical property parameters of the particles to obtain particle size distribution data. A particle sieving device is used, and the particle size screen specifications are set to a total of 8 sieve segments between 10-500μm.

[0042] Step S13: Construct a particle frequency ratio based on particle size distribution data, wherein the frequency ratio value is limited to within ±5% error range, and the distribution ratio of each particle size segment is controlled within the range of 5% to 30%.

[0043] Step S14: Perform three-dimensional packing simulation on the original explosive data using particle frequency ratio to generate a simulated packing configuration. The compaction degree of the simulated packing is limited to between 62% and 76%, and the size of the simulated unit is controlled within 10 cm. 3 within;

[0044] Step S15: Reconstruct a proportional simulated explosive model based on the simulated filler configuration and the basic physical properties of the particles.

[0045] In this embodiment, the physical properties of the original explosive sample were measured. A thermogravimetric analyzer (TGA) was used to obtain data such as the pyrolysis initiation temperature, mass retention ratio, and residual carbon component percentage of the explosive within the range of 25°C to 300°C under normal pressure. Simultaneously, a laser particle size analyzer (Malvern Mastersizer 3000) was used to measure the average particle size as 186 μm and the particle density as 1.73 g / cm³. 3 The static specific surface area is 0.98 m². 2The specific gravity was measured to be 1.81 using a helium hydrometer. The static friction coefficient of the particles was tested using the inclined plate method, and the result was 0.52. The hygroscopicity of the particles was tested using the vacuum drying method, and the moisture content after constant weight was 1.8%. The particle size of the above explosive particles was measured using a YBS-200 ultrasonic vibrating sieve separator. The sieve separator was equipped with standard sieves of 10μm, 30μm, 50μm, 75μm, 125μm, 200μm, 350μm and 500μm, a total of 8 sieves. The single sample feed amount was 150g, the amplitude was set to 3.5mm, the sieve time was set to 15min, and the motor frequency was 45Hz. After sieve separation, the residual particles on each sieve were weighed and the proportion of each particle size segment was recorded. The statistical results showed that the particle size was mainly distributed between 50μm and 350μm. The 125μm segment accounted for the highest proportion at 22.4%, while the 30μm segment accounted for the lowest at 5.1%. All sieving data were standardized and converted into frequency distribution curves. The particle size frequency ratio was constructed using numerical interpolation fitting with the SciPy library in Python. A continuous curve was fitted to the eight particle size segments using a cubic spline interpolation function, limiting the error fluctuation range to ±5%. A distribution restriction range of 5%–30% was set for each particle size segment, with automatic weighting adjustments for values ​​exceeding the limit to ensure the total particle size frequency sum equals 100%. The completed frequency ratio model was plotted using Matplotlib for visualization and verification of the ratio's accuracy and smoothness. The ratio values ​​were stored as structured JSON format as input templates for subsequent filler particle size simulations. The open-source particle packing simulation tool Package ItascaPFC3D (Particle Flow Code in 3 Dimensions) was used, with the simulation cell size set to 5cm × 2cm × 1cm, approximately 10cm. 3 The internal control requirements are as follows: the total number of particles is set to 3000; the aforementioned particle size frequency data is input; and the compaction parameter is set between 0.62 and 0.76 to automatically fit the compaction state. During the simulation, a periodic boundary condition is used to simulate the homogeneity of the packing environment. The particle collision parameters are set as follows: elastic modulus 2 GPa, Poisson's ratio 0.25, contact friction coefficient 0.5, and simulation duration 6000 steps. The final compaction uniformity is 0.69, and the spatial packing uniformity error is controlled within 3%. The output configuration file is a .obj 3D mesh model. Based on the aforementioned .obj file, the ANSYS SpaceClaim modeling environment is imported, and the particles in the configuration are marked as definable material domains and reassigned according to the original particle basic physical property parameters, including a density set to 1.73 g / cm³. 3The thermal conductivity was set to 0.24 W / (m·K), the elastic modulus to 1.3 GPa, the interparticle adhesion force to 0.2 N, and the interparticle thermal resistance to 2.1 K / W. After the model was built, it was transferred to the Digital Twin Studio platform environment. In the environment, a coordinate system and structural mapping relationship consistent with the real explosive sample were established. The twin mapping and debugging of the model were completed through the VRML interface, so that the simulated explosive model has the ability to be mapped to the same scale as the real model with complete physical, geometric and thermal properties. This outputs a digital twin explosive model data structure that can be used for subsequent molding process simulation and mechanical prediction.

[0046] Preferably, step S2 includes the following steps:

[0047] Step S21: Record the explosive forming process; sort the process nodes in the explosive forming process by time to generate an explosive process timeline, where the smallest unit of each node is 15 seconds, and the total recording period is not less than 180 minutes.

[0048] Step S22: Mark the nodes of the explosive process timeline to generate multiple stage division data, where the stage segment length is set between 5 and 30 minutes;

[0049] Step S23: Identify the working condition characteristics of each stage in the stage zoning data;

[0050] Step S24: Mark the explosive operating condition data of each stage of the explosive forming process according to the operating condition characteristics of each stage, wherein the data sampling frequency should be limited to 1-10Hz.

[0051] In this embodiment, the Moxa ioLogik E1260 industrial-grade remote data acquisition module is selected. This module supports the Modbus TCP communication protocol and can interface with the PLC in real time. It is installed at key control nodes of the molding production line, including the feeding device, mixing device, molding system, vacuum chamber, and temperature control module. The module is set to record process parameters every 15 seconds, including pressure (MPa), temperature (°C), rotation speed (RPM), humidity (%RH), and molding time (s). A synchronous timestamp method is used for unified time marking. The recording period lasts for 192 minutes, covering the entire explosive molding process. All data is connected to the OPC of the central processing control unit. The UA server interface uses a unified clock reference to form the original time series matrix of the process. The sampling interval for each time point is 15 seconds. The time series data is structured using the pandas and numpy libraries in Python. First, the time axis is tiled according to the chronological order. Then, it is divided and marked according to a set stage length range. The stage length is set to a fixed value of 20 minutes. The entire 192-minute process is divided into 9 complete stages and 1 tail stage. Each stage contains 80 15-second time points. The tail stage is 12 minutes long and contains 48 time points. Each data segment is named stage_1 to stage_10. The original time series labels are retained internally, and the zoning criteria are not interpolated or reconstructed. Only boundary identifiers are added to the segment positions for subsequent working condition identification and retrieval. Statistical analysis is performed on the time period of each stage, using standard deviation, maximum value, minimum value, and mean value as primary statistical features. At the same time, the time domain change rate is used as a dynamic feature indicator, where the temperature change rate ΔT is the temperature difference between adjacent time points divided by the time interval of 15 seconds, and the pressure change rate ΔP is processed similarly. Threshold identification criteria are set: if ΔT exceeds 0.3℃ / s, it is marked as a high temperature variable working condition; if ΔP exceeds 0.02MPa / s, it is marked as a high pressure variable working condition; if the temperature stability σT is less than 0.5℃ and the pressure stability σP is less than 0...If the pressure is 0.01 MPa, it is marked as a steady-state operating condition. The time points within each segment are characterized according to the above rules and output as a stage operating condition label sequence. The stage operating condition labels are divided into five categories: high-temperature variation, high-pressure variation, mixed vibration, vacuum holding, and steady-state pressure molding. The label structure maintains a one-to-one correspondence with the time points. This is achieved using NI... The cDAQ-9178 data acquisition chassis and the NI-9215 four-channel voltage input module perform high-frequency data acquisition. Within each stage, the sampling frequency is set to 5Hz, recording 5 time points per second. Five indicators—temperature, pressure, mold position, mixing motor current, and molding cavity pressure—are continuously recorded. All data sampling uses a LabVIEW environment to set the data flow path and simultaneously imports the aforementioned stage condition labels, associating each sampling point with its respective stage and condition type. Data recording continuously covers 192 minutes across 10 stage blocks, ultimately forming a dataset of the entire explosives process with a sampling frequency of 5Hz. This dataset contains over 57,600 process feature records, each including a timestamp, process segment number, condition type, 5 numerical indicators, and corresponding rate of change indicators. This completes the multi-stage condition data labeling for the entire explosives molding process. The measurement error of all acquisition modules does not exceed ±0.5%, the channel response time is less than 1ms, and the sampling consistency is controlled within ±2μs.

[0052] Preferably, step S3 includes the following steps:

[0053] Step S31: Divide the explosive working condition data into time-series windows to obtain explosive working condition time-series segments; construct a continuous working condition sequence based on the explosive working condition time-series segments;

[0054] Step S32: Identify the loading rate change sequence in the continuous working condition sequence; perform explosive stability analysis based on the loading rate change sequence, and extract real-time stability data to obtain real-time explosive stability data;

[0055] Step S33: Determine the steady-state region based on the real-time stability data of the explosive; perform field parameterization on the steady-state region to obtain field parameter data;

[0056] Step S34: Perform scene simulation reconstruction based on field parameters to build an explosive stability environment field.

[0057] In this embodiment, explosive forming condition data with a sampling frequency of 5Hz after preprocessing is selected. This data sequence includes six dimensions: timestamp, condition label, temperature, pressure, mixing motor current, cavity pressure, and mold displacement. A sliding window mechanism is used in the construction of the time sequence window. Each window is set to a length of 10 seconds, with 5 sampling points per second, so each window contains 50 data points. The sliding step is set to 2 seconds, that is, after every 10-second window, the window slides backward for 2 seconds to rebuild the next window. After sliding segmenting the entire 192-minute explosive condition data, approximately 5,700 explosive condition time sequence segments are constructed. Each segment has a uniform matrix structure with a dimension of 50 rows and 6 columns. At the same time, the time period information and condition label information corresponding to each segment are retained. Then, condition label splicing is performed on all segments. Adjacent segments with the same condition label in continuous segments are combined into a continuous condition sequence according to the time dimension. If two adjacent segments If the labels are the same and the time interval is less than 5 seconds, they are merged into a continuous working condition sequence; otherwise, they are marked as the starting point of a new sequence. Finally, a total of 1,240 non-overlapping continuous working condition sequences of explosives are generated. Each segment contains multiple time window segments. Based on the mold displacement data and timestamp data in each time segment, the finite difference method is used to calculate the change in mold feed distance per unit time, and then the corresponding explosive loading rate is estimated. The unit is set to mm / s. Each loading rate change sequence forms a change curve with equal time intervals as the horizontal axis and the loading rate in the current segment as the vertical axis. Then, a weighted smoothing filter algorithm is used to process the loading rate sequence. The filter window is set to 5 points, and the filter function is a weighted average weight function w = [1,2,3,2,1] to enhance the continuity of the change trend and eliminate pseudo-fluctuation interference. In the stability analysis, the first-order difference value of the loading rate is set to be below 0.1 mm / s as the stable section, and above 0.1 mm / s as the stable section.The 3 mm / s range represents the fluctuation zone, with intermediate values ​​indicating a transition zone. Following the aforementioned rules, stability zones are labeled for each change sequence, ultimately forming real-time explosive stability data. This data includes the loading rate values ​​at each time point, fluctuation zone category labels, and stability level identifiers. The stability level identifiers are divided into three categories: stable, critical, and unstable, based on the mean and maximum amplitude range of the change. A continuous stability level segment extraction operation is performed on the real-time explosive stability data using a sliding scan mechanism with a window length of 60 seconds. The frequency of stability level labels is counted in each continuous time slice. If the proportion of stable levels exceeds 80%, it is considered a steady-state region. Boundary information for each steady-state region is recorded according to the start and end times, and all stability feature data within the boundaries are statistically extracted. Specific extraction indicators include the average loading rate during the stable period, pressure stability σP, temperature stability σT, and current fluctuation rate αI. Stability is defined as the standard deviation of the indicator within the steady-state segment, and current fluctuation rate is defined as the ratio of the maximum current difference to the mean. After normalization of the extracted indicators, K-means clustering algorithm is used. Each steady-state region is clustered and classified, with the number of clusters K set to 3. Each steady-state region is ultimately assigned to its corresponding stable field category. Based on this, three types of field feature parameter templates are constructed to form field parameter data for simulation environment construction. This data includes steady-state average eigenvalues, stability variation range, boundary duration, and corresponding operating condition categories. The Unity3D simulation engine is used as the scene reconstruction platform, and a parameter injection module is built using the C# scripting language to inject each type of field parameter data into the digital twin scene model. The digital twin model is based on Solid. The skeleton of the 3D model of the explosive forming equipment built by Works was imported. The model includes a cavity volume change mechanism, a mold feeding system, and a dynamic temperature and pressure sensing module. Simulation scripts were built for each type of stability field, with a time step of 0.2 seconds per frame and a simulation cycle of 90 seconds. During the simulation, the filling status, pressure distribution, and mold response delay parameters inside the cavity were collected in real time, and the field map was visualized and output simultaneously. The field map was constructed based on a 3D heat map, using the filling density as the heat distribution benchmark, with the color range from blue (0 g / cm³). 3 From red to red (maximum filling density), a three-dimensional map of the stability environment is presented. All simulation results are recorded as discrete frame sequence data and spatial distribution matrix, which constitute the basic data for stability environment modeling required for subsequent digital twin prediction models.

[0058] Preferably, step S4, which involves simulating the stacking and storage of explosives based on the explosive stability environment field, includes:

[0059] Based on the explosive stability environment field, the simulated explosive model is mapped to stacked elements to obtain the explosive stacking elements;

[0060] Spatial layout mapping is performed based on the explosive stacking units to obtain the spatial layout structure;

[0061] Determine the size of the simulated explosive model;

[0062] Stacking path planning is performed based on the spatial layout structure and the size of the explosive model to obtain stacking path data;

[0063] Based on the stacking path data and the explosive model size, the simulated explosive model is stacked and stored to obtain the simulated explosive stacking data.

[0064] In this embodiment, the constructed explosive stability environment field is used as the background data source. This environment field is a three-dimensional continuous tensor with dimensions of 200×200×150 mm. An explosive simulation model entity is embedded within this space. The baseline size of the simulation model is set to 50 mm for the side length of each explosive unit. The simulated explosive model is based on a C4 explosive entity as its three-dimensional model, with a regular hexahedral structure. Within the environment field, the field is divided into stacked units of the same size through three-dimensional spatial block mapping. Each stacked unit is a 50 mm × 50 mm × 50 mm cube unit. The mapping algorithm is based on spatial orthogonal partitioning. A spatial index table is constructed sequentially along the x, y, and z axes, with the starting point of the three-dimensional rectangular coordinate axes as the origin. All non-edge... All boundary units participate in the formation of stacked units. A field stability label is attached to each stacked unit. The label comes from the stability level value of the corresponding position in the environmental field parameter tensor. The level is 0 to indicate extremely unstable and 3 to indicate highly stable. Finally, a cluster data of explosive stacked units containing a three-dimensional mapping of spatial location, unit size and stability label is formed. Spatial layout mapping operation is carried out. The spatial layout mapping is based on the above stacked unit cluster data and uses a three-dimensional stability priority algorithm to extract the layout region. First, it scans layer by layer from the lowest layer on the z-axis. All stacked units with a stability level of 3 constitute stable layout candidate blocks. Each candidate block is clustered by its xy plane projection range, and the minimum clustering area threshold is set to 0.For a 25-square-meter area, spatial closure correction was applied to the clustering results. All isolated stacked units that could not be connected were excluded using the eight-neighbor connectivity rule, retaining only continuous stacked blocks. Subsequently, all stackable regions were numbered, and three-dimensional spatial layout structure data was generated according to the numbering order. This structure data includes the volume information, starting coordinates, maximum stacking height, and average internal stability of each stackable unit block. This structure data will serve as the spatial input basis for subsequent stacking path planning and shape matching calculations. Based on the initial geometry of the explosive model, shape parameters were extracted from the simulated explosive model, including the base dimensions. The dimensions, total height, center of gravity position, outer box volume, and minimum circumscribed cube shape are determined. The base dimensions are automatically identified by the CAD modeling system using the longest and widest side of the explosive model's lower surface. The maximum side length of the base is set to not exceed three times the side length of the stacking unit, i.e., 150mm. The total height is measured based on the maximum distance along the z-axis and must not exceed the allowable stacking height in the stable region. The center of gravity position is calculated using the 3D model voxelization and uniform mass density to obtain the 3D coordinates. The outer box volume is obtained using a boundary envelope algorithm, constructing a minimum cuboid using the outermost pole of the explosive model. The minimum circumscribed cube is determined by identifying the optimal rotation angle using a rotation envelope method. The cube envelope obtained after the degree is calculated is used to determine whether each stackable region can completely accommodate the explosive model unit size based on the existing spatial layout structure and the explosive model's shape parameters. If the stackable region size meets the volume and shape requirements of the outer box, it is included in the path candidate set. A three-dimensional path search algorithm is used in the path planning, setting the initial point as the environmental field feeding channel position and the target point as the center point of the stacking region. During the search process, the cost function of each path step is compounded and weighted. The cost function is f(n) = g(n) + h(n) + s(n), where g(n) is the cumulative cost of the current path, and h(n) is the cumulative cost of the current path. The cost of Manhattan distance estimation is performed, where s(n) is a stability penalty term calculated from the average stability level of the area traversed by the path. The lower the stability level, the higher the penalty value. Finally, the path with the minimum total cost is selected as the recommended stacking path. The search results are output as a point sequence to form stacking path data. Each path node contains spatial coordinates and a path sequence number. The explosive stacking evolution environment is constructed using the Unity3D 3D simulation platform. The stacking path data and explosive model body parameters are imported, and the rigid body properties of the explosive model are set using the RigidBody physics engine module, including a density setting of 1.4 g / cm³. 3 The elastic modulus is set to 1.2 × 10⁻⁶. 7Pa, with the friction coefficient set to 0.35, the model placement process uses the stacking path sequence as input to control the model's movement trajectory, moving frame by frame according to the path node order, with each frame's displacement not exceeding 10mm. At the same time, the collision detection module is enabled to prevent models from intersecting. After stacking, the position of each explosive model is fixed, and the final center point coordinates, placement order, and stability level label of the contact surface below each model are recorded. The simulation platform simultaneously outputs the stacked structure image and dynamic data frames of the stacking process, ultimately forming complete simulated explosive stacking data.

[0065] Preferably, step S4, which involves inferring the pressure data of the explosive stack based on the simulated explosive stack data, includes:

[0066] Confirm the center of gravity of the simulated explosive model;

[0067] Based on the simulated explosive stacking data and the centroid of the explosive model, the interaction force data between adjacent simulated explosive models is calculated.

[0068] Determine the explosive stacking framework based on simulated explosive stacking data;

[0069] A stacked pressure conduction chain is constructed using the explosive stacking framework and the aforementioned interaction force data;

[0070] The pressure data of the explosive stack under pressure in each simulated explosive model are inferred from the stack pressure transmission chain.

[0071] In this embodiment, the center of gravity calculation is performed based on the constructed three-dimensional simulated explosive model data. The simulation model uses hexahedral C4 explosive units as the structural basis, and each model is divided into voxel units with a side length of 5mm by a three-dimensional mesh, with the voxel density set to 1.59g / cm³. 3 All voxel units were assigned equal mass values, and the spatial mass centroid of the 3D structure was solved using the voxel centroid stacking method. During the calculation, the OpenVDB voxel data processing tool was used, and the VDBVolumeToPoints module was used to convert all voxel positions and masses into a dense point cloud format. Then, a custom C++ program was used to multiply the x, y, and z coordinates of all point cloud data by the mass value and perform a weighted average, ultimately generating a 3D floating-point coordinate representing the mass centroid of the simulated explosive model. This centroid coordinate is represented in three-axis floating-point form, for example, (25.6, 31.2, 49.8) in millimeters. For each model entity in the simulated explosive stack data, its center coordinates, bottom contact surface index, and centroid position were extracted. The number of supporting units and their spatial positions were determined for the bottom contact surface using a surface contact recognition algorithm. A contact threshold was set as an effective supporting surface if the inter-surface distance was less than 1 mm and the angle between the normal directions was less than 5°. Based on the identified supporting surfaces, the static pressure value experienced by each explosive unit was calculated, with the gravitational acceleration set to 9.8 m / s².2 The weight of a unit is obtained by multiplying its mass by gravity. This weight is then evenly distributed to each contact point on the support surface, and the pressure value is used as the force exerted by that support point on the lower explosive unit. This process is implemented using the computeContactForces module in the PyBullet physics simulation engine. High-precision double-precision floating-point mode is used in the calculation to ensure data accuracy. Finally, the contact surface force data between each adjacent explosive unit is extracted into a triplet structure, containing the force-applying unit index, the force-receiving unit index, and the normal pressure value, in Newtons. A three-dimensional spatial stacking topology diagram is constructed based on the stacking data, with each explosive unit as a node in the diagram. All adjacent contact units are connected by edges to represent the stacking relationship. Each edge is assigned a contact type attribute, including three types: "face-to-face," "edge-to-face," and "point-to-face." "Face-to-face" is defined as a distance of ≥300mm between two explosive models. 2 The contact area is defined as follows: "edge-to-opposite" is defined as a single edge segment with a length ≥20mm in contact with the plane, and "point-to-opposite" represents a single-point support relationship. Based on the above structure, a topology graph data structure is constructed using an adjacency list. Nodes record the model index, coordinate position, and model quality, while edges record the contact type and contact surface size. After construction, the connected_components algorithm from the Boost Graph Library is used to verify the connectivity of the stacked framework, ensuring that all explosive units are in the same connection domain. Based on the aforementioned 3D stacked topology graph and interaction force triplet data, a direction-weighted directed graph model is used to model the pressure transmission path. All forces are set to point from upper-level model nodes to lower-level support nodes, with the pressure magnitude used as the edge weight. If a node has multiple lower-level support units, the applied weight is distributed proportionally according to the support area. For example, the contact area between the bottom surface of explosive model A and models B and C is 80cm². 2 and 120cm 2The allocation ratio is 2:3. Directed weighted edges are constructed for all support paths and added to the transmission graph. A DAG (Directed Acyclic Graph) structure is used to maintain the transmission direction and prevent loops. After construction, a hierarchical traversal algorithm is executed, transmitting pressure values ​​sequentially from the top-level model node down to the next level. The cumulative pressure value of each intermediate node is recorded. A stability label is added as a transmission damping factor in each propagation. If the stability level of the target node is low, a penalty coefficient is added to the corresponding path to reflect pressure deformation sensitivity. This transmission chain ultimately forms a complete pressure propagation spectrum. A top-down propagation path accumulation algorithm is used to calculate the pressure accumulation for each path from the top layer of the stack to any node. The final pressure value borne by a node is obtained by summing the pressures from all paths to that node. The unit is Newton. The transmission weight of each path is set by multiplying the gravity of the starting node by the weighting coefficient of the edge in the path. The weighting coefficient of the edge is determined by the contact area ratio and the stability level of the supporting node. For example, if the supporting surface of a certain supporting path accounts for 60% and the stability level is 3, then the weight of the edge is set to 0.6 × 1.0 = 0.6. If the stability level is 1, then it is multiplied by the stability decay factor of 0.75, resulting in an edge weight of 0.45. All path calculations are performed by the custom_weighted_path_sum function in the networkx graph calculation library. During the processing, it is ensured that there are no repetitions or loops in all transmission paths. Finally, each explosive model entity corresponds to a set of pressure data, which consists of three-dimensional coordinates and pressure values, with the unit uniformly set to Newton.

[0072] Of particular importance is the calculation of the interaction force data between adjacent simulated explosive models based on the simulated explosive stack data and the center of gravity of the explosive model, including:

[0073] Deconstruction and localization are performed based on simulated explosive stacking data to obtain the arrangement position of the simulated explosive model;

[0074] The arrangement of the simulated explosive model is transformed by projection, and the centroid direction vector of the simulated explosive model is analyzed.

[0075] Extract the adjacent boundaries of the arrangement position of the simulated explosive model and the centroid direction vector of the simulated explosive model;

[0076] Perform contact surface geometry reconstruction on adjacent boundaries and calculate the contact area of ​​the model boundaries;

[0077] The contact pressure is estimated based on the contact area at the model boundary, and force vector conversion is performed to obtain the interaction force data between adjacent simulated explosive models.

[0078] In this embodiment, the original 3D model index dataset from the simulated explosive stacking data is called. This dataset contains the model number, initial placement coordinates, and model rotation matrix of all explosive units. It originates from a 3D explosion scene stacking simulation environment. All explosive units are hexahedral C4 voxel structures, each voxel with a side length of 5mm. A spatial point cloud distribution map is constructed using Open3D (Open 3D Data Processing Library), and PCL (Point Cloud) is used. The VoxelGrid filter in the point cloud library performs spatial normalization on the point cloud data. Then, the local coordinates of each model are converted to their spatial position in the global 3D coordinate system using the rotation matrix of each model. The conversion process uses the simulation origin as a reference, employing matrix multiplication to multiply the model's local coordinates (x, y, z) with the rotation matrix to obtain the transformed position vector. This is then combined with a displacement vector to achieve coordinate mapping. The positions of all explosive models are output as 3D floating-point arrays, where the position of each explosive unit is represented by (xi, yi, zi) in millimeters. Using the 3D position coordinates of each explosive model as the center point, the model's center of mass data is retrieved. This center of mass position has been obtained using a voxel-weighted average method. The center of mass direction vector for each model is calculated by subtracting the bottom coordinates of the model from the center of mass coordinates. The reference point coordinates are obtained, and then a 3D projection transformation matrix is ​​established with the vertical direction as the principal axis. The model position and centroid coordinates are represented using a homogeneous coordinate system. A uniform affine projection transformation is applied to each model, projecting it from the 3D coordinate system to a 2D plane, where the 2D plane is taken as the horizontal base (xz plane). During the transformation, the `cv::projectPoints` function in OpenCV is used to convert the coordinates, obtaining 2D projection coordinate pairs (x_proj, z_proj). Simultaneously, the centroid direction vector is standardized to a unit length of 1. This vector is then decomposed into components on the 2D plane to obtain the horizontal and vertical projection angles in radians. A spatial index tree is constructed based on the set of 2D projection coordinates of all models, using a kd-tree (K-Dimensional Tree). The Tree structure accelerates the neighborhood search process. After construction, the adjacency radius is set to 60mm. An adjacency search is performed within this radius for each model, extracting the indices of all adjacent models that are in contact with it or whose distance is less than a set threshold. Then, for each pair of adjacent models, the angle between their centroid direction vectors is analyzed. Models with an angle less than 15 degrees and intersecting voxels at their boundaries are considered to have a true adjacency relationship. Boundary overlap detection is performed, and the Boolean intersection operation in MeshLab is used to calculate the intersection region between the two model meshes. If the projected area of ​​the intersection region is greater than 30mm... 2If a boundary is confirmed as a valid adjacent boundary, it is composed of two model number pairs and their overlapping region indices. All boundary data is output as triplets, including model A number, model B number, and overlapping region index number. The overlapping voxel region of each confirmed adjacent boundary pair is extracted. The Polyhedron_3 mesh Boolean operation in CGAL (Computational Geometry Algorithms Library) is called to perform precise polyhedral reconstruction of the overlapping region. This operation voxels the overlapping mesh regions of the two models, finds their intersection, and reconstructs them into a set of regular polygonal faces. The areas of all faces in this set are then summed to calculate the contact area, in square millimeters. During processing, the minimum side length of each triangular face is set to be no less than 1 mm to avoid mesh fragmentation causing error amplification. The area calculation is performed using the Heron formula, with double-precision floating-point numbers used for face accumulation in each calculation. All contact area data is stored with a correspondence between the adjacent model pair number and the area value. The contact area between each pair of models must be no less than 30 mm². 2 If the value is below this threshold, it is considered an ineffective support contact, and the boundary pair is excluded. The mass value of each model and its applied gravity are then retrieved. The mass value is calculated by multiplying the number of voxels by the mass per voxel (the mass per voxel is 1.59g, and the voxel volume is 125mm²). 3 The result is obtained by multiplying by 9.8 m / s. 2 Obtain the weight, then distribute the weight evenly across all contact boundaries, weighting them according to the contact area ratio. For example, if the total contact area of ​​model A is 1800 mm²... 2 The contact area with model B is 600 mm². 2 The pressure allocated to model B is (600 / 1800) * the weight of model A. This pressure value is the total force acting on the contact boundary, in Newtons. Then, the force vector is projected by combining the model's center of gravity direction vector. The pressure value is multiplied by the unit direction vector to obtain the three-dimensional component form of the force vector (Fx, Fy, Fz). All contact force vectors are processed in batches using NumPy array calculation. Finally, the interaction force data between each pair of adjacent models consists of the model A number, the model B number, and the corresponding three-dimensional force vector, in Newtons.

[0079] Preferably, step S5 includes the following steps:

[0080] Step S51: Determine the completion rate of the explosive working condition data based on the preset standard working condition data to obtain the explosive working condition completion rate data;

[0081] Step S52: Confirm the explosive pressure threshold based on the explosive condition completion data; conduct a real-time stacking safety assessment by using the explosive stacking pressure and the explosive pressure threshold to generate real-time explosive stacking safety data.

[0082] Step S53: Identify pressure change data during the stacking process based on the pressure data of the explosive stack;

[0083] Step S54: Infer the stacked pressure of each simulated explosive model segment by using pressure change data; calculate the cumulative stacked pressure of the simulated explosive model based on the simulated explosive model.

[0084] In this embodiment, a complete standard operating condition dataset is established. This dataset should originate from five basic variables recorded in a standard detonation test experiment conducted at a room temperature of 23 degrees Celsius and a relative humidity of 40%—detonation velocity, detonation pressure, temperature rise curve, stacking density, and stress changes. The standard data is grouped according to the explosive type. In each group, the detonation velocity is controlled within the range of 7000 to 8200 meters per second, the detonation pressure is 25 to 32 gigapascals (GPa), the detonation temperature rise does not exceed 480 degrees Celsius, the standard error of the stacking density does not exceed 0.2 grams per cubic centimeter, and the stress fluctuation frequency needs to be controlled within 200 Hz. The operating condition data of the explosive to be evaluated is input into a tensor model with the same structure as the standard operating condition data, using symmetric... The cosine similarity matching method checks the two tensors, setting a similarity threshold of 0.92. If the similarity between the explosive working condition data and the standard working condition data is greater than or equal to 0.92, the working condition completion rate is considered to be 100%. If the similarity is less than 0.92, a piecewise linear regression function is used to calculate the actual working condition completion rate, and the explosive working condition completion rate data is finally output. The corresponding working condition matching level is extracted from the aforementioned explosive working condition completion rate data as input conditions. An empirical mapping function is set to obtain the corresponding pressure-bearing threshold of the explosive. This function is constructed based on the empirical formula Pth=P0×(1+k×(1―Cm)), where Pth is the pressure-bearing threshold, P0 is the theoretical maximum static pressure of the material, and k is an empirical coefficient with a value of 0.4. Cm is the working condition completion ratio factor, ranging from 0 to 1. After calculating the pressure threshold, it is compared frame by frame with the pressure data acquired in real time during the explosive stacking process. This process uses a laser interferometric pressure sensor array (PDV array) to record the pressure changes during the stacking process at 2 nanosecond intervals. A sliding window mechanism is used to perform moving average processing on the time series data, with a window width of 200 nanoseconds and a step size of 20 nanoseconds. If the average pressure of any window exceeds the pressure threshold by more than 10%, it is determined to be a state of decreased stacking safety; otherwise, it is recorded as a state of normal stacking safety. Finally, the stacking safety status is output as real-time stacking safety data. The raw pressure data sequence was processed using multi-resolution wavelet transform, and a five-level decomposition was performed using the db6 wavelet function. Approximation coefficients and detail coefficients were obtained for each level. The detail coefficient sequences of the third and fourth levels were analyzed in detail. Critical pressure change points were identified using a variance mutation point extraction algorithm. The criterion was that the difference between the means of five consecutive data points was greater than three times the standard deviation of the previous mean. Data within 20 nanoseconds after each extracted mutation point was classified as a single pressure change event. By rearranging and classifying these single events in time sequence, a pressure change map was constructed from the pressure change data throughout the stacking process. Each event in the map includes the start time, end time, and maximum pressure. Five key parameters—intensity, rate of change, and duration of change—are used to input the pressure change map into a structure-mapping neural network model (SMNN). This model consists of a three-layer fully connected network and a bidirectional long short-term memory network (BiLSTM), with an input dimension of 5 and hidden layer neurons of 64, 128, and 64 respectively. The output is the pressure bearing capacity value of each segment of the simulated explosive model. This segmented structure is established according to a preset geometric segmentation strategy, with each segment's length not exceeding 1 / 15 of the total explosive length. Each segment's model number is consistent with the corresponding simulated geometric mesh mapping. The inferred result is the maximum pressure bearing capacity of each segment under a given stacking stress sequence. The pressure-bearing capacity data of each segment is input into the stacking path of the simulated explosive model. Based on the spatial hierarchy, the data from each segment is aggregated upwards using a weighted average formula. The weights are set according to the material density and volume ratio of each segment. The final output is the cumulative pressure data of the simulated explosive model under the complete stacking path. This data can be used to compare and fit with pressure monitoring data of real explosive structures, thereby completing the reverse optimization and calibration of the simulation model's accuracy. No thermal, entropy structure, or dynamic methods are introduced in any calculation process; modeling and inference are based solely on pressure, density, geometry, and stress temporal changes. All parameters are explicitly defined by experimental data and structural configuration.

[0085] Preferably, step S6, which involves predicting the sustained pressure deformation of the simulated explosive model based on the cumulative pressure borne by the stack, includes:

[0086] Material deconstruction is performed on the simulated explosive model to obtain material deconstruction data;

[0087] Pressure response analysis of the simulated explosive model is performed based on material deconstruction data to generate explosive pressure response data;

[0088] By accumulating pressure through stacking, the simulated explosive model is subjected to continuous pressure simulation, thereby generating continuous pressure simulation data.

[0089] Deformation response is predicted based on the pressure response data of explosives and the continuous pressure simulation data, thereby obtaining the predicted explosive deformation data.

[0090] In this embodiment, the simulated explosive model undergoes material deconstruction processing. This deconstruction process employs a layered mapping material identification mechanism. First, the three-dimensional simulated explosive model is voxelized at 1-millimeter intervals along the X, Y, and Z axes, resulting in voxel clusters composed of three-dimensional mesh units. Then, the corresponding material composition table is retrieved using the known original explosive formula number. The material composition table is obtained through X-ray diffraction (XRD) and scanning electron microscopy energy dispersive spectroscopy (SEM-EDS) experiments, with RDX mass fraction as the main variable, combined with the proportions of multiple components such as Al powder, paraffin, and glass fiber, according to different... The density and texture parameters of the same voxel region are matched against a material database. Each voxel region is assigned a specific material label and its set of microstructure parameters, which includes the particle size distribution range (controlled between 5 and 80 micrometers), average particle spacing (less than 3 micrometers), microcrack density (not higher than 0.01 microcracks per cubic millimeter), and binder coverage (greater than 95%). The final output material deconstruction data includes the material number, microparameter set, and macro density value of each voxel element, and a structure-material mapping table is established. The multiphysics coupled finite element tool ANSYS is used. Autodyn constructs a microstructure response model for explosive materials. This model uses the microparticle structure from material deconstruction data as input. For each material number, different pressure-strain curves and yield models are constructed. The RDX region is modeled using the Johnson-Cook yield criterion with parameters set as A = 240 MPa, B = 90 MPa, n = 0.26, C = 0.015, and m = 1.03. The glass fiber region uses a linear elastic-damage model with a Young's modulus of 70 GPa and a Poisson's ratio of 0.21. The binder region uses an elastoviscoplastic constitutive model with a viscous modulus of 25 MPa·s. Through a multibody interaction interface defined in the model, different material regions are coupled and simulated at the interface, utilizing compression loading edges. Boundary conditions were applied to the model using proportional pressure curves ranging from 0 to 200 MPa, with the simulation duration controlled within 2 microseconds. The maximum plastic strain, local compressibility, and residual deformation of each material region were calculated. The cumulative pressure data of the stack obtained in the previous stage was retrieved, and the pressure sequence was input into the material response model for time-stretched simulation. First, a continuous medium simulation model based on the Eulerian grid was constructed. The stack pressure signal was discretized into 500 nanosecond frames along the time dimension, with the total time length set to 5 milliseconds. The stack pressure was applied to the contact surface region of the simulated explosive model using a continuous loading method. The pressure value applied in each frame was equal to the weighted update amount of the pressure value in the previous frame. The weight was set as the ratio of the product of the local voxel strain gradient and its corresponding material impedance. The minimum element size was 0 during the simulation.A 25 mm mesh was used, and the Arbitrary Lagrangian-Eulerian (ALE) algorithm was enabled to prevent mesh collapse. The model recorded four indices in real-time each frame: local displacement, stress distribution, volume change rate, and strain rate. The final output of the sustained pressure simulation data consisted of the stress-strain data of each voxel node evolving over time, including the pressure field response sequence at consecutive time points in a three-dimensional spatial distribution. A spatial deformation propagation model based on a graph neural network (GNN) was constructed. This model used the simulated three-dimensional explosive structure as a node graph structure, with edges representing the mechanical coupling relationship between adjacent voxel units. The node feature vector consisted of the strain difference, displacement difference, and pressure difference of each node at any two adjacent time points in the sustained pressure simulation data. A three-layer graph convolutional network structure was used, with ReLU activation for the node update function in each layer. The first layer has an output dimension of 64, the second layer 128, and the third layer 64. A graph attention mechanism (GAT) is then used to enhance the identification of heterogeneous connected regions, converging the response features of each node into a global prediction function. Finally, the fully connected layer outputs the predicted deformation value of each voxel for the next 1 to 10 milliseconds, organized into complete predicted explosive deformation data in the form of a structural deformation field. This data is in a three-dimensional tensor format, where each voxel corresponds to the axial and radial deformation values ​​over multiple future time periods, in millimeters. During model training, high-frequency mechanical experimental data is used as the ground truth for fitting and optimization. The training batch size is set to 300 epochs, using the Adam optimizer with an initial learning rate of 0.0005 and a batch size of 64. The loss function is a dual constraint function of mean squared error and structural deformation consistency.

[0091] Of particular importance is the simulation of sustained pressure on the simulated explosive model through stacked and accumulated pressure, including:

[0092] The pressure data of the explosive stack is linearly expanded in segments to obtain the pressure distribution of the stack.

[0093] Based on the pressure distribution data of the stack, pressure mapping transformation is performed to generate surface pressure data of the simulated explosive model;

[0094] The nodal pressure input data is obtained by simulating the nodal pressure injection of the simulated explosive model using surface pressure data of the simulated explosive model.

[0095] The time step is loaded based on the node pressure input data to generate continuous pressure data for the simulated explosive model;

[0096] The pressure response trajectory of the simulated explosive model under continuous pressure is tracked, and iterative superposition is performed based on the pressure response trajectory to obtain continuous pressure simulation data.

[0097] In this embodiment, the raw pressure data is divided into multiple continuous segments along the time axis, with each segment corresponding to a physical loading cycle. In the experiment, each loading cycle is set to 2 milliseconds. The stacked pressure data is obtained from multiple loading experiments. Each experiment uses a piezoelectric sensor to record the force change curve at a time interval of 1 microsecond, forming a data segment of 500 points. The least squares method is used to perform linear fitting on each data segment, constructing a linear function P(t) = a*t + b for each segment, where a is the pressure increase rate and b is the initial pressure value. This linear expansion minimizes the squared residual. Using the difference as the objective, linear segments with fitting residuals less than 0.03 MPa are retained. For segments with residuals exceeding the threshold, a double-interval split is performed, and linear approximation is performed again. The final stacked pressure distribution is a linear increment sequence of pressure over multiple time intervals. Each sequence item includes the start and end points of the time interval, fitting parameters a and b, and its error limit. All linear segments are output in a structured manner. The time-pressure linear function is mapped onto the surface of the simulated explosive model. The operation first performs triangular meshing on the surface of the simulated explosive model, and uses the Marching Cubes algorithm to extract isosurfaces from the voxel structure to generate an outer shell model composed of patches with a patch size of approximately 0.With a mesh size of 5 mm and a corresponding grid point count controlled within 500,000, the linear segment pressure function is then distributed to the model shell. Using an area-weighted method, the loading value of each pressure segment is evenly distributed to each triangular facet element on the model's contact surface. The pressure increment per unit time is defined as ΔP = a × Δt, where a is the linear segment pressure increase rate and Δt is the time interval. Spatially, by correcting the normal vector direction of each facet, only facet elements pointing towards the external loading direction are retained for pressure calculation. The final generated simulated explosive model surface pressure data is a multi-dimensional array structure. Each frame records the loading start time, end time, pressure change function parameters, and pressure direction vector information for each facet. A node pressure injection framework corresponding to the simulation model mesh is constructed. First, the entire model is numbered, and each node records its spatial coordinates and the face it belongs to. Based on the panel number and material number, and through the pressure mapping relationship between the panels, the pressure function of each pressure-applied panel is applied to the set of nodes it covers. The projected area value of the stressed area is calculated for the local coordinates of each node. Local pressure interpolation is performed using the real-time calculated value of the pressure function and the ratio of the node area to obtain the pressure value of the node at time t: P_node(t) = P_face(t) * A_node / A_face, where P_face is the pressure per unit area of ​​the panel, A_node is the projected area value of the node, and A_face is the area of ​​the entire panel. A pressure input time series is generated for each node and sampled at a time resolution of 1 microsecond. The node pressure input data is ultimately represented as a three-dimensional array, with each dimension representing the node number, time step, and applied pressure value. The numerical unit is uniformly set to megapascals (MPA). The finite element calculation platform ANSYS is then used. The ExplicitDynamics module inputs the node pressure sequence into the loading control module to construct loading boundary conditions. A time-step control strategy is employed, with each step being 1 microsecond and a total loading time of 10 milliseconds. For each microsecond step, the control module reads the corresponding node pressure value array and applies it to the simulation model nodes. The simulation calculation uses an adaptive time integration method to ensure numerical stability, employing the Newmark-beta integration method for time progression. The control parameters β are set to 0.25 and γ to 0.5. Ensure that the numerical error is less than 10^-5 during each loading step. During the calculation, use the mesh element interpolation mapping method to map the nodal pressure values ​​to the element stress field. The structural solver outputs the stress, strain, and deformation data of the internal nodals of the model at each time step. This process is continuously executed until all time steps are completed, ultimately outputting the continuous pressure data of the simulated explosive model. This data structure includes the force path, total strain increment, and instantaneous displacement vector values ​​of all nodes throughout the entire loading sequence. A three-dimensional time-space pressure trajectory tensor is constructed, with dimensions X×Y×Z×T, where X, Y, and Z are the voxel mesh dimensions of the model, and T is the total number of time frames. The pressure values ​​of each voxel node during the simulation are then pieced together along the time axis to form a complete sequence. The response trajectory of each voxel is defined as P(x,y,z,t). Every 10 frames, the trajectory is locally differencing and multiplied by the impedance factor of the corresponding voxel in the material deconstruction data to update the local pressure residual at the current node. The iteration increment is set to the difference between the previous and current time, ΔP = P(t) - P(t-1). Multiple iterations are performed until the pressure change rate of all nodes is lower than 0.001 MPa per microsecond. The change path of each iteration is recorded, and the total pressure response value is accumulated. Finally, the final pressure of all voxel points is superimposed to form the continuous pressure simulation data. This data is a four-dimensional tensor, with each dimension being spatial coordinates X, Y, Z and time T, with the corresponding unit being MPa, and the data volume is not less than 10. 9 Each record represents a complete stress response history of a specific spatial node at a single time step.

[0098] Preferably, step S6 includes predicting the continuous stacking safety based on the predicted explosive deformation data and assessing the safety in conjunction with the real-time explosive stacking safety, including:

[0099] The simulated explosive model is subjected to continuous pressure based on the cumulative pressure of the stack, thereby observing the continuous deformation data of the explosive model.

[0100] Determine the explosive deformation threshold using continuous deformation data from an explosive model;

[0101] Determine the dangerous deformation distance based on the explosive deformation threshold and predicted explosive deformation data;

[0102] Continuous stacking safety is generated by predicting the distance of dangerous deformation.

[0103] By combining real-time explosive stack safety and real-time explosive stack safety, the safety of explosives is assessed, thereby obtaining explosive safety data.

[0104] In this embodiment, Finite Element Analysis (FEA) software is used to mesh the simulated explosive model, dividing it into small elements to accurately simulate the stress distribution and deformation response of each element. The input stacked cumulative pressure value is used as an external load applied to the model surface and interior. A time-stepping method is used to simulate the continuous application of pressure through multiple consecutive time steps, recording the strain and displacement changes of the explosive model within each time step. This yields the continuous deformation data of the explosive model throughout the entire pressure application period, including linear displacement, shear deformation, and volumetric change. The deformation measurement accuracy reaches the micrometer level, with multi-dimensional parameters and a pressure application time step set to the millisecond level. The pressure range is set to 10 to 100 MPa based on previous working condition data. Statistical analysis methods are used to perform peak analysis on continuous deformation data, extracting the maximum deformation value and its corresponding stress point. The critical deformation empirical value is used as a threshold benchmark. This threshold value is set based on the safe limit deformation obtained from previous laboratory blasting tests, and is usually set as a percentage of the model's maximum allowable deformation. Specifically, the threshold range is 85% to 95% of the elastic limit of the explosive material. The threshold determination process is verified through multiple simulation results. To accurately locate the deformation limit, outlier data points are excluded, and cubic spline interpolation is used to fit the deformation curve. The difference between the maximum deformation in the predicted deformation data and the deformation threshold is calculated, and the calculated difference is called the dangerous deformation distance. This distance represents the numerical range within which the predicted deformation exceeds the safety limit. The Euclidean distance method is used in the calculation, and vector difference operations are performed on the deformation of each node in the simulated explosive model space. The local deformation differences are accumulated through numerical integration to obtain the dangerous deformation distance of the overall model. The numerical range of the dangerous deformation distance is limited to 0 to 5 mm, and the larger the value, the higher the degree of exceeding the limit. The calculation process strictly considers the nonlinear characteristics of materials and boundary condition constraints, and maps the dangerous deformation distance to a safety level index. The safety level adopts a graded numerical system with a value range from 0 to 1, where 0 represents extreme danger and 1 represents complete safety. An inverse mapping relationship is established between the dangerous deformation distance and the safety level, that is, the larger the dangerous deformation distance, the lower the safety level. The mapping process uses an exponential decay function. After inputting the dangerous deformation distance into the function, the corresponding safety level value is output. The parameters of the exponential decay function are adjusted according to historical explosive failure data, and the parameter range is set to a decay rate of 0.5 to 1.5. The mapping results are normalized to ensure the output safety level is between 0 and 1. This safety level is the continuous stacking safety index. Real-time explosive stacking safety data is obtained by collecting real-time pressure and deformation information during explosive stacking through an online sensor system. This data is calculated using a preset safety threshold model, and the real-time safety index value also falls between 0 and 1. Subsequently, a weighted fusion algorithm is applied to the continuous stacking safety and real-time safety indices. Both are multiplied by weight values, with a total weight of 1. The weights are set as follows: real-time safety accounts for 0.6, and continuous safety accounts for 0.4. The fusion process uses a weighted average method, and the product results are summed to form a comprehensive safety index. The final safety data presents a continuous change curve, reflecting the safety level of the current explosive stacking state, which is used to guide subsequent safety monitoring and risk management. During the fusion process, the data of each indicator are aligned in time to eliminate the impact of data latency on the accuracy of the assessment, ensuring the real-time nature and reliability of the comprehensive safety assessment results.

[0105] The present invention also provides an explosive safety prediction system based on a digital twin model, used to execute the explosive safety prediction method based on a digital twin model as described above. The explosive safety prediction system based on a digital twin model includes:

[0106] The model simulation module is used to acquire raw explosive data; extract particle size distribution data from the raw explosive data; and construct a scaled-down simulated explosive model based on the particle size distribution data.

[0107] The working condition recording module is used to record the explosive forming process; extract the working condition characteristics in the explosive forming process, and annotate the explosive working condition data of each stage of the explosive forming process based on the working condition characteristics.

[0108] The stability analysis module is used to perform real-time stability analysis of explosives using explosive operating condition data, thereby obtaining real-time stability data of the explosives; and to construct the stability environment field of the explosives based on the real-time stability data of the explosives.

[0109] The stacking simulation module is used to simulate the stacking and storage of explosives based on the explosive stability environment field to obtain simulated explosive stacking data; and to infer the pressure data that the explosive stacks bear based on the simulated explosive stacking data.

[0110] The real-time safety assessment module is used to perform real-time stack safety assessment based on the pressure borne by the explosive stack and the explosive working condition data, so as to generate real-time explosive stack safety; and to calculate the cumulative pressure borne by the stack through the explosive stack pressure data.

[0111] The safety joint assessment module is used to predict the continuous pressure deformation of the simulated explosive model based on the cumulative pressure of the stack, so as to obtain the predicted explosive deformation data; based on the predicted explosive deformation data, the continuous stack safety is predicted, and the safety assessment is carried out in combination with the real-time explosive stack safety.

[0112] This invention achieves comprehensive acquisition of raw explosive data through a model simulation module, ensuring data integrity and accuracy. The extracted particle size distribution data provides a scientific basis for constructing the simulated explosive model. The generated scaled-down simulated explosive model realistically reproduces the explosive's characteristics, significantly improving the reliability and practicality of the simulation. The working condition recording module comprehensively records the explosive forming process, and the extracted working condition characteristics provide crucial support for subsequent stability analysis. The real-time stability analysis module enhances the dynamic assessment of explosive safety through real-time monitoring of explosive working condition data. The constructed stability environment field provides a scientific basis for simulating stacked storage. The stacking simulation module, combined with the environmental field, effectively reflects the pressure under different stacking conditions, and infers... Stack pressure data provides key parameters for safety assessment. The real-time safety assessment module combines operating data and pressure to conduct a comprehensive stack safety assessment and generate a real-time safety report. The calculation of cumulative stack pressure provides data support for long-term safety. The joint safety assessment module can predict continuous pressure deformation, providing early warning for the long-term storage and use of explosives. The resulting comprehensive safety assessment system promotes the intelligentization and modernization of explosives management, significantly improves the industry's safety standards and technical level, ensures the stability and safety of explosives in various environments, provides strong technical support for the safety management of future explosive applications, promotes the technological progress and standardized development of the explosives industry, and enhances the overall safety management capabilities and efficiency.

[0113] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0114] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for predicting the safety of explosives based on a digital twin model, characterized in that, Includes the following steps: Step S1: Obtain the raw explosive data; extract the particle size distribution data from the raw explosive data; construct a scaled-down simulation explosive model based on the particle size distribution data. Step S2: Record the explosive forming process; Extract the working condition characteristics in the explosive forming process, and label the explosive working condition data of each stage of the explosive forming process based on the working condition characteristics. Step S3: Perform real-time stability analysis of the explosive using explosive working condition data to obtain real-time stability data of the explosive; An explosive stability environment field was constructed based on real-time explosive stability data. Step S4: Simulate the stacking and storage of explosives based on the explosive stability environment field to obtain simulated explosive stacking data; Infer the pressure data of the explosive stack based on simulated explosive stack data; Step S5: Perform a real-time stacking safety assessment based on the pressure borne by the explosive stack and the explosive working condition data to generate a real-time explosive stacking safety data. The cumulative pressure borne by the stack of explosives can be estimated by using data on the pressure borne by the stack of explosives. Step S6: Based on the cumulative pressure borne by the stack, predict the continuous pressure deformation of the simulated explosive model to obtain the predicted explosive deformation data; The safety of continuous stacking is predicted based on the predicted explosive deformation data, and a safety assessment is performed in combination with the real-time explosive stacking safety.

2. The method for predicting the safety of explosives based on a digital twin model according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Obtain raw explosive data; extract the basic particle properties from the raw explosive data; Step S12: Perform particle size classification measurement based on the basic physical property parameters of the particles to obtain particle size distribution data. A particle sieving device is used, and the particle size screen specifications are set to a total of 8 sieve segments between 10-500μm. Step S13: Construct a particle frequency ratio based on particle size distribution data, wherein the frequency ratio value is limited to within ±5% error range, and the distribution ratio of each particle size segment is controlled within the range of 5% to 30%; Step S14: Perform three-dimensional packing simulation on the original explosive data using particle frequency ratio to generate a simulated packing configuration. The compaction degree of the simulated packing is limited to between 62% and 76%, and the size of the simulated unit is controlled within 10 cm. 3 within; Step S15: Reconstruct a proportional simulated explosive model based on the simulated filler configuration and the basic physical properties of the particles.

3. The method for predicting the safety of explosives based on a digital twin model according to claim 1, characterized in that, Step S2 includes the following steps: Step S21: Record the explosive forming process; sort the process nodes in the explosive forming process by time to generate an explosive process timeline, where the smallest unit of each node is 15 seconds, and the total recording period is not less than 180 minutes. Step S22: Mark the nodes of the explosive process timeline to generate multiple stage division data, where the stage segment length is set between 5 and 30 minutes; Step S23: Identify the working condition characteristics of each stage in the stage zoning data; Step S24: Mark the explosive operating condition data of each stage of the explosive forming process according to the operating condition characteristics of each stage, wherein the data sampling frequency should be limited to 1-10Hz.

4. The method for predicting the safety of explosives based on a digital twin model according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: Divide the explosive working condition data into time-series windows to obtain explosive working condition time-series segments; construct a continuous working condition sequence based on the explosive working condition time-series segments; Step S32: Identify the loading rate change sequence in the continuous working condition sequence; perform explosive stability analysis based on the loading rate change sequence, and extract real-time stability data to obtain real-time explosive stability data; Step S33: Determine the steady-state region based on the real-time stability data of the explosive; perform field parameterization on the steady-state region to obtain field parameter data; Step S34: Perform scene simulation reconstruction based on field parameters to build an explosive stability environment field.

5. The method for predicting the safety of explosives based on a digital twin model according to claim 1, characterized in that, Step S4, which involves simulating the stacking and storage of explosives based on the explosive stability environment field, includes: Based on the explosive stability environment field, the simulated explosive model is mapped to stacked elements to obtain the explosive stacking elements; Spatial layout mapping is performed based on the explosive stacking units to obtain the spatial layout structure; Determine the size of the simulated explosive model; Stacking path planning is performed based on the spatial layout structure and the size of the explosive model to obtain stacking path data; Based on the stacking path data and the explosive model size, the simulated explosive model is stacked and stored to obtain the simulated explosive stacking data.

6. The method for predicting the safety of explosives based on a digital twin model according to claim 1, characterized in that, Step S4 involves inferring the pressure data that the explosive stack can withstand based on the simulated explosive stack data, including: Confirm the center of gravity of the simulated explosive model; Based on the simulated explosive stacking data and the center of gravity of the explosive model, the interaction force data between adjacent simulated explosive models is calculated. Determine the explosive stacking framework based on simulated explosive stacking data; A stacked pressure conduction chain is constructed using the explosive stacking framework and the aforementioned interaction force data; The pressure data of the explosive stack under pressure in each simulated explosive model are inferred from the stack pressure transmission chain.

7. The method for predicting the safety of explosives based on a digital twin model according to claim 1, characterized in that, Step S5 includes the following steps: Step S51: Determine the completion rate of the explosive working condition data based on the preset standard working condition data to obtain the explosive working condition completion rate data; Step S52: Confirm the explosive pressure threshold based on the explosive condition completion data; conduct a real-time stacking safety assessment by using the explosive stacking pressure and the explosive pressure threshold to generate real-time explosive stacking safety data. Step S53: Identify pressure change data during the stacking process based on the pressure data of the explosive stack; Step S54: Infer the segmented stacking pressure of each simulated explosive model through pressure change data; calculate the cumulative stacking pressure of the simulated explosive model based on the simulated explosive model.

8. The method for predicting the safety of explosives based on a digital twin model according to claim 1, characterized in that, Step S6, which involves predicting the sustained pressure deformation of the simulated explosive model based on the cumulative pressure borne by the stack, includes: Material deconstruction is performed on the simulated explosive model to obtain material deconstruction data; Pressure response analysis of the simulated explosive model is performed based on material deconstruction data to generate explosive pressure response data; By accumulating pressure through stacking, the simulated explosive model is subjected to continuous pressure simulation, thereby generating continuous pressure simulation data. Deformation response is predicted based on the explosive pressure response data and the continuous pressure simulation data, thereby obtaining the predicted explosive deformation data.

9. The method for predicting the safety of explosives based on a digital twin model according to claim 1, characterized in that, Step S6 involves predicting the continuous stacking safety based on the predicted explosive deformation data, and conducting a safety assessment in conjunction with the real-time explosive stacking safety, including: The simulated explosive model is subjected to continuous pressure based on the cumulative pressure of the stack, thereby observing the continuous deformation data of the explosive model. Determine the explosive deformation threshold using continuous deformation data from an explosive model; Determine the dangerous deformation distance based on the explosive deformation threshold and predicted explosive deformation data; Continuous stacking safety is generated by predicting the distance of dangerous deformation. By combining real-time explosive stack safety and real-time explosive stack safety, the safety of explosives is assessed, thereby obtaining explosive safety data.

10. A system for predicting the safety of explosives based on a digital twin model, characterized in that, For executing the explosive safety prediction method based on a digital twin model as described in claim 1, the explosive safety prediction system based on a digital twin model comprises: The model simulation module is used to acquire raw explosive data; extract particle size distribution data from the raw explosive data; and construct a scaled-down simulated explosive model based on the particle size distribution data. The working condition recording module is used to record the explosive forming process; extract the working condition characteristics in the explosive forming process, and annotate the explosive working condition data of each stage of the explosive forming process based on the working condition characteristics. The stability analysis module is used to perform real-time stability analysis of explosives using explosive operating condition data, thereby obtaining real-time stability data of the explosives; and to construct the stability environment field of the explosives based on the real-time stability data of the explosives. The stacking simulation module is used to simulate the stacking and storage of explosives based on the explosive stability environment field to obtain simulated explosive stacking data; and to infer the pressure data that the explosive stacks bear based on the simulated explosive stacking data. The real-time safety assessment module is used to perform real-time stack safety assessment based on the pressure borne by the explosive stack and the explosive working condition data, so as to generate real-time explosive stack safety; and to calculate the cumulative pressure borne by the stack through the explosive stack pressure data. The safety joint assessment module is used to predict the continuous pressure deformation of the simulated explosive model based on the cumulative pressure of the stack, so as to obtain the predicted explosive deformation data; based on the predicted explosive deformation data, the continuous stack safety is predicted, and the safety assessment is carried out in combination with the real-time explosive stack safety.

Images