Method and system for preventing collapse of dangerous chemical goods shelf based on multi-sensor monitoring and medium

Abstract

The present application provides a kind of based on multi-sensor monitoring's dangerous chemical goods shelf anti-collapse regulation method, system and medium, belong to the field of stacking safety monitoring.The method comprises: obtaining first point cloud and second point cloud and carrying out weighted registration, obtains fusion point cloud;Determine the inclination rate of change based on fusion point cloud;According to the inclination rate of change and the material characteristics of dangerous chemical, estimate the risk occurrence time and risk occurrence position of dangerous chemical risk;And in response to risk occurrence time less than time threshold, determine the main swing characteristics of liquid in dangerous chemical stacking according to fusion point cloud;Based on main swing characteristics, generate time-varying driving signal;Based on time-varying driving signal, drive actuator to apply pulse force with the frequency, phase and amplitude indicated by time-varying driving signal to the shelf.The present application can improve the efficiency and accuracy of stacking monitoring, while the effectiveness and safety of emergency disposal are enhanced for emergency risk, and intelligent grading active regulation is realized.

Description

Technical Field

[0001] This invention relates to the field of stacking safety monitoring, and in particular to a method, system, and medium for controlling the collapse of hazardous chemical shelving based on multi-sensor monitoring. Background Technology

[0002] With the rapid development of smart warehousing, stacking safety monitoring technology has become a crucial link in ensuring the safety of warehousing operations. Traditional stacking monitoring mainly relies on manual inspections and basic sensors, which suffers from low monitoring efficiency and data lag. Modern monitoring systems employ technologies such as laser ranging and ultra-wideband positioning to monitor stacks at safe distances through three-dimensional grid monitoring systems. However, this method primarily focuses on external deformation and displacement changes, with limited ability to perceive the internal stress state of the stack. Acoustic emission technology, as a non-destructive testing method, can capture elastic wave signals generated by internal material fractures and has already been applied in fields such as rock stability monitoring and pressure vessel inspection. However, its integrated application in the field of stacking safety monitoring is still in the exploratory stage.

[0003] Therefore, it is necessary to provide a method, system, and medium for controlling the collapse of hazardous chemical shelving based on multi-sensor monitoring, which can improve the efficiency and accuracy of stacking monitoring, enhance the effectiveness and safety of emergency response for urgent risks, and realize intelligent hierarchical proactive control. Summary of the Invention

[0004] One embodiment of the present invention provides a method for controlling the collapse prevention of hazardous chemical shelving based on multi-sensor monitoring. The method includes: acquiring a first point cloud and a second point cloud, wherein the first point cloud is generated by scanning the surface of the hazardous chemical stack using a high-frequency millimeter-wave sensor, and the second point cloud is generated by scanning the hazardous chemical stack using a lidar; performing weighted registration of the first and second point clouds to obtain a fused point cloud; determining the tilt angle change rate based on the fused point cloud; estimating the risk occurrence time and location of the hazardous chemical based on the tilt angle change rate and the material characteristics of the hazardous chemical; and, in response to the risk occurrence time being less than a time threshold, determining the main sloshing characteristics of the liquid inside the hazardous chemical stack based on the fused point cloud, the main sloshing characteristics including a main sloshing frequency and a main sloshing phase; generating a time-varying drive signal based on the main sloshing characteristics, the time-varying drive signal having a frequency the same as the main sloshing frequency, a phase opposite to the main sloshing phase, and an amplitude determined based on the stack slippage; and driving an actuator to apply a pulse force to the shelf with the frequency, phase, and amplitude indicated by the time-varying drive signal based on the time-varying drive signal.

[0005] One embodiment of the present invention provides a multi-sensor monitoring-based anti-collapse control system for hazardous chemical shelving. The system includes: at least one processor; and at least one memory storing computer instructions. When the computer instructions are executed by the processor, the processor is configured to: acquire a first point cloud and a second point cloud, wherein the first point cloud is generated by scanning the surface of the hazardous chemical stack using a high-frequency millimeter-wave sensor, and the second point cloud is generated by scanning the hazardous chemical stack using a lidar; perform weighted registration of the first point cloud and the second point cloud to obtain a fused point cloud; determine the tilt angle change rate based on the fused point cloud; determine the main sloshing characteristics of the liquid inside the hazardous chemical stack based on the tilt angle change rate and the occurrence time of the hazardous chemical being less than a time threshold, the main sloshing characteristics including a main sloshing frequency and a main sloshing phase; generate a time-varying drive signal based on the main sloshing characteristics, the time-varying drive signal having a frequency the same as the main sloshing frequency, a phase opposite to the main sloshing phase, and an amplitude determined based on the stack slippage; and drive an actuator to apply a pulse force to the shelf with the frequency, phase, and amplitude indicated by the time-varying drive signal based on the time-varying drive signal.

[0006] One embodiment of the present invention provides a computer-readable storage medium storing computer instructions. When the computer instructions are executed on a computer, the computer causes the computer to perform a method for controlling the collapse of hazardous chemical shelving based on multi-sensor monitoring, as described in any embodiment of the present invention.

[0007] The beneficial effects of this invention include, but are not limited to: by fusing point cloud data from high-frequency millimeter-wave sensors and lidar, and performing weighted registration, the accuracy and robustness of sensing the stacked state of hazardous chemicals in complex environments such as industrial sites with smoke and moisture are significantly improved. Based on high-precision fused point cloud data, risks are dynamically predicted by analyzing the rate of change of tilt angle, realizing a shift from passive alarm to proactive prevention. For imminent risks, an innovative approach is taken to generate an inverse time-varying drive signal by analyzing the main sway characteristics, driving the actuator to apply a precise reverse pulse force. This pulse force provides physical support while actively counteracting the swaying inertial force of the liquid inside the hazardous chemical stack, greatly enhancing the initiative and effectiveness of emergency response, achieving intelligent hierarchical proactive control, and effectively preventing collapse accidents. Attached Figure Description

[0008] The present invention will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting, and in these embodiments, the same reference numerals denote the same structures, wherein:

[0009] Figure 1These are hardware / software schematic diagrams of exemplary computer devices according to some embodiments of the present invention; Figure 2 This is an exemplary flowchart of a multi-sensor monitoring-based method for preventing the collapse of hazardous chemical shelving, as illustrated in some embodiments of the present invention; and Figure 3 This is an exemplary schematic diagram of an updated twin model according to some embodiments of the present invention. Detailed Implementation

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some examples or embodiments of the present invention. For those skilled in the art, these drawings can be applied to other similar scenarios without creative effort. Unless obvious from the context or otherwise specified, the same reference numerals in the drawings represent the same structures or operations.

[0011] It should be understood that the terms “system,” “device,” “unit,” and / or “module” used herein are one way to distinguish different components, elements, parts, sections, or assemblies at different levels. However, if other terms can achieve the same purpose, they may be replaced by other expressions.

[0012] As indicated in this invention and the claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" do not specifically refer to the singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0013] This invention uses flowcharts to illustrate the operations performed by the system according to embodiments of the invention. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, the steps can be processed in reverse order or simultaneously. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0014] Figure 1 This is a hardware / software schematic diagram of an exemplary computer device according to some embodiments of the present invention. In some embodiments, the computer device may be a multi-sensor monitoring-based hazardous chemical shelf collapse prevention and control system, used to execute a multi-sensor monitoring-based hazardous chemical shelf collapse prevention and control method according to any embodiment of the present invention.

[0015] like Figure 1As shown, the computer device 100 includes at least one processor 110, a storage device 120, an input / output (I / O) interface 130, and a communication interface 140. The processor 110, storage device 120, and I / O interface 130 are connected via a system bus 150, and the communication interface 140 is connected to the system bus 150 via the I / O interface 130. The processor 110 provides computing and control capabilities. The storage device 120 includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer instructions, and a database. The internal memory provides an environment for the operation of the operating system and computer instructions stored in the non-volatile storage medium. The database is used for data, etc. The I / O interface 130 is used for exchanging information between the processor 110 and external devices. The communication interface 140 is used for communication with external terminals (e.g., user terminals) via a network connection. When the computer instructions are executed by the processor 110, they implement a method for preventing the collapse of hazardous chemical shelving based on multi-sensor monitoring provided in this embodiment of the invention (e.g., the flow 200 of the method for preventing the collapse of hazardous chemical shelving based on multi-sensor monitoring). The processor 110 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0016] Understandable. Figure 1 The structure shown is merely a block diagram of a portion of the structure related to the computer device described in this invention, and does not constitute a limitation on the computer device described in this invention. For example, the computer device may include components such as... Figure 1 The components shown may include more or fewer components (e.g., multiple processors, etc.), or combinations of certain components, or different component arrangements.

[0017] Figure 2 This is an exemplary flowchart of a multi-sensor monitoring-based method for preventing the collapse of hazardous chemical shelving, according to some embodiments of the present invention. In some embodiments, the flowchart 200 of the multi-sensor monitoring-based method for preventing the collapse of hazardous chemical shelving can be executed by a processor. Figure 2 As shown, the process 200 of the hazardous chemical shelving collapse prevention and control method based on multi-sensor monitoring may include the following steps:

[0018] Step 210: Obtain the first point cloud and the second point cloud.

[0019] In some embodiments, the first point cloud is generated by scanning the surface of the hazardous chemical stack using a high-frequency millimeter-wave sensor, and the second point cloud is generated by scanning the hazardous chemical stack using a lidar.

[0020] Hazardous chemical stacking refers to a collection of one or more containers (such as boxes, barrels, bags, etc.) containing hazardous chemicals stacked and arranged in a certain manner. For example, multi-layered chemical boxes stacked in a warehouse or multiple chemical storage tanks stored in the open can be considered hazardous chemical stacking.

[0021] The surface of a hazardous materials stack (or simply the stacking surface) refers to the external contour or boundary of a hazardous materials stack in three-dimensional space that can be detected by sensors. For example, for a hazardous materials stack consisting of multiple boxes, its surface includes the top and sides of the outermost box. The hazardous materials stacking surface is the part whose morphological changes are directly observed and measured by monitoring equipment (such as a laser scanner).

[0022] A high-frequency millimeter-wave sensor is a sensor device that uses millimeter waves (typically electromagnetic waves with frequencies in the range of 30 GHz to 300 GHz) for detection and sensing. For example, a high-frequency millimeter-wave sensor can be used to emit millimeter waves and receive reflected echoes from targets to generate point cloud data with strong environmental penetration.

[0023] LiDAR (Light Detection and Ranging) is a remote sensing device or technology that measures distance and creates a three-dimensional model of a target by emitting a laser beam toward it and analyzing the received reflected signal. For example, LiDAR can accurately measure the distance between itself and various points on a hazardous materials stack using the Time of Flight (ToF) or phase method, thereby generating high-precision point cloud data.

[0024] The first point cloud refers to a set of coordinate data used to represent the three-dimensional spatial information of a target object or scene. For example, in an embodiment of the present invention, the first point cloud may be a set of data points collected by a high-frequency millimeter-wave sensor that can reflect the approximate surface contour of a stack of hazardous chemicals. The first point cloud data has good integrity in environments such as smoke and dust.

[0025] The second point cloud refers to a set of data points that differ from the first point cloud in terms of acquisition time, acquisition perspective, or sensor source, but also represent a target object or scene in a three-dimensional coordinate system. For example, the second point cloud may be point cloud data acquired by the same sensor as the first point cloud at a different acquisition time, or point cloud data acquired by a different sensor at the same time as the first point cloud. In embodiments of the present invention, the second point cloud may be a high-density, high-precision set of data points acquired by lidar that can accurately reflect the three-dimensional structure of hazardous chemical stacks.

[0026] In some embodiments, the processor may fix one or more high-frequency millimeter-wave sensors (e.g., millimeter-wave radar operating at 77 GHz) above or to the side of the area where the hazardous materials are stacked. The high-frequency millimeter-wave sensors periodically emit millimeter-wave beams and receive reflected signals. By calculating the time of flight or phase difference of the signals, the distance and orientation of each point on the surface of the hazardous materials stack are determined, thereby generating a first point cloud that reflects the overall outline of the hazardous materials stack.

[0027] In some embodiments, the processor can mount a high-frequency millimeter-wave sensor on a mobile platform (such as an inspection robot or drone). The mobile platform circles or flies over the hazardous chemical stacks along a preset path, and the sensor continuously scans the surface of the hazardous chemical stacks during the movement. The processor can stitch and register data collected from different positions and angles to ultimately generate a first point cloud.

[0028] In some embodiments, the processor can deploy at least one 3D LiDAR (e.g., a 64-line mechanical LiDAR) within the monitored area. This 3D LiDAR rotates to perform a 360-degree scan of the entire scene, including the hazardous materials stack. After data acquisition, the processor uses algorithms to remove background point clouds unrelated to the hazardous materials stack, thus obtaining a second point cloud.

[0029] In some embodiments, the processor can utilize a drone equipped with a lidar (such as a solid-state lidar) and an inertial measurement unit (IMU) to perform aerial scanning of hazardous chemical stacks. The drone flies along a planned route, simultaneously recording lidar data and its own position and attitude data. Through the fusion and calculation of point cloud and pose data, a high-precision second point cloud with geographic coordinates is generated.

[0030] The processor can also acquire the first and second point clouds in other ways. For example, a high-frequency millimeter-wave sensor and a lidar can be integrated on the same gimbal or mobile platform for synchronous scanning to ensure good temporal and spatial consistency between the two point cloud data.

[0031] In some embodiments of this invention, by simultaneously acquiring a first point cloud generated by a high-frequency millimeter-wave sensor and a second point cloud generated by a lidar sensor, complementary advantages in data acquisition are achieved. High-frequency millimeter waves can penetrate smoke and dust, ensuring that the complete outline (first point cloud) of hazardous chemical stacks can still be obtained even in harsh environments such as fires, avoiding complete data loss due to environmental interference. Meanwhile, the high-precision second point cloud provided by lidar under normal conditions can accurately describe the detailed structure of the hazardous chemical stacks. This solution ensures that effective three-dimensional data of hazardous chemical stacks can be obtained regardless of environmental conditions, providing a reliable data foundation for subsequent status monitoring and safety early warning, and significantly improving the safety of hazardous chemical storage.

[0032] Step 220: Perform weighted registration on the first point cloud and the second point cloud to obtain the fused point cloud.

[0033] Weighted registration is a point cloud registration technique that assigns different weights to different data points or point correspondences during the alignment of two or more point clouds, and calculates the optimal spatial transformation relationship between the point clouds based on these weights. For example, during the registration process, a larger weight can be assigned to point pairs with higher confidence or more significant features, while a smaller weight can be assigned to point pairs that may be noise points or outliers, thereby improving the accuracy and robustness of the registration.

[0034] A fused point cloud refers to a single point cloud dataset generated by registering and fusing at least two point clouds. Fused point clouds can represent a target object or scene more comprehensively or accurately.

[0035] For example, a fused point cloud can contain all or part of the information from a first point cloud and a second point cloud. Fusion processing increases the density of the point cloud, reduces noise, or fills in missing parts due to occlusion. For instance, a fused point cloud can be formed by merging point cloud data collected simultaneously by multiple sensors (such as LiDAR) deployed at different locations in the stack, or by merging point cloud data collected at different times by a single sensor, thus providing a more comprehensive and accurate description of the surface morphology of the entire stack.

[0036] In some embodiments, the process of weighted registration to obtain a fused point cloud may include: First, calculating the optimal transformation matrix to transform the first point cloud into the coordinate system of the second point cloud using an Iterative Closest Point (ICP) algorithm. This matrix includes a rotation matrix R and a translation vector T. Then, spatially transforming the first point cloud according to this transformation matrix to achieve initial alignment with the second point cloud. Finally, for the overlapping regions of the two point clouds, performing a weighted average of the coordinates according to preset weights (e.g., based on point confidence or sensor accuracy settings); for non-overlapping regions, directly retaining the original point cloud data. After merging the weighted averaged point cloud with the point clouds from the non-overlapping regions, an attention mechanism can be used to process the merged point cloud to filter out potential noise points, thereby obtaining the final fused point cloud.

[0037] In some embodiments, the weighted registration process can also employ a feature matching-based method. For example, firstly, local feature descriptors such as Fast Point Feature Histograms (FPFHs) are extracted from the first and second point clouds, respectively. Then, corresponding point pairs are found by matching the feature descriptors in the two point clouds. Next, a robust transformation matrix can be estimated from these corresponding point pairs using the Random Sample Consensus (RANSAC) algorithm. After obtaining the transformation matrix, the subsequent weighted fusion and noise filtering steps are similar to those in the aforementioned embodiments.

[0038] In some embodiments of the present invention, by performing weighted registration of the first and second point clouds, overlapping areas can be accurately fused, improving the accuracy and detail of the fused point cloud. This scheme can supplement information missing due to occlusion from a single viewpoint, improving the integrity and coverage of the point cloud. The resulting fused point cloud has less noise and higher reliability, providing a high-quality data foundation for downstream sensing tasks.

[0039] In some embodiments, the processor may further: acquire a pre-trained weight model; determine a reliability weight sequence based on the smoke concentration data of the current environment and the pre-trained weight model; and perform weighted registration of the first point cloud and the second point cloud based on the reliability weight sequence to obtain a fused point cloud.

[0040] A weighting model is a mathematical or algorithmic model that outputs weights for point cloud weighted registration based on input environmental parameters. For example, a weighting model could be a pre-trained neural network that receives smoke concentration data as input and outputs corresponding reliability weights for millimeter-wave and lidar point clouds. The pre-trained weighting model is used to dynamically assign weights to point cloud data (first and second point clouds) from different sensors (such as millimeter-wave radar and lidar) based on environmental parameters (e.g., smoke concentration).

[0041] In some embodiments, the process of acquiring a pre-trained weight model may include the following steps: First, in a controlled experimental environment, a reference point cloud of a target object (such as a stack) is acquired using a high-precision 3D scanner. Then, under different smoke concentration conditions, a first point cloud (millimeter-wave point cloud) and a second point cloud (LiDAR point cloud) are simultaneously acquired. Next, the registration errors between the first and second point clouds and the reference point cloud are calculated at each smoke concentration. Based on the calculated registration errors, the error values ​​are converted into normalized reliability weights using a function (such as the Softmax function), ensuring that sensors with smaller registration errors receive higher reliability weights. Finally, using these "smoke concentration-reliability weight" data pairs, a machine learning model (such as a neural network) is trained, enabling the model to learn and fit the mapping relationship (i.e., the first mapping relationship) between smoke concentration and sensor reliability weights.

[0042] In some embodiments, the pre-trained weight model can also be a regression model. For example, by analyzing historically collected smoke concentration and sensor registration error data using multinomial regression, a mathematical function can be established that can directly calculate the corresponding reliability weight value based on the input smoke concentration.

[0043] In some embodiments, the input to a pre-trained weight model may include smoke concentration data of the current environment, and the output may include a reliability weight sequence.

[0044] Smoke concentration data refers to quantitative data characterizing the content or density of smoke in the environment. For example, smoke concentration data can be collected in real time by environmental monitoring equipment, and the unit is mg / m³ or ppm.

[0045] A reliability weight sequence is a set of values ​​that correspond to different point cloud data sources and are used to characterize the reliability or importance of each data source in a specific environment. Weight values ​​range from 0 to 1. For example, a reliability weight sequence may include a weight value assigned to the first point cloud (millimeter-wave point cloud weight value) and a weight value assigned to the second point cloud (LiDAR point cloud weight value), with the sum of these two weight values ​​being 1, used for subsequent weighted registration calculations.

[0046] In some embodiments, the process of determining the reliability weight sequence is as follows: First, real-time smoke concentration data is acquired using a smoke concentration monitoring device (such as an optical smoke sensor or a PM2.5 sensor) deployed in the working environment. Then, the acquired smoke concentration data is sent as input to a pre-trained weight model. The pre-trained weight model calculates based on an internal first mapping relationship and outputs a reliability weight sequence containing two weight values, corresponding to the first point cloud and the second point cloud, respectively.

[0047] For example, when a smoke concentration monitoring device detects a current smoke concentration of 10 mg / m³ (which is considered a low concentration), the pre-trained weighted model might output a reliability weight sequence of [0.2, 0.8], where 0.2 corresponds to the first point cloud (millimeter-wave point cloud) and 0.8 corresponds to the second point cloud (LiDAR point cloud). As another example, when a fire occurs or a large amount of smoke is generated during production, and the smoke concentration monitoring device detects a current smoke concentration of 150 mg / m³ (which is considered a high concentration), the pre-trained weighted model might output a reliability weight sequence of [0.85, 0.15], where 0.85 corresponds to the first point cloud (millimeter-wave point cloud) and 0.15 corresponds to the second point cloud (LiDAR point cloud).

[0048] In some embodiments, the determination of the reliability weight sequence can also incorporate other environmental parameters. For example, in addition to smoke concentration, environmental data such as humidity and dust concentration can be input into a more complex multi-input weight model to generate a weight sequence that is more adapted to the current overall environment.

[0049] In some embodiments, the first point cloud and the second point cloud are weighted and registered according to a reliability weight sequence to obtain a fused point cloud.

[0050] In some embodiments, weighted registration can be performed based on the ICP algorithm. Specifically, in each iteration of the ICP algorithm, when calculating the error between corresponding point pairs (such as the sum of squares of Euclidean distances), the error term for each point pair is multiplied by the reliability weight of its respective point cloud. In this way, during the optimization of the transformation matrix, the error of point pairs from point clouds with higher weights will have a greater impact on the final result, thereby guiding the registration process to align with a more reliable data source.

[0051] In some embodiments, weighted registration can also be achieved by modifying the NDT algorithm. In the NDT algorithm, the point cloud is represented as a set of Gaussian distributions. A reliability weight sequence can be used to adjust the influence of each sensor point cloud in the probability density function construction process. For example, a point cloud with a higher weight contributes more to the objective function, making the registration result more aligned with the probability distribution of that point cloud.

[0052] In some embodiments of this invention, dynamic adaptive adjustment of the weights of multi-sensor point clouds is achieved by introducing smoke concentration data and a pre-trained weight model. When the smoke concentration is low, higher weights are automatically assigned to the high-resolution lidar point cloud to ensure measurement accuracy; when the smoke concentration increases, the weights are dynamically transferred to the more penetrating millimeter-wave point cloud to ensure the continuity and reliability of data acquisition. This dynamic weighting strategy based on environmental perception significantly improves the quality and robustness of the fused point cloud in complex and variable environments, providing a highly reliable data foundation for subsequent accurate calculations and risk warnings.

[0053] In some embodiments, determining smoke concentration data may include: controlling a high-frequency millimeter-wave sensor and a lidar to detect a preset reference target, respectively, and acquiring first detection data and second detection data; determining an echo intensity ratio based on the first detection data and the second detection data; and determining smoke concentration data based on the echo intensity ratio and a pre-calibrated relationship curve.

[0054] A reference target is an object that is pre-set in the detection environment and used as a measurement benchmark. For example, a reference target can be an object that is inherent in the environment and has stable reflective properties, such as a column or a wall, or it can be a calibration plate specially set up for measurement purposes.

[0055] The first detection data refers to the data obtained after a high-frequency millimeter-wave sensor detects a reference target. For example, the first detection data can be the echo signal strength value received by the high-frequency millimeter-wave sensor, in dBm.

[0056] Secondary detection data refers to the data obtained after the lidar detects the reference target. For example, secondary detection data can be the intensity value of the echo signal received by the lidar. Secondary detection data is usually expressed as photon counts or voltage values ​​and converted into relative intensity units.

[0057] In some embodiments, the process of the processor controlling the sensor to detect and acquire data first requires a pre-defined reference target. This reference target typically has known structural features and stable surface reflectivity. For example, a pillar or wall in a stacked environment can be selected as the reference target. The processor records the coordinates of the reference target and instructs the high-frequency millimeter-wave sensor and lidar to periodically emit detection signals toward the target. Alternatively, a calibration plate with high reflectivity can be specifically set up within the detection area as a reference target to obtain a stronger and more stable echo signal.

[0058] In some embodiments, the process of acquiring the first detection data and the second detection data is as follows: A high-frequency millimeter-wave sensor emits a millimeter-wave signal toward a reference target, receives and processes the echo signal from the target, and obtains the echo signal intensity value (e.g., in dBm) as the first detection data. Simultaneously, a lidar emits a laser pulse toward the same reference target, receives and processes the echo signal, and converts the intensity of the echo signal (e.g., expressed as photon count or voltage value) into relative intensity units as the second detection data.

[0059] In some embodiments, after acquiring the first detection data and the second detection data, the processor can determine the echo intensity ratio based on the first detection data and the second detection data.

[0060] The echo intensity ratio is a value calculated based on the first and second detection data to characterize the differences in how different detection signals are affected by environmental factors (such as smoke attenuation). For example, the echo intensity ratio can be the ratio between the second detection data and the first detection data after normalization to a reference intensity.

[0061] In some embodiments, the echo intensity ratio (R) can be calculated using a normalization formula. This normalization formula utilizes baseline intensity values ​​pre-measured in clean air to eliminate the influence of differences in sensor performance. For example, the calculation formula could be: R = (Second detection data / Baseline intensity of the lidar) / (First detection data / Baseline intensity of the high-frequency millimeter-wave sensor). Here, the baseline intensities of the lidar and the high-frequency millimeter-wave sensor are the stable echo intensity values ​​obtained by the two sensors under clean air (smoke-free) conditions for detecting the same reference target; these values ​​are pre-measured and stored.

[0062] In some embodiments, after determining the echo intensity ratio, the processor can determine smoke concentration data based on the echo intensity ratio and a pre-calibrated relationship curve.

[0063] A pre-calibrated relationship curve is a function, model, or lookup table used to characterize the correspondence between echo intensity ratio and smoke concentration data. For example, this pre-calibrated relationship curve can be pre-established by conducting experimental measurements at different known smoke concentrations and fitting the measured echo intensity ratio and smoke concentration data.

[0064] In some embodiments, the pre-calibrated relationship curve is obtained through experimental calibration. The experimental calibration process can be carried out in a controlled smoke chamber, using a standard smoke concentration measuring instrument (such as a turbidimeter) to set and measure different levels of smoke concentration (C). At each smoke concentration, the method of this embodiment is run synchronously to measure and calculate the corresponding echo intensity ratio (R). After collecting multiple sets of (R,C) data pairs, a curve fitting method (e.g., exponential decay model) is used to establish... The functional relationship is such that the function curve is the pre-calibrated relationship curve.

[0065] In some embodiments, the process of determining smoke concentration data involves taking the echo intensity ratio R calculated in real time as input and substituting it into a pre-calibrated relational curve function. In this process, the current smoke concentration data C (e.g., in mg / m³) can be directly calculated. Alternatively, a pre-calibrated relationship curve can be stored as a lookup table. The echo intensity ratio R calculated in real time is searched in the lookup table to find the closest R value, and its corresponding smoke concentration data C is read. If the R value lies between two table points, the corresponding C value can be calculated using methods such as linear interpolation.

[0066] In some embodiments, the mapping relationship between the echo intensity ratio and smoke concentration can also be established by training a machine learning model (e.g., a neural network or support vector regression model). Alternatively, other numerical methods such as piecewise linear interpolation can be used to determine smoke concentration data.

[0067] In some embodiments of this invention, online, low-cost measurement of smoke concentration data is achieved by utilizing the ratio of echo intensity of high-frequency millimeter waves and lidar to the same reference target. This method eliminates the need for a dedicated smoke sensor, directly utilizing existing sensing equipment. Through relative measurement, it effectively eliminates the influence of environmental interference and sensor drift, improving the stability and reliability of smoke concentration measurement and providing accurate environmental parameter input for subsequent dynamic weight registration.

[0068] Step 230: Determine the tilt angle change rate based on the fused point cloud.

[0069] The rate of change of tilt angle is a physical quantity that characterizes how quickly the tilt angle of a stack surface relative to a reference plane (usually a horizontal plane) changes over time. The rate of change of tilt angle can be expressed as the change in the tilt angle of the stack's center of gravity or a specific location per unit time, and the unit can be degrees per second or radians per second. For example, if the tilt angle of a stack surface increases uniformly from 35 degrees to 35.2 degrees within 10 seconds of continuous monitoring, its rate of change of tilt angle can be determined as 0.02 degrees per second. This value can intuitively reflect the speed and stability trend of stack slippage.

[0070] In some embodiments, the process of determining the tilt angle change rate may include: First, within a preset time period (e.g., 10 seconds), continuously acquiring fused point clouds of the stacked surface at fixed time intervals (e.g., 1 second) to form a point cloud sequence. Then, for each fused point cloud in the sequence, principal component analysis (PCA) is used to calculate the normal vector of its best-fit plane. Specifically, PCA finds the eigenvector corresponding to the minimum eigenvalue by calculating the covariance matrix of the point cloud coordinates; this vector is the normal vector of the point cloud surface. The tilt angle at the current moment is calculated based on the angle between this normal vector and the vertical direction (e.g., the Z-axis). Finally, linear regression analysis is performed on all tilt angle values ​​obtained within the time period to calculate the slope of the fitted line; this slope is the tilt angle change rate.

[0071] In some embodiments, a least-squares plane fitting method can also be used to calculate the tilt angle. Specifically, for each fused point cloud in the point cloud sequence, a plane equation is fitted using the least-squares method, minimizing the sum of the squared distances from all points to the plane. The coefficients of this plane equation constitute the normal vector of the plane. The subsequent tilt angle calculation and the process of determining the tilt angle change rate through linear fitting are similar to the aforementioned embodiments.

[0072] In some embodiments of this invention, by determining the tilt change rate based on fused point clouds, the three-dimensional morphology of the stack surface can be comprehensively and accurately captured using multi-source or multi-temporal data. Compared to monitoring only the static tilt angle, calculating the tilt change rate, a dynamic indicator, can more effectively identify unstable trends such as slow creep or accelerated slippage of the stack. This enables the system to provide early warnings of potential landslide risks, improving the timeliness and accuracy of warnings, avoiding false alarms or missed alarms due to static thresholds, and enhancing the safety of yard operations.

[0073] In some embodiments, the processor may also determine the tilt rate of change by: acquiring an initial twin model and acoustic emission signals collected by acoustic sensors; updating the initial twin model based on the fused point cloud and acoustic emission signals to obtain an updated twin model; and determining the tilt rate of change based on the updated twin model.

[0074] An initial twin model refers to a digital representation of a physical entity (such as a stack) in its initial state or before monitoring begins. For example, an initial twin model can be a three-dimensional digital model based on the physical parameters of the stacked material (such as density, coefficient of friction, etc.) that can reflect the initial geometry, internal stress distribution, and other states of the stacked material.

[0075] An acoustic sensor is a device that can detect sound waves or acoustic emission phenomena and convert them into a usable output signal (usually an electrical signal). For example, an acoustic sensor can be a piezoelectric sensor used to capture the weak sound wave signals generated by materials during deformation under stress.

[0076] Acoustic emission signals are transient signals generated when a material experiences rapid localized stress release (such as microcrack propagation) within its interior, releasing energy in the form of elastic waves. For example, acoustic emission signals can include characteristic parameters such as amplitude, energy, frequency, and duration, used to reflect the evolution of damage within the material.

[0077] In some embodiments, obtaining an initial twin model can be achieved through a pre-established physical simulation model. For example, a three-dimensional digital model of the stacked material can be established using discrete element method (DEM) or finite element method (FEM) software based on the physical parameters of the stacked material (such as particle size, density, coefficient of friction, elastic modulus, etc.). This three-dimensional digital model serves as the basis for subsequent updates, i.e., as the initial twin model.

[0078] In some embodiments, acoustic emission signals are acquired by acoustic sensors deployed near or in contact with the stacked structure. For example, piezoelectric acoustic sensors are arranged on the retaining wall or bottom support structure of the stack to continuously acquire acoustic emission signals at a high-frequency sampling rate (e.g., 100kHz-1MHz) and extract characteristic parameters such as amplitude, energy, and frequency of the signals.

[0079] An updated twin model refers to a digital representation that reflects the current state of a physical entity (such as a stack) by incorporating real-time or near-real-time observation data and adjusting the state of the initial twin model or the twin model from the previous moment. For example, an updated twin model can be a digital model that is corrected by fusing point clouds and acoustic emission signals and using a data assimilation algorithm, and can more accurately characterize the current displacement field, stress field, or damage state of the stack.

[0080] In some embodiments, the process of updating the twin model can employ a data assimilation method. This data assimilation method typically includes two phases: a prediction phase and an update phase. In the prediction phase, based on the initial twin model and the current twin model (i.e., the twin model updated after the previous observation data (fusion of point cloud and acoustic emission signal)), the predicted state of the model at future times is derived by extrapolating forward using physical equations (such as discrete element or finite element equations). In the update phase, the predicted state of the model is corrected by using the fused point cloud (representing external morphology) and acoustic emission signal (representing internal stress activity) as observation data to obtain the updated twin model.

[0081] For example, the Kalman filter algorithm can be used to update the twin model. The state vector of the twin model (containing particle position, velocity, stress, etc.) is used as the system state, and the fused point cloud and acoustic emission signals are used as observation data. By minimizing the error between the model prediction and the actual observation through Kalman gain, the updated twin model with the optimal estimate is obtained.

[0082] For example, a particle filtering algorithm can be used to update the Siamese model. This particle filtering method uses a set of weighted random samples (particles) to represent the posterior probability distribution of the stack state. The particle weights are updated based on the degree of matching between the observed data and the predicted state of each particle; particles with higher weights represent states closer to the true state. Through a resampling process, an updated Siamese model that accurately reflects the current state of the stack is finally obtained.

[0083] In some embodiments, the step of determining the tilt angle change rate based on the updated twin model includes: extracting the three-dimensional coordinate time series of key stack locations (such as the center of gravity or the highest point at the top) from the updated twin model; calculating the sequence of stack tilt angle changes over time based on the three-dimensional coordinate time series, i.e., the tilt angle time series; and differentiating the tilt angle time series to obtain the tilt angle change rate.

[0084] For example, the processor can extract the three-dimensional coordinates of the heap centroid from the updated twin model. , , ); Calculate the displacement relative to the initial position. , and ; Calculate the tilt angle of the stack or ; regarding the angle of inclination By taking the derivative over time, we can obtain the rate of change of tilt angle. The unit is degrees per second. To eliminate noise, the processor can filter or smooth the calculation results.

[0085] In some embodiments, obtaining the three-dimensional coordinates of the stack's center of gravity can be achieved through geometric analysis of the updated twin model. Since the updated twin model contains precise geometric information (such as length, width, and height) and position and orientation information for each item in the stack, the processor can obtain the individual center of gravity coordinates and mass (which can be calculated based on preset material density and volume) of each item in the updated twin model, and then calculate the center of gravity coordinates of the entire stack using the center of gravity formula in physics (e.g., the weighted average method).

[0086] In some embodiments, the rate of change of tilt angle can also be determined by analyzing other physical quantities in the updated twin model. For example, the average angular velocity components of key region nodes can be extracted directly from the velocity field data of the updated twin model as a characterization of the rate of change of tilt angle. Alternatively, the rate of tilt intensification can be indirectly inferred by analyzing the evolution trend of stress concentration regions in the model.

[0087] In some embodiments of this invention, by fusing external point clouds and internal acoustic emission signals to update the digital twin model, multi-source sensing and dynamic updating of the internal and external states of the stack are achieved. Acoustic emission signals can capture micro-damage events such as particle friction and compression within the stack, compensating for the inability of point clouds to detect internal stress evolution. By fusing the fused point cloud and acoustic emission signals using a data assimilation algorithm, the twin model can reflect the complete state evolution of the stack from its external morphology to its internal stress field in real time. This combined internal and external monitoring method significantly improves the accuracy of tilt angle change rate calculation and prediction lead time, enabling earlier identification of instability risks, gaining valuable time for proactive control, and effectively avoiding sudden collapse accidents caused by internal stress concentration.

[0088] In some embodiments, the processor may: update the stress field of the initial twin model based on the acoustic emission signal; register the fused point cloud with the initial twin model after the stress field update; and update the registered initial twin model through an optimization algorithm to obtain the updated twin model.

[0089] A stress field is a physical field that describes the distribution of stress at various points inside an object. For example, in a twin model of a stack, the stress field can represent the magnitude and direction of the pressure and shear forces acting on different locations inside the stack.

[0090] In some embodiments, the specific implementation of updating the stress field is as follows: performing time-frequency analysis on the acquired acoustic emission signals to extract key feature parameters such as acoustic emission event rate and energy release rate; using multiple acoustic sensors deployed around the stack, and based on the time difference of the time when each sensor receives the same acoustic emission signal, calculating the spatial location of the source of the acoustic emission event using a time difference localization algorithm; converting the extracted key feature parameters into stress values ​​at the location of the acoustic emission source according to a pre-established "acoustic emission feature-stress" mapping relationship (i.e., the second mapping relationship); based on these discrete stress values, retrieving the stress field distribution of the entire stack through an interpolation algorithm (such as Kriging interpolation) or a stress-acoustic emission empirical model, and using this distribution to update the stress tensor field in the initial twin model.

[0091] Acoustic emission event rate refers to the number of acoustic emission events per unit time. Energy release rate refers to the cumulative acoustic emission energy per unit time.

[0092] The acoustic emission characteristic-stress mapping relationship can include the correspondence between key characteristic parameters and stress values. In some embodiments, the process of establishing a second mapping relationship can be completed under controlled laboratory conditions by performing loading experiments on a stacked model.

[0093] For example, establishing a second mapping relationship can be achieved through the following steps: First, a series of known, increasing loads are applied to one or more stacked models in a laboratory. During loading, acoustic emission signals and corresponding strain data are simultaneously acquired using acoustic emission sensors and strain gauges deployed on the surface of the stacked models. Then, acoustic emission features (such as amplitude and energy) are extracted from the acquired acoustic emission signals, and the strain data is converted into stress values, thus forming a calibration dataset containing "acoustic emission feature-stress value" data pairs. Finally, a machine learning model, such as a regressive neural network, is trained using this calibration dataset. The trained neural network model then constitutes the second mapping relationship reflecting the acoustic emission features and the internal stress state of the stack.

[0094] In some embodiments, a second mapping relationship can also be established based on physical laws such as the Kaiser effect. For example, by repeatedly stressing the stack model at different historical maximum stress levels and recording acoustic emission activities, the historical maximum stress values ​​corresponding to different acoustic emission activity spikes can be identified. In this way, a mapping relationship based on the acoustic emission activity threshold and corresponding to the historical maximum internal stress state of the stack is established, namely, the second mapping relationship.

[0095] In some embodiments, the stress field can also be updated using a machine learning model. For example, a deep neural network model can be constructed, taking the original waveform of the acoustic emission signal or its spectral characteristics over a period of time as input and outputting the three-dimensional mesh data of the stress field for the entire stack. By training on a dataset calibrated in a simulation environment or experiment, the model can learn and establish a complex nonlinear mapping relationship from the acoustic emission signal to the stress field distribution, thereby achieving rapid end-to-end stress field updates.

[0096] Registration refers to the process of aligning two or more datasets in space to unify them into the same coordinate system. For example, aligning the fused point cloud representing the actual surface shape of a stack with the surface mesh of the initial twin model in terms of spatial position and orientation.

[0097] In some embodiments, the registration process between the fused point cloud and the initial twin model after updating the stress field can employ the ICP algorithm. This ICP algorithm iteratively calculates and finds the optimal rotation and translation transformation matrix between the fused point cloud and the surface mesh of the initial twin model, thereby minimizing the distance between corresponding points and achieving precise spatial alignment.

[0098] Optimization algorithms are a class of mathematical methods used to find the optimal solution from all possible solutions under certain constraints. For example, genetic algorithms or particle swarm optimization algorithms can be used to iteratively calculate and adjust the parameters of the initial twin model to minimize the difference between the surface of the twin model and the registered fused point cloud, thereby obtaining a more accurate twin model.

[0099] In some embodiments, the update process using an optimization algorithm can employ a genetic algorithm (GA). Specifically, key geometric parameters of the initial twin model (such as stacking height, slope position, etc.) are used as individual genes in the genetic algorithm, and the Hausdorff distance or Chamfer distance between the twin model surface and the registered fused point cloud is used as the fitness function. By simulating selection, crossover, and mutation operations in biological evolution, the algorithm iteratively searches for the parameter combination that minimizes the fitness function, thereby updating the geometry of the initial twin model.

[0100] In some embodiments, particle swarm optimization (PSO) can also be used. Each particle represents a set of candidate parameters for the initial twin model. By tracking the individual optimal solution and the global optimal solution, it updates its own velocity and position (i.e., model parameters), eventually converging to the optimal parameters, thus completing the update of the initial twin model.

[0101] In some embodiments of this invention, acoustic emission signals and point clouds are fused to achieve coordinated internal and external updates of the initial twin model. Real-time sensing of the internal stress field via acoustic emission signals solves the problem of internal state monitoring; simultaneously, registration of the fused point clouds optimizes the external morphology of the twin model, ensuring a high degree of consistency between the twin model and the actual entity. This combined internal and external update mechanism significantly improves the accuracy of the updated twin model in reflecting the true state of the stack, thereby enhancing the accuracy of instability risk prediction and providing a reliable basis for safety monitoring and proactive control.

[0102] Figure 3 This is an exemplary schematic diagram of an updated twin model according to some embodiments of the present invention.

[0103] In some embodiments, such as Figure 3 As shown, the processor can: acquire the current operating condition 310; calculate the deviation 330 between the surface deformation predicted by the updated twin model 320 and the fused point cloud based on the current operating condition 310 and the updated twin model 320; when the deviation 330 exceeds a preset threshold, calculate the displacement field 340 of the hazardous chemical stacking surface through the updated twin model 320; and determine the tilt angle change rate 350 based on the slip surface position and slip amount in the displacement field 340.

[0104] Current operating conditions refer to the set of real-time environmental or physical condition parameters that affect the stacking status of hazardous chemicals. For example, current operating conditions may include parameters such as real-time temperature, ambient humidity, and ground vibration intensity at the stacking site.

[0105] In some embodiments, current operating conditions are obtained by deploying multiple sensors at the stacking site. For example, temperature sensors can monitor the temperature distribution on and inside the stacking surface, humidity sensors can monitor ambient humidity, and vibration sensors can monitor the intensity of ground vibrations caused by external vibration sources such as equipment operation or vehicle traffic. These sensors collect data in real time, which together constitute the parameter set of the current operating conditions.

[0106] In some embodiments, current operating conditions can also be obtained in conjunction with external data sources. For example, in addition to on-site sensors, regional temperature, humidity, and air pressure data can be obtained from public meteorological service systems, or geological vibration information can be obtained from regional earthquake monitoring networks to more comprehensively assess environmental impacts.

[0107] In some embodiments, the method of obtaining the current working conditions is not limited to this, and may also include manually entered information about specific events (such as nearby construction activities), or periodic environmental changes predicted by analyzing historical data.

[0108] Surface deformation refers to changes in the geometry of the surface of a stack of hazardous chemicals. For example, surface deformation can manifest as localized bulges, depressions, or overall subsidence of the stacked surface.

[0109] Bias refers to the difference between the surface morphology of hazardous chemical stacks predicted by the twin model and the surface morphology actually measured by sensors. For example, bias can be the spatial distance between the position of a point on the surface predicted by the twin model (updated twin model) and the corresponding point in the fused point cloud.

[0110] In some embodiments, the method for calculating the bias includes: spatially aligning and registering the surface deformation data predicted by the updated twin model with the fused point cloud data measured in reality; calculating the Euclidean distance between a point on the surface of the updated twin model and the nearest corresponding point in the fused point cloud, and using the average or maximum value of these distances as the final bias value.

[0111] In some embodiments, the bias can be determined by calculating volume differences. For example, the surface predicted by the updated twin model and the surface formed by the fused point cloud can be considered as two three-dimensional surfaces, and the volume of the space enclosed by these two surfaces can be calculated as a measure of bias.

[0112] In some embodiments, the method for calculating the bias may also employ other statistical indicators, such as root mean square error (RMSE), to comprehensively assess the consistency between model predictions and actual measurements.

[0113] A preset threshold is a pre-defined critical value used for judgment and decision-making. For example, when the deviation exceeds a preset threshold (such as 5 centimeters), the processor will trigger the next calculation or an alarm.

[0114] A displacement field is a set of vectors that describes the change in position of points on or within a stack of hazardous materials from one moment to another. For example, a displacement field can visually show which areas of the stack have moved, as well as the direction and distance of that movement.

[0115] In some embodiments, the method for calculating the displacement field includes: registering the fused point cloud acquired at the current moment with a stable reference base (e.g., the fused point cloud acquired at the initial moment). By comparing the spatial position differences of corresponding points in the two point clouds, a displacement vector can be generated for each point, and the set of these displacement vectors constitutes the displacement field of the entire stack surface.

[0116] In some embodiments, the displacement field can also be directly calculated from the updated twin model. By recording the position changes of particles or units within the updated twin model from the initial state to the current updated state, the three-dimensional displacement fields of the interior and surface can be directly generated.

[0117] In some embodiments, digital image correlation (DIC) technology can be combined to calculate the displacement field by analyzing continuously captured images of the stacked surface, and the result can be fused with the model calculation result to improve the accuracy of the displacement field calculation.

[0118] The slip surface location refers to the spatial location of the interface between different material parts that undergo relative sliding within a hazardous materials stack. For example, the slip surface location can be identified as the region in the displacement field where the displacement vector changes abruptly.

[0119] Slippage refers to the distance that two relatively sliding parts of material move along the sliding direction on a sliding surface. For example, slippage could be the amount of material above the sliding surface sliding 10 centimeters relative to the material below.

[0120] In some embodiments, the method for determining the slip surface location and slip amount includes: applying displacement gradient analysis or an edge detection algorithm to the calculated displacement field to find regions in the displacement field where the magnitude or direction of the displacement vector changes abruptly; these regions are identified as slip surface locations. Subsequently, on the identified slip surface, the displacement component in the sliding direction is calculated to determine the slip amount.

[0121] In some embodiments, a clustering algorithm can be used to identify slip surfaces. For example, by clustering points in the displacement field based on the similarity of their displacement vectors, the boundary regions between different clusters can be identified as slip surfaces. The amount of slip can then be obtained by calculating the average displacement difference between adjacent clusters.

[0122] In some embodiments, the step of determining the rate of change of the tilt angle may include: calculating the equivalent tilt angle θ(t) of the stack at the current moment based on the position of the slip surface and the amount of slip through geometric relationships, for example, the tilt angle can be estimated by the formula θ=arctan(amount of slip / height of slip surface); repeating the calculation at fixed time intervals to form a time series of the tilt angle changing with time; and calculating the rate of change by numerically differentiating the time series. The final rate of change of tilt angle is obtained.

[0123] In some embodiments, the method for determining the rate of change of tilt angle can also employ time-series data processing techniques such as Kalman filtering to smooth and predict the tilt angle time series in order to obtain more stable and accurate rate of change results.

[0124] In some embodiments of this invention, a closed-loop verification system is established between operating conditions, twin model predictions, and measured data to achieve accurate identification of stack slippage. This method triggers refined slippage analysis only when the deviation between model predictions and measured data exceeds a threshold, extracting the slippage surface position and slippage amount from the displacement field to determine the rate of change of tilt angle. This deviation-triggered intelligent analysis mechanism effectively avoids resource waste and data noise interference caused by continuous computation. While ensuring real-time monitoring, it significantly improves the accuracy and reliability of slippage early warning and reduces false alarms.

[0125] Step 240: Based on the rate of change of inclination angle and the material characteristics of the hazardous chemicals, predict the time and location of the risk of hazardous chemical occurrence.

[0126] Hazardous chemicals refer to chemicals that possess dangerous characteristics such as flammability, explosiveness, toxicity, toxicity, and corrosivity, and may cause damage to personnel, facilities, and the environment. For example, in this technical solution, hazardous chemicals may refer to sulfur, ammonium nitrate, or certain metal powders stored in a stacked manner.

[0127] Material properties refer to the inherent attributes of a substance that determine its physical or chemical behavior. For example, the material properties of hazardous chemicals may include particle size, density, coefficient of internal friction, angle of repose (or angle of relativity), etc.

[0128] Risk refers to the combination of the probability of a hazardous event occurring and its consequences in a specific environment. For example, in this technical solution, risk specifically refers to the collapse or slippage event caused by the instability of a hazardous chemical stack.

[0129] The risk occurrence time refers to the estimated point in time or period during which a risk event is expected to occur. For example, the risk occurrence time can be the predicted length of time from the current moment until the stack collapses or slips, and the unit can be seconds or minutes.

[0130] The location of a risk event refers to the specific spatial location or area where a predicted risk event is expected to occur. For example, the location of a risk event could be a stacked area where a collapse or slippage is anticipated, and this area could be represented by three-dimensional coordinates, grid numbers, or other area identifiers.

[0131] In some embodiments, the process of predicting the time and location of risk occurrence can be achieved by constructing and solving a dynamic model.

[0132] Specifically, in one embodiment, an analysis method based on the Coulomb friction model can be employed. This method takes the rate of change of inclination angle and material properties (such as the internal friction coefficient and density) as inputs to calculate the shear stress inside the stack. Through numerical integration, the specific time and location at which the shear stress exceeds the material's shear strength can be predicted, thereby estimating the time and location of the risk.

[0133] In one embodiment, the Discrete Element Method (DEM) can be used for simulation prediction. This DEM method models the hazardous chemical stack as a large number of discrete particles, simulating the movement and interaction of these particles based on input material properties and measured rates of change of tilt angle. Through simulation, particle regions where large-scale displacement (i.e., collapse) is imminent can be identified, thereby predicting the timing and location of the risk.

[0134] In some embodiments of this invention, the timing and location of hazardous chemical risks can be accurately predicted. Compared to traditional methods that can only determine whether a risk exists, this solution upgrades the early warning from "whether" to "when and where," providing a basis for decision-making on precise, time-based intervention measures (such as local reinforcement and zoned evacuation), significantly improving the initiative and effectiveness of risk management, and thus better protecting the safety of personnel and property.

[0135] Step 250: In response to the risk occurring less than a time threshold, perform the first operation.

[0136] A time threshold is a pre-defined time threshold used to determine whether a certain time value meets a preset condition. For example, the time threshold can be set to a specific time value, such as 5 seconds or 100 milliseconds, based on the application scenario or empirical data.

[0137] In some embodiments, the first operation includes: determining the main sloshing characteristics of the liquid inside the hazardous chemical stack based on the fused point cloud, the main sloshing characteristics including the main sloshing frequency and the main sloshing phase; generating a time-varying drive signal based on the main sloshing characteristics, the time-varying drive signal having a frequency the same as the main sloshing frequency, a phase opposite to the main sloshing phase, and an amplitude determined based on the stack slip; and driving an actuator to apply a pulsed force to the shelf with a frequency, phase, and amplitude indicated by the time-varying drive signal based on the time-varying drive signal.

[0138] The dominant sloshing characteristic refers to the set of parameters characterizing the motion properties of the liquid inside a hazardous chemical stack during the sloshing process. For example, the dominant sloshing characteristic may include the dominant sloshing frequency and the dominant sloshing phase.

[0139] The dominant sloshing frequency refers to the frequency component where energy is most concentrated during liquid sloshing, and it is the main parameter characterizing the period of liquid sloshing. The unit of dominant sloshing frequency is Hz.

[0140] The principal sloshing phase is a physical quantity used to describe the state of liquid sloshing at a certain moment, characterizing the instantaneous position of the sloshing waveform. The unit of the principal sloshing phase is radians.

[0141] In some embodiments, the processor initiates emergency intervention measures only when the time of the risk occurrence is less than a preset time threshold; that is, the process of determining the main sway characteristics can be triggered. The fused point cloud is first used to accurately locate the specific location and area of ​​the hazardous chemical stacks with instability risks.

[0142] In some embodiments, the process of determining the main sloshing characteristics includes scanning the risk area indicated by the fused point cloud using millimeter-wave radar. The sloshing of the liquid inside the stack causes minute vibrations on the container surface, resulting in a micro-Doppler effect in the millimeter-wave radar echo signal. The time-frequency spectrum of the signal can be obtained by performing time-frequency analysis (e.g., short-time Fourier transform or wavelet transform) on the echo signal containing the micro-Doppler frequency shift characteristics. The frequency component with the highest energy concentration in this time-frequency spectrum is the main sloshing frequency (typically in the range of 0.1-2 Hz), and the phase of this frequency component at the current moment is the main sloshing phase.

[0143] In some embodiments, the main sloshing characteristics can also be determined by installing high-sensitivity accelerometers at key locations in the hazardous materials stack (e.g., at the top or middle of the stack). The sensors acquire vibration acceleration signals from the containers, and a spectrum is obtained by performing a Fast Fourier Transform (FFT) on the time-series vibration acceleration signals. The peak frequencies in the spectrum correspond to the main sloshing frequencies of the liquid, and the main sloshing phase can be determined by analyzing the zero-crossing points or peak points of the original signal.

[0144] A time-varying drive signal is a signal used for drive control whose one or more parameters (such as frequency, phase, and amplitude) change over time. For example, the time-varying drive signal can be a sine wave signal whose amplitude is adjusted over time, with a specific frequency, phase, and amplitude.

[0145] Stack slip refers to the magnitude of displacement of a hazardous chemical stack or a portion of its surface relative to its initial position or reference datum.

[0146] In some embodiments, the process of determining the stacking slip includes: when the hazardous chemical stacking is in a stable state, acquiring the initial three-dimensional point cloud of the hazardous chemical stacking as a reference point cloud; registering and aligning the currently acquired fused point cloud with the reference point cloud (for example, the Iterative Closest Point (ICP) algorithm can be used); after registration, calculating the horizontal displacement of the current point cloud's coordinates in the risk area relative to the corresponding area coordinates of the reference point cloud, and this displacement is the stacking slip.

[0147] In some embodiments, the time-varying drive signal can be generated into a sinusoidal signal by a signal generator or processor based on the following parameters: the frequency of the time-varying drive signal is set to the main oscillation frequency determined in the previous step. The process is exactly the same; the phase of the time-varying drive signal is set to be opposite to the main oscillation phase φ determined in the previous step, that is, π radians (180 degrees) are added to the main oscillation phase; the amplitude of the time-varying drive signal is dynamically determined according to the stacking slip amount. For example, the amplitude and slip amount can be set to a linear relationship, that is, the larger the slip amount, the larger the signal amplitude, but a maximum amplitude upper limit will be set to protect the actuator.

[0148] For example, the generated time-varying driving signal A(t) can be expressed as: ,in, This indicates the magnitude determined based on the stack slippage. For example, This can be determined using a preset lookup table or piecewise function: when the slip is between 0-5mm, It is 20% of the maximum amplitude; when the slippage is 5-10mm, It is 60% of the maximum range.

[0149] An actuator is a device that converts control signals into specific physical actions (such as force, displacement, or torque). For example, an actuator may include a hydraulic servo valve, a servo motor, or a pneumatic cylinder. A hydraulic servo valve drives a hydraulic cylinder to generate thrust by controlling the flow and direction of hydraulic oil; a servo motor drives a lead screw or gear to produce linear or rotary motion by controlling the current.

[0150] A shelf is a rack-like structure or device used for supporting, placing, or storing goods. For example, a shelf can be a storage rack in a warehouse, a material rack on a production line, or a rack-like structure used for the turnover, display, or exhibition of goods. In this embodiment of the invention, the shelf can be a rack-like structure used for stacking hazardous chemicals.

[0151] A pulse force is a force that acts on an object for a short period of time. For example, a pulse force can manifest as a series of brief, periodic pushing or pulling forces.

[0152] In some embodiments, the actuator may be a hydraulic cylinder controlled by a hydraulic servo valve. A time-varying drive signal (typically a voltage or current signal) generated by the processor is sent to the hydraulic servo valve. The servo valve precisely controls the flow and direction of hydraulic oil entering the hydraulic cylinder according to the amplitude and polarity of the signal, thereby driving the piston rod to apply a periodically changing thrust or pull force to the rack corresponding to the signal waveform (frequency, phase, amplitude), forming a pulse force.

[0153] In some embodiments, the actuator may also be other types of drive devices, such as a ball screw mechanism driven by a servo motor, a pneumatic actuator, a piezoelectric ceramic actuator, or an electromagnetic actuator. The method of applying pulse force can also be diverse, not limited to directly pushing or pulling the shelf, but can also be achieved indirectly by controlling the rotation of eccentric wheels at the bottom of the shelf. This invention does not impose specific limitations in this regard.

[0154] In some embodiments, the method disclosed in this invention further includes a step of temperature control of objects such as stacks in emergency situations (e.g., when the risk occurs less than a preset time threshold).

[0155] In some embodiments, in response to a risk occurrence time less than a time threshold, the processor may: determine a temperature influence correlation coefficient based on the tilt angle change rate and a pre-stored mapping relationship (i.e., a third mapping relationship); in response to a temperature influence correlation coefficient greater than a temperature correlation threshold, determine a cooling power and a cooling location based on the current ambient temperature and the location of the risk occurrence; and control the cooling device to cool the cooling location according to the cooling power.

[0156] The third mapping relationship refers to a data structure or function used to establish the correspondence between the rate of change of tilt angle and temperature-related variables. For example, the third mapping relationship can be a pre-stored mapping table of "rate of change of tilt angle - temperature influence correlation coefficient" (i.e., the first mapping table), which records the corresponding temperature influence correlation coefficient values ​​under different rates of change of tilt angle.

[0157] The temperature effect correlation coefficient is a parameter used to quantify the degree of influence of temperature factors on the instability risk of objects (such as stacks). For example, the temperature effect correlation coefficient can be a dimensionless value between 0 and 1. The larger the value, the more significant the effect of temperature on stack instability, and the more sensitive the instability trend is to temperature factors.

[0158] In some embodiments, the process of determining the temperature effect correlation coefficient can be implemented using a lookup table. The processor pre-stores a first mapping table. Upon obtaining the current tilt angle change rate, the processor can directly look up the corresponding or closest tilt angle change rate value in the first mapping table and read its corresponding temperature effect correlation coefficient. For example, this first mapping table can be used to apply loads to the stack at different temperatures, record the tilt angle change rate and instability data, and fit a third mapping relationship using methods such as regression analysis.

[0159] In some embodiments, the first mapping table can be obtained experimentally. The experimental process includes: applying tilt loads to the stack under different ambient temperature gradients (e.g., 20℃, 30℃, 40℃, 50℃) until it becomes unstable, and simultaneously collecting data on the tilt angle change rate and the instability critical point; using methods such as multiple regression analysis to analyze the correlation between the tilt angle change rate, temperature, and stack stability, and calculating the temperature influence correlation coefficient; and establishing a third mapping relationship between the tilt angle change rate and the temperature influence correlation coefficient through repeated experiments and data fitting, and storing it as the first mapping table.

[0160] In some embodiments, the third mapping relationship can also be a pre-trained function model, such as a regression model. This function model takes the rate of change of tilt angle as input and directly calculates and outputs the correlation coefficient of temperature influence.

[0161] The temperature correlation threshold is a preset reference value used to determine whether the correlation coefficient of temperature influence reaches a level that requires the initiation of temperature control measures. For example, the temperature correlation threshold can be set to 0.7. When the calculated temperature influence correlation coefficient is greater than 0.7, it is determined that temperature is a significant factor causing the current risk, and cooling measures need to be initiated.

[0162] The current ambient temperature refers to the real-time temperature value of the environment in which an object (such as a stack) is located. For example, the current ambient temperature can be measured in real time by one or more temperature sensors (such as thermocouples or infrared thermometers) deployed around the stack.

[0163] Cooling power refers to the energy output or operating intensity set for performing a cooling operation. For example, cooling power can be a specific value, such as 500 watts (W), which is used to control the cooling intensity of the cooling device.

[0164] The cooling location refers to the target spatial location or area where cooling operations need to be performed. For example, the cooling location can be directly set to the same area as the location where the risk occurs.

[0165] In some embodiments, the cooling location can be determined directly using the risk location identified in the preceding steps. The processor can directly assign the three-dimensional coordinates or region identifier of the risk location to the cooling location to achieve precise cooling of the risk source.

[0166] In some embodiments, the cooling power can be determined using a preset calculation formula. For example, the cooling power can be calculated using the following formula: Cooling power = Power adjustment coefficient Temperature effect correlation coefficient (Current ambient temperature - Upper limit of safe storage temperature for hazardous chemicals) + Basic cooling power. The power adjustment coefficient, upper limit of safe storage temperature, and basic cooling power are all preset parameters.

[0167] In some embodiments, the determination of cooling power can also be based on a tiered strategy. The processor can preset multiple cooling power levels (such as low, medium, and high), and match a preset cooling power level to the risk based on the temperature influence correlation coefficient and the difference between the current ambient temperature and the safe temperature.

[0168] In some embodiments, the determination of cooling power and cooling location can also be dynamically adjusted by combining factors such as the thermal conductivity of the material and the efficiency of the cooling device, through a more complex physical model.

[0169] A cooling device is a piece of equipment or system used to perform cooling operations on a designated location or area. For example, a cooling device can be an air-cooled system, a water-cooled spray system, a thermoelectric cooler, or a dry ice spraying device.

[0170] In some embodiments, the method of controlling the cooling device depends on the specific type of device. For example, if the cooling device is an air-cooled system, the control command can adjust the fan speed and cooling status; if it is a water-cooled spray system, the control command can control the opening of specific nozzles and the spray flow rate or duration.

[0171] In some embodiments, the control system can interact with the cooling device via standard communication protocols (such as Modbus or CAN bus) to achieve closed-loop or open-loop control of the cooling process. Furthermore, the control method can be adapted to other types of cooling devices, such as controlling the input current of a thermoelectric cooler.

[0172] In some embodiments of this invention, by introducing a temperature-related control mechanism, the impact of temperature on instability risk can be quickly determined based on the rate of change of tilt angle in emergency situations. When the impact is significant, the system can automatically calculate the cooling power and location, and control the device to perform precise local cooling. This effectively suppresses the decrease in material strength or the acceleration of chemical reactions caused by high temperatures, buying time for emergency response and significantly improving the accuracy and safety of the response.

[0173] In some embodiments, the processor may perform a second operation in response to a risk occurring at a time threshold not less than a certain time.

[0174] In some embodiments, the second operation includes: determining the target barrier based on the location of the risk occurrence; sending a trigger command to the electromechanical drive unit corresponding to the target barrier to drive the target barrier to switch from a retracted state to an extended state; and generating an electronic early warning data packet containing the coordinates of the risk occurrence location and sending it to the central control system.

[0175] In some embodiments, when the processor determines that the time since the risk occurred is not less than a time threshold, it means there is sufficient time to execute proactive protective measures. At this point, the processor will determine the target isolation barrier that needs to be activated based on the location of the risk.

[0176] Targeted barriers are protective devices used to physically isolate specific areas to prevent the spread of risk or entry of personnel. For example, a targeted barrier can be an automatically retractable metal fence, protective netting, or barrier installed at critical points in roadways or tunnels.

[0177] In some embodiments, the process of determining the target barrier can be based on a pre-built fourth mapping relationship. For example, the fourth mapping relationship can be represented by a "risk occurrence location-barrier" mapping table (i.e., a second mapping table) stored internally by the processor. This second mapping table records the correspondence between each gridded protection area and a specific target barrier device ID. Once the coordinates of the risk occurrence location are obtained, the processor can directly determine the target barrier responsible for protecting that risk occurrence location by querying the second mapping table.

[0178] In some embodiments, the process of determining target barriers can also be implemented through real-time calculation. For example, after the processor obtains the coordinates of the location where the risk occurs, it calculates the spatial distance between that location and all deployed target barriers in real time. Subsequently, the processor selects one or more target barriers closest to the location where the risk occurs as the target barriers to be activated, in order to form the most effective isolation barrier.

[0179] In some embodiments, the determination of target barriers can also be based on preset logical rules, such as selecting one or more target barriers that can most effectively block the risk propagation path according to a risk diffusion model.

[0180] In some embodiments, after determining the target isolation barrier, the processor sends a trigger command to the electromechanical drive unit corresponding to the target isolation barrier to drive the target isolation barrier to quickly switch from the stored state to the unfolded state, so as to form physical isolation.

[0181] An electromechanical drive unit is a device that integrates a motor and mechanical structure to provide power to drive a target barrier to change its physical state. For example, an electromechanical drive unit may include a motor, a reducer, a rack and pinion mechanism, or a hydraulic / pneumatic actuator to realize the unfolding and retraction of the target barrier.

[0182] A trigger command is an electrical signal or data instruction used to start or trigger a specific device to perform a predetermined action. For example, a trigger command may be a specific digital code or level signal sent to an electromechanical drive unit to command it to immediately drive the target barrier to unfold, switching the target barrier from a retracted state (folded or retracted) to an unfolded state (upright or unfolded).

[0183] The "folded-up state" refers to the folded, retracted, or collapsed form of the target barrier when it is not in operation or on standby. In this state, it occupies less space and does not constitute an obstacle. For example, the folded-up state can be the grid structure of the target barrier folded tightly against the side wall of the alley, or the protective netting rolled up in a roll box.

[0184] The deployed state refers to the unfolded, extended, or erected state of the target barrier when performing its isolation function, in which it can effectively close or isolate a designated area. For example, the deployed state can be the target barrier's grid structure being fully extended and locked, spanning the entire width of an alley or passageway.

[0185] In some embodiments, the trigger command is sent via industrial Ethernet. The control unit sends the encapsulated command to the local controller of the target barrier. After parsing the command, the controller drives the motor and rack and pinion mechanism to unfold the target barrier along a predetermined track.

[0186] In some embodiments, the trigger command can also be sent wirelessly (such as LoRa or Zigbee). This method is suitable for scenarios where wiring is difficult, where the control unit wirelessly sends the command to the receiving module of the electromechanical drive unit to initiate the isolation action.

[0187] In some embodiments, in addition to sending trigger commands, the processor also generates electronic warning data packets containing the coordinates of the location where the risk occurs and sends them to the central control system for monitoring personnel to make decisions and record.

[0188] Risk location coordinates refer to one or more sets of values ​​used to accurately describe the location of a risk in a specific coordinate system. For example, risk location coordinates can be (X, Y, Z) values ​​represented in a three-dimensional Cartesian coordinate system, used to pinpoint the precise location of the risk point in space.

[0189] An electronic early warning data packet is a structured data unit that encapsulates information related to risk warnings according to a predetermined format, and is used for transmission between different systems or devices. For example, an electronic early warning data packet can be a data file in JSON or XML format, and its internal fields may include information such as the coordinates of the risk location, risk level, risk type, sensor data, and timestamps (such as the expected time of occurrence).

[0190] A central control system is a system used for centralized monitoring, management, and decision-making. It can receive data from front-end devices and send control commands to execution units. For example, a central control system can be a software platform deployed in a ground dispatch center or in the cloud, used to manage the stacking and shelving of hazardous chemicals, and has functions such as data display, alarm handling, log recording, and remote control.

[0191] For example, electronic early warning data packets can be encapsulated in JSON format and sent to the central control system via the TCP / IP protocol. Alternatively, they can be encapsulated in XML format to adapt to different system interface requirements. Furthermore, to improve transmission efficiency and real-time performance, a more compact binary protocol can be used for data encoding and transmission.

[0192] In some embodiments of this invention, by automatically deploying the target isolation barrier and simultaneously sending early warning data before a risk occurs, closed-loop control from risk prediction to proactive isolation and remote early warning is achieved. This method requires no manual intervention, greatly shortens emergency response time, effectively prevents the spread of risk, provides accurate decision-making basis for the central control system, and significantly improves operational safety.

[0193] The technical solutions provided in some embodiments of this invention significantly improve the accuracy and robustness of hazardous chemical stacking status perception in complex environments such as industrial sites with smoke, dust, and water vapor by fusing point cloud data from high-frequency millimeter-wave sensors and lidar and performing weighted registration. Based on high-precision fused point clouds, risks are dynamically predicted by analyzing the rate of change of tilt angle, realizing a shift from passive alarm to proactive prevention. For imminent risks, an innovative approach is taken to generate an inverse time-varying drive signal by analyzing the main sway characteristics, driving the actuator to apply a precise reverse pulse force. This pulse force provides physical support while actively counteracting the swaying inertial force of the liquid inside the hazardous chemical stack, greatly enhancing the initiative and effectiveness of emergency response, achieving intelligent hierarchical proactive control, and effectively preventing collapse accidents.

[0194] This invention also provides a computer-readable storage medium storing computer instructions. When the computer instructions are executed on a computer, the computer performs a method for controlling the collapse of hazardous chemical shelving based on multi-sensor monitoring, as described in any embodiment of this invention.

[0195] The basic concepts have been described above. It is clear that the detailed disclosure above is merely illustrative and does not constitute a limitation of the present invention. Although not explicitly stated herein, various modifications, improvements, and corrections may be made to the present invention by those skilled in the art. Such modifications, improvements, and corrections are suggested in this invention and therefore remain within the spirit and scope of the exemplary embodiments of the present invention.

[0196] Meanwhile, specific terms are used to describe embodiments of the invention. For example, "an embodiment," "one embodiment," and / or "some embodiments" refer to a particular feature, structure, or characteristic related to at least one embodiment of the invention. Therefore, it should be emphasized and noted that "an embodiment," "one embodiment," or "an alternative embodiment" mentioned twice or more in different locations in this invention do not necessarily refer to the same embodiment. Furthermore, certain features, structures, or characteristics in one or more embodiments of the invention can be appropriately combined.

[0197] Furthermore, unless expressly stated in the claims, the order of processing elements and sequences, the use of numbers and letters, or other names described in this invention are not intended to limit the order of the processes and methods of this invention. Although the foregoing disclosure has discussed some currently considered useful embodiments of the invention through various examples, it should be understood that such details are for illustrative purposes only, and the appended claims are not limited to the disclosed embodiments; rather, the claims are intended to cover all modifications and equivalent combinations that conform to the spirit and scope of the embodiments of this invention. For example, while the system components described above can be implemented by hardware devices, they can also be implemented solely by software solutions, such as installing the described system on existing servers or mobile devices.

[0198] Similarly, it should be noted that, in order to simplify the description of this invention and thus aid in the understanding of one or more embodiments, the foregoing description of the embodiments of this invention sometimes combines multiple features into a single embodiment, drawing, or description thereof. However, this disclosure method does not imply that the subject matter of the invention requires more features than those mentioned in the claims. In fact, the embodiments contain fewer features than all the features of the single embodiment disclosed above.

[0199] In some embodiments, numbers describing the quantity of components and attributes are used. It should be understood that such numbers used in the description of embodiments are modified in some examples with the terms "approximately," "approximately," or "generally." Unless otherwise stated, "approximately," "approximately," or "generally" indicates that the numbers are allowed to vary by ±20%. Accordingly, in some embodiments, the numerical parameters used in the specification and claims are approximate values, which may be changed depending on the characteristics required by individual embodiments. In some embodiments, numerical parameters should take into account specified significant digits and employ a general method of digit reservation. Although the numerical ranges and parameters used to confirm their breadth of range in some embodiments of the invention are approximate values, in specific embodiments, such values ​​are set as precisely as feasible.

[0200] For each patent, patent application, patent application publication, and other material, such as articles, books, specifications, publications, and documents, referenced in this invention, the entire contents of which are incorporated herein by reference. This excludes historical application documents that are inconsistent with or conflict with the content of this invention, as well as documents that limit the broadest scope of the claims of this invention (currently or subsequently appended to this invention). It should be noted that if there are any inconsistencies or conflicts between the descriptions, definitions, and / or terminology used in the appended materials of this invention and the content of this invention, the descriptions, definitions, and / or terminology used in this invention shall prevail.

[0201] Finally, it should be understood that the embodiments described in this invention are merely illustrative of the principles of the invention. Other modifications may also fall within the scope of this invention. Therefore, alternative configurations of the embodiments of this invention are considered as examples and not limitations, and are regarded as consistent with the teachings of this invention. Accordingly, the embodiments of this invention are not limited to those explicitly described and illustrated herein.

Claims

1. A method for controlling the collapse prevention of hazardous chemical shelving based on multi-sensor monitoring, characterized in that, include: A first point cloud and a second point cloud are obtained, wherein the first point cloud is generated by scanning the surface of the hazardous chemical stack by a high-frequency millimeter-wave sensor, and the second point cloud is generated by scanning the hazardous chemical stack by a lidar. The first point cloud and the second point cloud are weighted and registered to obtain a fused point cloud. The tilt angle change rate is determined based on the fused point cloud; Based on the rate of change of inclination angle and the material characteristics of the hazardous chemical, the timing and location of the risk of the hazardous chemical incident are estimated; and In response to the risk occurring less than a time threshold, The main sloshing characteristics of the liquid inside the hazardous chemical stack are determined based on the fused point cloud, and the main sloshing characteristics include the main sloshing frequency and the main sloshing phase; Based on the main sway characteristics, a time-varying driving signal is generated. The time-varying driving signal has the same frequency as the main sway frequency, a phase opposite to the main sway phase, and an amplitude determined based on the stacking slip amount. Based on the time-varying drive signal, the drive actuator applies a pulse force to the shelf with the frequency, phase, and amplitude indicated by the time-varying drive signal.

2. The method according to claim 1, characterized in that, The step of weighted registration of the first point cloud and the second point cloud to obtain the fused point cloud includes: Obtain a pre-trained weight model; Based on the current smoke concentration data and the pre-trained weight model, a reliability weight sequence is determined; and The first point cloud and the second point cloud are weighted and registered according to the reliability weight sequence to obtain the fused point cloud.

3. The method according to claim 1, characterized in that, The determination of the tilt angle change rate based on the fused point cloud includes: Acquire the initial twin model and the acoustic emission signals collected by the acoustic sensors; Based on the fused point cloud and the acoustic emission signal, the initial twin model is updated to obtain the updated twin model; and Based on the updated twin model, the rate of change of tilt angle is determined.

4. The method according to claim 3, characterized in that, Determining the rate of change of tilt angle based on the updated twin model includes: Obtain the current operating status; Based on the current operating conditions and the updated twin model, calculate the deviation between the surface deformation predicted by the updated twin model and the fused point cloud; When the deviation exceeds a preset threshold, the displacement field of the hazardous chemical stacking surface is calculated using the updated twin model; and The rate of change of tilt angle is determined based on the position of the slip surface and the amount of slip in the displacement field.

5. The method according to claim 1, characterized in that, The method further includes: In response to the risk occurring at a time not less than the time threshold, a target isolation fence is determined based on the location of the risk. Send a trigger command to the electromechanical drive unit corresponding to the target barrier to drive the target barrier to switch from a retracted state to an unfolded state; and An electronic early warning data packet containing the coordinates of the location where the risk occurred is generated and sent to the central control system.

6. A multi-sensor monitoring-based anti-collapse control system for hazardous chemical shelving, characterized in that, include: At least one processor; as well as At least one memory storing computer instructions, wherein when the computer instructions are executed by the processor, the processor is configured to: A first point cloud and a second point cloud are obtained, wherein the first point cloud is generated by scanning the surface of the hazardous chemical stack by a high-frequency millimeter-wave sensor, and the second point cloud is generated by scanning the hazardous chemical stack by a lidar. The first point cloud and the second point cloud are weighted and registered to obtain a fused point cloud. The tilt angle change rate is determined based on the fused point cloud; Based on the rate of change of inclination angle and the material characteristics of the hazardous chemical, the timing and location of the risk of the hazardous chemical incident are estimated; and In response to the risk occurring less than a time threshold, The main sloshing characteristics of the liquid inside the hazardous chemical stack are determined based on the fused point cloud, and the main sloshing characteristics include the main sloshing frequency and the main sloshing phase; Based on the main sway characteristics, a time-varying driving signal is generated. The time-varying driving signal has the same frequency as the main sway frequency, a phase opposite to the main sway phase, and an amplitude determined based on the stacking slip amount. Based on the time-varying drive signal, the drive actuator applies a pulse force to the shelf with the frequency, phase, and amplitude indicated by the time-varying drive signal.

7. The system according to claim 6, characterized in that, The processor is also configured to: Obtain a pre-trained weight model; Based on the current smoke concentration data and the pre-trained weight model, a reliability weight sequence is determined; and The first point cloud and the second point cloud are weighted and registered according to the reliability weight sequence to obtain the fused point cloud.

8. The system according to claim 6, characterized in that, The processor is also configured to: Acquire the initial twin model and the acoustic emission signals collected by the acoustic sensors; Based on the fused point cloud and the acoustic emission signal, the initial twin model is updated to obtain the updated twin model; and Based on the updated twin model, the rate of change of tilt angle is determined.

9. The system according to claim 6, characterized in that, The processor is also configured to: In response to the risk occurring at a time not less than the time threshold, a target isolation fence is determined based on the risk location; A trigger command is sent to the electromechanical drive unit corresponding to the target isolation barrier to drive the target isolation barrier to switch from the stored state to the unfolded state; as well as An electronic early warning data packet containing the coordinates of the risk location is generated and sent to the central control system.

10. A computer-readable storage medium, characterized in that, The storage medium stores computer instructions, which, when executed on a computer, cause the computer to perform a multi-sensor monitoring-based method for preventing the collapse of hazardous chemical shelving, as described in claims 1-5.

Images