Steel logistics intelligent access control system based on Internet of Things

Abstract

The invention relates to the technical field of logistics management and control, in particular to an intelligent steel logistics access control system based on the Internet of Things, which comprises a data processing module, a data processing module and a data processing module, chassis vibration time sequence data of an entering vehicle and a vehicle-mounted steel cargo stacking topology point cloud are obtained respectively; and performing frequency domain feature extraction and noise reduction processing on the chassis vibration time sequence data to obtain a chassis microcosmic frequency conversion domain feature spectrum. According to the method, the chassis microcosmic frequency conversion domain characteristic spectrum and the cargo stacking topological point cloud are fused, the chassis microcosmic frequency conversion domain characteristic spectrum and the cargo stacking topological point cloud are associated and coupled through Euler-Lagrange dynamics mapping, a load dynamic fingerprint rule with physical uniqueness is generated, the fingerprint distinguishes the total weight and recognizes the subtle difference between load distribution and a fixing mode, and the dynamic fingerprint identification accuracy is improved. And the possibility of logistics fraud through traditional means such as simple weight balancing, local hiding or license plate replacement is avoided.

Description

Technical Field

[0001] This invention relates to the field of logistics management and control technology, specifically to an intelligent access control system for steel logistics based on the Internet of Things. Background Technology

[0002] With the advancement of digital transformation and intelligent manufacturing in the steel industry, steel logistics, as a crucial link connecting production and the supply chain, directly impacts the operating costs of enterprises through its efficiency and security management. In large steel industrial parks, the frequency of daily vehicle traffic is extremely high, and the goods carried are typically heavy, high-value, and non-standardized in geometric shape. Therefore, logistics access control systems not only perform vehicle identification but also bear the important responsibilities of cargo verification, fraud prevention, and transportation safety assessment.

[0003] Existing steel logistics access control systems typically employ a combination of license plate recognition, weighbridge weighing, and manual inspection. This involves using cameras to identify license plates to confirm vehicle identity, using weighbridges to obtain the vehicle's total weight, and relying on manual visual inspection of the goods. However, with the advancement of logistics fraud methods and increasing demands for transportation security, traditional access control systems have gradually revealed a series of insurmountable technical flaws, specifically:

[0004] First, the detection methods are too static and isolated. Existing access control systems only focus on the static total weight and surface images of vehicles, and cannot perceive the actual physical response of the vehicle chassis when carrying heavy loads. The high density of steel goods will have a significant stress effect on the vehicle suspension system. Static weighing alone cannot determine whether the goods are securely fixed, whether the chassis suspension is within the safe working range, or identify potential transportation risks caused by suspension modification or aging.

[0005] Second, there are blind spots in the anti-fraud capabilities. Driven by profit, logistics fraud methods are emerging one after another, such as privately modifying water tanks to add water and increase weight, changing license plates or using fake plates, or smuggling heavy objects in empty vehicles. Existing weight detection cannot distinguish between the effective weight of goods and the fraudulent counterweights. Moreover, video recognition is limited by lighting and occlusion, making it difficult to accurately quantify the stacking topology of goods.

[0006] Third, the minute vibrations of the vehicle chassis under load, the deformation of the tires, and the hysteresis behavior of the suspension contain a wealth of load information. However, existing technologies have not effectively utilized IoT sensing technology to capture this dynamic time-series data, making it impossible for access control systems to determine from a physical perspective whether the current loading status conforms to historical compliance patterns, thus hindering the achievement of truly intelligent and precise control. Summary of the Invention

[0007] To achieve the above objectives, the present invention provides the following technical solution: an intelligent access control system for steel logistics based on the Internet of Things, comprising:

[0008] The data processing module is used to acquire the time-series vibration data of the chassis of the incoming vehicle and the topological point cloud of the stacked steel cargo on the vehicle through a piezoelectric vibration sensor array and a three-dimensional lidar scanning device deployed in the access control channel; and to perform frequency domain feature extraction and noise reduction processing on the chassis vibration time-series data to obtain the frequency domain feature spectrum of chassis micro-strain.

[0009] The fingerprint generation module is used to perform load-chassis stiffness coupling mapping on the frequency domain feature spectrum of the chassis micro-strain based on the stacked topology point cloud of the vehicle-mounted steel cargo to obtain stiffness coupling mapping data; and to perform nonlinear deformation hysteresis behavior learning on the stiffness coupling mapping data to generate load dynamic fingerprint patterns.

[0010] The access control module is used to construct a vehicle access security verification model based on the dynamic fingerprint pattern of the load, and obtain a security verification decision model; the security verification decision model is deployed to the intelligent access control terminal to execute the passage control of steel logistics vehicles.

[0011] Preferably, the chassis vibration time series data is subjected to frequency domain feature extraction and noise reduction processing to obtain the chassis micro-strain frequency domain feature spectrum, including:

[0012] The chassis vibration time series data were processed by Hilbert-Huang transform and time window slices were marked to construct a vibration mode time-frequency diagram;

[0013] The vibration mode time-frequency diagram is processed by background noise spectrum subtraction to obtain a clean vibration mode diagram.

[0014] The pure vibration mode spectrum is smoothed by Fast Fourier Transform to obtain the frequency domain characteristic spectrum of the chassis micro-strain.

[0015] Preferably, the method involves performing load-chassis stiffness coupling mapping on the frequency domain feature spectrum of the chassis micro-strain based on the stacked topological point cloud of the vehicle-mounted steel cargo to obtain stiffness coupling mapping data; and performing nonlinear deformation hysteresis behavior learning on the stiffness coupling mapping data to generate load dynamic fingerprint patterns, including:

[0016] Extract the geometric centroid distribution and edge contour curvature of the cargo in the topological point cloud of the vehicle-mounted steel cargo stack;

[0017] Based on the geometric centroid distribution of the cargo and the curvature of the edge contour, the frequency domain characteristic spectrum of the chassis micro-strain is subjected to Euler-Lagrange dynamic mapping to obtain dynamic response mapping data.

[0018] The dynamic response mapping data is subjected to nonlinear deformation hysteresis behavior learning, and the load response learning data under deformation hysteresis state is output.

[0019] The load response learning data is subjected to chassis-cargo coupled fingerprint quantization under the influence of gravity field to generate the dynamic fingerprint pattern of the load.

[0020] Preferably, the method for learning nonlinear deformation hysteresis behavior from the dynamic response mapping data and outputting load response learning data under deformation hysteresis state includes:

[0021] Extract the natural frequency offset and amplitude attenuation coefficient of the suspension system from the dynamic response mapping data, and obtain the vehicle self-weight calibration information;

[0022] Based on the vehicle self-weight calibration information and the amplitude attenuation coefficient, the elastic potential energy dissipation gradient under static-dynamic load conversion is calculated to obtain the elastic potential energy dissipation gradient.

[0023] Based on the elastic potential energy dissipation gradient, the transient impact response index between the stress wave transmission correlation is analyzed on the dynamic response mapping data to obtain the transient impact response index.

[0024] Based on the natural frequency offset of the suspension system and the transient impact response index, the tire-ground contact surface slip loss is derived to obtain slip loss data.

[0025] Based on the slip loss data, nonlinear deformation hysteresis behavior learning is performed, and the load response learning data under the deformation hysteresis state is output.

[0026] Preferably, the transient impact response index is obtained by analyzing the stress wave propagation correlation between the dynamic response mapping data based on the elastic potential energy dissipation gradient, including:

[0027] The interlayer frictional resistance torque of the leaf spring and the viscous resistance torque of the hydraulic damper are calculated based on the elastic potential energy dissipation gradient at different load-bearing stages.

[0028] The instantaneous restoring force required by the vehicle chassis system is analyzed based on the interlayer frictional resistance torque of the leaf spring and the viscous resistance torque of the hydraulic shock absorber.

[0029] Based on the instantaneous restoring force required by the vehicle chassis system, the dynamic response mapping data is coupled with the restoring force-deformation displacement to obtain restoring force-displacement correlation data;

[0030] The instantaneous phase plane energy loop is obtained by integrating the restoring force displacement correlation data in the phase plane.

[0031] The transient impact response index is obtained by analyzing the transient impact response index between stress wave transmission correlations based on the instantaneous phase plane energy loop.

[0032] Preferably, the tire-ground contact surface slip loss is derived based on the natural frequency offset of the suspension system and the transient impact response index to obtain slip loss data, including:

[0033] The dynamic stiffness hardening state of the tire sidewall is analyzed based on the natural frequency offset of the suspension system.

[0034] Based on the dynamic stiffness hardening state of the tire sidewall, nonlinear shrinkage analysis of the contact imprint area was performed to obtain imprint shrinkage data.

[0035] Analyze the trend of hysteresis loss tangent of tire rubber material based on the imprint shrinkage data;

[0036] Based on the trend of the hysteresis loss tangent angle and the transient impact response index, a rolling resistance regression analysis is performed to quantify the gradient of traction force transmission efficiency loss.

[0037] Based on the traction transmission efficiency loss gradient, the tire-ground contact surface slip loss is derived to obtain slip loss data.

[0038] Preferably, the load response learning data is subjected to chassis-cargo coupled fingerprint quantization under the influence of gravity field to generate the load dynamic fingerprint pattern, including:

[0039] Analyze the nonlinear damping evolution data of the suspension system under the influence of the gravity field in the load response learning data;

[0040] Chaotic attractor trajectory analysis is performed on the nonlinear damping evolution data to obtain damped chaotic trajectory data;

[0041] Based on the damped chaotic trajectory data, Lyapunov exponential convergence calculation is performed on the nonlinear damped evolution data to obtain stability convergence exponential data.

[0042] Based on the support vector machine algorithm and the stability convergence index data, the load fingerprint features driven by the gravity field dimension are quantified on the nonlinear damping evolution data to generate the dynamic fingerprint pattern of the load.

[0043] Preferably, the method involves constructing a vehicle access safety verification model based on the dynamic fingerprint pattern of the load to obtain a safety verification decision model; and deploying the safety verification decision model to an intelligent access control terminal to execute steel logistics vehicle access control, including:

[0044] Principal component analysis was performed to reduce the dimensionality of the dynamic fingerprint of the load, resulting in a high-dimensional fingerprint feature vector.

[0045] Historical compliance trajectory regression learning is performed on the high-dimensional fingerprint feature vector based on a long short-term memory network to obtain compliance confidence learning data.

[0046] The vehicle access safety verification model is constructed based on the compliance confidence learning data to obtain the safety verification decision model;

[0047] The security verification decision model is deployed to the intelligent access control terminal to receive real-time fingerprint data of vehicles for comparison and output gate opening and closing control commands to execute the passage control of steel logistics vehicles.

[0048] Preferably, historical compliance trajectory regression learning is performed on the high-dimensional fingerprint feature vector based on a long short-term memory network to obtain compliance confidence learning data, including:

[0049] The high-dimensional fingerprint feature vector is segmented into a time series sliding window, and the feature fluctuation sequence of each window is extracted to obtain the fingerprint fluctuation sequence.

[0050] The parameters of the forget gate, input gate, and output gate of the long short-term memory network are initialized based on the fingerprint fluctuation sequence to obtain the initial state matrix of the long short-term memory network.

[0051] The fingerprint fluctuation sequence is processed by bidirectional recursive propagation through the initial state matrix of the Long Short-Term Memory Network, and the cell state update and hidden layer output are calculated step by step to obtain fingerprint time-series dependent output data.

[0052] Historical compliance trajectory regression learning is performed based on the fingerprint time-series dependent output data to obtain the compliance confidence learning data.

[0053] Compared with existing technologies, the advantages of this invention are as follows: By fusing the frequency domain characteristic spectrum of chassis micro-strain and the topological point cloud of cargo stacking, and using Eulerian-Lagrange dynamic mapping to couple the two, a physically unique load dynamic fingerprint pattern is generated. This fingerprint is quantified based on the inherent physical characteristics of the nonlinear interaction between the chassis and cargo under a gravitational field. It not only distinguishes the total weight but also identifies subtle differences in load distribution and fixing methods, eliminating the possibility of logistics fraud through traditional methods such as simple counterweighting, partial concealment, or license plate replacement. Through nonlinear deformation hysteresis behavior learning, the elastic potential energy dissipation gradient and transient impact response index of the suspension system can be analyzed. This not only accurately assesses the impact of actual loads on the vehicle chassis stiffness but also identifies potential... Suspension system fatigue or abnormal modifications, especially through the derivation of tire-ground contact surface slip loss, quantify the gradient of traction transmission efficiency loss, providing microscopic data on chassis operating status that traditional systems cannot provide, and offering a deeper assessment basis for the dynamic transportation safety of vehicles; based on stability convergence index data obtained from chaotic attractor trajectory analysis and Lyapunov exponent convergence calculation, the stability of the vehicle-cargo coupling system is quantified from the perspective of nonlinear dynamics; using a long short-term memory network to perform historical compliance trajectory regression learning on high-dimensional fingerprint feature vectors, the system can memorize and learn the dynamic characteristics of compliant vehicles under different loads, thereby determining whether the current vehicle load state is a normal fluctuation on the compliance trajectory, thus achieving more intelligent access safety verification. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the overall system architecture in one embodiment of the present invention;

[0055] Figure 2 This is a flowchart of the overall system in one embodiment of the present invention. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] Please see Figure 1 and Figure 2 The IoT-based intelligent access control system for steel logistics includes the following modules:

[0058] The data processing module is used to acquire the time-series vibration data of the chassis of the incoming vehicle and the topological point cloud of the stacked steel cargo on the vehicle through the piezoelectric vibration sensor array and the three-dimensional lidar scanning device deployed in the access control channel; and to perform frequency domain feature extraction and noise reduction processing on the chassis vibration time-series data to obtain the frequency domain feature spectrum of chassis micro-strain.

[0059] In this embodiment of the invention, high-precision MEMS accelerometers are pre-embedded underground in the intelligent access control channel of the park. The sensors are arranged in a 4×4 matrix with a spacing of 0.5 meters to cover the chassis vibration area of ​​vehicles with different wheelbases. The sampling frequency is set to 2048Hz to ensure that it can capture the minute vibration details of the chassis. At the same time, a 64-line three-dimensional LiDAR is deployed on the gantry above the channel with a scanning frequency of 10Hz and a vertical angular resolution of 0.4 degrees to acquire point cloud data of vehicles and goods. The acquired data is transmitted to the edge computing gateway through optical fiber. The signal processing software performs Hilbert-Huang transform on the chassis vibration time series data, removes environmental white noise and power frequency interference through empirical mode decomposition, extracts the first 5 intrinsic mode functions, and performs fast Fourier transform to generate the chassis micro-strain frequency domain feature spectrum in the range of 0-1000Hz. This feature spectrum is stored in the form of a 500×1 dimension vector.

[0060] The fingerprint generation module is used to perform load-chassis stiffness coupling mapping on the frequency domain feature spectrum of chassis micro-strain based on the topological point cloud of the stacked steel cargo on the vehicle, and obtain stiffness coupling mapping data; nonlinear deformation hysteresis behavior learning is performed on the stiffness coupling mapping data to generate load dynamic fingerprint law.

[0061] In this embodiment of the invention, based on the topological point cloud of stacked steel cargo on a vehicle, geometric features are first extracted. A point cloud processing library is used to perform voxel grid downsampling and statistical filtering on the original point cloud to remove outliers. The ground and vehicle body are segmented using a random sample consensus algorithm. Euclidean clustering is used to identify the cargo region, and the geometric volume, stacking height, and centroid coordinates of the cargo are calculated to construct a load distribution model. Subsequently, a vehicle-road coupled dynamics model is introduced, using the load distribution model as the input boundary condition. The least squares method is used to establish the mapping relationship between the frequency domain feature spectrum of the chassis micro-strain and the stiffness coefficient of the vehicle suspension system, obtaining stiffness coupling mapping data. Based on this, considering the nonlinear characteristics of the heavy vehicle suspension system, an improved Bouc-Wen hysteresis model is introduced, whose core nonlinear restoring force... It can be represented as: ,in This is the proportionality coefficient of the stiffness before and after hysteresis. For initial stiffness, For displacement, For lagged variables, satisfying nonlinear differential equations The model utilizes genetic algorithms or particle swarm optimization algorithms to... , , , The parameters are identified, and the characteristics such as the area and energy dissipation coefficient of the hysteresis loop are analyzed. Finally, a load dynamic fingerprint law containing time-varying stiffness, damping ratio and hysteresis characteristics is generated. This load dynamic fingerprint law can uniquely characterize the physical response state of the current vehicle under a specific load.

[0062] The access control module is used to construct a vehicle access security verification model based on the dynamic fingerprint pattern of the load, and obtain a security verification decision model; the security verification decision model is deployed to the intelligent access control terminal to execute the passage control of steel logistics vehicles.

[0063] In this embodiment of the invention, a deep learning classification model is constructed based on the dynamic fingerprint pattern of the load. In this example, a bidirectional long short-term memory network combined with an attention mechanism is selected. This network consists of two layers of Bi-LSTM, each containing 128 neurons, with a hidden state dimension of... The system first uses a Bi-LSTM layer followed by a 32-dimensional fully connected layer, then an attention mechanism layer, and finally a fully connected layer with a Sigmoid activation function to output the confidence probability between 0 and 1. First, the dynamic fingerprint data of historical compliant vehicle loads is labeled to construct a positive sample set. Then, a negative sample set is generated by simulating abnormal loading, such as overloading, partial loading, and mixed empty loading. The dataset is divided into training and validation sets. The model is trained using the Adam optimizer with a binary cross-entropy loss function. The trained safety verification decision model can output the confidence probability of vehicle loading compliance. After training, the model is converted to the lightweight ONNX format and distributed to the intelligent access control terminal deployed at the access control site via a security protocol. The terminal runs the model in real time. When the confidence level is higher than a set threshold, compliance is determined and the gate is opened; otherwise, an audible and visual alarm is triggered and the data is uploaded to the cloud platform for manual review.

[0064] It should be noted that MEMS accelerometers belong to the category of microelectromechanical systems sensors, characterized by their small size, high precision, and strong vibration resistance, making them suitable for underground installations to collect minute vibrations. The Hilbert-Huang transform (HWH) for chassis vibration time-series data is a time-frequency analysis method applicable to nonlinear and non-stationary signals. Empirical mode decomposition (EMD) is the core component of the HWH, used to decompose complex non-stationary signals into a finite number of eigenmode functions. The Fast Fourier Transform (FFT) is an efficient algorithm for calculating the Discrete Fourier Transform (DFT), used here to convert the time-domain signal to the frequency domain and extract its energy spectrum characteristics. An improved Bouc-Wen hysteresis model is introduced, a mathematical model of a nonlinear suspension system used to describe energy dissipation and hysteresis characteristics under periodic loads. The parameters in its formula... Represents the ratio of elastic stiffness to hysteresis stiffness. and Used to control the shape and size of the hysteresis loop, This controls the smoothness of the hysteresis loop; after the model is trained, it is converted to the lightweight ONNX format, which is an open neural network exchange format that is easy to deploy efficiently on edge computing devices.

[0065] In an optional embodiment, frequency domain feature extraction and noise reduction are performed on the chassis vibration time series data to obtain the frequency domain feature spectrum of chassis micro-strain, including:

[0066] The chassis vibration time series data were processed by Hilbert-Huang transform and time window slices were marked to construct the vibration mode time-frequency diagram;

[0067] The background noise spectrum was subtracted from the vibration mode time-frequency diagram to obtain a clean vibration mode diagram.

[0068] The pure vibration mode spectrum was smoothed by Fast Fourier Transform to obtain the frequency domain characteristic spectrum of chassis micro-strain.

[0069] In this embodiment of the invention, a digital signal processing chip is used to perform Hilbert-Huang transform on the original chassis vibration time series data. Specifically, the operation includes: fitting the upper and lower envelopes of the signal using cubic spline interpolation, calculating the mean curve, subtracting the mean curve from the original signal to obtain the intermediate signal, and repeating the filtering process until the stopping criterion is met (standard deviation SD < 0.2), thereby decomposing 10 intrinsic mode function components from high to low frequencies. For environmental noise such as wind noise and distant background noise, an adaptive spectral subtraction method is used. The system automatically collects environmental background noise and updates the noise estimation model during periods without vehicles, and subtracts noise from the source signal when vehicles pass. The power spectrum of the noise signal is subtracted from the estimated noise power spectrum. The spectral subtraction factor is set to 2.0 and the smoothing factor to 0.1 to prevent the generation of musical noise, thus obtaining pure vibration mode data. Finally, the first 5 natural mode function components in the energy concentration are selected, and a fast Fourier transform is performed with a Blackman window function to calculate the energy spectral density at each frequency point. The 0-1000Hz frequency band is divided into 500 characteristic frequency bands, and the energy values ​​of each frequency band are extracted to form the chassis micro-strain frequency domain characteristic spectrum. This characteristic spectrum is saved in CSV format, containing timestamps, frequency indices, and corresponding energy amplitudes, as the benchmark input for subsequent physical modeling.

[0070] It should be noted that the vibration mode time-frequency plot is a two-dimensional image showing the changes in signal frequency and amplitude over time after the signal is decomposed into intrinsic mode functions using the Hilbert-Huang transform and then sliced ​​into time windows; adaptive spectral subtraction is a commonly used noise reduction method that estimates the power spectrum of ambient background noise in real time or periodically and subtracts it when processing noisy signals, thus preserving the pure signal components; musical noise is a residual noise artifact that may be generated after spectral subtraction processing, sounding like rapidly changing pitches; the Blackman window function is a time-domain window function that weights the signal segment before performing a fast Fourier transform to reduce spectral leakage and improve the accuracy of frequency domain feature extraction.

[0071] In an optional embodiment, the method involves performing load-chassis stiffness coupling mapping on the frequency domain feature spectrum of chassis micro-strain based on the topological point cloud of the vehicle-mounted steel cargo stack, to obtain stiffness coupling mapping data; and performing nonlinear deformation hysteresis behavior learning on the stiffness coupling mapping data to generate load dynamic fingerprint patterns, including:

[0072] Extract the geometric centroid distribution and edge contour curvature of the stacked steel cargo from the topological point cloud of the vehicle-mounted cargo;

[0073] In this embodiment of the invention, a three-dimensional LiDAR scanning device installed at the top of the access control channel acquires the topological point cloud of the stacked steel cargo on the vehicle as it passes through. A random sample consensus algorithm is used to remove the ground and vehicle body point clouds, retaining only the cargo point cloud data. For the retained cargo point cloud, K-Means clustering is used to identify and segment independent cargo units. For each cargo unit, the minimum bounding box algorithm is used to determine its geometric boundary, and its geometric centroid coordinates are calculated. To describe the stability of cargo stacking, Gaussian curvature and mean curvature at cargo edge points are calculated using normal vector estimation and principal curvature analysis. In particular, attention is paid to the contact surface edges between stacked layers to extract curvature abrupt change points, forming a feature set of cargo geometric centroid distribution and edge contour curvature. This feature set is stored in JSON format and used as input for the Euler-Lagrange dynamics mapping.

[0074] Based on the geometric centroid distribution and edge contour curvature of the cargo, the frequency domain characteristic spectrum of the chassis micro-strain is subjected to Euler-Lagrange dynamic mapping to obtain dynamic response mapping data.

[0075] In this embodiment of the invention, a coupled dynamic model of the vehicle-cargo-suspension system is established based on Newton's second law and the Lagrange equation. The cargo stack is considered as a rigid or semi-rigid body. The extracted cargo center of mass distribution and edge curvature are used as mass distribution constraints and geometric boundary conditions of the model. The energy amplitude in the frequency domain characteristic spectrum of the chassis micro-strain is regarded as the external excitation input of the system. A virtual coupled model is established through a finite element analysis software platform, and the dynamic response of the suspension system to chassis micro-strain excitation under a specific load distribution is solved in the frequency domain. The core is to solve the Lagrange function in generalized coordinates. ,in As the system's kinetic energy, Potential energy, generalized coordinates The instantaneous vertical displacement of the chassis and cargo is selected. yaw angle and pitch angle The output results are dynamic response mapping data including suspension displacement, velocity, acceleration, and internal stress wave transfer function. This data is stored in a mixed time and frequency domain format for subsequent nonlinear analysis.

[0076] Nonlinear deformation hysteresis behavior learning is performed on the dynamic response mapping data, and the load response learning data under deformation hysteresis state is output, including:

[0077] Extract the natural frequency offset and amplitude attenuation coefficient of the suspension system from the dynamic response mapping data, and obtain the vehicle self-weight calibration information;

[0078] Based on the vehicle's self-weight calibration information and amplitude attenuation coefficient, the elastic potential energy dissipation gradient under static-dynamic load conversion is calculated to obtain the elastic potential energy dissipation gradient.

[0079] The transient impact response index is obtained by analyzing the stress wave propagation correlation between the dynamic response mapping data based on the elastic potential energy dissipation gradient, including:

[0080] The interlayer frictional resistance torque of the leaf spring and the viscous resistance torque of the hydraulic damper are calculated based on the elastic potential energy dissipation gradient at different load stages.

[0081] The instantaneous restoring force required by the vehicle chassis system is analyzed based on the interlayer frictional resistance torque of the leaf spring and the viscous resistance torque of the hydraulic shock absorber.

[0082] Based on the instantaneous restoring force required by the vehicle chassis system, the dynamic response mapping data is coupled with the restoring force-deformation displacement to obtain the restoring force-displacement correlation data;

[0083] Phase plane trajectory integration is performed on the restoring force displacement correlation data to obtain the instantaneous phase plane energy loop;

[0084] The transient impact response index is obtained by analyzing the stress wave transmission correlation based on the instantaneous phase plane energy loop.

[0085] In this embodiment of the invention, firstly, based on the calculated elastic potential energy dissipation gradient, the dry friction resistance torque between the leaf spring layers is calculated using the Coulomb friction model, and the viscous resistance torque of the hydraulic shock absorber is calculated using fluid dynamics formulas. Next, the aforementioned resistance torque is combined with the elastic potential energy gradient, and the instantaneous restoring force required by the vehicle chassis system at the current moment is calculated using the force balance equation. Subsequently, a correlation coupling model between the instantaneous restoring force and the dynamic response mapping data is established, and trajectory mapping is performed on the phase plane. A closed integral is performed on this trajectory, and the area enclosed by it is calculated. This area is the instantaneous phase plane energy ring, representing the energy dissipation under a single impact. Finally, the area value of this energy ring is compared and normalized with the theoretical energy value under a standard unit impact, thereby obtaining the dimensionless transient impact response index.

[0086] Based on the natural frequency offset and transient impact response index of the suspension system, the tire-ground contact surface slip loss is derived, and the slip loss data is obtained, including:

[0087] Analysis of tire sidewall dynamic stiffness hardening state based on suspension system natural frequency offset;

[0088] Based on the dynamic stiffness hardening state of the tire sidewall, nonlinear shrinkage analysis of the contact imprint area was performed to obtain imprint shrinkage data.

[0089] Analysis of the hysteresis loss tangent variation trend of tire rubber materials based on imprint shrinkage data;

[0090] Based on the trend of hysteresis loss tangent angle and transient impact response index, rolling resistance regression analysis is performed to quantify the gradient of traction force transmission efficiency loss.

[0091] Based on the gradient of traction transmission efficiency loss, the tire-ground contact surface slip loss is derived, and slip loss data is obtained.

[0092] In this embodiment of the invention, firstly, based on the natural frequency offset of the suspension system, the dynamic stiffness hardening state of the tire sidewall under high load is identified, i.e., the characteristic of stiffness increasing nonlinearly with load. Based on this hardening state, the nonlinear shrinkage of the tire-ground contact area is analyzed using a contact mechanics model to obtain the imprint shrinkage data. Then, the viscoelastic changes of the tire rubber material under the imprint shrinkage state, especially the trend of the hysteresis loss tangent angle with frequency and temperature, are analyzed. This trend and the obtained transient impact response index are input into a pre-trained rolling resistance regression model to quantify the efficiency loss gradient of traction force during transmission. Finally, based on this loss gradient, the specific slip loss data of the tire-ground contact surface is derived and calculated, where the slip loss data includes longitudinal slip and lateral creep loss.

[0093] Nonlinear deformation hysteresis behavior learning is performed based on slip loss data, and load response learning data under deformation hysteresis state is output.

[0094] In this embodiment of the invention, the slip loss data obtained from the aforementioned steps is used as a non-conservative force work term and introduced into the Preisach or generalized Maxwell hysteresis model of the suspension system. Specifically, the slip loss data is used to correct the total energy dissipation equation in the model, so that the output load response learning data can simultaneously characterize the vertical hysteresis characteristics of the suspension system and the longitudinal coupling slip characteristics of the tire-ground, thereby more comprehensively reflecting the dynamic fingerprint of the vehicle under load.

[0095] It should be noted that the cargo geometric centroid distribution refers to the set of positions of the geometric centers of all cargo units in three-dimensional space calculated based on point cloud data, reflecting the spatial arrangement of the loads; the edge profile curvature quantifies the local bending degree of the cargo stack surface by analyzing the principal curvature of the cargo point cloud edges, and is used to evaluate stacking stability; the Eulerian-Lagrange dynamics mapping is a mathematical method based on energy principles to establish the equations of motion of a physical system, used to transform the static load distribution into the dynamic response of the chassis system; the phase plane trajectory integral is a nonlinear system analysis technique that calculates the energy dissipation of the system over one cycle by performing closed-loop integration on the system's motion trajectory in the phase plane; and the instantaneous phase plane energy loop... The area enclosed by the phase plane trajectory integral is a key physical characteristic for quantifying the hysteresis energy dissipation of suspension systems or materials. The hysteresis loss tangent is a key parameter for viscoelastic materials, used to quantify the material's energy dissipation capacity, and is a core input feature in rolling resistance regression analysis. Rolling resistance regression analysis is a statistical analysis method based on experience or physical models, used to quantify the loss of traction efficiency caused by factors such as hysteresis loss during tire rolling. The Preisach model and the generalized Maxwell model are both classic hysteresis mathematical models, which respectively simulate the energy dissipation and nonlinear deformation behavior of materials such as rubber suspensions under periodic stress through magnetic elements or spring-damper series combinations.

[0096] The load response learning data is subjected to chassis-cargo coupled fingerprint quantization under the influence of gravity field to generate dynamic load fingerprint patterns, including:

[0097] Analyze the nonlinear damping evolution data of the suspension system driven by the gravity field dimension in the load response learning data;

[0098] Chaotic attractor trajectory analysis was performed on the nonlinear damping evolution data to obtain damped chaotic trajectory data;

[0099] Based on the damped chaotic trajectory data, Lyapunov exponential convergence calculation was performed on the nonlinear damped evolution data to obtain the stability convergence exponent data.

[0100] Based on the support vector machine algorithm and stability convergence index data, load fingerprint features driven by the gravity field dimension are quantified in the nonlinear damping evolution data to generate dynamic load fingerprint patterns.

[0101] In this embodiment of the invention, firstly, based on load response learning data, the analysis is performed in a gravitational field. The nonlinear damping evolution data of the undercarriage system is treated as a high-dimensional time series. The phase space is reconstructed using the embedding theorem, and the Grassberger-Procaccia algorithm is employed to calculate the correlation dimension and analyze the chaotic attractor trajectory. Subsequently, the chaotic characteristics and long-term stability of the system are quantified by calculating the Lyapunov exponent, yielding stability convergence exponent data. Finally, using a support vector machine (SVM) algorithm, these stability and chaotic features are used as input to quantify the load fingerprint features driven by the gravity field dimension under supervised learning of the nonlinear damping evolution data. The SVM employs radial basis functions and kernel functions to distinguish the chaotic attractor shapes under different loads. The quantization result is the load dynamic fingerprint pattern, a high-dimensional feature vector used for subsequent safety verification.

[0102] It should be noted that the Lyapunov exponent is used to quantify the degree of chaos in a dynamic system. A positive value indicates that the system has chaotic characteristics and is sensitive to initial conditions, while a negative value indicates that the system is stable. Here, it is used to quantify the stability and uniqueness of the load fingerprint. When reconstructing the phase space, Takens' embedding theorem is used to determine the embedding dimension of the reconstruction space. and time delay These two parameters determine whether the reconstructed phase space can preserve the topology of the original system; the correlation dimension calculated by the Grassberger-Procaccia algorithm. It is a feature quantity describing the complexity of a chaotic attractor, and can be regarded as a non-integer description of the system's degrees of freedom; the radial basis function kernel function used by the support vector machine is a non-linear kernel function, whose role is to efficiently map the original feature vectors to a high-dimensional space, making samples that are inseparable in the low-dimensional space linearly separable in the high-dimensional space. It can be represented as

[0103] , where parameters This controls the model's sensitivity to individual training samples.

[0104] In an optional embodiment, the method for constructing a vehicle access safety verification model based on the dynamic fingerprint pattern of the load to obtain a safety verification decision model, and deploying the safety verification decision model to an intelligent access control terminal to execute steel logistics vehicle access control, includes:

[0105] Principal component analysis is used to reduce the dimensionality of the dynamic fingerprint of the load to obtain a high-dimensional fingerprint feature vector.

[0106] In this embodiment of the invention, since the dynamic fingerprint pattern of the load generated in the above steps is a high-dimensional feature vector with redundancy and collinearity, principal component analysis is used for dimensionality reduction to improve model efficiency and robustness. First, the fingerprint pattern data is Z-score normalized, and then its covariance matrix is ​​calculated. Eigenvalue decomposition is performed on the covariance matrix to extract the principal component with the largest eigenvalue. The cumulative contribution rate threshold is set to 95%, that is, the principal component that can explain 95% of the variance of the original data is retained. After dimensionality reduction, the 500-dimensional fingerprint pattern is projected onto a new low-dimensional space to obtain a high-dimensional fingerprint feature vector.

[0107] Historical compliance trajectory regression learning is performed on high-dimensional fingerprint feature vectors based on long short-term memory networks to obtain compliance confidence learning data;

[0108] A vehicle access safety verification model is constructed based on compliance confidence learning data to obtain a safety verification decision model;

[0109] In this embodiment of the invention, Bayesian discriminant analysis is used to construct the decision logic. Although the Long Short-Term Memory network outputs the compliance confidence score, to improve the robustness of the decision, the model does not solely rely on a hard threshold. Instead, it constructs prior probabilities for two distributions—compliant and non-compliant—based on historical data, calculates the likelihood ratio of the current confidence score under the two distributions, and sets a safety threshold when the posterior probability of the compliant class is significantly higher than that of the non-compliant class. When the compliance confidence level is greater than or equal to When the vehicle is deemed compliant, it is deemed compliant; when it is less than [a certain size] When a vehicle is deemed non-compliant, this discrimination model is the final safety verification decision model, which maps compliance confidence to a binary passage decision.

[0110] The security verification decision model is deployed to the intelligent access control terminal to receive real-time fingerprint data of vehicles for comparison and output gate opening and closing control commands to execute the passage control of steel logistics vehicles.

[0111] In this embodiment of the invention, the obtained security verification decision model is encapsulated in a lightweight ONNX format and deployed to an intelligent access control terminal. The terminal receives real-time fingerprint data collected and processed when a vehicle passes through via an IoT interface. The terminal performs model comparison. If the decision model output is compliant, an opening command is sent to the gate controller via the 485 bus. If the output is non-compliant, a keep-close / alarm command is sent. In addition, non-compliant data will trigger alarm indicator lights and voice prompts, and will be uploaded to the cloud database for manual review.

[0112] It should be noted that Z-Score normalization of fingerprint pattern data is a data preprocessing method that eliminates the influence of dimensions by transforming the data into a distribution with a mean of 0 and a standard deviation of 1. Principal component analysis (PCA) dimensionality reduction is a statistical method used to transform a set of potentially related variables into a set of linearly unrelated variables through orthogonal transformation. These new variables are called principal components, and the purpose is to reduce the dimensionality of the data while retaining its main information. Bayesian discriminant analysis is a classification method based on Bayes' theorem that calculates the posterior probability of different categories under given features to make decisions about the loading status of vehicles. The 485 bus refers to the RS-485 serial communication standard, a commonly used physical layer standard in industrial control. It supports differential signal transmission and has the characteristics of strong anti-interference ability, long transmission distance, and support for multi-point communication. Here, it is used to reliably send control commands between the intelligent terminal and the gate controller.

[0113] In an optional embodiment, historical compliance trajectory regression learning is performed on the high-dimensional fingerprint feature vector based on a long short-term memory network to obtain compliance confidence learning data, including:

[0114] The high-dimensional fingerprint feature vector is segmented into a time series sliding window, and the feature fluctuation sequence of each window is extracted to obtain the fingerprint fluctuation sequence.

[0115] The parameters of the forget gate, input gate, and output gate of the Long Short-Term Memory Network are initialized based on the fingerprint fluctuation sequence to obtain the initial state matrix of the Long Short-Term Memory Network.

[0116] The fingerprint fluctuation sequence is processed by bidirectional recursive propagation through the initial state matrix of the long short-term memory network. The cell state update and hidden layer output are calculated step by step to obtain the fingerprint time-dependent output data.

[0117] Historical compliance trajectory regression learning is performed based on fingerprint time-series dependent output data to obtain compliance confidence learning data.

[0118] In this embodiment of the invention, the output high-dimensional fingerprint feature vector is first segmented by a time-series sliding window with a window size of T=10 time steps and a step size of S=1. The fingerprint fluctuation sequence is extracted. A bidirectional long short-term memory network is used as the core model. Before training, the weight parameters of the forget gate, input gate, and output gate of the bidirectional long short-term memory network are initialized according to Glorot initialization or He initialization to obtain the initial state matrix of the long short-term memory network. During the training phase, the model receives the fingerprint fluctuation sequence and performs bidirectional recursive transmission processing, calculating the cell state update and hidden layer output step by step to obtain the fingerprint time-series dependent output data. Finally, a fully connected layer is connected to the output layer of the bidirectional long short-term memory network for regression analysis. Based on historical compliant fingerprint data, a compliance confidence score between 0 and 1 is output, which is the compliance confidence learning data. The closer the confidence score is to 1, the better the dynamic fingerprint of the payload matches the historical compliance trajectory.

[0119] It should be noted that Glorot initialization or He initialization is a commonly used deep learning weight initialization method. It calculates the distribution range of the initial weights based on the number of neurons, aiming to ensure the stability of activation values ​​and gradients during network training and avoid gradient vanishing or exploding. Long Short-Term Memory (LSTM) networks are a special type of recurrent neural network that uses three gate structures—forget gate, input gate, and output gate—to control the flow of information. This effectively solves the problem of gradient vanishing or exploding when traditional recurrent neural networks process long sequence data, making them very suitable for learning time-dependent features such as dynamic fingerprints of payloads.

[0120] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. An intelligent access control system for steel logistics based on the Internet of Things, characterized in that, include: The data processing module is used to acquire the time-series vibration data of the chassis of the incoming vehicle and the topological point cloud of the stacked steel cargo on the vehicle through a piezoelectric vibration sensor array and a three-dimensional lidar scanning device deployed in the access control channel; and to perform frequency domain feature extraction and noise reduction processing on the chassis vibration time-series data to obtain the frequency domain feature spectrum of chassis micro-strain. The fingerprint generation module is used to perform load-chassis stiffness coupling mapping on the frequency domain feature spectrum of the chassis micro-strain based on the stacked topology point cloud of the vehicle-mounted steel cargo, and obtain stiffness coupling mapping data. The stiffness coupling mapping data is subjected to nonlinear deformation hysteresis behavior learning to generate load dynamic fingerprint patterns; The access control module is used to construct a vehicle access security verification model based on the dynamic fingerprint pattern of the load, and obtain a security verification decision model; the security verification decision model is deployed to the intelligent access control terminal to execute the passage control of steel logistics vehicles.

2. The intelligent access control system for steel logistics based on the Internet of Things as described in claim 1, characterized in that, Frequency domain feature extraction and noise reduction are performed on the chassis vibration time series data to obtain the chassis micro-strain frequency domain feature spectrum, including: The chassis vibration time series data were processed by Hilbert-Huang transform and time window slices were marked to construct a vibration mode time-frequency diagram; The vibration mode time-frequency diagram is processed by background noise spectrum subtraction to obtain a clean vibration mode diagram. The pure vibration mode spectrum is smoothed by Fast Fourier Transform to obtain the frequency domain characteristic spectrum of the chassis micro-strain.

3. The intelligent access control system for steel logistics based on the Internet of Things according to claim 1, characterized in that, The load-chassis stiffness coupling mapping is performed on the frequency domain feature spectrum of the chassis micro-strain based on the topological point cloud of the vehicle-mounted steel cargo stack to obtain stiffness coupling mapping data. The stiffness coupling mapping data is subjected to nonlinear deformation hysteresis behavior learning to generate load dynamic fingerprint patterns, including: Extract the geometric centroid distribution and edge contour curvature of the cargo in the topological point cloud of the vehicle-mounted steel cargo stack; Based on the geometric centroid distribution of the cargo and the curvature of the edge contour, the frequency domain characteristic spectrum of the chassis micro-strain is subjected to Euler-Lagrange dynamic mapping to obtain dynamic response mapping data. The dynamic response mapping data is subjected to nonlinear deformation hysteresis behavior learning, and the load response learning data under deformation hysteresis state is output. The load response learning data is subjected to chassis-cargo coupled fingerprint quantization under the influence of gravity field to generate the dynamic fingerprint pattern of the load.

4. The IoT-based intelligent access control system for steel logistics according to claim 3, characterized in that, This is used to perform nonlinear deformation hysteresis behavior learning on the dynamic response mapping data, and output load response learning data under deformation hysteresis state, including: Extract the natural frequency offset and amplitude attenuation coefficient of the suspension system from the dynamic response mapping data, and obtain the vehicle self-weight calibration information; Based on the vehicle self-weight calibration information and the amplitude attenuation coefficient, the elastic potential energy dissipation gradient under static-dynamic load conversion is calculated to obtain the elastic potential energy dissipation gradient. Based on the elastic potential energy dissipation gradient, the transient impact response index between the stress wave transmission correlation is analyzed on the dynamic response mapping data to obtain the transient impact response index. Based on the natural frequency offset of the suspension system and the transient impact response index, the tire-ground contact surface slip loss is derived to obtain slip loss data. Based on the slip loss data, nonlinear deformation hysteresis behavior learning is performed, and the load response learning data under the deformation hysteresis state is output.

5. The IoT-based intelligent access control system for steel logistics according to claim 3, characterized in that, Based on the elastic potential energy dissipation gradient, the transient impact response index is analyzed between the stress wave propagation correlation and the dynamic response mapping data to obtain the transient impact response index, including: The interlayer frictional resistance torque of the leaf spring and the viscous resistance torque of the hydraulic damper are calculated based on the elastic potential energy dissipation gradient at different load-bearing stages. The instantaneous restoring force required by the vehicle chassis system is analyzed based on the interlayer frictional resistance torque of the leaf spring and the viscous resistance torque of the hydraulic shock absorber. Based on the instantaneous restoring force required by the vehicle chassis system, the dynamic response mapping data is coupled with the restoring force-deformation displacement to obtain restoring force-displacement correlation data; The instantaneous phase plane energy loop is obtained by integrating the restoring force displacement correlation data in the phase plane. The transient impact response index is obtained by analyzing the transient impact response index between stress wave transmission correlations based on the instantaneous phase plane energy loop.

6. The IoT-based intelligent access control system for steel logistics according to claim 4, characterized in that, Based on the natural frequency offset of the suspension system and the transient impact response index, the tire-ground contact surface slip loss is derived, yielding slip loss data, including: The dynamic stiffness hardening state of the tire sidewall is analyzed based on the natural frequency offset of the suspension system. Based on the dynamic stiffness hardening state of the tire sidewall, nonlinear shrinkage analysis of the contact imprint area was performed to obtain imprint shrinkage data. Analyze the trend of hysteresis loss tangent of tire rubber material based on the imprint shrinkage data; Based on the trend of the hysteresis loss tangent angle and the transient impact response index, a rolling resistance regression analysis is performed to quantify the gradient of traction force transmission efficiency loss. Based on the traction transmission efficiency loss gradient, the tire-ground contact surface slip loss is derived to obtain slip loss data.

7. The IoT-based intelligent access control system for steel logistics according to claim 3, characterized in that, The load response learning data is subjected to chassis-cargo coupled fingerprint quantization under the influence of gravity field to generate the dynamic fingerprint pattern of the load, including: Analyze the nonlinear damping evolution data of the suspension system under the influence of the gravity field in the load response learning data; Chaotic attractor trajectory analysis is performed on the nonlinear damping evolution data to obtain damped chaotic trajectory data; Based on the damped chaotic trajectory data, Lyapunov exponential convergence calculation is performed on the nonlinear damped evolution data to obtain stability convergence exponential data. Based on the support vector machine algorithm and the stability convergence index data, the load fingerprint features driven by the gravity field dimension are quantified on the nonlinear damping evolution data to generate the dynamic fingerprint pattern of the load.

8. The intelligent access control system for steel logistics based on the Internet of Things according to claim 1, characterized in that, This is used to construct a vehicle access safety verification model based on the dynamic fingerprint pattern of the load, thereby obtaining a safety verification decision model; Deploying the security verification decision model to an intelligent access control terminal to perform access control for steel logistics vehicles includes: Principal component analysis was performed to reduce the dimensionality of the dynamic fingerprint of the load, resulting in a high-dimensional fingerprint feature vector. Historical compliance trajectory regression learning is performed on the high-dimensional fingerprint feature vector based on a long short-term memory network to obtain compliance confidence learning data. The vehicle access safety verification model is constructed based on the compliance confidence learning data to obtain the safety verification decision model; The security verification decision model is deployed to the intelligent access control terminal to receive real-time fingerprint data of vehicles for comparison and output gate opening and closing control commands to execute the passage control of steel logistics vehicles.

9. The intelligent access control system for steel logistics based on the Internet of Things according to claim 8, characterized in that, Historical compliance trajectory regression learning is performed on the high-dimensional fingerprint feature vector based on a long short-term memory network to obtain compliance confidence learning data, including: The high-dimensional fingerprint feature vector is segmented into a time series sliding window, and the feature fluctuation sequence of each window is extracted to obtain the fingerprint fluctuation sequence. The parameters of the forget gate, input gate, and output gate of the long short-term memory network are initialized based on the fingerprint fluctuation sequence to obtain the initial state matrix of the long short-term memory network. The fingerprint fluctuation sequence is processed by bidirectional recursive propagation through the initial state matrix of the Long Short-Term Memory Network, and the cell state update and hidden layer output are calculated step by step to obtain fingerprint time-series dependent output data. Historical compliance trajectory regression learning is performed based on the fingerprint time-series dependent output data to obtain the compliance confidence learning data.

Images