Automobile driver dummy collision analysis system based on six-dimensional force sensor

Abstract

The invention discloses an automobile driver dummy collision analysis system based on a six-dimensional force sensor, which relates to the field of dummy collision analysis and comprises the six-dimensional force sensor, a processing module, an optimization positioning module, an assessment map module, a model construction module and an early warning module. The method comprises the following steps: performing synchronization and tensor construction analysis on a force signal, a torque signal, a temperature signal and a vibration signal to obtain a dynamic net feature tensor, constructing a coupling equation to perform topological derivative analysis to obtain a damage hot spot coordinate and multi-dimensional potential energy topological derivation distribution, and creating a partial differential equation model of the neck injury to obtain a partial differential model of the neck injury. The method comprises the following steps: carrying out time discretization through an implicit Euler method, analyzing to obtain accumulated potential energy extension damage energy of each area, carrying out transient impact vibration extreme risk grade analysis, extracting contour surfaces, carrying out stereoscopic three-dimensional damage topology analysis to obtain a stereoscopic three-dimensional damage topology visual map, generating corresponding signals, and carrying out neck injury risk judgment based on the signals. And the risk level of the driver dummy can be accurately judged.

Description

Technical Field

[0001] The present invention relates to the field of dummy collision analysis, in particular to a car driver dummy collision analysis system based on a six-dimensional force sensor. Background Art

[0002] In automobile collision safety testing, driver dummy neck injury assessment is a key step, but existing collision analysis systems have the following bottlenecks: 1. Using only traditional filtering (such as low-pass filtering) to process noise cannot separate physiological temperature rise from collision-induced temperature rise, resulting in the masking of damage-related features in the temperature signal. Vibration signal analysis relies on Fourier transform, which cannot effectively analyze nonlinear frequency modulation components and loses the time-frequency distribution characteristics of high-frequency impact energy.

[0003] 2. The existing tensor model only considers time, space and signal type, without introducing derivative information and historical frame comparison, resulting in incomplete feature expression and difficulty in distinguishing normal fluctuations from damage-related features.

[0004] 3. The existing finite element method requires repeated iterations when dealing with thermal-mechanical-vibration coupling problems, which takes too long to calculate, is highly sensitive to grids in high-gradient areas, and has poor numerical stability.

[0005] 4. Existing damage assessments mostly rely on two-dimensional cloud maps or single parameter thresholds, which cannot intuitively display the energy propagation path in three-dimensional space; risk level determination lacks multi-dimensional data fusion, making it difficult to achieve real-time automated early warning.

[0006] In order to solve the above-mentioned defects, a technical solution is now provided. Summary of the Invention

[0007] In order to solve the technical problems raised by the above background technology, the present invention is proposed. The embodiment of the present invention provides a car driver dummy collision analysis system based on a six-dimensional force sensor.

[0008] The purpose of the present invention can be achieved through the following technical solutions: a car driver dummy collision analysis system based on a six-dimensional force sensor, including a processing module, an optimization positioning module, an evaluation map module, a model construction module and an early warning module, and also including an intelligent six-dimensional force sensor installed on the neck of the collision dummy body; the intelligent six-dimensional force sensor includes a lower plate; a group of force beams are provided at the top end of the outer wall of the lower plate; a group of strain gauges are provided on the outer wall of the force beam; an upper plate is provided at the top end of the outer wall of a group of the force beams; the outer wall of the upper plate is provided with a hinge hole; the lower plate is provided with an optical fiber temperature sensor and a piezoelectric vibration sensor, and the force signal, torque signal, temperature signal and vibration signal collected by the processing module are synchronized, tensor construction and analysis are performed to obtain a dynamic net characteristic tensor; The optimization positioning module constructs coupling equations based on the dynamic net characteristic tensor and performs topological derivative analysis to obtain the coordinates of the damage hotspot and the multi-dimensional potential energy topological derivative distribution; The model building module creates a partial differential equation model of neck injury based on the multi-dimensional potential energy topological derivative distribution, and uses the implicit Euler method for time discretization to analyze and obtain the cumulative potential energy derivative injury energy of each region; The assessment map module analyzes the instantaneous impact vibration risk level based on damage hotspot coordinates, multi-dimensional potential energy topological derivative distribution, and cumulative potential energy derivative damage energy. It also extracts isosurfaces and three-dimensional damage topology analysis to obtain a three-dimensional damage topology visualization map. The early warning module generates corresponding signals for the three-dimensional damage topology visualization map and determines the risk of neck injury based on the signals.

[0009] Furthermore, the steps of dynamic net feature tensor analysis are as follows: Based on the frequency modulation interference cancellation tensor, the K-SVD algorithm is used to train a sparse dictionary, screen the characteristic atoms related to damage, and reconstruct the tensor using the sparse coefficients to obtain a sparse feature canonical tensor. Construct a multimodal graph from the sparse feature canonical tensor, aggregate features through a graph convolutional network, generate global features through attention pooling, and obtain a graph fusion pooled feature vector; The generator G and discriminator D in the generative adversarial network are used to learn the normal feature distribution and distinguish anomalies of the image fusion pooling feature vector respectively, and the anomalies are eliminated and repaired through the dynamic threshold to obtain the dynamic net feature tensor.

[0010] Furthermore, the steps of analyzing the frequency modulation interference cancellation tensor are as follows: The force signal and torque signal collected by the strain gauge, the temperature signal and vibration signal collected by the optical fiber temperature sensor and the piezoelectric vibration sensor respectively; The generalized cross-correlation algorithm is used to synchronize temperature, vibration, force, and torque signals at the microsecond level to obtain multi-source synchronous alignment signals. Based on multi-source synchronization, the temperature and vibration signals in the signals are mapped to the mechanical reference system of the six-dimensional force sensor through the geometric transformation matrix to construct a spatiotemporal anchored five-dimensional tensor; The temperature signal in the spatiotemporal anchored five-dimensional tensor is subjected to morphological open-close filtering, the vibration signal is subjected to variational nonlinear frequency modulation mode decomposition, and the six-dimensional force and torque signals are subjected to principal component whitening to obtain the frequency modulation interference cancellation and buffering tensor.

[0011] Furthermore, the damage hotspot coordinates and multi-dimensional potential energy topological derivative distribution analysis steps are as follows: Based on the dynamic net characteristic tensor, a nonlocal thermoelastic-vibration coupling equation is constructed. The stress field is modified by the nonlocal kernel function to obtain the nonlocal multi-field coupling equation. The damage potential energy functional is defined based on the non-local multi-field coupling equation, and the weight is dynamically adjusted according to the signal energy ratio to obtain the weighted damage potential energy functional. A Lagrangian function is constructed for the weighted damage potential energy functional, and the material parameters are updated iteratively through gradient descent. Through topological derivative analysis, the high gradient area is located, and the coordinates of the damage hotspot and the multi-dimensional potential energy topological derivative distribution are obtained.

[0012] Furthermore, the steps of analyzing the three-dimensional damage topology visualization map are as follows: Based on the coordinates of the damage hotspot and the multidimensional potential energy topological derivative distribution normalized to energy density, the extreme vibration intensity values ​​of the hotspot area are extracted. The dynamic threshold correction model is derived by combining the instantaneous impact energy core value and the extreme vibration intensity value. The model classifies the risk level of the local potential energy density Ψhots and the cumulative potential energy topological derivative damage energy, and obtains the instantaneous impact vibration extreme risk level matrix. The Marching Cubes algorithm is used to extract the isosurface of the potential energy distribution. A Gaussian distribution highlight area is generated within the isosurface with the damage hotspot coordinates xhots as the center. The color mapping is dark red. The regional transparency is adjusted according to the instantaneous impact vibration risk level matrix, and the interactive parameters are displayed in floating mode accordingly. Streamlines are drawn from the damage hotspot coordinates along the gradient direction of the multidimensional potential energy topological derivative distribution to generate a three-dimensional damage topology visualization map.

[0013] Furthermore, the cumulative potential energy extension damage energy analysis steps are as follows: Based on the topological derivative distribution of multidimensional potential energy Create a partial differential equation model of neck injury that includes: , where D represents the energy diffusion coefficient, k represents the energy dissipation rate, S(x,t) represents the external excitation source term, which is determined by the instantaneous load during the collision process. Specifically, S(x,t)=F(t)×δ(x-ximpact), where F(t) is the six-dimensional force vector, δ(*) represents the Dirac function, and ximpact represents the collision point; The implicit Euler method is used to time discretize the partial differential equation model of neck injury. The specific implicit Euler method is: , where Lij is the Laplace matrix element, i and j represent the node numbers, Δt represents the time interval, and n represents the time step number. By iteratively solving the linear equations, the multidimensional potential energy topological derivative distribution of each time step is obtained. , and sum the time steps and multi-dimensional potential energy topological derivative distributions of nodes in each region to obtain the cumulative potential energy topological derivative damage energy E of each region.

[0014] Furthermore, the weighted damage potential energy functional analysis steps are as follows: The damage potential energy functional Ψ is defined based on the nonlocal stress field: ,in represents the strain tensor, k represents the thermal conductivity, and Φ represents the vibration damping coefficient, where represents the mechanical energy, represents the heat conduction energy, Represents vibration dissipation energy, and the weight is dynamically adjusted based on the signal energy ratio: , wF represents the weight of mechanical energy. Similarly, the heat conduction energy weight wΔT and the vibration dissipation energy weight wV are defined to dynamically adjust the weights to obtain the weighted damage potential energy functional , represents the square of the L2 norm, F represents the six-dimensional force signal in the dynamic net characteristic tensor, V represents the vibration signal in the dynamic net characteristic tensor, and ΔT represents the temperature signal in the temperature channel of the dynamic net characteristic tensor.

[0015] Furthermore, the analysis steps of the non-local thermoelasticity-vibration coupling equation are as follows: The dynamic net characteristic tensor is based on the nonlocal thermoelasticity theory and constructs the nonlocal thermoelasticity-vibration coupling equation: stress tensor ; Among them, S ijkl represents the fourth-order stiffness tensor, describing the elastic response of the material, represents the strain component, ζ represents the thermal expansion coefficient, ΔT represents the temperature signal in the temperature channel of the dynamic net characteristic tensor, represents the fractional derivative, V(τ) is the vibration signal in the dynamic net characteristic tensor, τ is the integral variable, Ω represents the vibration-stress coupling coefficient, represents the strain tensor, δkl represents the Kronecker symbol, k and l are indices used to identify the tensor subscript, when k = 1, the value is equal to 1, otherwise it is equal to zero.

[0016] Furthermore, the neck injury risk determination steps are as follows: According to the color intensity of each area in the three-dimensional damage topology visualization map, the area corresponds to the high-risk area, the interactive parameters displayed in suspension, and the multi-dimensional potential energy topological derivative distribution gradient, the corresponding signals one, two and three are emitted, and the distribution status of signals one, two and three are statistically analyzed to correspondingly emit high risk, medium risk and low risk of neck injury.

[0017] Furthermore, the steps of constructing the Lagrangian function are as follows: Constructing the Lagrangian function for the weighted damage potential energy functional Ψ′ , represents the material parameter constraints, represents geometric conservation, represents the Lagrange multiplier corresponding to the inequality constraint, represents the Lagrange multiplier corresponding to the equality constraint, and θ represents the vector of material parameters to be optimized; Material parameter constraint g i (θ) and geometric conservation h j Specific definition of (θ): ; , where ζmax represents the maximum value of the set vibration coupling coefficient, represents the initial volume of the neck, V represents the volume of the neck, represents the neck node displacement, ubouy represents the prescribed displacement at the boundary, and i and j represent indices.

[0018] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention sets an intelligent six-dimensional force sensor on the neck of the collision dummy body, and sets an optical fiber temperature sensor and a piezoelectric vibration sensor. The collected force signal, torque signal, temperature signal and vibration signal are synchronized and tensor constructed and analyzed to obtain a dynamic net characteristic tensor. Based on the dynamic net characteristic tensor, a coupling equation is constructed and a topological derivative analysis is performed to obtain the coordinates of the damage hotspot and the multi-dimensional potential energy topological derivative distribution. Based on the multi-dimensional potential energy topological derivative distribution, a partial differential equation model of neck injury is created, and time discretization is performed through the implicit Euler method to analyze the cumulative potential energy of each area and the damage energy. Based on the damage hotspot, the partial differential equation model of neck injury is created. The instantaneous impact vibration risk level analysis is carried out based on the coordinates, multi-dimensional potential energy topological derivative distribution and cumulative potential energy derivative damage energy, and the isosurface and three-dimensional damage topology analysis are extracted to obtain a three-dimensional three-dimensional damage topology visualization map. Through synchronization and tensor construction analysis, the physiological temperature rise and collision-induced temperature rise can be separated, the nonlinear frequency modulation component can be effectively analyzed, the time-frequency distribution characteristics of high-frequency impact energy can be ensured, the derivative information and historical frame comparison can be introduced, the feature expression is complete, and normal fluctuations and damage-related characteristics can be distinguished. The coupling equation is constructed by tensor and the partial differential equation model of neck injury is created, which is highly sensitive to the high-gradient area grid and has good numerical stability.

[0019] 2. The present invention generates corresponding signals from the three-dimensional damage topology visualization map, and determines the risk of neck injury based on the signals. It can intuitively display the energy propagation path in three-dimensional space. The risk level determination can use multi-dimensional data fusion to achieve real-time automatic early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. The following drawings are not intentionally scaled to the actual size, and the focus is on illustrating the main purpose of the present invention.

[0021] Figure 1 is a system block diagram of the present invention; Figure 2 Schematic diagram of the six-dimensional force sensor structure of the present invention; Figure 3 This is a top view of the six-dimensional force sensor of the present invention. DETAILED DESCRIPTION

[0022] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts also fall within the scope of protection of the present invention.

[0023] like Figure 1 As shown, the automobile driver dummy collision analysis system based on the six-dimensional force sensor includes an intelligent six-dimensional force sensor installed on the neck of the collision dummy body, a processing module, an optimization positioning module, an evaluation map module, a model building module and an early warning module.

[0024] like Figure 2 、 Figure 3 As shown, the intelligent six-dimensional force sensor includes a lower plate 1; a group of force beams 3 are provided at the top of the outer wall of the lower plate 1; a group of strain gauges 4 are provided on the outer wall of the force beams 3; an upper plate 2 is provided at the top of the outer wall of the group of force beams 3; a hinge hole 5 is provided on the outer wall of the upper plate 2; an optical fiber temperature sensor and a piezoelectric vibration sensor are provided on the lower plate 1; Specifically, the intelligent six-dimensional force sensor can be conveniently hinged at the hinge between the upper neck and head of the dummy and the hinge between the lower neck and torso when in use. The intelligent six-dimensional force sensor has also become the core sensor of the humanoid robot - the six-dimensional force / torque sensor. The intelligent six-dimensional force sensor is made of a special alloy elastomer. The elastomer has a two-layer structure with a hollow middle. The special material and structural design of the elastomer give it ideal elastic modulus and Poisson's ratio physical properties. On an outer cylindrical elastomer, four slightly rectangular stress beams are evenly distributed in four directions. Each beam has four stress surfaces, including a lateral plane and a symmetrical plane with it, a frontal plane and a symmetrical plane with it. The stress areas of the frontal plane and the lateral plane are not equal. Each surface is used to attach a strain gauge. The strain gauge 4 consists of three layers of stacked sheets at 45 degrees and two layers of stacked sheets at 90 degrees. In the figure, Fx, Fy, and Fz are the three force directions, and Mx, My, and Mz are the three torque directions. Under the action of force and torque, the elastic body undergoes slight deformation, causing the strain gauge 4 to bend and causing its resistance to change. By collecting the changes in the output electrical signal, the force and torque information of the intelligent six-dimensional force sensor in three different directions can be obtained.

[0025] The force signal, torque signal, temperature signal and vibration signal collected by the processing module are synchronized and tensor constructed and analyzed to obtain the dynamic net characteristic tensor; The optimization positioning module constructs coupling equations based on the dynamic net characteristic tensor and performs topological derivative analysis to obtain the coordinates of the damage hotspot and the multi-dimensional potential energy topological derivative distribution; The model building module creates a partial differential equation model of neck injury based on the multi-dimensional potential energy topological derivative distribution, and uses the implicit Euler method for time discretization to analyze and obtain the cumulative potential energy derivative injury energy E of each area; The assessment map module analyzes the instantaneous impact vibration risk level based on damage hotspot coordinates, multi-dimensional potential energy topological derivative distribution, and cumulative potential energy derivative damage energy. It also extracts isosurfaces and three-dimensional damage topology analysis to obtain a three-dimensional damage topology visualization map. The early warning module generates corresponding signals for the three-dimensional damage topology visualization map and determines the risk of neck injury based on the signals.

[0026] Specifically, the analysis steps of the processing module are as follows: The force signal and torque signal collected by the strain gauge, the temperature signal and vibration signal collected by the optical fiber temperature sensor and the piezoelectric vibration sensor respectively; The generalized cross-correlation algorithm is used to synchronize temperature, vibration, force, and torque signals at the microsecond level to obtain multi-source synchronous alignment signals. Based on multi-source synchronization, the temperature and vibration signals in the signals are mapped to the mechanical reference system of the six-dimensional force sensor through the geometric transformation matrix to construct a spatiotemporal anchored five-dimensional tensor; Morphological open-close filtering is performed on the temperature signal in the spatiotemporal anchored five-dimensional tensor, variational nonlinear frequency modulation mode decomposition is performed on the vibration signal, and principal component whitening is performed on the six-dimensional force and torque signals to obtain the frequency modulation interference cancellation tensor. Based on the frequency modulation interference cancellation tensor, the K-SVD algorithm is used to train a sparse dictionary, screen the characteristic atoms related to damage, and reconstruct the tensor using the sparse coefficients to obtain a sparse feature canonical tensor. Construct a multimodal graph from the sparse feature canonical tensor, aggregate features through a graph convolutional network, generate global features through attention pooling, and obtain a graph fusion pooled feature vector; The generator G and discriminator D in the generative adversarial network are used to learn the normal feature distribution and distinguish abnormalities of the image fusion pooling feature vector respectively. The abnormalities are removed and repaired through dynamic thresholds to obtain a dynamic net feature tensor. Specifically, a generalized cross-correlation algorithm is used to synchronize the temperature, vibration, force, and torque signals at the microsecond level to eliminate clock drift and obtain a multi-source synchronization alignment signal. The original position coordinates Kraw of the temperature and vibration signals in the multi-source synchronization signal are converted into the coordinates Knew in the mechanical reference system through the geometric transformation matrix L. Specifically: Knew = L × Kraw; Constructing a space-time anchored five-dimensional tensor , which includes 1) time series t, 2) sensor signals, specifically temperature, vibration, force in three directions, and torque in three directions, 3) spatial coordinates (x, y, z), 4) signal channels, which are original values ​​and first-order derivatives, 5) time window identifiers, which are current frames and historical frames; Morphological open-close filtering is performed on the temperature signal of the spatiotemporally anchored five-dimensional tensor to separate the slowly varying physiological temperature rise from the impulse noise: ΔTclean=Close(Open(ΔT,Sker),Sker) Where ΔT represents the original temperature change signal, Sker represents the flat structure element used for morphological operation, Open(*) represents the morphological opening operation, Close(*) represents the morphological closing operation, and ΔTclean represents the temperature signal processed by morphological open-close filtering; The vibration signal of the spatiotemporally anchored five-dimensional tensor is decomposed using variational nonlinear frequency modulation mode, and the decomposed signal is divided into frequency modulation and amplitude modulation components: , where V(t) represents the original vibration signal that changes with time t, represents the amplitude modulation function of the kth component, describing the change of the amplitude of the component over time, represents the frequency modulation function of the kth component, describing the change of instantaneous frequency with time τ, retaining the main frequency band, specifically the 50-800 Hz band; Perform principal component whitening on the six-dimensional force and torque signals of the five-dimensional space-time anchored tensor to eliminate coupling interference: , where F represents the original six-dimensional force and torque signal vector, ɑ represents the mean value of F, U represents the orthogonal matrix obtained by principal component analysis, Λ represents the diagonal matrix composed of the eigenvalues ​​of F, Fwhite represents the mechanical signal after principal component whitening to eliminate coupling interference, T represents the transpose of the matrix, and then the processed signals are merged to obtain the frequency modulation interference elimination buffer tensor Gfilte; Using K-SVD algorithm to train sparse dictionary based on frequency modulation interference cancellation tensor , where m represents the number of dimensions of the signal features, K represents the number of FM-AM components obtained after the variational nonlinear FM mode decomposition, and the objective function is optimized: , where X represents the sparse coefficient matrix, γ represents the regularization parameter, represents the square of the Frobenius norm, Represents the L1 norm, screens the characteristic atoms related to damage, and uses the sparse coefficient Gsparse=Dselecte×Xselecte to obtain the sparse feature canonical tensor Gsparse, where Dselecte represents the subset screened out from the sparse dictionary D, retaining only the characteristic atoms related to neck injury, and Xselecte represents the sparse coefficient matrix corresponding to the screened damage-related characteristic atoms; Each non-zero feature slice of the sparse feature canonical tensor Gsparse, such as the time-space-channel combination, is a node vi∈V of the multimodal graph. Each node contains a feature vector hi. For example, at a certain time point tk, the sensor type is force (Fx, Fy, Fz), the spatial coordinates are (x, y, z), and the channel is the original value. Then the slice corresponds to a node, and the edge weight wij is calculated by the normalized mutual information between the node features. , where I(Xi;Xj) represents the mutual information between the features of node vi and node vj, H(Xi) represents the entropy of the feature of node vi, if the edge weight wij is less than the set threshold wth, it is removed, and a sparse adjacency matrix C is generated based on the removed edge weight. The features are aggregated through the graph convolutional network update rule, and the update rule is: ,in Represents the node feature matrix of the lth layer, D represents the degree matrix, diagonal matrix, represents the trainable parameter matrix of the lth layer, σ represents the ReLU nonlinear activation function, represents an adjacency matrix with self-loops, , I represents the unit matrix, and the updated node feature matrix is ​​obtained ; Generate global feature vector through attention pooling , where the attention coefficient , where softmax represents the normalization function, represents the feature of the i-th node in the second layer, and q represents the learning query vector; The global feature vector r is reconstructed using a generative adversarial network to calculate the reconstruction error Score, and abnormal data is removed through a dynamic threshold Threshold to obtain a dynamic net feature tensor Grepair. Specifically: , Threshold = μScore + 3σScore, where G(r) represents the generator G trying to reconstruct r, and D(r) represents the discriminator D distinguishing whether the input feature is a real feature or a feature generated by the generator. represents the square of the L2 norm, μScore represents the mean of the historical reconstruction error, and σScore represents the standard deviation of the historical reconstruction error; Specifically, the analysis steps of the optimization positioning module, evaluation map module, and model building module are as follows: Based on the dynamic net characteristic tensor, a nonlocal thermoelastic-vibration coupling equation is constructed. The stress field is modified by the nonlocal kernel function to obtain the nonlocal multi-field coupling equation. The damage potential energy functional is defined based on the non-local multi-field coupling equation, and the weight is dynamically adjusted according to the signal energy ratio to obtain the weighted damage potential energy functional. A Lagrangian function is constructed for the weighted damage potential energy functional, and the material parameters are updated iteratively through gradient descent. Through topological derivative analysis, the high gradient area is located, and the coordinates of the damage hotspot and the multi-dimensional potential energy topological derivative distribution are obtained. Based on the coordinates of the damage hotspot and the multidimensional potential energy topological derivative distribution normalized to energy density, the extreme vibration intensity values ​​of the hotspot area are extracted. The dynamic threshold correction model is derived by combining the instantaneous impact energy core value and the extreme vibration intensity value. The model classifies the risk level of the local potential energy density Ψhots and the cumulative potential energy topological derivative damage energy, and obtains the instantaneous impact vibration extreme risk level matrix. The Marching Cubes algorithm is used to extract the isosurface of the potential energy distribution. A Gaussian distribution highlight area is generated within the isosurface with the damage hotspot coordinates xhots as the center. The color mapping is dark red. The regional transparency is adjusted according to the instantaneous impact vibration risk level matrix, and the interactive parameters are displayed in floating mode accordingly. Streamlines are drawn from the damage hotspot coordinates along the gradient direction of the multidimensional potential energy topological derivative distribution to generate a three-dimensional damage topology visualization map.

[0027] Specifically, the dynamic net characteristic tensor is based on the nonlocal thermoelasticity theory to construct the nonlocal thermoelasticity-vibration coupling equation:

[0028] in, represents the fourth-order stiffness tensor, describing the elastic response of the material, represents the strain component, which is calculated by inverse calculation of the six-dimensional force and torque signals in the dynamic net characteristic tensor through Hooke's law, ζ represents the thermal expansion coefficient, ΔT represents the temperature signal in the temperature channel in the dynamic net characteristic tensor, represents the fractional derivative, which is used to describe the cumulative effect of high-frequency vibration energy. V(τ) is the vibration signal in the dynamic net characteristic tensor, τ is the integral variable, and Ω represents the vibration-stress coupling coefficient. It represents the strain tensor, which describes the degree of deformation of the material after being subjected to stress. δkl represents the Kronecker symbol. k and l are indices used to identify the tensor subscript. When k=1, the value is equal to 1, and otherwise it is equal to zero. Further introduction of kernel function Modified stress tensor σij, ,in is the nonlocal kernel function, the spatial positions x and x′, Я represents the integration region, and the nonlocal multi-field coupling equation σ′ij is obtained; The damage potential energy functional Ψ is defined based on the nonlocal stress field, which integrates vibration dissipation, heat conduction energy and mechanical strain energy: ,in represents the strain tensor, k represents the thermal conductivity, and Φ represents the vibration damping coefficient, where represents the mechanical energy, represents the heat conduction energy, Represents vibration dissipation energy, and the weight is dynamically adjusted based on the signal energy ratio: , wF represents the weight of mechanical energy. Similarly, the heat conduction energy weight wΔT and the vibration dissipation energy weight wV are defined to dynamically adjust the weights to obtain the weighted damage potential energy functional , represents the square of the L2 norm, and F represents the six-dimensional force signal; Weighted Damage Potential Energy Functional Constructing the Lagrangian function , represents the material parameter constraints, represents geometric conservation, Represents the Lagrange multiplier corresponding to the inequality constraint, which adjusts the influence of the inequality constraint. represents the Lagrange multiplier corresponding to the equality constraint, and θ represents the vector of material parameters to be optimized, such as stiffness tensor, thermal expansion coefficient, and vibration coupling coefficient; Material parameter constraint g i (θ) and geometric conservation h j Specific definition of (θ): ; , where ζmax represents the maximum value of the set vibration coupling coefficient, represents the initial volume of the neck, V represents the volume of the neck, represents the neck node displacement, ubouy represents the prescribed displacement at the boundary, ensuring that the displacement boundary conditions at the connection between the neck and the head and torso match, and i and j represent indices; Iteratively update the material parameters θ by gradient descent, , where n represents the number of iterations, records the optimization process, and η represents the learning rate. Denotes the weighted damage potential energy functional for parameter θ The gradient of , through topological derivative analysis, locate the high gradient area, , the variational derivative of the damage potential energy with respect to the region Ω, locate the local maximum point of the potential energy gradient, determine the damage hotspot coordinates xhots and the multi-dimensional potential energy topological derivative distribution ; Based on the topological derivative distribution of multidimensional potential energy Create a partial differential equation model of neck injury that includes: , where D represents the energy diffusion coefficient, reflecting the energy transfer efficiency of the neck tissue, k represents the energy dissipation rate, indicating the ability of the tissue to absorb energy, S(x, t) represents the external excitation source term, which is determined by the instantaneous load during the collision. Specifically, S(x, t) = F(t) × δ (x-ximpact), where F(t) is a six-dimensional force vector, δ (*) represents the Dirac function, and ximpact represents the collision point; The implicit Euler method is used to time discretize the partial differential equation model of neck injury. The specific implicit Euler method is: , where Lij is the Laplace matrix element, describing the energy diffusion relationship between nodes i and j, i and j represent the node numbers, Δt represents the time interval, and n represents the time step number. By iteratively solving the linear equations, the multidimensional potential energy topological derivative distribution of each time step is obtained. , and sum the time steps and multi-dimensional potential energy topological derivative distributions of nodes in each region to obtain the cumulative potential energy topological derivative damage energy E of each region; Specifically, a partial differential equation model of neck injury is created based on the multi-dimensional potential energy topological derivative distribution. Through the energy diffusion coefficient D, the energy dissipation rate k and the external excitation source term S(x,t) containing the Dirac function, it quantifies the energy transfer, absorption and instantaneous load of the neck tissue in the collision, realizes the mathematical description of the dynamic evolution of potential energy, and provides a theoretical framework for the energy accumulation analysis of the damaged area. The implicit Euler method serves as a bridge connecting the theoretical model and numerical calculation. Through the discretization of the time domain, the partial differential equation is converted into a linear equation system containing the Laplace matrix. Under the premise of ensuring numerical stability, the potential energy distribution of each time step is iteratively solved, and the cumulative potential energy-derived damage energy E is obtained by summing the potential energy of the time steps in the region, so that the abstract physical model can be converted into computable discrete data, supporting the subsequent quantitative assessment and spatial positioning of injury risks.

[0029] Distribution derived from damage hotspot coordinates xhots and multidimensional potential energy topology , calculate the local potential energy density in the hotspot area , where Vhots represents the volume of the hotspot area, Ωhots represents the spherical area centered on xhots, and the extreme value of the vibration intensity of the hotspot area is extracted , where V(xhots, t) represents the vibration intensity at the hotspot coordinate xhots that changes with time t, and the local potential energy density Ψhots and the cumulative potential energy extension damage energy E in the hotspot area are divided into risk levels through the dynamic threshold correction model. If it is greater than or equal to the dynamic threshold Ψcrit and the cumulative potential energy extension damage energy E is greater than the set threshold, it corresponds to a high-risk area, and the others are low-risk areas. The instantaneous impact vibration risk level matrix Risk is obtained, where the dynamic threshold correction model includes: , where Eimp represents the instantaneous impact energy kernel value, λ0 represents the model parameter, which adjusts the weight of the global collision influence, and represents the model parameter, which controls the influence of the vibration intensity on the threshold. Indicates dynamic weight, which is adjusted according to the proportion of hotspot potential energy. represents the maximum value of the multidimensional potential energy topological derivative distribution, and Ψcrit represents the dynamic threshold; The specific analysis steps of the instantaneous impact energy nuclear value Eimp are as follows: , where F is the six-dimensional force vector, v represents the velocity vector, and τ represents the integral variable; Based on the instantaneous impact vibration risk level matrix Risk, damage hot spot coordinates xhots and multi-dimensional potential energy topological derivative distribution , using the Marching Cubes algorithm to extract the multidimensional potential energy topological derivative distribution The damage boundary is defined by the isosurface. Within the isosurface, a Gaussian distribution highlight area is generated with the damage hotspot coordinate xhots as the center. The color mapping is dark red, and the intensity decays with distance. The color intensity of each area is obtained: ,in Represents the color mapping function of the hotspot area, r represents the Gaussian kernel radius, which is used to control the diffusion range of the highlighted area, exp represents the exponential function with the natural constant e as the base, X represents (x, y, z), the coordinates of the current spatial point, which is used to locate any point in the three-dimensional space. The regional transparency is adjusted according to the instantaneous impact vibration risk level matrix Risk. The transparency of the high-risk area is set to 30%, and the transparency of the low-risk area is set to 60%. Interactive parameters are displayed in the high-risk area, specifically the local potential energy density, the cumulative potential energy extension damage energy E and the vibration intensity extreme value. Starting from the damage hotspot coordinates, the distribution gradient is derived along the multidimensional potential energy topology. Draw streamlines in different directions to reveal the propagation path of damage from hot spots to surrounding tissues, and obtain a three-dimensional damage topology visualization map, including different areas of color transparency changes, gradient streamlines, and interaction parameters; Specifically, the analysis steps of the early warning module are as follows: If the color intensity of each area in the three-dimensional three-dimensional damage topology visualization map is greater than the set threshold, signal one is issued, otherwise signal three is issued. If the area in the three-dimensional three-dimensional damage topology visualization map corresponds to a high-risk area and the floating display interactive parameters are all higher than the corresponding set parameters, signal one is issued, otherwise signal two is issued. If the area corresponds to a high-risk area, signal three is issued. If the multi-dimensional potential energy topological derivative distribution gradient in the three-dimensional three-dimensional damage topology visualization map is greater than the set threshold, signal one is issued, otherwise signal three is issued. The proportion of signal one, signal two and signal three received by the system per unit time is counted. If the proportion of signal one and signal two is greater than the threshold xT1, the corresponding neck injury risk is high. If the proportion of signal one and signal two is equal to the threshold xT1, the corresponding neck injury risk is medium. If the proportion of signal one and signal two is less than the threshold xT1, the corresponding neck injury risk is low. Specifically, a neck injury assessment system was constructed. A generalized cross-correlation algorithm was used to synchronize multi-source signals in microseconds. A dynamic net feature tensor was constructed by combining geometric transformations with deep learning (K-SVD, graph convolutional networks, and generative adversarial networks), achieving signal noise purification and damage feature enhancement. A coupled equation was constructed based on nonlocal thermoelasticity theory, dynamically locating damage hotspots and potential energy distribution through topological derivative analysis and Lagrangian optimization. A partial differential equation model with energy diffusion, dissipation, and transient loads was created based on the multidimensional potential energy distribution. The implicit Euler method was used to solve the problem in space and time to obtain the cumulative damage energy. A dynamic threshold correction model was integrated with the Marching Cubes algorithm to generate a three-dimensional damage topology visualization map with Gaussian highlights, gradient streamlines, and interactive parameters, visualizing the risk distribution. Multidimensional signals were generated based on map color intensity, parameter thresholds, and gradient features, and the neck injury risk level was intelligently determined based on signal proportion. The entire system, powered by data-driven and physical modeling, achieves intelligent and visual processing from signal acquisition to risk warning, providing accurate judgment for neck injury assessment.

[0030] The above is an illustration of the present invention and should not be considered as limiting thereof. Although several exemplary embodiments of the present invention have been described, it will be readily understood by those skilled in the art that many modifications may be made to the exemplary embodiments without departing from the novel teachings and advantages of the present invention. Therefore, all such modifications are intended to be included within the scope of the present invention as defined by the claims. It should be understood that the above is an illustration of the present invention and should not be considered as being limited to the specific embodiments disclosed, and modifications to the disclosed embodiments and other embodiments are intended to be included within the scope of the appended claims. The present invention is defined by the claims and their equivalents.

Claims

1. A six-dimensional force sensor-based automobile driver dummy collision analysis system, comprising an intelligent six-dimensional force sensor, a processing module, an optimization positioning module, an evaluation map module, a model building module, and an early warning module, characterized in that: The processing module collects force signals, torque signals, temperature signals and vibration signals and performs synchronization and tensor construction analysis to obtain dynamic net feature tensors; The optimization positioning module constructs coupling equations based on the dynamic net characteristic tensor and performs topological derivative analysis to obtain the coordinates of the damage hotspot and the multi-dimensional potential energy topological derivative distribution; The model building module creates a partial differential equation model of neck injury based on the multi-dimensional potential energy topological derivative distribution, and uses the implicit Euler method for time discretization to analyze and obtain the cumulative potential energy derivative injury energy of each region; The assessment map module analyzes the instantaneous impact vibration risk level based on damage hotspot coordinates, multi-dimensional potential energy topological derivative distribution, and cumulative potential energy derivative damage energy. It also extracts isosurfaces and three-dimensional damage topology analysis to obtain a three-dimensional damage topology visualization map. The early warning module generates corresponding signals for the three-dimensional damage topology visualization map and determines the risk of neck injury based on the signals.

2. The automobile driver dummy collision analysis system based on a six-dimensional force sensor according to claim 1 is characterized in that: The intelligent six-dimensional force sensor is installed on the neck of the collision dummy body, comprising a lower plate (1); a group of force beams (3) are provided at the top of the outer wall of the lower plate (1); a group of strain gauges (4) are provided on the outer wall of the force beams (3); an upper plate (2) is provided at the top of the outer wall of the group of force beams (3); a hinge hole (5) is provided on the outer wall of the upper plate (2); an optical fiber temperature sensor and a piezoelectric vibration sensor are provided on the lower plate (1); The steps of dynamic net feature tensor analysis are as follows: Based on the frequency modulation interference cancellation tensor, the K-SVD algorithm is used to train a sparse dictionary, screen the characteristic atoms related to damage, and reconstruct the tensor using the sparse coefficients to obtain a sparse feature canonical tensor. Construct a multimodal graph from the sparse feature canonical tensor, aggregate features through a graph convolutional network, generate global features through attention pooling, and obtain a graph fusion pooled feature vector; The generator G and discriminator D in the generative adversarial network are used to learn the normal feature distribution and distinguish anomalies of the image fusion pooling feature vector respectively, and the anomalies are eliminated and repaired through the dynamic threshold to obtain the dynamic net feature tensor.

3. The automobile driver dummy collision analysis system based on a six-dimensional force sensor according to claim 2 is characterized in that: The steps of analyzing the frequency modulation interference cancellation tensor are as follows: The force signal and torque signal collected by the strain gauge, the temperature signal and vibration signal collected by the optical fiber temperature sensor and the piezoelectric vibration sensor respectively; The generalized cross-correlation algorithm is used to synchronize temperature, vibration, force, and torque signals at the microsecond level to obtain multi-source synchronous alignment signals. Based on multi-source synchronization, the temperature and vibration signals in the signals are mapped to the mechanical reference system of the six-dimensional force sensor through the geometric transformation matrix to construct a spatiotemporal anchored five-dimensional tensor; The temperature signal in the spatiotemporal anchored five-dimensional tensor is subjected to morphological open-close filtering, the vibration signal is subjected to variational nonlinear frequency modulation mode decomposition, and the six-dimensional force and torque signals are subjected to principal component whitening to obtain the frequency modulation interference cancellation and buffering tensor.

4. The automobile driver dummy collision analysis system based on a six-dimensional force sensor according to claim 1 is characterized in that: The steps for analyzing the damage hotspot coordinates and multi-dimensional potential energy topological derivative distribution are as follows: Based on the dynamic net characteristic tensor, a nonlocal thermoelastic-vibration coupling equation is constructed. The stress field is modified by the nonlocal kernel function to obtain the nonlocal multi-field coupling equation. The damage potential energy functional is defined based on the non-local multi-field coupling equation, and the weight is dynamically adjusted according to the signal energy ratio to obtain the weighted damage potential energy functional. A Lagrangian function is constructed for the weighted damage potential energy functional, and the material parameters are updated iteratively through gradient descent. Through topological derivative analysis, the high gradient area is located, and the coordinates of the damage hotspot and the multi-dimensional potential energy topological derivative distribution are obtained.

5. The automobile driver dummy collision analysis system based on a six-dimensional force sensor according to claim 1 is characterized in that: The steps of analyzing the three-dimensional damage topology visualization map are as follows: Based on the coordinates of the damage hotspot and the multidimensional potential energy topological derivative distribution normalized to energy density, the extreme vibration intensity values ​​of the hotspot area are extracted. The dynamic threshold correction model is derived by combining the instantaneous impact energy core value and the extreme vibration intensity value. The model classifies the risk level of the local potential energy density Ψhots and the cumulative potential energy topological derivative damage energy, and obtains the instantaneous impact vibration extreme risk level matrix. The Marching Cubes algorithm is used to extract the isosurface of the potential energy distribution. A Gaussian distribution highlight area is generated within the isosurface with the damage hotspot coordinates xhots as the center. The color mapping is dark red. The regional transparency is adjusted according to the instantaneous impact vibration risk level matrix, and the interactive parameters are displayed in floating mode accordingly. Streamlines are drawn from the damage hotspot coordinates along the gradient direction of the multidimensional potential energy topological derivative distribution to generate a three-dimensional damage topology visualization map.

6. The automobile driver dummy collision analysis system based on a six-dimensional force sensor according to claim 5 is characterized in that: The steps of cumulative potential energy extension damage energy analysis are as follows: Based on the topological derivative distribution of multidimensional potential energy Create a partial differential equation model of neck injury that includes: , where D represents the energy diffusion coefficient, k represents the energy dissipation rate, S(x,t) represents the external excitation source term, which is determined by the instantaneous load during the collision process. Specifically, S(x,t)=F(t)×δ(x-ximpact), where F(t) is the six-dimensional force vector, δ(*) represents the Dirac function, and ximpact represents the collision point; The implicit Euler method is used to time discretize the partial differential equation model of neck injury. The specific implicit Euler method is: , where Lij is the Laplace matrix element, i and j represent the node numbers, Δt represents the time interval, and n represents the time step number. By iteratively solving the linear equations, the multidimensional potential energy topological derivative distribution of each time step is obtained. , and sum the time steps and multi-dimensional potential energy topological derivative distributions of nodes in each region to obtain the cumulative potential energy topological derivative damage energy E of each region.

7. The automobile driver dummy collision analysis system based on a six-dimensional force sensor according to claim 4 is characterized in that: The steps of weighted damage potential energy functional analysis are as follows: The damage potential energy functional Ψ is defined based on the nonlocal stress field: ,in represents the strain tensor, k represents the thermal conductivity, and Φ represents the vibration damping coefficient, where represents the mechanical energy, represents the heat conduction energy, Represents vibration dissipation energy, and the weight is dynamically adjusted based on the signal energy ratio: , wF represents the weight of mechanical energy. Similarly, the heat conduction energy weight wΔT and the vibration dissipation energy weight wV are defined to dynamically adjust the weights to obtain the weighted damage potential energy functional , represents the square of the L2 norm, F represents the six-dimensional force signal in the dynamic net characteristic tensor, V represents the vibration signal in the dynamic net characteristic tensor, and ΔT represents the temperature signal in the temperature channel of the dynamic net characteristic tensor.

8. The automobile driver dummy collision analysis system based on a six-dimensional force sensor according to claim 4 is characterized in that: The analysis steps of the nonlocal thermoelasticity-vibration coupling equation are as follows: The dynamic net characteristic tensor is based on the nonlocal thermoelasticity theory and constructs the nonlocal thermoelasticity-vibration coupling equation: stress tensor ,in, represents the fourth-order stiffness tensor, describing the elastic response of the material, represents the strain component, ζ represents the thermal expansion coefficient, ΔT represents the temperature signal in the temperature channel of the dynamic net characteristic tensor, represents the fractional derivative, V(τ) is the vibration signal in the dynamic net characteristic tensor, τ is the integral variable, Ω represents the vibration-stress coupling coefficient, represents the strain tensor, δkl represents the Kronecker symbol, k and l are indices used to identify the tensor subscript, when k = 1, the value is equal to 1, otherwise it is equal to zero.

9. The automobile driver dummy collision analysis system based on a six-dimensional force sensor according to claim 1 is characterized in that: The steps for determining the risk of neck injury are as follows: According to the color intensity of each area in the three-dimensional damage topology visualization map, the area corresponds to the high-risk area, the interactive parameters displayed in suspension, and the multi-dimensional potential energy topological derivative distribution gradient, the corresponding signals one, two and three are emitted, and the distribution status of signals one, two and three are statistically analyzed to correspondingly emit high risk, medium risk and low risk of neck injury.

10. The automobile driver dummy collision analysis system based on six-dimensional force sensor according to claim 4 is characterized in that: The steps of constructing the Lagrangian function are as follows: Weighted Damage Potential Energy Functional Constructing the Lagrangian function , represents the material parameter constraints, represents geometric conservation, represents the Lagrange multiplier corresponding to the inequality constraint, represents the Lagrange multiplier corresponding to the equality constraint, and θ represents the vector of material parameters to be optimized; Material parameter constraint g i (θ) and geometric conservation h j Specific definition of (θ): ； , where ζmax represents the maximum value of the set vibration coupling coefficient, represents the initial volume of the neck, V represents the volume of the neck, represents the neck node displacement, ubouy represents the prescribed displacement at the boundary, and i and j represent indices.

Images