Nonlinear decoupling method and system of three-axis force sensor

Abstract

The invention relates to the technical field of force sensors, and discloses a nonlinear decoupling method and system for a three-axis force sensor, and the method comprises the steps: collecting multi-dimensional original data, and carrying out the adaptive noise reduction, dynamic selection of a wavelet basis, and construction of a filtering matrix, and obtaining a noise reduction signal through calculation; constructing a dynamic PINN proxy model, fusing material parameters and environment parameters, outputting a stiffness matrix, establishing an online parameter updating mechanism, calculating the obtained optimization parameters, and constructing an enhanced dynamic compensation matrix; the output stiffness matrix and the enhanced dynamic compensation matrix are fused, and a main singular value is reserved; reconstructing to obtain a reconstructed strain field, establishing a multi-physics field coupling model to compensate quantum-thermal coupling errors, and compensating wire inductive coupling under high frequency; constructing a multi-scale graph structure, strengthening a loss function through physical constraint, performing crystal plasticity hysteresis compensation, fusing multi-scale features, and calculating and outputting a decoupling force value; and an intelligent calibration decision is established, the drift trend is predicted, and the calibration period is dynamically adjusted.

Description

Technical Field

[0001] This invention relates to the field of force sensor technology, and specifically to a nonlinear decoupling method and system for a triaxial force sensor. Background Technology

[0002] Triaxial force sensors are precision instruments that simultaneously measure forces in two orthogonal directions. They are widely used in industrial automation, robotics, medical equipment, and fitness equipment. When an external force is applied to an elastic body, the strain gauge changes resistance as it deforms. The tiny resistance change is converted into a voltage signal by a bridge circuit. Some high-end products use the piezoelectric principle, which achieves dynamic force measurement through the positive piezoelectric effect of a piezoelectric quartz crystal and combines it with a charge amplifier to improve signal stability.

[0003] In existing technologies, linear models or simple polynomial expansions are used to handle nonlinear problems within a specific range. However, under high loads or complex operating conditions, higher-order nonlinear effects can still lead to significant errors. Traditional strain gauge layouts or bridge circuits can only partially suppress cross-interference through hardware structures, resulting in large coupling errors under dynamic forces. Existing decoupling methods rely on a large amount of static calibration data and require offline training, leading to long calibration cycles and high costs. After long-term use, parameter drift can cause the decoupling model to fail, and existing technologies lack online self-calibration capabilities. Automotive powertrain testing requires sensors with dynamic response bandwidths that are typically higher than those of traditional strain gauge sensors, and phase distortion exists. Hardware decoupling schemes often sacrifice sensitivity or structural strength. Pin-type sensors have grooves on the surface to enhance stress, but this weakens overload capacity. MEMS sensors have achieved miniaturization, but the decoupling structure is limited by the manufacturing process, resulting in insufficient lateral stiffness.

[0004] Therefore, there is a need to provide a nonlinear decoupling method and system for a triaxial force sensor. Summary of the Invention

[0005] The purpose of this invention is to provide a nonlinear decoupling method and system for a triaxial force sensor. To solve the aforementioned problems in the prior art, this invention achieves this through the following technical solution: In a first aspect, the present invention provides a nonlinear decoupling method for a triaxial force sensor, which specifically includes the following steps: Step 1: Collect multidimensional raw data, perform adaptive noise reduction through spatiotemporal correlation, dynamically select wavelet basis, introduce spatiotemporal correlation constraints, construct filter matrix and calculate the noise-reduced signal; Step 2: Based on the obtained denoised signal, construct a dynamic PINN proxy model, integrate material parameters and environmental parameters, output stiffness matrix, establish an online parameter update mechanism, calculate the obtained optimized parameters, and construct an enhanced dynamic compensation matrix; fuse the output stiffness matrix and the enhanced dynamic compensation matrix to retain principal singular values ​​and solve matrix ill-conditioning problems. Step 3: Based on the dynamic PINN proxy model, super-resolution strain field reconstruction is performed to obtain the reconstructed strain field. A multiphysics coupling model is established to compensate for quantum-thermal coupling errors and to compensate for wire inductive coupling at high frequencies. Step 4: Construct a multi-scale graph structure based on the reconstructed strain field, strengthen the loss function through physical constraints, perform crystal plastic hysteresis compensation, integrate multi-scale features, and calculate the output decoupling force value; Step 5: Based on the obtained decoupling force value, establish intelligent calibration decision, predict the drift trend, and dynamically adjust the calibration cycle.

[0006] Furthermore, the method for constructing the filter matrix is ​​as follows: Multimodal data is collected synchronously, and multidimensional raw data is acquired synchronously based on a preset sampling rate; Based on the obtained multidimensional raw data, adaptive noise reduction is performed through spatiotemporal correlation; Real-time detection of the dominant frequency of the signal using FFT Dynamically select the wavelet basis to obtain the optimized wavelet basis. According to the dominant frequency of the signal Dynamic switching; Based on the obtained optimized wavelet basis, spatiotemporal correlation constraints are introduced to construct the filtering matrix, and the following formula is used: The noise-reduced signal was calculated. ,in, To optimize the wavelet basis, The original input signal, For spatial gradient operators, Spatiotemporal correlation factor The infrared temperature field matrix, Here is the strain field matrix. It is a matrix direct product; Furthermore, the method for outputting the stiffness matrix is ​​as follows: A dynamic PINN proxy model is constructed, and a 4-layer, 128-node neural network is designed to integrate material parameters and environmental parameters, outputting a stiffness matrix. : ; in, For neural network modules, For Young's modulus, Poisson's ratio, For geometric tolerance, The network training parameters are obtained through pre-training with entity labeling data. Humidity coupling factor, For temperature coupling factor, For ambient humidity, The ambient temperature; Furthermore, the method for establishing the online parameter update mechanism is as follows: An online parameter update mechanism is established through the dynamic PINN proxy model; The model is pre-tuned every 1000 sets of measured data, optimizing parameters through gradient descent, using the formula: Calculate the optimization parameters ,in, For historical model parameters, For learning rate, The loss function with respect to the model parameters gradient vector, The L2 norm squared error between the model's predicted and measured values. To output the stiffness matrix, For the measured response signal, For standard force input; Furthermore, the method for solving the matrix ill-conditioned problem is as follows: Construct an enhanced dynamic compensation matrix; Through enhanced dynamic compensation matrix: ; Add a coupling term between the humidity change rate and the temperature gradient, where, The vibration attenuation coefficient is... For temperature gradient, , and These represent the accelerations of the beam along the X-axis, Y-axis, and Z-axis, respectively. Let be the derivative of ambient humidity with respect to time, i.e., the rate of change of humidity. , , The humidity change rate and humidity coupling coefficient are the humidity coupling coefficients. , , The vibration coupling coefficient is... , , This is the temperature coupling coefficient; The fusion matrix of the output stiffness matrix and the enhanced dynamic compensation matrix Perform truncated singular value decomposition (TSVD) and retain the first 6 principal singular values. in, It is a left singular matrix. It is a right singular matrix. It is a principal singular value diagonal matrix; Furthermore, the method for super-resolution strain field reconstruction is as follows: Ultrasonic surface waves are excited, and the time of flight of the sound waves is collected using four acoustic emission sensors. Combining the piezoresistive impedance and the time of flight of the sound waves, the strain field is reconstructed using a time-inversion reconstruction operator to improve spatial resolution, through the formula: Obtain the reconstructed strain field ,in, For time-reversal reconstruction operators, For piezoresistive impedance, For the time of sound wave flight, Use a Gaussian smoothing kernel; Compensating for quantum-thermal coupling errors: Establishing a multiphysics coupling model for the resistance of silver nanowires : ; in, Temperature coefficient of resistance For temperature gradient, The quantum tunneling coefficient, For the variation in the spacing between the silver lines, For the absolute value of strain, Thermo-humid coupling coefficient, The triple integral term characterizes the spatial coupling effect of strain, temperature, and humidity. Furthermore, the method for compensating for inductive coupling in conductors at high frequencies is as follows: To compensate for inductive coupling of conductors at high frequencies, for an X-axis beam, the formula is: The compensation voltage of the X-axis beam was calculated. ,in, The original voltage of the X-axis beam is... The distributed capacitance of the kth adjacent channel. For time steps, The inductance is distributed for the kth adjacent channel. The current is distributed in the kth adjacent channel. This provides an index for adjacent channels and compensates for high-frequency conductor inductive coupling on the Y-axis and Z-axis beams. Furthermore, the method for compensating for crystal plastic hysteresis is as follows: Construct a multi-scale graph structure, which includes: atomic layer, micro layer and macro layer; The loss function is strengthened by physical constraints, incorporating curl, divergence, and causal constraints, as shown in the formula: ; Obtain the enhancement loss function ,in, For physical constraints, For causal constraints, For data loss; Crystal plastic hysteresis compensation: Introducing a fractional-order hysteresis operator based on dislocation dynamics, using the formula: The hysteresis compensation strain was calculated. ,in, Shear modulus For the Bergman vector, For dislocation density, For fractional order, The fractional derivative of the strain force. For strain force; Furthermore, the method for calculating the output decoupling force value is as follows: Multi-scale features are fused using an 8-head image attention network (GAT), ultimately outputting a decoupling force value, expressed by the formula: The decoupling force value was calculated. ,in, For multi-scale features, These are the node feature vectors at the atomic, micro, and macro levels, respectively. The edge features from the atomic layer to the microscopic layer, The edge features from the micro-level to the macro-level. To compensate for the hysteresis strain, For strain force, It is an inertial force.

[0007] Secondly, the nonlinear decoupling system for a triaxial force sensor provided in this embodiment of the invention specifically includes the following modules: Data acquisition module: Collects multidimensional raw data, performs adaptive noise reduction through spatiotemporal correlation, dynamically selects wavelet basis, introduces spatiotemporal correlation constraints, constructs filter matrix to calculate the noise-reduced signal; Matrix enhancement module: Based on the obtained denoised signal, a dynamic PINN proxy model is constructed, material parameters and environmental parameters are fused, a stiffness matrix is ​​output, an online parameter update mechanism is established, the optimized parameters are calculated, and an enhanced dynamic compensation matrix is ​​constructed; the output stiffness matrix and the enhanced dynamic compensation matrix are fused to retain the principal singular values ​​and solve the matrix ill-conditioning problem; Reconstruction and compensation module: Based on the dynamic PINN proxy model, the super-resolution strain field is reconstructed to obtain the reconstructed strain field. A multi-physics coupling model is established to compensate for quantum-thermal coupling errors and to compensate for wire inductive coupling at high frequencies. Fusion output module: Based on the reconstructed strain field, a multi-scale graph structure is constructed. The loss function is strengthened through physical constraints, crystal plastic hysteresis compensation is performed, multi-scale features are fused, and the output decoupling force value is calculated. Intelligent calibration module: Based on the obtained decoupling force value, it establishes intelligent calibration decisions, predicts drift trends, and dynamically adjusts the calibration cycle.

[0008] The beneficial effects of this invention are: 1. By combining fractal geometry with multimodal sensing, the uniformity of strain transmission is optimized through fractal iteration. At the same time, the spatiotemporal coupling characteristics of infrared temperature field and strain field are utilized to improve the signal-to-noise ratio and break through the limitations of traditional single strain sensing. This enables the collaborative sensing of multiple parameters such as force, temperature and humidity, providing rich original information dimensions for decoupling. It also breaks through the bottleneck of traditional static calibration models, constructs a dynamic physical information neural network proxy model that integrates material parameters and environmental parameters, and introduces an online parameter update mechanism and an enhanced dynamic compensation matrix to solve the model mismatch problem caused by sensor aging and environmental changes, thus achieving dynamic self-adaptation. 2. A three-level multi-scale graph structure is constructed to associate material lattice properties, micro-strain unit characteristics, and macroscopic sensing data. By combining physical constraints and causal inference loss functions, the true and false correlations between physical fields are distinguished, overcoming the limitation of traditional macroscopic modeling that ignores the microscopic properties of materials. Atomic-level elastic parameters, crystal plastic hysteresis compensation, and macroscopic force decoupling are combined to achieve deep fusion of cross-scale physical information. An adaptive excitation strategy driven by deep reinforcement learning is introduced, and the calibration cycle is dynamically adjusted by combining LSTM to predict drift trends. The calibration mechanism breaks through the passivity of traditional fixed-cycle calibration and achieves active calibration and maintenance cost optimization under all working conditions. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart of the steps of a nonlinear decoupling method for a triaxial force sensor provided in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the nonlinear decoupling system of a triaxial force sensor provided in Embodiment 2 of the present invention. Detailed Implementation

[0011] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0012] Example 1: As Figure 1As shown in the figure, the nonlinear decoupling method for a triaxial force sensor provided by this embodiment of the invention specifically includes the following steps: Step 1: Collect multidimensional raw data, perform adaptive noise reduction through spatiotemporal correlation, dynamically select wavelet basis, introduce spatiotemporal correlation constraints, construct filter matrix and calculate the noise-reduced signal; The fusion of fractal microchannels and multimodal sensing enhances strain transfer uniformity through an H-shaped tree-like branching structure, introduces a spatiotemporal correlation filtering mechanism, and utilizes the spatiotemporal coupling characteristics of infrared temperature field and strain field to improve the dynamic signal-to-noise ratio. In a specific embodiment, the main material of the triaxial force sensor is an aluminum alloy beam, which has both high strength and good elastic recovery performance, ensuring good deformation linearity within the preset load range. use The ceramic structure forms a ceramic isolation layer, which effectively blocks the transmission of thermal strain and reduces the impact of temperature changes on the measurement. based on The ceramic isolation layer, constructed of ceramic, uses copper foil of a predetermined thickness to include the ceramic substrate on which the ceramic isolation layer is located, forming an electromagnetic shielding barrier that attenuates the intensity of external electromagnetic interference to within a predetermined range. Design the geometry of a triaxial force sensor, which includes an X-axis beam, a Y-axis beam, and a Z-axis beam; Design a fractal sensing array, laser-etch H-shaped tree-like branched microchannels on the surface of an elastomer, preset the main channel width and the branch channel width, and iterate the fractal three times; The integration of the fractal sensing array includes: 3-axis nano-silver wire strain gauges distributed at the bifurcation nodes, 8-point thin-film platinum resistance thermometers distributed along the depth direction of the flow channel, and a micro infrared thermopile array, to realize two-dimensional temperature field distribution monitoring. Multimodal data is acquired synchronously using light pulse-triggered sampling, and multidimensional raw data is obtained synchronously based on a preset sampling rate. The multidimensional raw data includes: infrared temperature field matrix Strain gauge output voltage Triaxial acceleration Ambient humidity and 8-point thin-film platinum resistance temperature ; Based on the obtained multidimensional raw data, adaptive noise reduction is performed through spatiotemporal correlation; Specifically, the dominant frequency of the signal is detected in real time using FFT. Dynamically select the wavelet basis using the formula: ; The optimized wavelet basis is calculated. According to the dominant frequency of the signal Dynamic switching is used for subsequent wavelet transform denoising of the original signal, wherein... It uses the Daubechies 8th order wavelet, which is superior for time-frequency decomposition of high-frequency signals. It is a Symlets 6th order wavelet, balancing the smoothness and orthogonality of low-frequency signals; Based on the obtained optimized wavelet basis, spatiotemporal correlation constraints are introduced to construct the filtering matrix, and the following formula is used: ; The noise-reduced signal was calculated. ,in, To optimize the wavelet basis, The original input signal, For spatial gradient operators, Spatiotemporal correlation factor The infrared temperature field matrix, Here is the strain field matrix. The matrix direct product is used to further suppress noise by utilizing the spatial correlation between the temperature field and the strain field. Step 2: Based on the obtained denoised signal, construct a dynamic PINN proxy model, integrate material parameters and environmental parameters, output stiffness matrix, establish an online parameter update mechanism, calculate the obtained optimized parameters, and construct an enhanced dynamic compensation matrix; fuse the output stiffness matrix and the enhanced dynamic compensation matrix to retain principal singular values ​​and solve matrix ill-conditioning problems. The PINN surrogate model of physical information neural network is combined with online multidimensional raw data feedback to realize real-time dynamic parameter updates, solving the problem of insufficient environmental adaptability of traditional static calibration models; a three-dimensional coupling term of humidity-temperature-force is introduced to construct a full-condition calibration matrix, improving the decoupling accuracy in complex environments; In a specific embodiment, a dynamic PINN proxy model is constructed, a 4-layer, 128-node neural network is designed, material parameters and environmental parameters are fused, and the stiffness matrix is ​​output. : ; in, For neural network modules, For Young's modulus, Poisson's ratio, For geometric tolerance, The network training parameters are obtained through pre-training with entity labeling data. Humidity coupling factor, For temperature coupling factor, For ambient humidity, The ambient temperature; It should be noted that the material parameters include: Young's modulus, Poisson's ratio, and geometric tolerances; the environmental parameters include: ambient humidity and ambient temperature. An online parameter update mechanism is established based on the dynamic PINN agent model; The model is pre-tuned every 1000 sets of measured data, optimizing parameters through gradient descent, using the formula: ; Calculate the optimization parameters ,in, For historical model parameters, For learning rate, The loss function with respect to the model parameters gradient vector, The L2 norm squared error between the model's predicted and measured values. To output the stiffness matrix, For the measured response signal, The standard force input ensures that the model dynamically adapts as the sensor ages; Based on the calculated optimization parameters, an enhanced dynamic compensation matrix is ​​constructed. Through enhanced dynamic compensation matrix: ; Add a coupling term between the humidity change rate and the temperature gradient, where, This is the vibration attenuation coefficient, with a preset value of 0.02. For temperature gradient, , and These represent the accelerations of the beam along the X-axis, Y-axis, and Z-axis, respectively. Let be the derivative of ambient humidity with respect to time, i.e., the rate of change of humidity. , , The humidity change rate and humidity coupling coefficient are the humidity coupling coefficients. , , The vibration coupling coefficient is... , , This is the temperature coupling coefficient; The fusion matrix of the output stiffness matrix and the enhanced dynamic compensation matrix Perform truncated singular value decomposition (TSVD) and retain the first 6 principal singular values. : ; in, It is a left singular matrix. It is a right singular matrix. Using a principal singular value diagonal matrix to solve matrix ill-conditioning problems; Step 3: Based on the dynamic PINN proxy model, super-resolution strain field reconstruction is performed to obtain the reconstructed strain field. A multiphysics coupling model is established to compensate for quantum-thermal coupling errors and to compensate for wire inductive coupling at high frequencies. In practice, the strain field super-resolution reconstruction fused with acoustic emission-impedance tomography is combined with time reversal algorithm to improve spatial resolution; the quantum tunneling-thermal coupling compensation model reduces the nonlinear error of the silver nanowire strain gauge under micro-strain. Super-resolution strain field reconstruction: Ultrasonic surface waves are excited, and the time-of-flight of the sound waves is collected using four acoustic emission sensors. Combining piezoresistive impedance and the time-of-flight of the sound waves, a time-reversal reconstruction operator is used to reconstruct the strain field and improve spatial resolution, using the formula: ; Obtain the reconstructed strain field ,in, For time-reversal reconstruction operators, For piezoresistive impedance, For the time of sound wave flight, Gaussian smoothing kernel, standard deviation The default value is 0.05; Compensating for quantum-thermal coupling errors: Establish a multiphysics coupling model of the resistance of silver nanowires : ; in, Temperature coefficient of resistance For temperature gradient, The quantum tunneling coefficient, For the variation in the spacing between the silver lines, For the absolute value of strain, Thermo-humid coupling coefficient, The triple integral term characterizes the spatial coupling effect of strain, temperature, and humidity. Compensate for inductive coupling in conductors at high frequencies and eliminate conductor crossing interference: Based on transmission line theory, to compensate for inductive coupling of conductors at high frequencies, for an X-axis beam, the formula is: ; The compensation voltage of the X-axis beam was calculated. ,in, The original voltage of the X-axis beam is... The distributed capacitance of the kth adjacent channel. For time steps, The inductance is distributed for the kth adjacent channel. The current is distributed in the kth adjacent channel. This provides an index for adjacent channels and compensates for high-frequency conductor inductive coupling on the Y-axis and Z-axis beams. Step 4: Construct a multi-scale graph structure based on the reconstructed strain field, strengthen the loss function through physical constraints, perform crystal plastic hysteresis compensation, integrate multi-scale features, and calculate the output decoupling force value; The multi-scale graph attention network MS-GAT integrates atomic-micro-macro structural information to solve the problem of traditional graph networks ignoring the microscopic properties of materials; it introduces a causal inference loss function to distinguish between causal and spurious correlations between physical fields, thereby improving the robustness of the decoupling model. Construct a multi-scale graph structure, which includes: atomic layer, micro layer and macro layer; Atomic layer: using lattice nodes of the elastomer material as units, eigenvectors , atomic density, It is the atomic-level elastic modulus. Poisson's ratio; Micro-layer: with micro-strain elements as nodes, feature vectors ,in, For micro-strain, For strain gradient, For microscopic temperature; Macroscopic layer: using physical nodes of the sensor array as units, feature vectors ,in, For macroscopic strain, The infrared temperature field matrix, For ambient humidity, the interlayer edge feature is defined as the scale transformation coefficient. The atomic-microscopic preset value is The micro-macro preset values ​​are ; The loss function is strengthened by physical constraints, incorporating curl, divergence, and causal constraints, as shown in the formula: ; Obtain the enhancement loss function ,in, For physical constraints, The curl of the strain force. For strain force, For theoretical curl, Let be the divergence of the Cauchy stress tensor. For Cauchy stress tensor, This is the vector of external forces; For causal constraints, Let KL divergence function be used. Let be the probability distribution function. For strain force, For ambient temperature, For ambient humidity, the conditional independence of force and environmental parameters is measured using the KL divergence measure. For data loss, To predict strain, To measure the strain force, The influence coefficient is preset to a value of 10. Perform crystal plastic hysteresis compensation: A fractional-order hysteresis operator based on dislocation dynamics is introduced, through the formula: ; The hysteresis compensation strain was calculated. ,in, Shear modulus For the Burgers vector, The dislocation density is estimated in real time using the PINN model. For fractional order, The fractional derivative of the strain force. For strain force; Decoupling calculation process: Multi-scale features are fused using an 8-head image attention network (GAT), ultimately outputting a decoupling force value, expressed by the formula: ; The decoupling force value was calculated. ,in, For multi-scale features, These are the node feature vectors at the atomic, micro, and macro levels, respectively. The edge features from the atomic layer to the microscopic layer, The edge features from the micro-level to the macro-level. To compensate for the hysteresis strain, For strain force, It is an inertial force; Step 5: Based on the obtained decoupling force value, establish intelligent calibration decision-making, predict the drift trend, and dynamically adjust the calibration cycle; A deep reinforcement learning (DQN)-driven adaptive excitation strategy, combined with laser thermal excitation and piezomechanical excitation, enables calibration without blind spots under all operating conditions; based on LSTM drift trend prediction, the calibration cycle is dynamically adjusted to reduce maintenance costs. Establish intelligent calibration decisions driven by deep reinforcement learning (DQN); Constructing the state space: ; in, For drift amplitude, Let be the derivative of ambient temperature with respect to time, i.e., the rate of temperature change. Signal-to-noise ratio (SNR) For ambient humidity; Construct a motion space, which includes: step mechanical excitation, frequency sweep mechanical excitation, random mechanical excitation and laser thermal excitation; Construct the reward function: ,in, To incentivize energy consumption, an ε-greedy strategy is used to select the optimal action; Based on the obtained motion space, the laser thermal excitation is calibrated: A controllable thermal strain field is generated by local heating using a fiber laser. : ; in, The coefficient of thermal expansion is... For laser power, The radius of the light spot; The thermal conductivity of TC4 is... The coordinates of the heating center are... The spatial coordinates of thermal strain; Predict drift trends and dynamically adjust calibration cycles; Specifically, the LSTM network is trained to predict the drift amount over the next N hours, and the calibration period is dynamically adjusted. For example, if the absolute value of the drift in the next N hours is less than 0.3%. If the calibration period is extended to 1.2 times the current period, the calibration period will remain unchanged.

[0013] Example 2: Figure 2 As shown in the figure, the nonlinear decoupling system for a triaxial force sensor provided in this embodiment of the invention specifically includes the following modules: Data acquisition module: Collects multidimensional raw data, performs adaptive noise reduction through spatiotemporal correlation, dynamically selects wavelet basis, introduces spatiotemporal correlation constraints, constructs filter matrix to calculate the noise-reduced signal; Matrix enhancement module: Based on the obtained denoised signal, a dynamic PINN proxy model is constructed, material parameters and environmental parameters are fused, a stiffness matrix is ​​output, an online parameter update mechanism is established, the optimized parameters are calculated, and an enhanced dynamic compensation matrix is ​​constructed; the output stiffness matrix and the enhanced dynamic compensation matrix are fused to retain the principal singular values ​​and solve the matrix ill-conditioning problem; Reconstruction and compensation module: Based on the dynamic PINN proxy model, the super-resolution strain field is reconstructed to obtain the reconstructed strain field. A multi-physics coupling model is established to compensate for quantum-thermal coupling errors and to compensate for wire inductive coupling at high frequencies. Fusion output module: Based on the reconstructed strain field, a multi-scale graph structure is constructed. The loss function is strengthened through physical constraints, crystal plastic hysteresis compensation is performed, multi-scale features are fused, and the output decoupling force value is calculated. Intelligent calibration module: Based on the obtained decoupling force value, it establishes intelligent calibration decisions, predicts drift trends, and dynamically adjusts the calibration cycle.

[0014] The above provides a detailed description of one embodiment of the present invention, but the content described is only a preferred embodiment of the present invention and should not be considered as limiting the scope of the present invention. The above formulas are all dimensionless numerical calculations, and the formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world situation. The preset parameters in the formulas are set by those skilled in the art based on actual conditions and historical experience, and can be adjusted according to actual conditions. The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. All equivalent changes and improvements made in accordance with the scope of the present invention should still fall within the patent coverage of the present invention.

Claims

1. A nonlinear decoupling method for a triaxial force sensor, characterized in that, Includes the following steps: Multidimensional raw data is collected, adaptive noise reduction is performed through spatiotemporal correlation, wavelet basis is dynamically selected, spatiotemporal correlation constraints are introduced, and a filter matrix is ​​constructed to calculate the noise-reduced signal. Based on the obtained denoised signal, a dynamic PINN proxy model is constructed, which integrates material parameters and environmental parameters, outputs the stiffness matrix, establishes an online parameter update mechanism, calculates the obtained optimization parameters, and constructs an enhanced dynamic compensation matrix. The output stiffness matrix and the enhanced dynamic compensation matrix are fused to preserve the principal singular values ​​and solve the matrix ill-conditioning problem; The reconstructed strain field is obtained by super-resolution strain field reconstruction based on the dynamic PINN proxy model. A multi-physics coupling model is established to compensate for quantum-thermal coupling error and to compensate for wire inductive coupling at high frequency. A multi-scale graph structure is constructed based on the reconstructed strain field. The loss function is enhanced by physical constraints, crystal plastic hysteresis compensation is performed, multi-scale features are integrated, and the decoupling force value is calculated and output. Intelligent calibration decisions are established based on the obtained decoupling force values, and drift trends are predicted and calibration cycles are dynamically adjusted.

2. The nonlinear decoupling method for a triaxial force sensor according to claim 1, characterized in that, The method for constructing the filter matrix is ​​as follows: Multimodal data is collected synchronously, and multidimensional raw data is acquired synchronously based on a preset sampling rate; Based on the obtained multidimensional raw data, adaptive noise reduction is performed through spatiotemporal correlation; Real-time detection of the dominant frequency of the signal using FFT Dynamically select the wavelet basis to obtain the optimized wavelet basis. According to the dominant frequency of the signal Dynamic switching; Based on the obtained optimized wavelet basis, spatiotemporal correlation constraints are introduced to construct the filtering matrix, and the following formula is used: The noise-reduced signal was calculated. ,in, To optimize the wavelet basis, The original input signal, For spatial gradient operators, Spatiotemporal correlation factor The infrared temperature field matrix, Here is the strain field matrix. It is a matrix direct product.

3. The nonlinear decoupling method for a triaxial force sensor according to claim 1, characterized in that, The method for outputting the stiffness matrix is ​​as follows: A dynamic PINN proxy model is constructed, and a 4-layer, 128-node neural network is designed to integrate material parameters and environmental parameters, outputting a stiffness matrix. : ； in, For neural network modules, For Young's modulus, Poisson's ratio, For geometric tolerances, The network training parameters are obtained through pre-training with entity labeling data. Humidity coupling factor, For temperature coupling factor, For ambient humidity, The ambient temperature.

4. The nonlinear decoupling method for a triaxial force sensor according to claim 1, characterized in that, The method for establishing the online parameter update mechanism is as follows: An online parameter update mechanism is established through the dynamic PINN proxy model; The model is pre-tuned every 1000 sets of measured data, optimizing parameters through gradient descent, using the formula: Calculate the optimization parameters ,in, For historical model parameters, For learning rate, The loss function with respect to the model parameters gradient vector, The L2 norm squared error between the model's predicted and measured values. To output the stiffness matrix, For the measured response signal, This is the standard force input.

5. The nonlinear decoupling method for a triaxial force sensor according to claim 1, characterized in that, The method for solving the ill-conditioned matrix problem is as follows: Construct an enhanced dynamic compensation matrix; Through enhanced dynamic compensation matrix: ； Add a coupling term between the humidity change rate and the temperature gradient, where, The vibration attenuation coefficient is... For temperature gradient, , and These represent the accelerations of the beam along the X-axis, Y-axis, and Z-axis, respectively. Let be the derivative of ambient humidity with respect to time, i.e., the rate of change of humidity. , , The humidity change rate and humidity coupling coefficient are the humidity coupling coefficients. , , The vibration coupling coefficient is... , , This is the temperature coupling coefficient; The fusion matrix of the output stiffness matrix and the enhanced dynamic compensation matrix Perform truncated singular value decomposition (TSVD) and retain the first 6 principal singular values. in, It is a left singular matrix. It is a right singular matrix. It is a main singular value diagonal matrix.

6. The nonlinear decoupling method for a triaxial force sensor according to claim 1, characterized in that, The method for super-resolution strain field reconstruction is as follows: Ultrasonic surface waves are excited, and the time of flight of the sound waves is collected using four acoustic emission sensors. Combining the piezoresistive impedance and the time of flight of the sound waves, the strain field is reconstructed using a time-inversion reconstruction operator to improve spatial resolution, through the formula: Obtain the reconstructed strain field ,in, For time-reversal reconstruction operators, For piezoresistive impedance, For the time of sound wave flight, Use a Gaussian smoothing kernel; Compensating for quantum-thermal coupling errors: Establishing a multiphysics coupling model for the resistance of silver nanowires : ； in, Temperature coefficient of resistance For temperature gradient, The quantum tunneling coefficient, For the variation in the spacing between the silver lines, For the absolute value of strain, Thermo-humid coupling coefficient, The triple integral term characterizes the spatial coupling effect of strain, temperature, and humidity.

7. The nonlinear decoupling method for a triaxial force sensor according to claim 1, characterized in that, The method for compensating for inductive coupling in conductors at high frequencies is as follows: To compensate for inductive coupling of conductors at high frequencies, for an X-axis beam, the formula is: The compensation voltage of the X-axis beam was calculated. ,in, The original voltage of the X-axis beam is... The distributed capacitance of the kth adjacent channel. For time steps, The inductance is distributed for the kth adjacent channel. The current is distributed in the kth adjacent channel. It serves as an index for adjacent channels and simultaneously compensates for conductor inductive coupling at high frequencies on the Y-axis and Z-axis beams.

8. The nonlinear decoupling method for a triaxial force sensor according to claim 1, characterized in that, The method for compensating for crystal plastic hysteresis is as follows: Construct a multi-scale graph structure, which includes: atomic layer, micro layer and macro layer; The loss function is strengthened by physical constraints, incorporating curl, divergence, and causal constraints, as shown in the formula: ； Obtain the enhancement loss function ,in, For physical constraints, For causal constraints, For data loss; Crystal plastic hysteresis compensation: Introducing a fractional-order hysteresis operator based on dislocation dynamics, using the formula: The hysteresis compensation strain was calculated. ,in, Shear modulus For the Burgers vector, For dislocation density, For fractional order, The fractional derivative of the strain force. For strain force.

9. The nonlinear decoupling method for a triaxial force sensor according to claim 1, characterized in that, The method for calculating the output decoupling force value is as follows: Multi-scale features are fused using an 8-head image attention network (GAT), ultimately outputting a decoupling force value, expressed by the formula: The decoupling force value was calculated. ,in, For multi-scale features, These are the node feature vectors at the atomic, micro, and macro levels, respectively. The edge features from the atomic layer to the microscopic layer, The edge features from the micro-level to the macro-level. To compensate for the hysteresis strain, For strain force, It is an inertial force.

10. A nonlinear decoupling system for a triaxial force sensor, the system being used to execute the decoupling method according to any one of claims 1-9, characterized in that, include: Data acquisition module: Collects multidimensional raw data, performs adaptive noise reduction through spatiotemporal correlation, dynamically selects wavelet basis, introduces spatiotemporal correlation constraints, constructs filter matrix to calculate the noise-reduced signal; Matrix enhancement module: Based on the obtained denoised signal, a dynamic PINN proxy model is constructed, material parameters and environmental parameters are fused, a stiffness matrix is ​​output, an online parameter update mechanism is established, the optimized parameters are calculated, and an enhanced dynamic compensation matrix is ​​constructed. The output stiffness matrix and the enhanced dynamic compensation matrix are fused to preserve the principal singular values ​​and solve the matrix ill-conditioning problem; Reconstruction and compensation module: Based on the dynamic PINN proxy model, the super-resolution strain field is reconstructed to obtain the reconstructed strain field. A multi-physics coupling model is established to compensate for quantum-thermal coupling errors and to compensate for wire inductive coupling at high frequencies. Fusion output module: Based on the reconstructed strain field, a multi-scale graph structure is constructed. The loss function is strengthened through physical constraints, crystal plastic hysteresis compensation is performed, multi-scale features are fused, and the output decoupling force value is calculated. Intelligent calibration module: Based on the obtained decoupling force value, it establishes intelligent calibration decisions, predicts drift trends, and dynamically adjusts the calibration cycle.

Images