A flexible three-dimensional force sensor based on orthogonal orientation and decoupling method

Abstract

The application discloses a flexible three-dimensional force sensor based on orthogonal oriented fiber layers and a decoupling method, and is used for solving the problems of serious force component coupling, complex structure and decoupling depending on a complex algorithm in the prior flexible three-dimensional force sensor. The sensor comprises two layers of orthogonally oriented conductive fiber piezoresistive layers and one layer of a random three-dimensional porous conductive fiber piezoresistive layer, and selectively responds to two tangential forces and a normal force. By utilizing the direction selectivity of the fiber orientation structure and the porous network, a three-dimensional force response matrix with a diagonal dominant characteristic is constructed, and three-dimensional force linear decoupling is realized based on a fixed sensitivity matrix. The sensor does not need a complex differential structure and a data driven model, has the advantages of simple structure, high decoupling precision, low calibration cost and the like, and can be used in the fields of robot tactile perception, human-computer interaction and flexible intelligent equipment.

Description

Technical Field

[0001] This invention belongs to the field of flexible tactile sensing technology, specifically relating to a flexible three-dimensional force sensor based on orthogonally oriented fiber layers and a decoupling method. Background Technology

[0002] Three-dimensional force (normal force and two orthogonal tangential forces) sensing is a fundamental requirement for robot dexterity, prosthetic tactile feedback, and human-computer interaction. The core challenge of flexible three-dimensional force sensors lies in signal decoupling: the responses of a single sensing layer to the three force components are intertwined, requiring complex structural differential, multi-mechanism fusion, or data-driven algorithms for separation, resulting in high calibration costs, poor cross-device consistency, and complex calculations.

[0003] At the device level, existing three-dimensional force sensing solutions mainly rely on geometric layout and force transmission structure to achieve directional differentiation. For example, US9797791B2 discloses a multi-axis soft artificial skin with liquid metal microchannels in an elastomer and a three-dimensional rigid force transmission component, where directional selectivity comes from the contact relationship between the microchannel geometry and the force transmission component; US11422045B2 discloses a flexible multi-axis force sensor with antagonistic sensitive elements arranged radially around a central raised area, detecting tangential and normal loads through differential resistance of orthogonal antagonistic pairs; WO2022035659A1 discloses a flexible tactile array that calculates shear force using an orthogonal combination of four piezoelectric sensitive elements. In the above solutions, each sensitive element is usually omnidirectional, and directional information depends on the spatial layout of the sensitive elements and differential calculations. The unit structure is complex, and the normal load still acts on all sensitive elements. The elimination of cross-coupling depends on back-end calibration and calculation.

[0004] At the materials level, the anisotropic piezoresistive effect of oriented conductive fiber networks has been studied: highly oriented carbon nanofiber membranes exhibit significant differences in piezoresistive sensitivity along the fiber direction and perpendicular to it; orthogonally stacking two oriented membranes can identify the in-plane strain direction; electrospun oriented cellulose / polypyrrole composite nanofibers also show direction-dependent resistance response; related reviews have summarized anisotropic conductive networks as a decoupling mechanism for multidimensional sensing. However, the above works have all stopped at identifying the in-plane strain direction and have not constituted a three-dimensional force sensor: firstly, they lack an independent sensing channel for normal force and cannot output complete three-dimensional force components; secondly, they do not involve the transmission relationship of surface contact force to the sensitive layer through the force transmission structure, and the oriented membrane directly bears in-plane tension rather than surface tangential load; thirdly, they have not established a quantifiable low-crosstalk sensitivity matrix relationship between the three-layer output and the three force components. In addition, existing studies have used the directional anisotropy of double-layer braided polymer optical fibers to achieve three-dimensional force detection and linear decoupling. However, these studies are based on the optical fiber loss mechanism, require a light source and photoelectric detection, and the normal force and tangential force share the same sensitive structure. They do not have an independent normal sensitive layer with clear suppression characteristics for in-plane loads.

[0005] In summary, current technology lacks a flexible three-dimensional force sensor that can simultaneously meet the following requirements: First, each of the three force components has its own dedicated sensitive material layer, and the directional selectivity comes from the response characteristics of the material layer itself rather than the spatial differential layout of the sensitive elements; Second, the sensitive layer responsible for the normal force has clear suppression characteristics against in-plane tension and shear, and the two layers responsible for the tangential force have clear suppression characteristics against each other in orthogonal directions and thickness compression, so that the three-dimensional sensitivity matrix satisfies the quantifiable diagonal dominance condition; Third, all sensitive layers adopt the same resistive readout mechanism, and only one fixed linear calibration is required to decouple the three-dimensional force, without the need for data-driven models such as neural networks, resulting in low calibration cost and good cross-device consistency. Summary of the Invention

[0006] To address the aforementioned shortcomings of existing technologies, the present invention aims to provide a flexible three-dimensional force sensor and decoupling method based on orthogonally oriented fiber layers. Through the collaborative design of two types of in-plane direction-selective piezoresistive layers and one type of thickness direction-selective piezoresistive layer, the three-dimensional sensitivity matrix satisfies the specified diagonal dominance condition, and three-dimensional force decoupling is achieved using only a fixed linear matrix.

[0007] To achieve the above objectives, the present invention adopts the following technical solution.

[0008] A flexible three-dimensional force sensor includes a first piezoresistive layer, a second piezoresistive layer, and a third piezoresistive layer arranged sequentially from bottom to top and electrically insulated from each other, and a common elastic force-transmitting protrusion covering the effective sensitive areas of the three layers; the first and second piezoresistive layers are both oriented conductive fiber network layers with a main orientation direction, and their main fiber orientation directions intersect each other, forming direction-selective sensitive layers in two orthogonal directions in the plane; the third piezoresistive layer is a random three-dimensional porous conductive fiber network layer, forming a thickness-direction-selective sensitive layer; under the action of external force transmitted by the protrusion, the output of the three piezoresistive layers forms a sensitivity matrix relative to the two orthogonal tangential force components and the normal force component, wherein the sensitivity of each piezoresistive layer to the stress component is greater than a specified multiple of the sensitivity of the layer to the other two force components, so that the sensitivity matrix satisfies the diagonal dominance condition.

[0009] Furthermore, the angle between the main fiber orientation directions of the first piezoresistive layer and the second piezoresistive layer is 80° to 100°, preferably 88° to 92°; the two layers are assembled by cutting the same oriented fiber membrane and rotating it 90°.

[0010] Furthermore, both the first piezoresistive layer and the second piezoresistive layer are conductive composite nanofiber membranes collected by electrospinning orientation, with conductive fillers blended within the fibers. The conductive fillers are one or more of carbon nanotubes, silver nanowires, carbon black, and MXene. The ratio of the tangential force sensitivity of a single layer along the main orientation direction to that in the perpendicular direction is not less than 5, preferably not less than 8.

[0011] Furthermore, the third piezoresistive layer is a three-dimensional loose conductive fiber network of random electrospun fibers with a porosity of not less than 70%; the ratio of its thickness-direction compressive sensitivity to its in-plane tensile sensitivity is not less than 5.

[0012] Furthermore, an elastic insulating film is provided between adjacent piezoresistive layers. The thickness of the elastic insulating film is 5 to 50 μm, and its elastic modulus is not higher than that of the piezoresistive layer.

[0013] Furthermore, the boss is an elastic hemisphere or cylinder, and its bottom surface projection covers the effective sensitive area of ​​the three piezoresistive layers. The ratio of the boss height to the bottom diameter is 0.3 to 1.

[0014] Furthermore, the diagonal dominance condition of the sensitivity matrix is ​​that the sum of the absolute values ​​of the diagonal elements accounts for a proportion of the sum of the absolute values ​​of all elements in the matrix not less than 0.80, and the ratio of any off-diagonal element to the diagonal element in the same row does not exceed 0.20.

[0015] Furthermore, the sensor also includes a temperature reference unit located in the strain isolation region. The temperature reference unit is made of the same material as the three piezoresistive layers, and its output is used for common-mode temperature compensation of the three-layer signals.

[0016] The present invention also provides a three-dimensional force decoupling method for the above-mentioned flexible three-dimensional force sensor, comprising the following steps: (1) acquiring the normalized resistance changes of the three piezoresistive layers respectively, and performing common-mode differential with the output of the temperature reference unit; (2) obtaining a fixed 3×3 sensitivity matrix through single-axis and combined load calibration experiments, and verifying that it satisfies the diagonal dominance condition; (3) multiplying the three differential signals by the inverse matrix or generalized inverse matrix of the sensitivity matrix to recover the two orthogonal tangential force components and normal force components; the decoupling process does not use a nonlinear model based on sample iterative training.

[0017] Furthermore, when the load exceeds the specified linear range, the decoupling method also includes a residual compensation step: the three-dimensional force component obtained by linear decoupling and the three normalized resistance signals are input into a pre-trained residual compensation model, and the model outputs the correction amount of the three-dimensional force component. After correction, the three-dimensional force component under the extended range is obtained. The residual compensation model only learns the difference between the linear decoupling result and the measured force value, and its training data is the loading data that exceeds the linear range in the calibration experiment.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0019] (1) The directional selectivity is borne by the material layer itself, and the unit structure is simple. Each of the three force components has its own dedicated sensitive layer. The directional information comes from the intrinsic response characteristics of the orientation fiber network and the porous network. There is no need for radial differential layout, multi-mechanism fusion or complex force transmission geometry. The sensitive layer can be prepared, cut and assembled in whole, and the process is fully compatible with the electrospinning platform.

[0020] (2) Three-layer synergy achieves quantified low crosstalk. Unlike single-layer anisotropic films that only demonstrate strain direction identification, this invention achieves device-level three-dimensional sensitivity matrix that satisfies the diagonal dominance condition through the synergy of layer sequence design, interlayer insulation, boss stress transfer and third-layer intralayer suppression characteristics, and crosstalk is suppressed within a specified ratio.

[0021] (3) Decoupling requires only one fixed linear calibration, and lightweight residual compensation can be selected for extended range. The near diagonal matrix degenerates the solution into a 3×3 linear operation. Within the normal range, no neural network training or a large number of calibration samples are required. It has good real-time performance, physical interpretability, and high consistency across devices. In the extended range, a lightweight compensation model that only learns linear residuals can be selected. Since the sensitivity matrix is ​​diagonally dominant, the residuals are small and low-dimensional. The number of parameters and training data requirements of this model are reduced by about an order of magnitude compared with the end-to-end learning scheme. This is an additional benefit brought by the three-layer orthogonal structure of this invention.

[0022] (4) All three layers use resistive readout. Unlike optical fiber and piezoelectric solutions that require dedicated readout, all three layers and the temperature reference unit use a unified resistive acquisition circuit, which simplifies the system and suppresses temperature drift with common-mode differential of the reference unit. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the layered structure of the flexible three-dimensional force sensor in an embodiment of the present invention;

[0024] Figure 2 These are morphological and orientation characterization diagrams of the oriented fiber layer and the random fiber layer in an embodiment of the present invention;

[0025] Figure 3 This is a normalized sensitivity matrix diagram of the three piezoresistive layers measured in Example 1;

[0026] Figure 4 The graph shows the response curves of the three piezoresistive layers under uniaxial loading in Example 1.

[0027] Figure 5 This is a comparison diagram of the decoupling errors of the present invention and various control schemes in Example 2;

[0028] Figure 6 This is a comparison chart of errors between linear decoupling and residual compensation under extended range in Example 2.

[0029] Reference numerals in the attached figures: 1-First piezoresistive layer; 2-Second piezoresistive layer; 3-Third piezoresistive layer; 4-Elastic insulating film; 5-Elastic force transmission boss; 6-Temperature reference unit; 7-Leading electrode; 8-Elastomer substrate. Detailed Implementation

[0030] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. The following embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0031] Specific Implementation Method 1: This embodiment describes a flexible three-dimensional force sensor with the following structure: from bottom to top, it consists of an elastic substrate, a first piezoresistive layer, an elastic insulating film, a second piezoresistive layer, an elastic insulating film, a third piezoresistive layer, and a common elastic force-transmitting protrusion; the first and second piezoresistive layers are oriented conductive fiber network layers with mutually intersecting main orientation directions, and the third piezoresistive layer is a random three-dimensional porous conductive fiber network layer; the sensitivity of each piezoresistive layer to stress components is greater than a specified multiple of the sensitivity to the other two force components, and the sensitivity matrix satisfies the diagonal dominance condition; during decoupling, the three normalized resistance signals are subjected to temperature common-mode differential, and then multiplied by the inverse matrix of the calibrated sensitivity matrix to recover the three-dimensional force components.

[0032] Specific Implementation Method Two: This implementation method differs from Specific Implementation Method One in that the included angle between the main orientation directions of the fibers in the first and second piezoresistive layers is 80° to 100°, and the two layers are assembled by rotating 90° after being cut from the same oriented fiber membrane. Other structures and parameters are the same as in Specific Implementation Method One.

[0033] Specific Implementation Method 3: This implementation method differs from Specific Implementation Method 1 or 2 in that the oriented conductive fiber network layer is a conductive composite nanofiber membrane collected by electrospinning orientation, and the conductive filler is one or more of carbon nanotubes, silver nanowires, carbon black, and MXene. The sensitivity ratio of the main orientation direction to the perpendicular direction of a single layer is not less than 5. Other structures and parameters are the same as in Specific Implementation Method 1 or 2.

[0034] Specific Implementation Method Four: This implementation method differs from Specific Implementation Methods One to Three in that the porosity of the third piezoresistive layer is not less than 70%, and the ratio of thickness compression sensitivity to in-plane tensile sensitivity is not less than 5. Other structures and parameters are the same as in Specific Implementation Methods One to Three.

[0035] Specific Implementation Method Five: This implementation method differs from one of Specific Implementation Methods One to Four in that the diagonal dominance condition is that the absolute value ratio of diagonal elements is not less than 0.80 and the maximum off-diagonal crosstalk ratio does not exceed 0.20. Other structures and parameters are the same as in one of Specific Implementation Methods One to Four.

[0036] Specific Implementation Method Six: This implementation method differs from Specific Implementation Methods One to Five in that it includes a temperature reference unit located in the strain isolation region, using the same material system as the piezoresistive layer. The three signals are first decoupled via common-mode differential switching. Other structures and parameters are the same as in Specific Implementation Methods One to Five.

[0037] Specific Implementation Method Seven: This implementation method differs from Specific Implementation Methods One through Six in that residual compensation is performed when the load exceeds the specified linear range. Using the linear decoupling result and three normalized resistance signals as inputs, a pre-trained lightweight residual compensation model outputs a correction value. This model only learns the difference between the linear decoupling result and the measured force value. Other structures and parameters are the same as in Specific Implementation Methods One through Six.

[0038] Example 1 (Sensor Fabrication and Sensitivity Matrix Calibration): A spinning solution was prepared using TPU as the fiber matrix and multi-walled carbon nanotubes (1.0 wt%) as the conductive filler. An oriented nanofiber membrane was obtained by collecting the solution using a high-speed drum at 2000 rpm, with a two-dimensional Fourier transform orientation degree of 0.82. Two pieces of the same membrane were cut and rotated 90° to serve as the first and second piezoresistive layers, respectively. A random, fluffy fiber membrane with a porosity of approximately 78% was obtained by static, low-speed collection. The three layers were insulated by a pure TPU electrospun membrane with a thickness of approximately 20 μm. The top was bonded with a PDMS hemispherical protrusion with a diameter of 5 mm and a height of 2 mm using pressure-sensitive adhesive. A temperature reference unit of the same system was set in the edge strain isolation region. Single-layer characterization: The tangential sensitivity ratio between the main orientation direction and the perpendicular direction of the oriented layer was 12.6; the sensitivity ratio between thickness compression and in-plane tension of the random layer was 8.4. The calibration was performed by applying single-axis and combined loads on a triaxial loading stage. The measured proportion η of the diagonal elements of the 3×3 sensitivity matrix was 0.87, and the maximum off-diagonal crosstalk ratio was 0.13, which met the diagonal dominance condition.

[0039] Example 2 (Decoupling Accuracy and Comparative Verification): The sensor from Example 1 was tested after a single linear calibration. The determination coefficient R² for the reconstruction of each component of the three-dimensional force was 0.985, the full-range error was 4.6%FS, and the force direction angle error under combined loading was 3.8°. Using another device from the same batch, the error using the original calibration matrix was 6.2%FS, verifying cross-device consistency. Four control groups were set up: Control A consisted of three random fiber membranes; Control B had first and second layers that were oriented but parallel and not orthogonal; Control C had only two orthogonally oriented layers and no third layer; Control D used an end-to-end neural network instead of linear decoupling. Results: The sensitivity matrices of Controls A and B were severely off-diagonal (η was 0.41 and 0.56 respectively), with linear decoupling errors reaching 18.5%FS and 12.7%FS respectively; Control C, lacking an independent normal sensitive layer, could not independently recover the normal component; Control D required 500 sets of calibration data to achieve an accuracy comparable to the single linear calibration of this invention. Furthermore, the linear decoupling results of this invention were cascaded with a lightweight residual compensation model (two-layer fully connected network, approximately 1200 parameters): within the conventional range (0–5 N), the pure linear decoupling error was 4.6% FS; in the extended range (up to 10 N), due to enhanced material nonlinearity, the linear decoupling error increased to 6.8% FS, but after cascaded residual compensation, it decreased to 3.2% FS. This compensation model converged after training with only 50 sets of calibration data in the superlinear region, requiring an order of magnitude less training data than the control model D. The above comparison demonstrates the necessity of the three-layer directional selective division of labor and the design of the diagonally dominant matrix, as well as the dimensionality reduction benefits of structural orthogonalization for the residual learning task.

[0040] Example 3 (Temperature Robustness Verification): During temperature variation testing within the range of 15–60°C, after enabling the common-mode differential temperature reference unit, the three-dimensional force zero-point drift decreased from 6.5% of the full scale to 1.2%. The same calibration matrix was used throughout the entire temperature range, and the error remained within 5.0%FS.

[0041] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A flexible three-dimensional force sensor based on orthogonally oriented fiber layers, characterized in that, The system comprises a first piezoresistive layer, a second piezoresistive layer, and a third piezoresistive layer, arranged sequentially from bottom to top and electrically insulated from each other, and a common elastic force-transmitting protrusion covering the effective sensitive areas of the three layers. The first and second piezoresistive layers are both oriented conductive fiber network layers with a main orientation direction, and their main fiber orientation directions intersect each other, forming two orthogonal direction-selective sensitive layers in the plane. The third piezoresistive layer is a random three-dimensional porous conductive fiber network layer, forming a thickness-direction-selective sensitive layer. Under the action of the external force transmitted by the protrusion, the output of the three piezoresistive layers forms a sensitivity matrix relative to the two orthogonal tangential force components and the normal force component. The sensitivity of each piezoresistive layer to the stress component is greater than a specified multiple of the sensitivity of the layer to the other two force components, so that the sensitivity matrix satisfies the diagonal dominance condition.

2. The flexible three-dimensional force sensor based on orthogonally oriented fiber layers according to claim 1, characterized in that, The angle between the main fiber orientation directions of the first piezoresistive layer and the second piezoresistive layer is 80° to 100°. The two layers are assembled by cutting the same oriented fiber membrane and rotating it 90°.

3. The flexible three-dimensional force sensor based on orthogonally oriented fiber layers according to claim 1, characterized in that, Both the first piezoresistive layer and the second piezoresistive layer are conductive composite nanofiber membranes collected by electrospinning orientation, with conductive fillers blended inside the fibers. The conductive fillers are one or more of carbon nanotubes, silver nanowires, carbon black, and MXene. The ratio of the tangential force sensitivity of a single layer along the main orientation direction to that in the perpendicular direction is not less than 5.

4. The flexible three-dimensional force sensor based on orthogonally oriented fiber layers according to claim 1, characterized in that, The third piezoresistive layer is a three-dimensional loose conductive fiber network of random electrospun fibers with a porosity of not less than 70% and a ratio of thickness-direction compressive sensitivity to in-plane tensile sensitivity of not less than 5.

5. The flexible three-dimensional force sensor based on orthogonally oriented fiber layers according to claim 1, characterized in that, An elastic insulating film is provided between adjacent piezoresistive layers. The thickness of the elastic insulating film is 5 to 50 μm, and the elastic modulus is not higher than that of the piezoresistive layer. The boss is an elastic hemisphere or cylinder, and its bottom surface projection covers the effective sensitive area of ​​the three piezoresistive layers. The ratio of the boss height to the bottom diameter is 0.3 to 1.

6. The flexible three-dimensional force sensor based on orthogonally oriented fiber layers according to claim 1, characterized in that, The diagonal dominance condition is as follows: the sum of the absolute values ​​of the diagonal elements of the sensitivity matrix accounts for a proportion of the sum of the absolute values ​​of all elements of the matrix not less than 0.80, and the ratio of any off-diagonal element to the diagonal element in the same row does not exceed 0.

20.

7. The flexible three-dimensional force sensor based on orthogonally oriented fiber layers according to claim 1, characterized in that, It also includes a temperature reference unit located in the strain isolation region. The temperature reference unit is made of the same material as the three piezoresistive layers, and its output is used for common-mode temperature compensation of the signals of the three piezoresistive layers.

8. A three-dimensional force decoupling method for a flexible three-dimensional force sensor according to any one of claims 1 to 7, characterized in that, The process includes the following steps: (1) Collecting the normalized resistance changes of the three piezoresistive layers respectively, and performing common-mode differential analysis using the temperature reference unit output; (2) Obtaining a fixed 3×3 sensitivity matrix through uniaxial and combined load calibration experiments, and verifying that it satisfies the diagonal dominance condition; (3) Multiplying the three differential signals by the inverse matrix or generalized inverse matrix of the sensitivity matrix to recover the two orthogonal tangential force components and normal force components; The decoupling process does not use a nonlinear model based on sample iterative training.

9. The three-dimensional force decoupling method according to claim 8, characterized in that, When the load exceeds the specified linear range, a residual compensation step is also included: the three-dimensional force component obtained by linear decoupling and the three normalized resistance signals are input into the pre-trained residual compensation model, and the model outputs the correction amount of the three-dimensional force component. After correction, the three-dimensional force component under the extended range is obtained. The residual compensation model only learns the difference between the linear decoupling result and the measured force value, and its training data is the loading data that exceeds the linear range in the calibration experiment.

10. The three-dimensional force decoupling method according to claim 9, characterized in that, The residual compensation model is a fully connected neural network with no more than 10,000 parameters.

Images