A planar layered integrated FBG six-dimensional force sensor and a measuring method

Abstract

The application discloses a planar layered integrated FBG six-dimensional force sensor and a measuring method, and the sensor comprises a shell, an elastic body, a plurality of FBG sensitive elements, a plurality of optical fibers and a six-dimensional force decoupling module; the elastic body is an integrated planar symmetrical structure and is fixedly installed in the shell; the plurality of FBG sensitive elements are fixedly attached to strain concentration areas of the functional beam and are arranged in a differential measurement channel in a symmetrical differential mode; each FBG sensitive element is connected to the six-dimensional force decoupling module through an independent optical fiber and a light signal output hole of the optical fiber, and transmits the light signal to the six-dimensional force decoupling module; the input end of the six-dimensional force decoupling module is connected to the light signal output end of each optical fiber, used for receiving and processing the light signal, and outputting six-dimensional force and torque components. The six-dimensional force / torque measurement with high precision, low coupling and high reliability can be realized.

Description

Technical Field

[0001] This invention relates to the field of fiber optic sensing and multidimensional force measurement technology, and in particular to a planar layered integrated FBG six-dimensional force sensor and measurement method. Background Technology

[0002] With the development of minimally invasive surgical robots, precision assembly robots, and intelligent operating systems, the three-dimensional forces acting on end effectors ( , , ) and three-dimensional torque ( , , The real-time, high-precision measurement of surgical instruments has placed higher demands on them. Especially in the field of surgical robots, the force information at the instrument end is directly related to tissue safety, operational precision, and tactile feedback accuracy, and is a key foundation for realizing closed-loop force control and intelligent assisted operation.

[0003] Existing six-dimensional force sensor technologies mainly include resistance strain gauge, piezoelectric, capacitive, and fiber optic solutions. Among these, resistance strain gauge six-dimensional force sensors suffer from electromagnetic interference sensitivity, large signal drift, and limitations in miniaturization; piezoelectric sensors are difficult to use for static force measurement; capacitive structures have insufficient impact resistance and weak environmental adaptability; traditional fiber Bragg grating (FBG) six-dimensional force sensors often employ spatial three-dimensional distributed structures or series fiber structures, resulting in complex structures, high processing and assembly difficulties, challenges in temperature drift compensation, and channel faults affecting the stability of the entire signal. Furthermore, existing multi-dimensional force sensing structures mostly employ three-dimensional beam structures, which have the following drawbacks: large structural volume, unsuitable for integration into slender instruments; severe coupling between components, leading to complex calibration; insufficient structural symmetry, resulting in a high decoupling matrix condition number and decreased stability; and the series fiber optic deployment means that damage to a single point can affect the signal acquisition of the entire fiber. In terms of decoupling algorithms, although the traditional linear matrix method has a simple structure, it cannot effectively compensate for the systematic deviations caused by material nonlinearity, assembly errors and processing errors; although the pure neural network method has nonlinear fitting ability, it has problems such as large training data requirements, poor physical interpretability and insufficient model generalization stability.

[0004] Therefore, there is an urgent need to propose a sensing structure and measurement method that is planar, hierarchical, and highly symmetrical, with independently deployed optical fibers, strong anti-interference capabilities, a physical model foundation, and a nonlinear compensation mechanism, and can achieve high-precision six-dimensional force decoupling. Summary of the Invention

[0005] To address the problems of existing six-dimensional sensors, such as complex structural spatial layout, severe three-dimensional component coupling, limited miniaturization integration, insufficient reliability of traditional serial fiber optic deployment, and the difficulty of effectively compensating for structural nonlinearity and assembly errors by relying solely on linear calibration models in existing decoupling methods, as well as the insufficient physical interpretability and significant impact of random initialization on stability when relying solely on data-driven models, this invention provides a planar layered integrated FBG six-dimensional force sensor and its measurement method to achieve high-precision, low-coupling, and high-reliability six-dimensional force / torque measurement.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows.

[0007] In a first aspect, the present invention provides a planar layered integrated FBG six-dimensional force sensor, comprising: a shell, an elastomer, multiple FBG sensing elements, multiple optical fibers, and a six-dimensional force decoupling module; The elastomer is a planar symmetrical structure integrally machined and installed inside the housing. It includes a load-bearing boss at the center, four modified Maltese cross beams, four I-beam elastic beams, and four fixing blocks. Each of the four sides of the load-bearing boss is connected to a corresponding I-beam elastic beam via a modified Maltese cross beam, and the four connected components are orthogonally symmetrically distributed. The fixing blocks are connected to the grooves of adjacent I-beam elastic beams via horizontal force-sensing elastic beams to sense in-plane force components. , The improved Maltese cross beam includes a central beam that serves as a connector, and horizontal and vertical beams orthogonally connected to both sides of the central beam. Grooves for fiber optic winding and placement are provided at the contact points between the fixing blocks and the outer shell. The horizontal and vertical beams are used to achieve strain conversion of vertical force and moment components, while the vertical beam is used to sense torque. The horizontal beam is used to sense force. and torque , .

[0008] Multiple FBG sensing elements are attached and fixed to the strain concentration area of ​​each functional beam and arranged in an orthogonal symmetrical differential manner to form six sets of differential measurement channels; each FBG sensing element transmits optical signals to the six-dimensional force decoupling module through an independent optical fiber with lead-out holes; the functional beams include the horizontal force-sensing elastic beam between the fixed block and the bottom of the groove of the adjacent I-shaped elastic beam, and the horizontal and vertical beams in the improved Maltese cross beam;

[0009] The input end of the six-dimensional force decoupling module is connected to the optical signal output end of each optical fiber to receive and process optical signals and output a six-dimensional force and torque load vector.

[0010] Furthermore, there are a total of 11 FBG sensitive elements: An XY coordinate system is constructed based on the plane of four orthogonally symmetrically distributed I-beams. In each of the four I-beams, a horizontal force-sensing elastic beam is selected from the groove and the fixed block, with the coordinate axes of the horizontal force-sensing elastic beams on opposite sides facing each other. One FBG sensitive element is attached to the strain concentration area of ​​each functional beam, for a total of four. In the four improved Maltese cross beams, two have FBG sensitive elements attached to the strain concentration areas of both the horizontal and vertical beams, for a total of four. One horizontal beam has FBG sensitive elements attached to both its upper and lower surfaces, for a total of two. The remaining improved Maltese cross beam has an FBG sensitive element attached only to its upper surface, for a total of one.

[0011] Furthermore, the sensor employs a differential measurement method to construct a six-dimensional differential wavelength output vector; at any loading moment, the first... The wavelength offset of each fiber Bragg grating is defined as the difference between its real-time center wavelength and its initial center wavelength, and six differential wavelength outputs are constructed according to the following pairing relationship: ; ; ; ; ; ; in, , , These represent the three force components respectively. , , Differential wavelength output corresponding to the measurement channel; , , They represent circumference respectively. , , Axial torque components , , Differential wavelength output corresponding to the measurement channel. ( ) is numbered The change in the Bragg center wavelength of the fiber Bragg grating sensing unit; the preset correspondences are as follows: subscripts 7 and 8 correspond to Channel, subscripts 9 and 10 correspond Channel, subscripts 3 and 10 correspond Channel, subscripts 3 and 4 correspond Channels, subscripts 1 and 2 correspond Channel, subscripts 5 and 6 correspond Channel; differential operations are obtained by the difference in wavelength changes between two FBGs.

[0012] Based on this, a six-dimensional differential wavelength output vector is constructed. : ; The six-dimensional differential wavelength output vector serves as the input feature quantity for six-dimensional force decoupling; In the differential measurement method, the temperature drift terms of two fiber Bragg gratings symmetrically arranged in the same temperature field cancel each other out, thereby obtaining a differential wavelength output related to the corresponding load component and realizing temperature self-compensation.

[0013] Furthermore, within the overall elastic operating range of the sensor structure, the six-dimensional differential wavelength output vector and the six-dimensional force / torque load vector satisfy a linear mapping relationship: ; in, This is a six-dimensional differential wavelength output vector; The decoupling matrix to be calibrated is... This is a six-dimensional force / moment load vector; the six-dimensional force / moment load vector is solved using the least squares method as follows: ; in, For the estimated six-dimensional force / torque load vector, It is the minimum norm generalized inverse of the decoupling matrix, used to achieve linear decoupling; Among them, the decoupling matrix to be calibrated It was obtained through a standard load calibrating experiment, that is, by performing least squares fitting on each group of known standard loads and the corresponding differential wavelength output data.

[0014] Furthermore, the six-dimensional force decoupling module adopts an Extreme Learning Machine (DOA-ELM) model optimized based on the dream optimization algorithm. The DOA-ELM model takes a six-dimensional differential wavelength output vector as input and a six-dimensional force / torque load vector as output. The dream optimization algorithm is used to optimize the connection weights from the input layer to the hidden layer and the hidden layer bias of the Extreme Learning Machine. The output weights from the hidden layer to the output layer of the Extreme Learning Machine are obtained by solving the Moore-Penrose generalized inverse.

[0015] Furthermore, the hidden layer output matrix of the Extreme Learning Machine satisfies: ; in, For a six-dimensional differential wavelength input sample matrix, This is the connection weight matrix from the input layer to the hidden layer. This is the hidden layer bias vector. For activation function, This is the hidden layer output matrix; The output weights from the hidden layer to the output layer of the Extreme Learning Machine satisfy the following: ; in, Hidden layer output matrix Moore-Penrose generalized inverse, Output sample matrix for the corresponding six-dimensional force / torque load. This is the output layer weight matrix.

[0016] Secondly, the present invention provides a planar layered integrated FBG six-dimensional force measurement method, wherein the measurement method is based on the sensor described above and includes: Obtain the initial center wavelength of each FBG under no external load condition and the real-time center wavelength during the loading process, and calculate the wavelength drift of each FBG. Based on the construction of the FBG wavelength drift-strain-temperature coupling model and the sensitive beam load-strain response model, the differential output is obtained by adopting an orthogonal symmetrical differential arrangement for the same load component and mapped to the differential wavelength. Based on the preset differential pairing relationship, a six-dimensional differential wavelength output vector is obtained by differential combination; The six-dimensional differential wavelength output vector is input into the trained DOA-ELM model to predict and output the final six-dimensional force and torque load vector.

[0017] Furthermore, the FBG wavelength drift-strain-temperature coupling model is as follows: ; in, For the first The wavelength shift of a fiber Bragg grating; For the first The center wavelength of a fiber Bragg grating; The photoelastic coefficient; For the first Axial strain at the location of each fiber Bragg grating; The coefficient of thermal expansion of optical fiber. Thermo-optic coefficient, This refers to the change in temperature. The specific load-strain response model for sensitive beams is as follows: External load When applied, the axial strain at the sensitive location of the horizontal force-sensing elastic beam is measured. for: ; in, The maximum bending moment at the sensitive section; The moment of inertia of the cross section; Surface distance; The elastic modulus of the elastomer material; The bending moment correction factor for a horizontally force-sensing elastic beam; For along Load distribution coefficient among force-sensing elastic beams in multiple horizontal directions under directional external loads; To act on the overall structure of the sensor External input force component in the direction; The equivalent force arm length of the horizontal force-sensing elastic beam from the force transmission position to the sensitive section position; The cross-sectional thickness of the elastic beam that senses horizontal force in the plane corresponding to the bending direction; The cross-sectional width corresponding to the bending direction in the plane of the elastic beam that senses horizontal force; External load The situation and Same type, make The load distribution factor in the direction is Then external load When applied, the axial strain at the sensitive location of the horizontal force-sensing elastic beam is measured. for: ; axial force Axial strain at the sensitive location of the horizontal beam during operation for: ; in, Equivalent force borne by a single horizontal beam This is the effective length of the horizontal beam; The width of the horizontal beam; The thickness of the horizontal beam; This is the sharing factor; This indicates the number of the four horizontal beams; Around Axial torque Equivalent to the lever arm A pair of opposing axial forces at the location Strain at sensitive location of horizontal beam during operation for: ; in, To bypass Axial torque The contribution factor during operation; Around Axial torque Strain at sensitive location of horizontal beam during operation and Same type; Around Shaft torque Equivalent to acting on radius Tangential force at the point Strain at sensitive locations of the vertical beam during operation satisfy: ; in, To bypass Shaft torque The factor of contribution when it is applied.

[0018] Furthermore, the DOA-ELM model is obtained by training an extreme learning machine DOA-ELM optimized based on the dream optimization algorithm, with a six-dimensional differential wavelength output vector as input and a six-dimensional force / torque load vector as output. The training process of the DOA-ELM model includes: encoding the input layer weight matrix and hidden layer bias vector of the extreme learning machine as variables to be optimized, using the prediction error of the validation set as the fitness function, and using the dream optimization algorithm to find the optimal variables; after obtaining the optimal input layer weight matrix and the optimal hidden layer bias, constructing the hidden layer output matrix and solving the output layer weights to establish a nonlinear mapping model from six-dimensional differential wavelength features to six-dimensional force / torque output, until the prediction error meets the preset conditions, thus obtaining the trained DOA-ELM model.

[0019] Furthermore, during training, the weight matrix of the input layer of the Extreme Learning Machine is... With hidden layer bias vector The variables to be optimized are encoded, and the validation set prediction error is used as the fitness function. The model initialization parameters are optimized through a combination of global search and local exploitation, specifically including: During the initialization phase of DOA, the population is randomly generated. ; in, For the first Individual, and These are the lower and upper bounds of the search space, respectively. For population size, for Random vectors on; During the exploration phase, the number of iterations ranges from [number missing]. to DOA first divides the population based on differences in memory ability, and then allows individuals within a group to inherit the position of the group's historical best individual, that is: ; in, For the first Group 1 in The optimal individual in the next iteration. For the first The individual in the first The position vector at the next iteration. The threshold number of iterations used to delineate the exploration and development phases is used to divide the algorithm into the global search phase and the local development phase. Its value satisfies the following: ; Subsequently, in a randomly selected forgetting dimension The above implements a forgetting and replenishment strategy, and its update formula is: ; in, For the first The value of the j-th dimension component of each individual at the (t+1)-th iteration. Let be the component value in the j-th dimension of the optimal individual in the q-th group during the t-th iteration. Let be the lower bound of the j-th dimension of the search space. Let be the upper bound of the j-th dimension of the search space. For the first The forgetting dimension of the group For a randomly selected set of forgetting dimensions, This represents the maximum number of iterations. During the exploration phase, DOA also executes a dream-sharing strategy in parallel, that is, for any forgetting dimension... Its update formula is: ; in, From A natural number randomly selected from the list; During the development phase, the number of iterations ranges from [number missing]. to DOA no longer groups individuals; instead, it updates all individuals around the globally best historical individual. Its memory strategy is as follows: ; The corresponding forgetting and replenishment strategy update formula is: ; in, This represents the number of forgetting dimensions during the development phase. The DOA parameter settings satisfy the following: ; ; ; in, Let the dimension of the problem to be optimized be the dimension of the individual position vector. A function for generating random integers. It is a rounding function; And through preset parameter values Control the execution ratio of forgetting and replenishment strategies and dream sharing strategies during the exploration phase; when random numbers If the forgetting and replenishment strategy is applied, the dream sharing strategy is applied otherwise. For boundary handling, when optimizing the problem dimension When the value is less than or equal to the boundary preset value, a random backfill method is used: ; when When the value exceeds the preset boundary value, a reassignment method similar to the dream-sharing strategy used during the development phase is adopted: .

[0020] This invention proposes a planar layered integrated FBG six-dimensional force sensor and measurement method. Compared with the prior art, this invention has the following advantages:

[0021] 1. This invention adopts a planar layered integrated elastic body structure, which integrates the central load-bearing boss, the outer I-shaped flexible beam, the inner modified Maltese cross beam and the outer fixing block into a single design and integral molding. The structure is compact and the thickness dimension is controllable, making it suitable for integration into a six-dimensional force measurement system in slender and space-constrained scenarios. It has good advantages in miniaturization and engineering integration.

[0022] 2. The present invention uses the orthogonal symmetrical arrangement of four sets of outer I-shaped flexible beams and inner improved Maltese cross beams to form distinguishable strain responses in the corresponding sensitive areas for different load components. This helps to reduce the cross coupling between the six-dimensional force / moment components and improve the structural decoupling stability and measurement accuracy.

[0023] 3. This invention adopts independent optical fiber lead-out and symmetrical differential measurement method. On the one hand, it avoids the cascading effect of single-point damage on the entire signal acquisition caused by traditional series optical fiber structure, thus improving system reliability. On the other hand, it can effectively suppress the influence of temperature changes and other common-mode interference on the measurement results, thereby enhancing the environmental adaptability and long-term stability of the sensing system.

[0024] 4. Based on structural mechanics modeling and linear decoupling, this invention introduces an Extreme Learning Machine (DOA-ELM) optimized by the Dream Optimization Algorithm to establish a nonlinear mapping relationship between six-dimensional differential wavelength features and six-dimensional force / torque output. This not only retains the advantages of the linear decoupling model, such as clear physical meaning and fast solution speed, but also improves the accuracy, stability and generalization ability of the nonlinear decoupling model. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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.

[0026] Figure 1 This is an overall flowchart of the planar layered integrated FBG six-dimensional force measurement method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the overall structure of the planar layered integrated FBG six-dimensional force sensor provided in an embodiment of the present invention; wherein, Figure 2 (a) is an exploded view of the overall structure. Figure 2 (b) is a structural diagram of the elastomer; Figure 3 This is a schematic diagram showing the three views and key dimensions of an elastomer provided in an embodiment of the present invention; wherein, Figure 3 (a) is the top view. Figure 3 (b) is the bottom view. Figure 3 (c) is the front view; Figure 4 This is a schematic diagram of the layout of FBG on an elastomer provided in an embodiment of the present invention; wherein, Figure 4 (a) is a 3D schematic diagram. Figure 4 (b) is a plan view; Figure 5 This is a finite element simulation cloud diagram of the elastic body strain field when each component is loaded individually in an embodiment of the present invention; wherein, Figure 5 (a) for applying Equivalent strain distribution diagram on the elastic body at 5N. Figure 5 (b) for applying Equivalent strain distribution diagram on the elastic body at 5N; Figure 6 This is a finite element simulation cloud diagram of the elastic body strain field when each component is loaded individually in an embodiment of the present invention; wherein, Figure 6 (a) for applying Equivalent strain distribution diagram on the elastic body at 5N. Figure 6 (b) for applying Equivalent strain distribution diagram on the elastic body when N·mm = 100 N·mm; Figure 7 This is a finite element simulation cloud diagram of the elastic body strain field when each component is loaded individually in an embodiment of the present invention; wherein, Figure 7 (a) for applying Equivalent strain distribution diagram on the elastic body when N·mm = 100 N·mm Figure 7 (b) for applying Equivalent strain distribution diagram on the elastic body when N·mm = 100 N·mm; Figure 8 The flowchart of the DOA-ELM decoupling algorithm proposed in this embodiment of the invention is shown below. In the diagram: 1-Upper end cap, 2-Lower end cap, 3-Elastomer, 4-Horizontal force-sensing elastic beam, 5-Improved Maltese cross beam, 51-Horizontal beam, 52-Vertical beam, 53-Intermediate beam, 6-I-shaped elastic beam, 7-Fiber optic cable, 8-Fixing block, 9-Through hole, 10-Load-bearing boss. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other implementation methods obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0028] Example 1

[0029] like Figure 2 As shown, this embodiment provides a planar layered integrated FBG six-dimensional force sensor, including: a shell, an elastomer 3, multiple FBG sensing elements, multiple optical fibers 7, and a six-dimensional force decoupling module; The elastomer 3 is a planar symmetrical structure integrally machined and fixedly installed inside the outer shell. It includes a load-bearing boss 10 located at the center, four improved Maltese cross beams 5, four I-beam elastic beams 6, and four fixing blocks 8. Each of the four sides of the load-bearing boss 10 is connected to an I-beam elastic beam 6 via an improved Maltese cross beam 5, and the four connected structures are orthogonally symmetrically distributed. The fixing blocks are connected to the grooves of adjacent I-beam elastic beams 6 via horizontal force-sensing elastic beams 4, which are used to sense the force components in the plane. , The improved Maltese cross beam includes a central beam 53, and horizontal beams 51 and vertical beams 52 connecting the two sides of the central beam. The horizontal beams 51 and vertical beams 52 are orthogonal and are used to realize the strain conversion of vertical force and moment components. The vertical beams 52 are used to sense torque. The horizontal beam is used to sense force. and torque , The fixing block 8 has a groove at its contact point with the outer shell for winding and laying the optical fiber 7. In specific implementation, such as... Figure 2 As shown in (a), the outer shell is a hollow cylinder, which can be disassembled into an upper end cap 1 and a lower end cap 2. An elastomer 3 is embedded within the cylinder. The elastomer 3 is made of PEEK (polyetheretherketone) material and is formed using a high-precision 3D integrated printing process. PEEK material was chosen because it has high specific strength, good biocompatibility, low water absorption, and excellent fatigue resistance, making it suitable for integrated applications in space-constrained scenarios such as the end caps of surgical robots. The dimensions of each part of the elastomer structure are shown in the attached figure. Figure 3 As shown, where Figure 3 (a) is the front view. Figure 3 (b) is the rear view. Figure 3 (c) is a side view. The overall thickness of the elastic body 3 is 9mm. The outermost fixing block 8 is a fan-shaped block with a radius of 76mm. The load-bearing boss 10 at the center is a square column with a bottom side length of 6.4mm. In the outer I-shaped elastic beam 6 structure, the bottoms of the two grooves on the left and right sides form a space between them and the fixing block 8 for measurement. , There are eight component-sensitive beams, namely horizontal force-sensing elastic beams. Each horizontal force-sensing elastic beam 4 has an effective length of 14mm, a width of 1.5mm, and a thickness of 3.5mm. The intermediate beam of the improved Maltese cross beam 5 has dimensions of 8mm × 5mm × 2.2mm. Both the horizontal beam 51 and the vertical beam 52 are rectangular prisms with dimensions of 1.5mm × 8mm × 5mm. The cross-sections of the two rectangular prisms are orthogonally arranged on opposite sides of the intermediate beam, as shown below. Figure 2 As shown in (b), an arc-shaped groove is provided in the middle of the contact area between the external fixing block 8 and the outer shell for fiber winding and laying. The depth of the arc-shaped groove is 1.5mm.

[0030] Multiple FBG sensing elements are attached and fixed to the strain concentration area of ​​the functional beam and arranged in an orthogonal symmetrical differential manner to form six sets of differential measurement channels. Each FBG sensing element transmits optical signals to the six-dimensional force decoupling module through an independent optical fiber 7 with lead-out holes. In specific implementation, the diameter of the through holes on the improved Maltese cross beam 5 is 0.7 mm, and the diameter of the through holes on the horizontal force-sensing elastic beam formed between the left and right grooves of the outer I-shaped elastic beam 6 and the fixing block 8 is 1.2 mm. The functional beam includes the horizontal force-sensing elastic beam 4 between the fixing block 8 and the bottom of the groove of the adjacent I-shaped elastic beam 6, and the horizontal beam 51 and vertical beam 52 in the improved Maltese cross beam.

[0031] Specifically, there are 11 FBG sensitive elements, which are constructed into 6 sets of differential measurement features using a 6-channel independent lead-out method. One set of differential features is realized by multiplexing FBG3. Taking the XY coordinate system constructed on the plane of four orthogonally symmetrically distributed I-shaped elastic beams 6 as an example, in each of the four I-shaped elastic beams 6, a horizontal force-sensing elastic beam 4 formed between its groove and the fixing block 8 is selected, and the coordinate axes of the horizontal force-sensing elastic beams 4 on opposite sides are opposite (i.e., if one side selects the positive x-axis direction, the opposite side selects the negative x-axis direction; if one side selects the positive y-axis direction, the opposite side selects the negative y-axis direction). One FBG sensitive element is attached to the strain concentration area of ​​each functional beam, for a total of 4. In the four improved Maltese cross beams 5, FBG sensitive elements are attached to the strain concentration areas of the horizontal beams 51 and vertical beams of two of them, for a total of 4. FBG sensitive elements are attached to both the upper and lower surfaces of the horizontal beam 51 of one of them, for a total of 2. FBG sensitive elements are attached to the upper surface of the horizontal beam 51 of the remaining improved Maltese cross beam, for a total of 1. This implementation method aims to reduce the number of sensing elements and improve space utilization while maintaining structural symmetry and the integrity of differential features. Each FBG is connected to the spectral demodulation system via an independent lead-out method to reduce the impact of single-point failure on the overall measurement. In specific implementation, 11 FBG sensing elements are glued to predetermined strain positions using epoxy resin adhesive (pre-stretching is required), and connected to the spectral demodulator after the adhesive layer has cured.

[0032] The input end of the six-dimensional force decoupling module is connected to the optical signal output end of each optical fiber to receive and process optical signals and output six-dimensional force and torque components.

[0033] Under load, each functional beam undergoes distinguishable bending and tensile-compressive combined deformation, causing a shift in the center wavelength of the fiber Bragg grating. After establishing a load-wavelength mapping relationship through a six-dimensional force decoupling module, six-dimensional force and torque components are output. This structure is symmetrical overall, has a small axial dimension, is suitable for integration in space-constrained scenarios, and helps reduce coupling between components and improve decoupling stability.

[0034] Example 2

[0035] Based on the sensor described in Example 1, this embodiment provides a planar layered integrated FBG six-dimensional force measurement method. Specifically, in implementation... Figure 4 The XY coordinate plane constructed in (a) will be described in detail, including: S1: Obtain the initial center wavelength of each FBG under no external load conditions and the real-time center wavelength during the loading process, and calculate the wavelength drift of each FBG.

[0036] S11: Obtain the initial center wavelength of each fiber Bragg grating under no external load conditions. Establish a reference wavelength vector : ; In this specific implementation, the initial center wavelength of each FBG is preferably located in the range of 1548nm to 1555nm.

[0037] S12: Under known six-dimensional standard loads A dimensional loading calibration experiment was conducted under the action of [unclear] to obtain the real-time center wavelength of each FBG sensitive element. Calculate wavelength shift Forming wavelength drift vector Among them, wavelength shift The calculation formula is: ; in, For the first The center wavelength of a fiber Bragg grating.

[0038] More specifically, the dimension-by-dimensional loading calibration includes: respectively for , , , , , The six components are loaded individually, and each component is set with multiple loading amplitudes. The corresponding differential wavelength output data is recorded, which aims to build a complete calibration dataset for solving linear decoupling matrices and training DOA-ELM models.

[0039] S2: Based on the construction of the FBG wavelength drift-strain-temperature coupling model and the sensitive beam load-strain response model, the differential output is obtained by orthogonal symmetric differential arrangement for the same load component and mapped to differential wavelength.

[0040] S21: Based on the condition satisfied by the center wavelength of the fiber Bragg grating in the FBG sensing element and the axial strain of the grating With temperature change The relative wavelength drift caused by the interaction of factors is used to construct a wavelength drift-strain-temperature coupling model for FBG.

[0041] The condition that the center wavelength of the fiber Bragg grating in the FBG sensing element must satisfy is: ; in, The center wavelength, For effective refractive index, The period is the grating period.

[0042] FBG wavelength drift-strain relationship: Under axial strain, the first Wavelength shift of a fiber Bragg grating satisfy: ; in, For the first The center wavelength of a fiber Bragg grating The photoelastic coefficient, For the first Axial strain at the location of each fiber Bragg grating.

[0043] When the grating is subjected to axial strain With temperature change When they act together, their relative wavelength shift is: ; in, For the effective photoelastic coefficient, The coefficient of thermal expansion of optical fiber. Thermo-optic coefficient, This refers to the change in temperature. Further refinement yields the FBG wavelength drift-strain-temperature coupling model: .

[0044] S22: Constructing a Sensitive Beam Load-Strain Response Model: For the horizontal force-sensing elastic beam between a fixed block and an I-beam, and the horizontal and vertical beams in the improved Maltese cross beam, the correspondence between each load component and the axial strain at the sensitive location is constructed. Combined with the FBG wavelength drift-strain relationship, a sensitive beam load-strain response model for six-dimensional load decoupling is constructed. The specific process is as follows: S221: General strain modeling relation: For any sensitive elastic beam, the surface normal strain at its sensitive section under external load satisfies: ; in, For FBG axial strain, For the normal stress on the surface of the elastic beam, The elastic modulus of the elastomer material. The bending moment at the sensitive section, Let be the distance from the surface where the FBG is located to the neutral axis. Let be the moment of inertia of the beam section about the bending axis.

[0045] To reflect the load distribution among multiple beams and the influence of boundary constraints on the peak bending moment, the equivalent load acting on the target beam will be... Represented as: ; in, For load distribution factor, This refers to the external input load acting on the overall structure of the sensor; And the maximum bending moment at the sensitive section Represented as: ; in, These are the boundary constraint coefficients. The effective length of the target beam. In practical implementation, the load distribution factor... With boundary constraint coefficients It can be determined through finite element analysis or calibration tests to adapt to different structural parameters and assembly boundary conditions.

[0046] S222: Horizontal force-sensing elastic beam (used for measurement) between the outer ring fixing block and the bottom of the groove of the adjacent I-beam elastic beam. , Strain model of ) When external load When the force is applied, the horizontally oriented force-sensing elastic beam primarily undergoes in-plane bending deformation, and its moment of inertia... Distance from surface Pick: ; in, The cross-sectional thickness of the elastic beam that senses horizontal force in the plane corresponding to the bending direction; The cross-sectional width corresponding to the bending direction in the plane of the elastic beam that senses the horizontal force.

[0047] With equivalent lateral force Characterizes the load components borne by the beam, and in Characterizing the peak bending moment at the sensitive section, the axial strain at the sensitive location of the horizontal force-sensing elastic beam is then determined. satisfy: ; in, The bending moment correction factor for a horizontally force-sensing elastic beam; The load distribution coefficient among the force-sensing elastic beams under external loads along the x-direction in multiple horizontal directions; To act on the overall structure of the sensor External input force component in the direction; The equivalent force arm length of the horizontal force-sensing elastic beam from the force transmission position to the sensitive section position; External load The situation and For the same type, let the load distribution factor in the y direction be... ,but: .

[0048] S223: Inner ring modified Maltese cross beam (with) , , For example, among which, and Strain models of the same type: In this embodiment, the improved Maltese cross beam in the inner ring consists of a horizontal beam and a vertical beam. Preferably, the FBG is adhered to the horizontal beam in the strain concentration region. Let the effective length of the horizontal beam be... Width is Thickness is .

[0049] (1) Axial force Strain model (out-of-plane bending);

[0050] Axial force The load is shared by the horizontal beams in four improved Maltese cross beams, assuming an equivalent load on each individual beam. for: ; in, This is the sharing coefficient.

[0051] Horizontal beam under axial force Under the influence of the action, the bending is mainly outward-facing, and its moment of inertia is Distance from surface Pick: ; Strain at sensitive locations satisfy: ; (2) Torque The strain model (equivalent couple); Around Axial torque Equivalent to the lever arm A pair of opposing axial forces at the location ,satisfy: ; Considering the effects of moment sharing and structural coupling, these are combined into a sharing factor. The strain at the sensitive location of the horizontal beam satisfy: ; (3) Torque Strain model (uniform tangential force): Around Shaft torque Equivalent to acting on radius Tangential force at the point The vertical beams in the four improved Maltese cross beams are evenly distributed: ; Tangential force causes in-plane bending of the horizontal beam, and its moment of inertia Distance from surface Pick: ; The distribution of tangential force and the effects of coupling are combined into Then the strain at the sensitive location satisfy: .

[0052] S224: Constructing a differential temperature compensation model: The FBG wavelength drift-strain relationship in S21 is introduced into the strain model in S22. For the first... Each FBG sensitive element has a wavelength shift amount satisfy: ; in, For the effective photoelastic coefficient, The coefficient of thermal expansion of optical fiber. For the first The grating of the FBG sensing element is subjected to axial strain. Thermo-optic coefficient, This represents the change in temperature.

[0053] In differential measurement, the temperature drift terms of two fiber Bragg gratings (FBGs) symmetrically arranged in the same temperature field cancel each other out, thereby obtaining a differential wavelength output related to the corresponding load component and achieving temperature self-compensation. Specifically, by employing an orthogonal symmetrical differential arrangement for the same load component, ensuring that the two FBGs of the differential pair are in the same temperature region, the temperature terms can be canceled out, resulting in a differential output. : ; in, To determine the initial Bragg center wavelength of the two FBGs under reference temperature and strain-free conditions for differential measurement; To perform differential measurement of the axial strain generated along the fiber axis by the tension side of the FBG; This represents the axial strain generated along the fiber axis by the FBG on the pressure side of the differential measurement. The equation is a differential output model that suppresses the temperature cross-sensitivity of the fiber Bragg grating under the isothermal assumption, demonstrating the mechanism of temperature self-compensation achieved by differential measurement.

[0054] When the differential pairs are arranged in tension / compression symmetrical positions on the same beam, Thus, the above , , , , It is directly mapped to the differential wavelength output of the corresponding channel, thereby improving sensitivity and suppressing temperature drift.

[0055] S3: Perform differential combination based on the preset differential pairing relationship to obtain a six-dimensional differential wavelength output vector.

[0056] The initial center wavelength of each FBG sensing unit was obtained under the initial state without external load. and the real-time center wavelength at any loading moment And define the first by the difference between the two. Wavelength offset of each FBG : ; To suppress the influence of common-mode disturbances such as temperature changes on the measurement results, a differential method is used to perform differential processing on the outputs of two FBGs within the same measurement channel, resulting in differential wavelength outputs for six load component channels. In this specific implementation, the FBG sensing elements are arranged as follows... Figure 4 As shown, where Figure 4 (a) is a 3D schematic diagram. Figure 4 (b) is a front view diagram, and its pairing relationship is defined as follows: ; in, , , These represent the three force components respectively. , , Differential wavelength output corresponding to the measurement channel; , , They represent circumference respectively. , , Axial torque components , , Differential wavelength output corresponding to the measurement channel. ( ) is numbered The change in the Bragg center wavelength of the fiber Bragg grating sensing unit is caused by the combined effect of axial strain induced by external load and changes in ambient temperature. Each differential channel consists of two FBGs arranged symmetrically in a spatially symmetrical, isothermal field, and their corresponding relationships are as follows: correspond aisle, correspond aisle, correspond aisle, correspond aisle, correspond aisle, correspond Channel; differential operations are obtained by the difference in wavelength changes between two FBGs.

[0057] Based on this, a differential wavelength output vector is constructed. for: .

[0058] S4: Input the six-dimensional differential wavelength output vector into the trained DOA-ELM model to predict the final six-dimensional force and torque components.

[0059] Specifically, the DOA-ELM model outputs a six-dimensional differential wavelength vector. As input, a six-dimensional force / moment load vector As the output, the Extreme Learning Machine (DOA-ELM) optimized based on the dream optimization algorithm is trained. In this embodiment, the training of DOA-ELM can be implemented as follows: the calibration data is divided into a training set, a validation set, and a test set, where the training set is used to solve the output layer weights and participate in the DOA search, the validation set is used to calculate the fitness function, and the test set is used to evaluate the model's generalization ability. The training process of the DOA-ELM model includes: encoding the input layer weight matrix and hidden layer bias vector of the Extreme Learning Machine as variables to be optimized, using the validation set prediction error as the fitness function, and using the dream optimization algorithm to find the variables to be optimized; after obtaining the optimal input layer weight matrix and the optimal hidden layer bias, constructing the hidden layer output matrix and solving the output layer weights to establish a nonlinear mapping model from six-dimensional differential wavelength features to six-dimensional force / torque output.

[0060] Among them, the six-dimensional force / torque load vector in the training data The acquisition process is as follows: Define a six-dimensional force / torque load vector for: ; Within the structural elastic operating range, the differential wavelength output and the six-dimensional load satisfy an approximately linear mapping relationship, and a linear decoupling model is established: ; in, The decoupling matrix is ​​to be determined. To obtain the decoupling matrix, a calibration experiment is performed: applying... The group has a known standard load and synchronously acquires differential wavelength outputs. The differential wavelength output vector of the calibration data is denoted as The six-dimensional load vector measured according to the corresponding standard is denoted as .Will The data sets are stacked column-wise and written in matrix form. The decoupling matrix is ​​estimated using the least squares criterion and the minimum norm solution, resulting in: ; in, This is the optimal estimate of the decoupling matrix obtained by the least squares criterion; This is an observation data matrix composed of column-stacked differential wavelength output vectors acquired during calibration experiments.

[0061] Alternatively, the generalized inverse form can be used in the online solution: ; in, The results are linear estimates of the six-dimensional force / torque to be measured. It is the Moore–Penrose generalized inverse of the decoupling matrix.

[0062] Specifically, the Extreme Learning Machine (DOA-ELM) model optimized based on the dream optimization algorithm is introduced to establish a nonlinear mapping model between six-dimensional differential wavelength features and six-dimensional force / torque outputs. The Extreme Learning Machine part still adopts a single hidden layer feedforward network structure, and its hidden layer output matrix is: ; The output layer weight matrix satisfies: ; The final nonlinear decoupling output is: ; in, The input sample matrix is ​​the differential wavelength. This is the connection weight matrix from the input layer to the hidden layer. This is the hidden layer bias vector. For activation function, This is the hidden layer output matrix. Output sample matrix for the corresponding six-dimensional force / torque load. Let H be the Moore–Penrose generalized inverse matrix of the hidden layer output matrix. This is the output layer weight matrix. This is the six-dimensional force / torque output vector predicted by the extreme learning machine model optimized based on the Dream Optimization Algorithm (DOA).

[0063] Furthermore, the Dream Optimization Algorithm (DOA) is inspired by the mechanisms of "memory retention, partial forgetting, self-organization and replenishment, and information sharing" in human dreams, and incorporates the input layer weight matrix of the Extreme Learning Machine. With hidden layer bias vector The variables to be optimized are encoded, and the prediction error on the validation set is used as the fitness function. The model initialization parameters are optimized through a combination of global search and local exploitation. The initialization order of DOA is generated using the following random population generation method: ; in, For the first Individual, and These are the lower and upper bounds of the search space, respectively. For population size, for A random vector on.

[0064] During the exploration phase, the number of iterations ranges from [number missing]. to DOA first divides the population into 5 groups based on differences in memory ability, and then allows individuals within a group to inherit the position of the group's historical best individual, i.e.: ; in, For the first Group 1 in The optimal individual in the next iteration. For the first The individual in the first The position vector at the next iteration. The threshold number of iterations used to delineate the exploration and development phases is used to divide the algorithm into the global search phase and the local development phase. Its value satisfies the following: .

[0065] Subsequently, in a randomly selected forgetting dimension The above implements a forgetting and replenishment strategy, and its update formula is: ; in, For the first The value of the j-th dimension component of each individual at the (t+1)-th iteration. Let be the component value in the j-th dimension of the optimal individual in the q-th group during the t-th iteration. Let be the lower bound of the j-th dimension of the search space. Let be the upper bound of the j-th dimension of the search space. For the first The forgetting dimension of the group For a randomly selected set of forgetting dimensions, This represents the maximum number of iterations. This formula reflects the process of gradually transitioning from global search to local search during the exploration phase.

[0066] During the exploration phase, DOA also executes a dream-sharing strategy in parallel to enhance its ability to escape local optima. For any forgetting dimension... Its update formula is: ; in, From The natural number is randomly selected from the population. This strategy allows individuals to randomly absorb positional information from other individuals along the forgetting dimension, thereby enhancing population diversity and the ability to escape local extrema.

[0067] During the development phase, the number of iterations ranges from [number missing]. to DOA no longer groups individuals; instead, it updates all individuals around the globally best historical individual. Its memory strategy is as follows: ; The corresponding forgetting and replenishment strategy update formula is: ; in, This represents the forgetting dimension during the development phase. This phase primarily involves a fine-grained local search around the current global optimum to improve the accuracy of parameter optimization.

[0068] DOA parameter settings meet the following requirements: ; ; ; in, Let be the dimension of the problem to be optimized, i.e., the dimension of the individual position vector, corresponding to the total number of all parameters to be optimized in the extreme learning machine (input layer weights and hidden layer biases). This is a random integer generation function used to uniformly generate integer values ​​within a given interval. This is the floor function, used to map a real number to the smallest integer not less than the given value.

[0069] And through preset parameter values Control the execution ratio of forgetting and replenishment strategies and dream sharing strategies during the exploration phase; when random numbers If the dream is active, a forgetting and replenishment strategy is executed; otherwise, a dream-sharing strategy is executed, with preference given to the latter. .

[0070] For boundary handling, when optimizing the problem dimension At that time, a random filling method will be used: ; when At that time, a reassignment method similar to the dream-sharing strategy used in the development phase is adopted: ; This ensures that all individuals remain within the search boundary and enhances the global search capability in high-dimensional optimization problems. In this specific implementation, the boundary preset value is set to 15.

[0071] In this embodiment, DOA performs the adjustment of the ELM input layer weight matrix. With hidden layer bias vector Optimization of output weights from hidden layer to output layer in Extreme Learning Machine The Moore–Penrose generalized inverse analytical solution is still used. This approach retains the advantages of ELM in terms of fast training speed and simple structure, while avoiding the instability caused by the random initialization of traditional ELM, making the nonlinear mapping between the six-dimensional differential wavelength features and the six-dimensional force / torque output more accurate and robust.

[0072] To verify the rationality of the strain distribution and the feasibility of the differential strain design of the designed planar layered integrated elastomer structure, a three-dimensional model of the elastomer was preferably established in finite element analysis software, and simulation analysis was performed based on the structural dimensions, material parameters, and boundary constraints of the embodiment. In the simulation model, the elastomer material parameters were set as follows: elastic modulus of 3 × 10⁻⁶. 9 Pa, Poisson's ratio is 0.43, and density is The external fixed frame is set as a fully constrained boundary, and the central load-bearing boss serves as the input location for the external load. Load conditions include... , , , , , The six components are loaded individually, and the simulated loading range of each load component is preferably consistent with the subsequent physical calibration range.

[0073] The purposes of simulation analysis include: First, to verify the rationality of the structural design of the elastic beam between the two grooves of the outer I-shaped elastic beam and the fixed block, which mainly corresponds to the in-plane force component response, and the inner ring improved Maltese cross beam, which mainly corresponds to the vertical force and moment component response. Secondly, the strain concentration and tensile-compressive symmetrical distribution characteristics at each FBG bonding area were verified. Third, verify the feasibility of the differential distribution design method for temperature common-mode suppression and component differentiation; Fourth, it provides a reference for determining the subsequent load distribution coefficient, boundary constraint coefficient, and sensitivity matrix parameters.

[0074] Simulation results can be displayed through strain contour plots, displacement contour plots, and strain extraction results at feature points. Figure 5 (a) Figure 5 (b) respectively applied =5N、 Equivalent strain distribution diagram on the elastic body at 5N; Figure 6 (a) Figure 6 (b) respectively applied =5N、 Equivalent strain distribution diagram on the elastic body when N·mm = 100 N·mm; Figure 7 (a) Figure 7 (b) respectively applied =100 N·mm Equivalent strain distribution diagram on the elastic body at a strength of 100 N·mm. In the specific implementation process, differential feature construction and linear decoupling are achieved: Based on the constructed wavelength-strain-temperature coupling model, structural strain-load relationship model, and differential temperature compensation model, six sets of differential wavelength features are used as decoupling inputs, instead of directly using the original wavelength drift of all FBGs as the input to the nonlinear model. The real-time center wavelength of each FBG is acquired at any loading moment, and the wavelength offset is defined as the difference between this and the initial center wavelength. Subsequently, a 6-dimensional differential wavelength feature vector is constructed according to a preset differential pairing relationship. Channel and The channels are configured into differential pairs by multiplexing FBG4. Therefore, although this embodiment only uses 11 FBG sensitive units, it can still form 6 sets of differential measurement features.

[0075] Within the structural elastic operating range, based on the linear decoupling model, the decoupling matrix is ​​obtained through dimension-by-dimensional calibration, and the least squares method is used to complete the linear online solution of the six-dimensional load. For each set of standard loading conditions, differential wavelength eigenvectors are collected, which, together with the corresponding standard loads, constitute a calibration dataset used to obtain the linear decoupling matrix.

[0076] To further improve the accuracy of six-dimensional force / torque measurement, based on the above linear decoupling, an Extreme Learning Machine (DOA-ELM) optimized by the Dream Optimization Algorithm is introduced to establish a nonlinear mapping model between the six-dimensional differential wavelength features and the six-dimensional force / torque output. The specific process is as follows: Figure 8 As shown.

[0077] In this embodiment, the input to the DOA-ELM model is a 6-dimensional differential wavelength eigenvector, and the output is a 6-dimensional force / torque load vector. The Extreme Learning Machine (ELM) employs a single hidden layer feedforward network structure, and the solution for the hidden layer output matrix and output layer weights can follow the definitions described in the aforementioned invention. The Dream Optimization Algorithm (DOA) is used to optimize the input layer weight matrix and hidden layer bias vector of the ELM, while the output layer weights are still solved analytically using the Moore-Penrose generalized inverse method.

[0078] In this embodiment, the optimization process of DOA is as follows: First, the input layer weight matrix and hidden layer bias vector are encoded as variables to be optimized, and an initial population is generated using a random population initialization method. ; in, For the first Individual, and These are the lower and upper bounds of the search space, respectively. Population size.

[0079] During the exploration phase, the number of iterations was [number missing]. to DOA divides the population into 5 groups and uses an in-group optimal memory strategy, a forgetting and replenishment strategy, and a dream sharing strategy to update the variables to be optimized; the corresponding update mechanism and parameter settings can adopt the formulas in the aforementioned invention. Specifically, DOA retains the historical optimal solution within the group through the memory strategy, achieves a balance between global search and local convergence through the forgetting and replenishment strategy, and enhances the ability to escape local optima through the dream sharing strategy.

[0080] During the development phase, the number of iterations was [number missing]. to DOA no longer groups data but instead focuses on local optimization around the globally best historical individual, further improving optimization accuracy. The preferred parameter settings for DOA are: ; The number of forgetting dimensions can be set according to the problem dimensions. and Parameters used in the exploration phase Adjust the execution ratio of the "forgetting and replenishment strategy" and the "dream sharing strategy" to optimize the latter. .

[0081] To further verify the sensor's six-dimensional force measurement performance under actual working conditions, a six-dimensional sequential loading calibration experiment can be conducted after sample processing, FBG bonding, and optical path connection are completed. Specifically, this can be achieved by... , , , , , The six components are loaded individually, with multiple loading points set for each component. Wavelength data is collected after the system has stabilized at each loading point for a period of time. Preferably, 100 sets of data are sampled at each loading point and the average value is taken to reduce the influence of noise and improve calibration stability.

[0082] In subsequent physical verification, the following two types of results can be compared: (1) The error between the output of the linear decoupling model and the standard load; (2) Error between DOA-ELM nonlinear decoupling output and standard load.

[0083] Evaluation indicators can include root mean square error, mean absolute error, maximum absolute error, or correlation coefficient.

[0084] In summary, this embodiment provides a complete implementation path for the planar layered integrated FBG six-dimensional force sensor of the present invention, covering structural design, material selection, sensor assembly, simulation implementation, linear decoupling, DOA-ELM nonlinear decoupling, and subsequent physical calibration. Those skilled in the art can make equivalent substitutions or improvements to the structural dimensions, material parameters, load range, and decoupling algorithm parameters without departing from the overall concept of the present invention; all such substitutions and improvements should fall within the protection scope of the present invention.

[0085] It is understood that the same or similar parts in the above embodiments can be referred to each other, and the contents not described in detail in some embodiments can be referred to the same or similar contents in other embodiments.

[0086] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A planar layered integrated FBG six-dimensional force sensor, characterized in that, include: The enclosure, elastomer, multiple FBG sensitive elements, multiple optical fibers, and a six-dimensional force decoupling module; The elastomer is a planar symmetrical structure integrally machined and installed inside the outer shell. It includes a load-bearing boss at the center, four modified Maltese cross beams, four I-beam elastic beams, and four fixing blocks. Each of the four sides of the load-bearing boss is connected to an I-beam elastic beam via a modified Maltese cross beam, and the four connected components are orthogonally symmetrically distributed. An XY coordinate system is constructed based on the plane containing the four orthogonally symmetrically distributed I-beam elastic beams. The projection of each I-beam elastic beam onto the XOY plane is I-shaped, comprising two oppositely arranged transverse beams and a connecting section. The intermediate connecting part of each horizontal beam; and the I-shaped elastic beam is a hollow structure, specifically, the two oppositely arranged horizontal beams and the intermediate connecting part form an I-shaped hollow cavity; the two sides of the intermediate connecting part form oppositely arranged open concave cavities, so that each I-shaped elastic beam has two grooves, and the fixing block is connected to the bottom of the groove of the adjacent I-shaped elastic beam through the horizontal force sensing elastic beam; the improved Maltese cross beam includes an intermediate beam that plays a connecting role, and horizontal beams and vertical beams that are orthogonal to each other on both sides of the intermediate beam; the fixing block is provided with a groove for optical fiber winding at the contact point with the outer shell; Multiple FBG sensing elements are attached and fixed to the strain concentration area of ​​each functional beam and arranged in an orthogonal symmetrical differential manner to form six sets of differential measurement channels; each FBG sensing element transmits optical signals to the six-dimensional force decoupling module through an independent optical fiber with lead-out holes; the functional beams include a horizontal force-sensing elastic beam, a horizontal beam and a vertical beam in the improved Maltese cross beam; The input end of the six-dimensional force decoupling module is connected to the optical signal output end of each optical fiber to receive and process optical signals and output a six-dimensional force and torque load vector. There are a total of 11 FBG sensitive elements: An XY coordinate system is constructed based on the plane of four orthogonally symmetrically distributed I-beams. In each of the four I-beams, a horizontal force-sensing elastic beam is selected between the bottom of its groove and the fixed block, with the coordinate axes of the horizontal force-sensing elastic beams on opposite sides facing opposite directions. One FBG sensitive element is attached to the strain concentration area of ​​each functional beam, for a total of four. In the four improved Maltese cross beams, two have FBG sensitive elements attached to the strain concentration areas of both the horizontal and vertical beams, for a total of four. One horizontal beam has FBG sensitive elements attached to both its upper and lower surfaces, for a total of two. The remaining improved Maltese cross beam has an FBG sensitive element attached only to the upper surface of its horizontal beam, for a total of one. The sensor employs a differential measurement method to construct a six-dimensional differential wavelength output vector; at any loading moment, the first... The wavelength offset of each fiber Bragg grating is defined as the difference between its real-time center wavelength and its initial center wavelength, and six differential wavelength outputs are constructed according to the following pairing relationship: ； ； ； ； ； ； in, , , These represent the three force components respectively. , , Differential wavelength output corresponding to the measurement channel; , , They represent circumference respectively. , , Axial torque components , , Differential wavelength output corresponding to the measurement channel; For the number The change in the center wavelength of the fiber Bragg grating sensing unit. The preset correspondences are as follows: subscripts 7 and 8 correspond to... Channel, subscripts 9 and 10 correspond Channel, subscripts 3 and 11 correspond Channel, subscripts 3 and 4 correspond Channels, subscripts 1 and 2 correspond Channel, subscripts 5 and 6 correspond aisle; Based on this, a six-dimensional differential wavelength output vector is constructed. : ； The six-dimensional differential wavelength output vector serves as the input feature quantity for six-dimensional force decoupling; In the differential measurement method, the temperature drift terms of two fiber Bragg gratings symmetrically arranged in the same temperature field cancel each other out, thereby obtaining a differential wavelength output related to the corresponding load component and realizing temperature self-compensation.

2. The sensor according to claim 1, characterized in that, Within the overall elastic operating range of the sensor structure, the six-dimensional differential wavelength output vector and the six-dimensional force / torque load vector satisfy a linear mapping relationship: ； in, This is a six-dimensional differential wavelength output vector; The decoupling matrix to be calibrated is... This is a six-dimensional force / moment load vector; the six-dimensional force / moment load vector is solved using the least squares method as follows: ； in, For the estimated six-dimensional force / torque load vector, The minimum norm of the decoupling matrix is ​​the generalized inverse. Among them, the decoupling matrix to be calibrated It was obtained through a standard load calibrating experiment, that is, by performing least squares fitting on each group of known standard loads and the corresponding differential wavelength output data.

3. The sensor according to claim 1, characterized in that, The six-dimensional force decoupling module adopts the Extreme Learning Machine (DOA-ELM) model optimized by the dream optimization algorithm. The DOA-ELM model takes a six-dimensional differential wavelength output vector as input and a six-dimensional force / torque load vector as output. The dream optimization algorithm is used to optimize the connection weights from the input layer to the hidden layer and the hidden layer bias of the Extreme Learning Machine. The output weights from the hidden layer to the output layer of the Extreme Learning Machine are obtained by Moore-Penrose generalized inverse solving.

4. The sensor according to claim 3, characterized in that, The hidden output matrix of the Extreme Learning Machine satisfies: ； in, For a six-dimensional differential wavelength input sample matrix, This is the connection weight matrix from the input layer to the hidden layer. This is the hidden layer bias vector. For activation function, This is the hidden layer output matrix; The output weights from the hidden layer to the output layer of the Extreme Learning Machine satisfy the following: ； in, Hidden layer output matrix Moore-Penrose generalized inverse, Output sample matrix for the corresponding six-dimensional force / torque load. This is the output layer weight matrix.

5. A planar layered integrated FBG six-dimensional force measurement method, wherein the measurement method is based on the sensor described in any one of claims 1-4, characterized in that, include: Obtain the initial center wavelength of each FBG under no external load condition and the real-time center wavelength during the loading process, and calculate the wavelength drift of each FBG. Based on the constructed FBG wavelength drift-strain-temperature coupling model and sensitive beam load-strain response model, an orthogonal symmetrical differential arrangement is adopted for the same load component to obtain differential output and map it to differential wavelength; Based on the preset differential pairing relationship, a six-dimensional differential wavelength output vector is obtained by differential combination; The six-dimensional differential wavelength output vector is input into the trained DOA-ELM model to predict and output the final six-dimensional force and torque load vector.

6. The method according to claim 5, characterized in that, The FBG wavelength drift-strain-temperature coupling model is as follows: ； in, For the first The wavelength shift of a fiber Bragg grating; For the first The center wavelength of a fiber Bragg grating; The photoelastic coefficient; For the first Axial strain at the location of each fiber Bragg grating; The coefficient of thermal expansion of optical fiber. Thermo-optic coefficient, This refers to the change in temperature. The specific load-strain response model for sensitive beams is as follows: External load When applied, the axial strain at the sensitive location of the horizontal force-sensing elastic beam is measured. for: ； in, The maximum bending moment at the sensitive section; The moment of inertia of the cross section; Surface distance; The elastic modulus of the elastomer material; The bending moment correction factor for a horizontally force-sensing elastic beam; For along Load distribution coefficient among force-sensing elastic beams in multiple horizontal directions under directional external loads; To act on the overall structure of the sensor External input force component in the direction; The equivalent force arm length of the horizontal force-sensing elastic beam from the force transmission position to the sensitive section position; The cross-sectional thickness of the elastic beam that senses horizontal force in the plane corresponding to the bending direction; The cross-sectional width corresponding to the bending direction in the plane of the elastic beam that senses horizontal force; External load The situation and Same type, make The load distribution factor in the direction is Then external load When applied, the axial strain at the sensitive location of the horizontal force-sensing elastic beam is measured. for: ； axial force Axial strain at the sensitive location of the horizontal beam during operation for: ； in, Equivalent force borne by a single horizontal beam This is the effective length of the horizontal beam; The width of the horizontal beam; The thickness of the horizontal beam; This is the sharing factor; This indicates the number of the four horizontal beams; Around Axial torque Equivalent to the lever arm A pair of opposing axial forces at the location Strain at sensitive location of horizontal beam during operation for: ； in, To bypass Axial torque The contribution factor during operation; Around Axial torque Strain at sensitive location of horizontal beam during operation and same type; Around Shaft torque Equivalent to acting on radius Tangential force at the point Strain at sensitive locations of the vertical beam during operation satisfy: in, To bypass Shaft torque The factor of contribution when it is applied.

7. The method according to claim 5, characterized in that, The DOA-ELM model is obtained by training an extreme learning machine DOA-ELM based on a dream optimization algorithm with a six-dimensional differential wavelength output vector as input and a six-dimensional force / torque load vector as output. The training process of the DOA-ELM model includes: encoding the input layer weight matrix and hidden layer bias vector of the extreme learning machine as variables to be optimized, using the prediction error of the validation set as the fitness function, and using the dream optimization algorithm to find the optimal variables; after obtaining the optimal input layer weight matrix and the optimal hidden layer bias, constructing the hidden layer output matrix and solving the output layer weights to establish a nonlinear mapping model from six-dimensional differential wavelength features to six-dimensional force / torque output, until the prediction error meets the preset conditions, thus obtaining the trained DOA-ELM model.

8. The method according to claim 7, characterized in that, During training, the input layer weight matrix of the extreme learning machine is... With hidden layer bias vector The variables to be optimized are encoded, and the validation set prediction error is used as the fitness function. The model initialization parameters are optimized through a combination of global search and local exploitation, specifically including: During the initialization phase of DOA, the population is randomly generated. ； in, For the first Individual, and These are the lower and upper bounds of the search space, respectively. For population size, for Random vectors on; During the exploration phase, the number of iterations ranges from [number missing]. to DOA first divides the population based on differences in memory ability, and then allows individuals within a group to inherit the position of the group's historical best individual, that is: ； in, For the first Group 1 in The optimal individual in the next iteration. For the first The individual in the first The position vector at the next iteration. The threshold number of iterations used to delineate the exploration and development phases is used to divide the algorithm into the global search phase and the local development phase. Its value satisfies the following: ； Subsequently, in a randomly selected forgetting dimension The above implements a forgetting and replenishment strategy, and its update formula is: ； in, For the first The value of the j-th dimension component of each individual at the (t+1)-th iteration. Let be the component value in the j-th dimension of the optimal individual in the q-th group during the t-th iteration. Let be the lower bound of the j-th dimension of the search space. Let be the upper bound of the j-th dimension of the search space. For the first The forgetting dimension of the group For a randomly selected set of forgetting dimensions, This represents the maximum number of iterations. During the exploration phase, DOA also executes a dream-sharing strategy in parallel, that is, for any forgetting dimension... Its update formula is: ； in, From A natural number randomly selected from the list; During the development phase, the number of iterations ranges from [number missing]. to DOA no longer groups individuals; instead, it updates all individuals around the globally best historical individual. Its memory strategy is as follows: ； The corresponding forgetting and replenishment strategy update formula is: ； in, This represents the number of forgetting dimensions during the development phase. The DOA parameter settings satisfy the following: ； ； ； in, Let the dimension of the problem to be optimized be the dimension of the individual position vector. A function for generating random integers. It is a rounding function; And through preset parameter values Control the execution ratio of forgetting and replenishment strategies and dream sharing strategies during the exploration phase; when random numbers If the forgetting and replenishment strategy is applied, the dream sharing strategy is applied otherwise. For boundary handling, when optimizing the problem dimension When the value is less than or equal to the boundary preset value, a random backfill method is used: when When the value exceeds the preset boundary value, a reassignment method similar to the dream-sharing strategy used during the development phase is adopted: 。

Images