EP4548061A1

EP4548061A1 - Multiple-axis force sensor

Info

Publication number: EP4548061A1
Application number: EP23738087.8A
Authority: EP
Inventors: Brian SHEIL; Jack TEMPLEMAN
Original assignee: Oxford University Innovation Ltd
Current assignee: Oxford University Innovation Ltd
Priority date: 2022-06-30
Filing date: 2023-06-23
Publication date: 2025-05-07
Also published as: WO2024003532A1; GB202209626D0

Abstract

A method of assembling a multiple-axis force sensor (2) comprises mounting a plurality of pivot points (18a, 18b, 18c) at respective locations on the surface of a sensor structure body (4). An optical fibre (6), comprising a plurality of strain sensors (8), is wound around the sensor structure body (4) in a serpentine path (10a, 10b) that connects the pivot points (18a, 18b, 18c). The method further includes fixing the optical fire (6) to the surface of the sensor structure body (4).

Description

Multiple-axis Force Sensor

This invention relates to a multiple-axis force sensor and methods of designing and assembling such a sensor.

A force sensor or load cell is a type of sensor that provides a measurement of forces and/or moments that are applied to it. Multiple-axis force sensors are capable of measuring force and moment components in multiple spatial directions. They have been widely used in industrial and research disciplines such as robotics, manufacturing, aerospace, medical applications and civil engineering.

Conventional multiple-axis force sensors comprise a number of strain-sensitive transducers mounted on a compliant structure, which is designed to deform elastically under a strain field induced by an applied load. The relationship between the strain transducer outputs and the applied forces is typically obtained by calibration, from which the applied forces can be predicted for a set of measured strain outputs in practice. Both the sensor structure and strain transducers require careful consideration to achieve appropriate measurement range, sensitivity, isotropy and stiffness and to minimise undesirable coupling (i.e. cross-talk) between the outputs.

The vast majority of existing commercial load cells use electrical strain measurement techniques, e.g. using electrical resistance strain gauges. Strain gauges are a well-proven technology but have several limitations. First, the use of strain gauges relies on accurate measurement of very small changes in resistance, which means they are vulnerable to electromagnetic interference. Strain gauges are also susceptible to permanent damage upon exposure to water, thereby limiting their use in many applications. Furthermore, although individual gauges can be placed accurately, the instrumentations process (involving bonding multiple gauges to a sensor structure and soldering their electrical connections) is slow and labour- intensive.

The use of optical strain measurement can partly overcome some of the limitations of electrical systems. Fibre Bragg gratings (FBGs) have become the most popular optical strain measurement method for discrete sensor applications. An FBG is an optical reflector created within the core of an optical fibre. When illuminated with a broadband light source, the FBG reflects a narrow band of light, referred to as a “peak”. FBGs can be exploited in sensor applications by bonding the optical fibre to a sensor structure and measuring the peak wavelength, which is shifted under the action of mechanical strain and/or temperature change.

Although FBGs offer significant advantages over electrical strain gauges, the process of bonding FBGs to sensor structures is complex, time-consuming and expensive. A plurality of FBGs are typically multiplexed in a single optical fibre, and a great deal of care must be taken during the assembly process to ensure that the optical fibre carrying the FBGs is not damaged, and that the FBGs are positioned and oriented correctly, in order to provide the required accuracy in the strain measurements. Furthermore, FBGs themselves are expensive in comparison to electrical strain gauges.

It is therefore an object of the present invention to provide an improved multipleaxis force sensor and methods for designing and assembling a multiple-axis force sensor that address the above-mentioned drawbacks.

When viewed from a first aspect, the invention provides a method of assembling a multiple-axis force sensor comprising: mounting a plurality of pivot points at respective locations on the surface of a sensor structure body; winding an optical fibre, comprising a plurality of strain sensors, in a serpentine path around the sensor structure body that connects the plurality of pivot points; and fixing the optical fibre to the surface of the sensor structure body.

Thus, the assembly method of the present invention comprises winding an optical fibre, comprising a plurality of strain sensors, between a plurality of pivot points that are mounted on the surface of a sensor structure body. Doing so results in the formation of a serpentine path for the optical fibre, as a result of which the strain sensors contained within the optical fibre can be arranged at different positions and orientations across the surface of the sensor structure body. The optical fibre is then fixed to the surface of the sensor structure body so as to set the respective positions of each of the plurality of strain sensors.

It will be appreciated that the use of pivot points mounted on the surface of the sensors structure body can greatly improve the ease and repeatability with which the multiple-axis force sensor can be assembled, as each of the apex portions of the serpentine path of the optical fibre can be held in place by a respective pivot point while the rest of the optical fibre is wound around the sensor structure body. In particular, the assembly of the force sensor by hand can be significantly simplified and without compromising on the accuracy of the sensor positioning. This means that assembly of the force sensor may not require a high level of skill to perform, and may be possible to perform more quickly. This can help to reduce the cost of manufacture.

A naive approach to mounting the optical fibre on the surface of the sensor structure body would be to wrap the optical fibre around the entire sensor structure body, using only the geometry of the sensor structure body to facilitate changes in direction in the optical fibre path. However, this is often inefficient and cumbersome, and can substantially reduce the range of strain sensor locations and orientations that can be implemented. Furthermore, this approach can limit the repeatability of the process.

In some embodiments, the force sensor comprises a cavity. Preferably the cavity is shaped to receive an insert. Preferably the cavity extends from an exterior surface of the sensor structure body. This can allow the insert to be inserted and removed from the sensor structure body after the sensor structure body has been manufactured, without requiring disassembly or destruction of the sensor structure body itself. As a result, a single insert can be reused to assemble a plurality of force sensors, thereby improving the efficiency of the assembly process, reducing waste, and improving repeatability.

Preferably the cavity has a constant cross-section. This can help to facilitate the insertion of the insert during assembly of the multiple-axis force sensor. The cavity may comprise a longitudinal bore. In some embodiments, the cavity is annular, e.g. extending around the perimeter of the sensor structure body. The sensor body may be made from any suitable or desired material. The sensor structure body is preferably elastically deformable. The sensor structure body preferably has a substantially prismatic shape. The sensor structure body may comprise a wall between the surface of the sensor structure body on which the optical fibre is wound and the cavity.

Preferably the width of the wall is between 0.1 mm and 2 mm, e.g. between 0.3 mm and 1 mm, e.g. approximately 0.5 mm. The Applicant has identified that this is a suitable thickness for providing strain measurements that are accurate within a useful range and with a desirable sensitivity (e.g. when the force sensor has a columnar construction). In some embodiments, e.g. when the force sensor has an E-type construction, the thickness of the wall may be approximately 1.5 mm.

Preferably the method comprises positioning an insert within the cavity. In preferred embodiments, the insert comprises a plurality of locating points for indicating the respective locations for the pivot points on the surface of the sensor structure body. The insert may be any suitable or desired material. In preferred embodiments, the insert is not magnetic. In some embodiments, the insert is made from plastics. This can make it easier for pivot magnets to be positioned on the surface of the sensor structure body at the desired location, i.e. the location of the respective locating magnets.

The locating points may provide any suitable or desired indication of the intended location of a pivot point on the surface of the sensor structure body. Preferably, each of the locating points comprises a respective locating magnet. Each locating magnet may be arranged on an (e.g. outer) surface of the insert. Preferably, each of the locating magnets is fixedly mounted to (e.g. the outer surface of) the insert. The locating magnets may be neodymium magnets. The locating magnets may be welded to the insert. In some embodiments, each of the locating magnets is mounted to the insert by an adhesive. The insert is preferably shaped such that, when the insert is positioned within the cavity, the locating point is adjacent the wall of the sensor structure body. The pivot points may be manufactured by CNC machining, 3D printing or stamping. Preferably the method comprises using the plurality of locating points to mount the plurality of pivot points at the respective locations on the surface of the sensor structure body. The pivot points may be mounted to the sensor by a clamp or jig. The pivot points may be mounted to the sensor by means of an (e.g. temporary) adhesive.

In preferred embodiments, each of the pivot points comprises a pivot magnet. Preferably mounting the plurality of pivot points comprises positioning each of the pivot magnets on the surface of the sensor structure body adjacent to a respective locating magnet, such that the pivot magnet is held in position by the magnetic interaction between the pivot magnet and the locating magnet.

The use of magnets in this way can allow an assembly worker to easily position each of the pivot points at their respective desired location. Furthermore, as the pivot points can be held in position by the force of the magnets alone, the fixation of the pivot points to the surface of the sensor structure body can be temporary, e.g. for the purpose of the sensor assembly alone. This helps to provide a greater flexibility in the design of the force sensor, as the geometry of the sensor structure body can be designed without protrusions or similar features for receiving a portion of the optical fibre during assembly.

Each of the plurality of pivot points preferably comprises a retaining portion for receiving an optical fibre. The retaining portion may comprise a groove or a notch. In some embodiments, the retaining portion is shaped such that, when the pivot point is mounted on the surface of the sensor structure body, the retaining portion and the surface of the sensor structure body together define a groove or notch for receiving an optical fibre. In some embodiments, each of the pivot points comprises a hook for receiving an optical fibre.

The retaining portion is preferably arcuate. The retaining portion preferably has a radius of curvature that is greater than or equal to the minimum bend radius of the optical fibre, i.e. the minimum radius of the curve into which an optical fibre can safely be bent (i.e. without snapping or otherwise being damaged). It is desirable to avoid damaging the optical fibre, as this can cause a loss of signal from the strain sensors.

Preferably the radius of curvature of the retaining portion is at least 2 mm, e.g. at least 3 mm, e.g. at least 6 mm. By shaping the retaining portion of the pivot points to have a radius of curvature equal to or greater than the minimum bend radius of the optical fibre, the optical fibre can be more easily wound around the surface of the sensor structure body without damaging the optical fibre, as the retaining portion acts to enforce a bend radius that is at least equal to the minimum bend radius of the optical fibre.

Each of the pivot points preferably comprises a mounting surface arranged to abut the surface of the sensor structure body when the pivot point is mounted on the sensor structure body. In some embodiments, the mounting surface is shaped to conform to the surface of the sensor structure body. Each of the pivot points may comprise a deformable (e.g. rubber) sheet that defines the mounting surface. This can allow the pivot points to be more easily mounted on the surface of the sensor structure body. In some embodiments, the mounting surface is curved.

Preferably the serpentine path comprises a plurality of straight portions extending between respective pairs of pivot points. The serpentine path may comprise a plurality of apex (curved) portions. Preferably each of the apex portions extends around (e.g. is formed around) a respective pivot point. It will be appreciated that, in embodiments in which the engagement surfaces of the pivot points have a radius of curvature that is equal to or greater than the minimum bend radius of the optical fibre, an assembly worker can easily wind the optical fibre in the serpentine path such that each of the apex portions of the serpentine path has a radius of curvature that is equal to or greater than the minimum bend radius of the optical fibre. This can help to ensure that the optical fibre is not damaged during assembly of the force sensor.

Preferably, each of the plurality of strain sensors is arranged within a respective straight portion of the path. This can help to ensure that the strain sensors provide an accurate strain measurement. This can also help to ensure that the strain sensors are precisely mounted to the surface of the sensor structure body. The strain sensors are preferably optical strain sensors. The strain sensors may be Fabry-Perot interferometers. The strain sensors may be Rayleigh- or Brillouin- scattering based strain sensors. Preferably, the strain sensors are Fibre Bragg Grating (FBG) sensors. FBG sensors have high sensitivity, immunity to electromagnetic interference and water damage, and can be multiplexed in large numbers along a single fibre, which helps to simplify sensor connectivity.

Preferably the optical fibre comprises at least seven strain sensors. Preferably the method comprises winding the optical fibre around the sensor structure body such that none of the strain sensors overlap. In some embodiments, the method comprises winding the optical fibre around the sensor structure body such that the optical fibre does not overlap.

Winding the optical fibre around the sensor structure body may comprise applying a pre-tension to the optical fibre while the optical fibre is being wound around the body. This can help to ensure that the straight portions of the path of the optical fibre are formed, rather than allowing the optical fibre to be continuously curved, e.g. owing to the stiffness of the optical fibre, which can reduce the accuracy of the strain sensors.

In some embodiments, fixing the optical fibre to the surface of the sensor structure body comprises applying an adhesive to at least a portion of the optical fibre. The method may comprise applying an adhesive to the entire length of the optical fibre. Preferably, the method comprises applying adhesive to one or more of the strain sensors of the optical fibre.

In some embodiments, the method comprises applying adhesive only to one or more strain sensors of the optical fibre, i.e. not applying adhesive to portions of the optical fibre not containing a strain sensor. By fixing the optical fibre to the sensor structure body at the locations of the strain sensors alone, the efficiency of the assembly process can be increased whilst helping to ensure that the strain sensors are accurately positioned. Preferably the method comprises removing the pivot points from the surface of the sensor structure body after the adhesive has cured. A multiple-axis force sensor comprising an optical fibre mounted in this way is considered to be novel and inventive in its own right. Thus, when viewed from a second aspect, the invention provides a multiple-axis force sensor comprising: a sensor structure body; and an optical fibre, comprising a plurality of strain sensors, extending around the sensor structure body in a serpentine path; wherein: the optical fibre is mounted such that the optical fibre comprises a plurality of apex portions and a plurality of straight portions, wherein each of the straight portions connects a respective pair of apex portions; and the straight portions of the optical fibre are fixedly mounted to the sensor structure body but the plurality of apex portions are moveable relative to the sensor structure body.

The Applicant has identified that this arrangement can significantly improve the ease with which the multiple-axis force sensor can be assembled, without limiting the performance of the sensor.

The strain sensors are preferably optical strain sensors, e.g. FBGs. The multipleaxis force sensor may comprise a light source for introducing light into the optical fibre. The light source is preferably a broadband light source. The multiple-axis force sensor may comprise a photodetector for detecting one or more wavelength(s) of light reflected by one or more of the strain sensors.

Preferably the assembly method of the present invention further comprises removing the insert from the cavity after fixing the optical fibre to the surface of the sensor structure body. Preferably removing the insert from the cavity does not involve destruction of part of the insert and/or part of the sensor structure body. It will be appreciated that a single insert can be used to assemble a plurality of force sensors, as the assembly method does not require destruction of the insert.

Furthermore, as the insert is separable from the force sensor, the design of the force sensor is not required to include features used solely for the assembly of the force sensor, such as pivot points that are integral with the sensor structure body. Thus, preferably the surface of the sensor structure body (e.g. at least the surface of the sensor structure body adjacent each of the apex portions of the optical fibre when the optical fibre is fixed to the surface of the sensor structure body) is devoid of protrusions (i.e. the optical fibre is not wrapped around protrusions on the structure surface).

In some embodiments, the force sensor has a columnar structure. Such a structure is typically very stiff in the longitudinal axis of the sensor, meaning that a force sensor of this type can be useful in applications with high axial forces but lower non-axial (e.g. shear) forces. In such embodiments, the force sensor may comprise an upper flange and a lower flange, connected by a central cylindrical column.

Preferably the cavity is defined within the central cylindrical column. Preferably the cavity extends longitudinally through the force sensor from the upper flange to the lower flange, through the central cylindrical column. Preferably the cavity comprises a longitudinal bore. The length of the column may be between 50 mm and 100 mm, e.g. between 60 mm and 80 mm, e.g. approximately 70 mm. The (outer) diameter of the column may be between 30 mm and 60 mm, e.g. between 40 mm and 50 mm, e.g. approximately 55 mm.

In some embodiments, the force sensor has an E-type structure. A force sensor with an E-type structure is typically more flexible in the axial direction than a columnar force sensor, so can be suitable for low axial loads for which a greater sensitivity is required. In such embodiments the force sensor may comprise an upper plate and a lower plate, wherein the cavity is defined between the upper plate and the lower plate. The upper plate and the lower plate may be connected by a (e.g. central) column. In such embodiments, the cavity may be annular, e.g. extending between the upper plate and the lower plate around the (e.g. central) column. The planes of the upper plate and the lower plate may be parallel. The upper plate and/or the lower plate may comprise a circular cross-section.

The upper plate and/or the lower plate may define a (e.g. cylindrical) recess. The cylindrical recess may be provided centrally in the upper and/or lower plate so as to define an annular flange extending around the perimeter of the recess. The force sensor may comprise an upper flange, provided on the upper plate, and/or a lower flange, provided on the lower plate. It will be appreciated that the provision of a flange to any of the force sensors described herein can facilitate the process of mounting the force sensor to a test structure in order to obtain strain measurements.

In some embodiments, the method of assembling the multiple-axis force sensor comprises mounting the plurality of pivot points at respective locations on the surface of the recess of the upper plate and/or of the lower plate. The (e.g. central) column may comprise a bore. The bore may connect the recess of the upper plate with the recess of the lower plate.

In some embodiments, the method of assembling the multiple-axis force sensor further comprises passing the optical fibre through the bore of the (e.g. central) column. This can allow a single optical fibre to be mounted on the sensor structure body such that strain sensors are located on both the upper plate and the lower plate.

The diameter of the upper and/or lower plate may be between 70 mm and 150 mm, e.g. between 90 and 120 mm, e.g. approximately 108 mm. The diameter of the recess of the upper and/or lower plate bay be between 50 mm and 120 mm, e.g. between 70 and 100 mm, e.g. approximately 80 mm. The length of the E-type force sensor may be between 20 mm and 50 mm, e.g. between 30 mm and 40 mm, e.g. approximately 36 mm. The height of the cavity (e.g. the separation between the upper plate and the lower plate and, e.g. the height of the insert) may be between 2 mm and 10 mm, e.g. between 4 mm and 8 mm, e.g. approximately 6 mm.

In some embodiments, the force sensor has a box-type (e.g. cuboidal) structure. The sensor structure body may have a substantially square cross-section. The force sensor may comprise a removable lid (e.g. a face plate). The removable lid may be arranged to cover the cavity of the sensor structure body.

It will be appreciated that a force sensor in accordance with the present invention may have any suitable or desired geometry. Different force sensor geometries may be appropriate for different use cases, e.g. requiring different overall measurement ranges, or different levels of sensitivity or precision. The geometry of the force sensor can be an important factor in determining the optimal or desired positions for the strain sensors, as different geometries may result in different deformations in the force sensor, thereby necessitating alternative strain sensor layouts. This means that the serpentine path of the optical fibre may be different depending on the geometry of the multiple-axis force sensor.

When viewed from a third aspect, the invention provides a (e.g. computer implemented) method of designing a multiple-axis force sensor comprising a sensor structure body and an optical fibre mounted on the sensor structure body, wherein the optical fibre comprises a plurality of strain sensors, the method comprising: receiving geometry of a sensor structure body; receiving parameters representative of: a target orientation of one or more strain sensors in an optical fibre when the optical fibre is mounted on a sensor structure body; a target location on a sensor structure body of one or more strain sensors in an optical fibre when the optical fibre is mounted on a sensor structure body; and a target minimum bend radius for the path of an optical fibre when the optical fibre is mounted on a sensor structure body; using the parameters to determine a set of apex locations on the sensor structure body such that, when an optical fibre is wrapped around the sensor structure body in a serpentine path such that each apex of the serpentine path is coincident with a respective apex location of the set of apex locations, the serpentine path has a minimum bend radius greater than the target minimum bend radius; and outputting the set of apex locations.

This aspect of the present invention provides a (e.g. computer implemented) method of designing a multiple-axis force sensor that may be particularly suited to assembly by means of the first aspect of the present invention. The design method disclosed herein generates a set of apex locations that can be used as the locations of respective pivot points for mounting an optical fibre to the surface of a sensor structure body. It will be appreciated that one or more (e.g. all) of the steps of the method of designing the multiple-axis force sensor, as well as the optional steps outlined herein, may be implemented on a (e.g. suitably programmed) computer. Furthermore, the design method of the present invention comprises determining the apex locations using parameters representative of a target location and orientation of one or more of the strain sensors, as well as a target minimum bend radius for the optical fibre. Thus, the set of apex locations can be generated to take account of both an optimal strain sensor layout and the physical constraints of the optical fibre.

In some embodiments, receiving geometry of the sensor structure body comprises receiving one or more of a length, radius and a wall thickness of the sensor structure body. Receiving geometry of the sensor structure body may comprise determining the geometry of the sensor structure body.

The method may comprise receiving a target load capacity for the multiple-axis force sensor. The target load capacity may comprise a required or desired capacity for each load component to be applied to the multiple-axis force sensor. Preferably the method comprises using the target load capacity to determine the geometry of the sensor structure body.

The method may comprise selecting a geometry of the sensor structure body from a set of predetermined geometries. The method may comprise receiving a material of the sensor structure body. The method may comprise receiving one or more dimensions (e.g. a length) of the one or more strain sensors.

In some embodiments, the method comprises receiving a parameter representative of a target number of strain sensors. The method may comprise receiving a parameter representative of a separation distance between a respective pair of strain sensors in an optical fibre along the length of an optical fibre.

In some embodiments, at least one of the target orientations is approximately 0°, 45° or 90° to the longitudinal axis of the force sensor. Preferably the method comprises determining the target orientation and/or the target location of the one or more strain sensors using an unconstrained optimisation process. The unconstrained optimisation process may be used to perform a preliminary assessment of possible layouts for the path of the optical fibre and/or the plurality of strain sensors. Preferably the unconstrained optimisation process comprises determining a respective performance metric for a plurality of candidate strain sensor layouts. Preferably each candidate strain sensor layout defines a plurality of respective locations on the sensor structure body for a predetermined number of strain sensors.

The predetermined number of strain sensors may be the minimum number of strain sensors that are required to be mounted on the sensor structure body in order for the multiple-axis force sensor to measure a desired set of force components and/or moments. The predetermined number of strain sensors may be six. It will be appreciated that six is the minimum number of strain sensors required in order for the force sensor to be able to determine the six degrees of freedom load components (F_x, F_y, F_z, M_x, M_y, M_z). The predetermined number of strain sensors may be seven. It will be appreciated that this is the minimum number of strain sensors required to allow the force sensor to additionally determine a temperature change AT.

The unconstrained optimisation process may comprise using an optimisation algorithm to determine an optimum candidate layout, from the plurality of candidate strain sensor layouts, that results in a maximum value of the performance metric. The optimisation algorithm may be a grid search algorithm, a genetic algorithm or Bayesian optimisation. The performance metric may be an R-squared coefficient.

In some embodiments, the method comprises determining, from the optimum candidate layout, a set of optimum apex locations. The method may comprise, after using the parameters to determine a set of apex locations, comparing the set of apex locations with the set of optimum apex locations. The method may comprise adjusting one or more of the apex locations of the set of apex locations based on said comparison.

Preferably, using the parameters to determine the set of apex locations comprises using a constrained optimisation process. The constraints of the optimisation process preferably comprise the target minimum bend radius for the path of the optical fibre, the (respective) target location of the one or more strain sensors, and the (respective) target orientation of the one or more strain sensors.

In some embodiments, the constraints of the optimisation process further comprise a target distribution of strain sensors across a surface area of the sensor structure body. The constraints may comprise a spacing of strain sensors along an optical fibre, i.e. a value representative of a length of optical fibre between two or more adjacent strain sensors. The constraints may further comprise a target sensitivity to a particular strain component, e.g. axial or circumferential. The optimisation process may comprise increasing the number of strain sensors that are positioned with the same orientation in order to increase the sensitivity in a direction parallel to this orientation.

In preferred embodiments, the method further comprises using the set of apex locations to determine a respective location and/or respective orientation for each of a plurality of strain sensors of an optical fibre when the optical fibre is mounted on the sensor structure body.

In some embodiments, the method further comprises generating a computer model of the force sensor.

In some embodiments, the modelled force sensor comprises a plurality of modelled strain sensors, each located at a respective location on the modelled force sensor. The modelled force sensor may comprise a modelled optical fibre mounted on the sensor structure body of the force sensor such that the optical fibre follows the serpentine path. The modelled optical fibre may extend between each of the set of apex locations on the surface of the modelled force sensor. However, in some embodiments, the optical fibre itself is not modelled. For example, the plurality of strain sensors may be modelled in isolation from an optical fibre.

The method preferably comprises extracting, from the computer model, an estimated strain measurement for each of the modelled strain sensors. These strain estimates can be used to indicate the measurements expected to be obtained by a respective strain sensor positioned at each of the plurality of locations on a (real- world) force sensor. The method preferably further comprises introducing a random change to the respective position and/or respective orientation of one or more of the plurality of strain sensors. This may comprise introducing a random change to the position of one or more of the modelled apex locations.

Preferably the method comprises modelling the performance of the force sensor to generate a performance metric, e.g. after introducing the random change. This may be repeated for a plurality of random changes, in order to determine a sensitivity of the force sensor to tolerances in the positioning of the strain sensors. In some embodiments, the method comprises using the performance metric to adjust one of more of the apex locations.

In some embodiments, the method comprises calculating a value representative of the variation in the respective orientations of the plurality of strain sensors. The value representative of the variation may be compared to a threshold variation. Preferably the method comprises, in response to determining that the value representative of the variation is less than the threshold variation, adjusting one or more of the apex locations. Preferably the one or more apex locations are adjusted so as to increase the value representative of the variation.

Once a multiple-axis force sensor according to an embodiment of the present invention has been assembled, a supervised machine learning model can be used to map the outputs of the strain sensors to applied load components. This supervised machine learning process may comprise applying a set of calibration loads, for which the values of the constituent load components are known, to the assembled force sensor and measuring the outputs of the strain sensors, i.e. the wavelengths (or wavelength shifts) detected by each of the sensors.

The process may further comprise applying one or more known calibration temperature changes to the force sensor. The set of calibration loads may be incremental. However, owing to the use of a supervised machine learning model, which does not require the mapping between the outputs of the strain sensors and the applied load components to be determined manually from individual gradients, the use of incremental calibration loads may not be a requirement. This can facilitate more flexible and/or automated means of calibrating the sensor, e.g. through the use of robotic manipulators.

The known load component values and the measured wavelengths (or wavelength shifts) may be input to the machine learning model as training outputs and training inputs respectively. The set of calibration temperature changes may (e.g. also) be input to the machine learning model as training outputs. The machine learning model is preferably configured to determine a relationship between the training inputs and the training outputs. Preferably determining the relationship comprises determining a set of coefficients/weights.

The training process may comprise applying a set of testing loads to the assembled force sensor, after the relationship has been determined. The set of testing loads can comprise more complex, combined load cases which may not form part of the set of calibration loads. This can be used to assess the performance of the assembled sensor and/or the machine learning model. The set of calibration loads and/or the set of testing loads may be generated by a computer model of the multiple-axis force sensor. The computer model of the multiple-axis force sensor may be based on Euler-Bernoulli beam theory or may be a 3D finite element model.

The trained machine learning model is preferably configured to determine, from a set of one or more measured wavelengths (or wavelength shifts), an estimated set of load components using the determined relationship between the training inputs and the training outputs, using the measured data from the multiple-axis force sensor.

Thus, when viewed from a further aspect, the invention provides a method of using a multiple-axis force sensor to determine an estimate of one or more load components and/or a temperature change applied to the multiple-axis force sensor, wherein the multiple-axis force sensor comprises: a sensor structure body; and an optical fibre, comprising a plurality of strain sensors, extending around the sensor structure body in a serpentine path; wherein: the optical fibre is mounted such that the optical fibre comprises a plurality of apex portions and a plurality of straight portions, wherein each of the straight portions connects a respective pair of apex portions; and the straight portions of the optical fibre are fixedly mounted to the sensor structure body but the plurality of apex portions are moveable relative to the sensor structure body; and wherein the method comprises: transmitting light from a light source into the optical fibre of the multiple-axis force sensor; detecting a wavelength of light reflected by the strain sensors; and using the wavelength of the reflected light to determine an estimate of one or more load components and/or a temperature change applied to the multiple-axis force sensor.

Preferably the method comprises transmitting light from a light source. The light source may be a light-emitting diode (LED). The light source is preferably a broadband light source. The method may comprise detecting the wavelength of the reflected light using a photodetector.

It will be appreciated that any aspect of the invention disclosed herein may (and preferably does) include one or more (e.g. all) of the disclosed optional and preferred features, as appropriate.

Furthermore, in embodiments, the multiple-axis force sensor (or a system comprising the multiple-axis force sensor) may comprise a processor for determining, from a set of one or more measured wavelengths (or wavelength shifts), an estimated set of load components using the determined relationship between the training inputs and the training outputs. The processor may be arranged to receive the wavelength(s) (or wavelength shift(s)) of the reflected light. The processor may be arranged to determine the wavelength shift(s) of the reflected light from the measured wavelength(s).

Certain preferred embodiments of the invention will now be described, by way of example only, with reference to the accompanying drawings in which: Figure 1 A shows a multiple-axis force sensor in accordance with an embodiment of the present invention;

Figure 1 B shows a schematic diagram of the optical components of the multiple-axis force sensor of Figure 1A;

Figure 2 shows a flowchart of a method of assembling a multiple-axis force sensor in accordance with an embodiment of the present invention;

Figures 3A to 3F show the process of assembling the multiple-axis force sensor of Figure 1A;

Figures 4A and 4B respectively show perspective and cross-sectional views of a multiple-axis force sensor in accordance with a further embodiment of the present invention;

Figures 4C to 4F show the process of assembling the multiple-axis force sensor of Figures 4A and 4B;

Figures 5A to 5C show the process of assembling a multiple-axis force sensor in accordance with another embodiment of the present invention;

Figure 6 shows a flowchart of a method of designing a multiple-axis force sensor in accordance with an embodiment of the present invention; and

Figures 7A to 7D show a process of selecting a layout of strain sensors for a cylindrical column sensor in accordance with an embodiment of the present invention.

Multiple-axis force sensors are used to measure force and moment components in multiple spatial directions. They are used widely in industrial and research disciplines such as robotics, manufacturing, aerospace, medical applications and civil engineering. Embodiments of a multiple-axis force sensor will now be described.

Figure 1 A shows a multiple-axis force sensor 2 in accordance with an embodiment of the present invention. The sensor 2 has a columnar structure comprising a cylindrical sensor body 4, an upper flange 3a and a lower flange 3b. The upper flange 3a is mounted at a first end of the cylindrical sensor body 4, and the lower flange 3b is mounted at a second (opposite) end of the cylindrical sensor body 4. The upper and lower flanges 3a, 3b extend substantially perpendicularly to the main cylindrical axis of the in cylindrical sensor body 4, each in an annulus around the respective ends of cylindrical sensor body 4. The force sensor 2 further comprises an optical fibre 6, mounted on an outer surface of the cylindrical sensor body 4. The optical fibre 6 comprises a plurality of Fibre Bragg Grating (FBG) strain sensors 8, four of which are visible in Figure 1A. The optical fibre 6 extends around an outer surface of the cylindrical sensor body 4 in a serpentine path comprising a plurality of curved (apex) portions 10a and a plurality of straight portions 10b. Each of the FBGs 8 is located on a straight portion 10b of the path of the optical fibre 6.

The sensor 2 comprises a cylindrical central bore 12, extending longitudinally through the upper flange 3a, the sensor body 4, and the lower flange 3b. The outer diameter of the cylindrical sensor body 4 is 55 mm, and the diameter of the central bore 12 is 54 mm, meaning that the thickness of the sensor body 4 between the outer surface of the sensor body and the inner surface of the bore 12 is 0.5 mm. The length of the cylindrical sensor body 4 (i.e. excluding the upper flange 3a and the lower flange 3b) is 70 mm.

Figure 1 B shows a schematic diagram of the optical components of the force sensor 2 of Figure 1A. In use, the optical fibre 6 is connected to a broadband light source 5 that is arranged to introduce light 7a into the optical fibre 6. The light 7a propagates through the optical fibre 6 and through each of the FBGs 8 connected in series. The FBGs 8 reflect a narrow band 7b of the incident light at a particular wavelength, while transmitting light at all other wavelengths. The wavelength of the reflected light 7b depends on the deformation of the optical fibre 6 resulting from strain in the force sensor 2, or depending on a change in temperature, at the locations of the FBGs 8. A photodetector 9 is arranged to detect the wavelength(s) of the reflected light.

The force sensor 2 is calibrated so that the outputs of the FBG strain sensors 8 (i.e. the detected wavelengths of reflected light) can be used to determine estimates of the load components applied to the force sensor 2.

In traditional sensor calibration, the individual linear relationships between strain transducer outputs and applied load components are determined manually. However, the use of a supervised machine learning model to map FBG 8 measurements to applied loads allows the entire mapping to be determined automatically, while taking into consideration the effects of temperature variation.

The sensor model takes an input vector of n strain measurements, w, which comprise the peak wavelength shifts of the FBGs 8 bonded to the sensor body 4, AA^: u = [AAi AA₂ AA₃ ... AA_n]^T (1)

The model returns a vector/ of m = 7 predicted sensor outputs comprising the full six DOF load components, as well as temperature change AT: f = [F_x F_y F_z M_x M_y M_z AT]⁷ (2)

In equation 2, F_z and M_z denote axial force and torsional moment respectively, F_x and F_y denote mutually orthogonal lateral forces, M_x and M_y denote mutually orthogonal lateral moments, and AT denotes a change in temperature. A multioutput linear regression (LR) model is used to predict the outputs in f from the measurements taken by the FBGs 8 mounted on the sensor body 4 using the relationship: f = tVu (3) where W is an m x n matrix of weights, or coefficients. As the FBG peak wavelength shifts in u are calculated from a predefined reference ‘zero’ state (measured after sensor initialisation), no intercept term is needed in this linear relationship.

The matrix of weights W is obtained automatically by training the LR model. This involves applying a plurality of calibration load cases to a sensor 2, and obtaining the peak wavelength shifts u measured by each of the FBGs 8. The peak wavelength shifts u are used as training inputs to the LR model, and the known load and temperature components f are used as the training outputs.

For the calibration load cases, each load component may be individually incremented in the conventional approach to allow the individual terms in W to be determined. However, this calibration process can be time consuming, and hence expensive, for multiple-axis sensors. The use of machine learning can relax the requirement for load cases to be applied in separate increments, thereby giving rise to the possibility of more flexible, automated means of calibrating multiple-axis sensors, e.g. using robotic manipulators.

In traditional sensors, temperature is not included in f and the predicted outputs are designed to be independent of temperature effects. This is typically achieved through the careful and painstaking placement of a large number of strain transducers on the sensor body 4 such that combining their signals produces an output that is temperature compensated.

However, in accordance with the method of the present invention, temperature effects are included in the matrix of weights IV by performing specific temperature calibration tests. The LR model is trained to identify the relationship between AT and the peak wavelength shifts in u directly. As a result, effective temperature compensation can be achieved without relying on strict positioning of individual strain transducers. This helps to simplify the instrumentation process by allowing the positions of the FBGs 8 to be altered to meet the physical restraints imposed by the optical fibre 6.

Furthermore, in comparison with using algebraic combinations, a significant reduction in the number of FBGs 8 can be achieved, e.g. to a minimum of seven (the number of outputs in f). This can have significant cost-saving implications.

A linear model is adopted under the assumption of there being a linear relationship between the FBG peak wavelengths and the applied loads. This reflects the expectation that the sensor structure will undergo linear elastic deformation and that the FBG strain dependence is known to be linear. Nevertheless, the proposed framework is not limited to linear models. It can be readily applied using alternative data-driven models, such as higher order polynomial-based models or even neural networks, to handle possible measurement non-linearities (e.g. non-linear sensor material behaviour). Figure 2 shows a flowchart of a method of assembling a multiple-axis force sensor

2 (e.g. as shown in Figure 1 A) in accordance with an embodiment of the present invention. The steps of the method are illustrated in Figures 3A to 3F.

Figure 3A shows, on the left-hand side of Figure 3A, a hollow cylindrical insert 14 for inserting into the central bore 12 of the sensor 2 of Figure 1 A. The cylindrical insert 14 comprises a plurality of locating magnets 16a, 16b, 16c fixedly mounted (e.g. glued) on the outer surface of the insert 14. Each of the locating magnets 16a, 16b, 16c is positioned at the location of a desired bend in the optical fibre 6 that is to be mounted to the outer surface of the cylindrical sensor body 4 of the sensor 2.

In a first step S100, the insert 14 is inserted into the central bore 12 of the sensor 2, as illustrated on the right-hand side of Figure 3A. The insert 14 is dimensioned such that each of the locating magnets 16a, 16b, 16c is arranged to abut the inner surface of the bore 12.

In a second step S102, a plurality of pivot magnets 18a, 18b, 18c are positioned on the outer surface of the cylindrical sensor body 4. Each of the pivot magnets 18a, 18b, 18c is positioned at the same axial and circumferential position as a respective locating magnet 16a, 16b, 16c. It will be appreciated that the interaction between the magnetic fields generated by each of the locating-pivot magnet pairs allows the pivot magnets 18a, 18b, 18c to be removably held in position on the outer surface of the sensor body 4. Figure 3B shows the co-location of the pivot magnets 18a, 18b, 18c, on the outer surface of the sensor body 4, and the respective locating magnets 16a, 16b, 16c on the insert 14, within the central bore 12.

In a third step S104, an optical fibre 6 is wound around the circumference of the outer surface of the sensor body 4, from one pivot magnet 18a to another 18b, 18c, so as to form a serpentine path comprising a plurality of curved portions 10a and a plurality of straight portions 10b. As can be seen in Figure 3C, the FBGs 8 are spaced along the length of the optical fibre 6 such that they are each arranged at a respective straight portion 10b of the serpentine path when the optical fibre 6 is wound around the pivot magnets 18a, 18b, 18c on the sensor body 4. The curved portions 10a of the serpentine path are located about the pivot magnets 18a, 18b, 18c. The provision of pivot magnets 18a, 18b, 18c to provide turning points for the optical fibre 6 in this way allows an assembly worker to position the optical fibre 6 and, thus, each of the FBGs 8 on the sensor body 4 accurately and straightforwardly, thereby greatly improving the ease and efficiency of the assembly process.

In a fourth step S106, shown in Figure 3D, adhesive 20 is applied to each of the FBGs 8, and to the adjacent straight portions 10b of the optical fibre 6, in order to fix the FBGs 8 in place.

In a fifth step S108, shown in Figure 3E, all of the pivot magnets 18a, 18b, 18c are removed from the outer surface of the sensor body 4. At this stage, the radii of curvature of the optical fibre 6 at each of the curved portions 10a of the serpentine path may increase slightly. However, it will be appreciated that the respective positions and orientations of the FBGs 8 will not change, as they are held in place by adhesive.

In a sixth step S110, shown in Figure 3F, the insert 14 is removed from the central bore 12 of the sensor 2.

Figures 4C to 4F illustrate the process for assembly of an E-type sensor 102, in accordance with a further embodiment of the present invention. The E-type sensor 102 is shown in Figures 4A and 4B. Reference is also made to the flowchart of Figure 2, the steps of which are equally applicable to the assembly of the E-type sensor 102.

Figures 4A and 4B show respectively a perspective view and a cross-sectional view of the sensor body 104 of an E-type sensor 102. The sensor body 104 comprises a circular upper plate 104a and a circular lower plate 104b, connected by a cylindrical central column 104c. The central column 104c defines a central bore 107. The upper plate 104a and the lower plate 104b are spaced apart by the central column 104c so as to define an annular space 105 therebetween. The upper plate 104a defines a central cylindrical recess 103a extending into the upper plate 104a from an upper surface of the upper plate 104a, so as to define a thin wall 113a between the recess 103a and the lower surface of the upper plate 104a. The lower plate 104b defines a central cylindrical recess 103b extending into the lower plate 104b from a lower surface of the lower plate 104b, so as to define a thin wall 113b between the recess 103b and the lower surface of the lower plate 104b.

An upper flange 111a is provided around the circumference of the central cylindrical recess 103a of the upper plate 104a (i.e. so as to define the recess 103a). A lower flange 111 b is provided around the circumference of the central cylindrical recess 103b of the lower plate 104b (i.e. so as to define the recess 103b). In use, the upper and lower flanges 111a, 111b, which are relatively thick in comparison to the thin walls 113a, 113b of the sensor 102, can be used to mount the sensor 102 securely to a test structure (not shown).

In the first step S100 of the process for assembling the E-type sensor 102 in accordance with an embodiment of the invention, as shown in Figure 4C, a circular insert 114 is inserted into the space 105 between the upper plate 104a and the lower plate 104b. The circular insert 114 comprises two substantially semi-circular pieces that are separable such that a first of the pieces can be inserted on a first side of the central column 104c, and a second of the pieces can be inserted on a second (opposite) side of the central column 104c.

The circular insert 114 comprises a plurality of locating magnets 116a, 116b, 116c fixedly mounted (e.g. glued) on both the upper and lower surfaces of the insert 114. Only the upper surface of the insert 114 is visible in Figure 4C. Each of the locating magnets 116a, 116b, 116c is positioned at the location of a desired bend in the optical fibre 106 that is to be mounted to the sensor 102.

In the second step S102, shown in Figure 4D, pivot magnets 118a, 118b, 118c are positioned on the surface of the sensor body 104. Each of the pivot magnets 118a, 118b, 118c is positioned at the same radial and circumferential position as a respective locating magnet 116a, 116b, 116c. As can be seen in Figure 4D, pivot magnets 118a, 118b, 118c are positioned within the central cylindrical recess 103a of the upper plate 104a. Although not visible in Figure 4D, pivot magnets are also positioned within the central cylindrical recess 103b of the lower plate 104b, at the same radial and circumferential positions as the locating magnets fixed to the lower surface of the insert 114.

In the third step S104, shown in Figure 4E, an optical fibre 106 is wound around each of the pivot magnets 118a, 118b, 118c, extending via the bore 107 in the central column 104c between the pivot magnets 118a, 118b, 118c on the upper plate 104a and the pivot magnets (not shown) on the lower plate 104b. The optical fibre 106 is wound between the pivot magnets 118a, 118b, 118c so as to form a serpentine path comprising a plurality of curved portions 110a and a plurality of straight portions 110b. A plurality of FBGs 108 are spaced along the length of the optical fibre 106 such that they are arranged at a respective straight portion 110b of the serpentine path when the optical fibre 106 is wound around the pivot magnets 118a, 118b, 118c on the sensor body 104.

The curved portions 110a of the serpentine path are located about the pivot magnets 118a, 118b, 118c.

In a fourth step S106, shown in Figure 4F, adhesive 120 is applied to each of the FBGs 108, and to the adjacent straight portions 110b of the optical fibre 106, in order to fix the FBGs 108 in place. In fifth and sixth steps S108, S110 the pivot magnets 18a, 18b, 18c and the insert 114 are removed from the sensor body 4.

Figures 5A to 5C show a box-type force sensor 202, in accordance with a further embodiment of the present invention. The box-type sensor 202 has a substantially square cross-section and comprises a removable lid 204a and a hollow body 204b defining a central bore 212 (shown in Figure 5B).

An optical fibre 206 comprising a plurality of FBGs (not shown) is mounted to the outer surface of the hollow body 204b in a serpentine path, as shown in Figure 5A. The optical fibre 206 is fixed to the hollow body 204b by adhesive 220.

In accordance with the assembly process described above, an insert 214 comprising a plurality of locating magnets (not shown) may be inserted into the central bore 212 of the sensor 202, after removing the lid 204a. As shown in Figure 5C, a plurality of pivot magnets 218a, 218b, 218c may then be positioned on the outer surface of the hollow body 204b. It will be appreciated that the pivot magnets 218a, 218b, 218c can be straightforwardly positioned at the same axial and circumferential positions as respective locating magnets, in a similar manner to that described above with reference to the columnar and E-type force sensors 2, 102. This allows the optical fibre 206 to be wound around the outer surface of the hollow body 204b.

Figure 6 shows a flowchart of a method of designing a multiple-axis force sensor in accordance with an embodiment of the present invention.

In a first step S200, the geometry of the sensor structure is selected. For example, the sensor structure may be columnar (as shown in Figure 1A), E-type (as shown in Figures 4A to 4F) or box-type (as shown in Figures 5A to 5C). One or more desired dimensions, or dimensions dictated by the sensor’s intended application, may also be selected at this stage and taken into consideration as constraints of the design process.

The optimised layout of FBGs 8 on the sensor body 4 is selected based on a compromise between several competing factors: a) Maximising sensitivity to the different load components; b) Minimising cross-talk between the components, in particular, the effect of temperature changes on the load predictions; c) Minimising the number of FBGs 8 used to reduce overall sensor cost; d) Ensuring the layout accounts for the physical constraints of the optical fibre 6 and is straightforward to implement in practice.

First, the sensor surface is parameterised to allow the layout of the FBGs 8 to be defined. For example, for the columnar sensor 2 shown in Figure 1A, this includes the number of FBGs 8 used and three positional parameters per FBG 8, i.e. the axial position, the circumferential position and the orientation angle.

By adopting the machine learning framework described above, layouts comprising only a small number of FBGs 8 can be sought, and these can be positioned on the sensor body 4 with much more freedom than with traditional approaches. However, there are nevertheless a number of physical constraints to be taken into account. For example, the sensor structure geometry, the FBG spacing along the optical fibre 6 and the number of FBGs 8 used are all highly coupled parameters that influence the choice of layout.

In order to select the layout of FBGs 8 on the surface of a sensor body 4, a ‘digital twin’ of the sensor body 4 is created. The purpose of this is to simulate the response of the sensor structure under a selection of applied load components and temperature changes. This is used to generate synthetic data representative of the surface strain on the sensor structure that the FBGs 8 would be exposed to in a real-world situation.

The digital twin of the sensor body may be used as the test sensor for the purposes of training the LR model. In other words, the digital twin may be used to model a set of calibration load cases f , with the modelled outputs of the FBGs 8 being used as the training inputs for the LR model.

The digital twin may also be used to generate more complex testing datasets, including complex combined load cases that are not typically included in training datasets. The testing datasets can be used to provide an indication of the performance of the trained LR model in operation, allowing different layouts and numbers of FBGs 8 to be explored and optimised.

During a preliminary layout assessment stage, the physical constraints of the optical fibre and the relative positioning of the FBGs 8 are neglected. In step S202 of the process, a numerically optimum layout of the minimum number (i.e. seven) FBGs 8 is identified using an optimisation algorithm, such as a grid search or genetic algorithm or Bayesian optimisation, to seek a layout which returns the maximum performance. For example, the digital twin of the sensor may be modified to model different sensor layouts, and its performance may be analysed using test data. This allows an optimum layout (e.g. one that provides the highest R² score) to be determined. Figures 7A to 7D illustrate the process of selecting a suitable layout of FBGs for a cylindrical column sensor. The circumferential position is indicated by the horizontal axis and the axial position is indicated by the vertical axis of each of Figures 7A to 7D. Figure 7A shows the results of an optimisation process to determine a set 308a of seven optimum FBG positions and orientations (shown in bold) on the cylindrical column sensor from a plurality of sets of seven candidate FBG locations 309.

In step S204, the digital twin is used to model a large number (e.g. more than 1000) of further layouts, comprising more than the minimum number of FBGs (e.g. greater than or equal to 7), and corresponding performance data is collected for each of the further layouts.

Figure 7B shows an example of a randomly generated redundant layout of a set 308b of n = 11 FBGs on the same cylindrical column sensor.

These modelling steps S202, S204 are used to inform the selection of a realistic layout, as they provide an indication of the maximum performance achievable, as well as the impact on performance of using a greater number of FBGs or of different FBG layouts. For example, these steps may provide information regarding a desirable distribution of FBGs around the perimeter of the sensor body, or regarding a set of particularly desirable orientations for one or more of the FBGs.

In a practical design stage of the process, a layout is sought that is practically implementable, taking into consideration the physical constraints of the optical fibre. In step S206, a crude candidate serpentine pattern 310 is identified based on the selected sensor structure geometry, the spacing of the FBGs 308c along the optical fibre 306, and any desirable characteristics identified during the preliminary layout assessment stage. An exemplary candidate pattern 310 of eight FBGs 308c is shown in Figure 7C. The selection of the candidate pattern may also be informed by physical prototyping and hands-on testing.

In step S208, computer-aided design (CAD) software is used to adjust the candidate pattern 310 so as to comply with the physical requirements of the optical fibre 306. In particular, a minimum bend radius in the path of the optical fibre 306 is introduced to ensure that the optical fibre 306 is not damaged when it is wound around each of the pivot magnets 318 during the assembly process. As shown in Figure 7D, tangency between all straight portions 310b and curved portions 310a of the path of the optical fibre 306 is also enforced, and the chainage along the geometry of the sensor body is calculated in order to track the positions of the FBGs 308c. This calculation may be performed using a computer program, and comprises calculating the lengths of each of the curved portions 310a and straight portions 310b and using the known spacing of the FBGs 308c along the optical fibre 306 to determine the respective positions and orientations of each of the FBGs 308c.

In the final analysis stage of the process, step S210, the optimised layout of the FBGs 8 is tested using the digital twin model. The performance of the layout is tested with different levels of artificial noise added to the peak wavelength shifts of the FBGs 8, as well as with random changes to the position and/or orientation of one or more of the FBGs 8. This helps to simulate typical positional accuracy that may be achieved in practice.

The practical design stage and the final analysis stages may be performed iteratively in order to find a suitable solution.

Claims

1. A method of assembling a multiple-axis force sensor comprising: mounting a plurality of pivot points at respective locations on the surface of a sensor structure body; winding an optical fibre, comprising a plurality of strain sensors, in a serpentine path around the sensor structure body that connects the plurality of pivot points; and fixing the optical fibre to the surface of the sensor structure body.

2. The method as claimed in claim 1, wherein the force sensor comprises a cavity, and wherein the method further comprises: positioning an insert within the cavity, the insert comprising a plurality of locating points for indicating the respective locations for the pivot points on the surface of the sensor structure body; and using the plurality of locating points to mount the plurality of pivot points at the respective locations on the surface of the sensor structure body.

3. The method as claimed in claim 2, wherein the force sensor has a columnar structure, and the cavity comprises a longitudinal bore.

4. The method as claimed in claim 2, wherein the force sensor has an E-type structure comprising an upper plate and a lower plate, wherein the upper plate and the lower plate are connected by a column and wherein the cavity is defined between the upper plate and the lower plate.

5. The method as claimed in claim 2, 3 or 4, wherein each of the locating points comprises a locating magnet fixedly mounted to the insert and each of the pivot points comprises a pivot magnet.

6. The method as claimed in claim 5, wherein mounting the plurality of pivot points comprises positioning each pivot magnet on the surface of the sensor structure body such that the pivot magnet is held at the respective location by the magnetic force of a corresponding locating magnet.

7. The method as claimed in any one of claims 2 to 6, further comprising removing the insert from the cavity after fixing the optical fibre to the surface of the sensor structure body.

8. The method as claimed in any one of the preceding claims, further comprising removing the plurality of pivot points from the surface of the sensor structure body after fixing the optical fibre to the surface of the sensor structure body.

9. The method as claimed in any one of the preceding claims, wherein the serpentine path comprises a plurality of straight portions extending between respective pairs of pivot points, wherein each of the plurality of strain sensors is arranged within a respective straight portion of the path.

10. The method as claimed in any one of the preceding claims, wherein fixing the optical fibre to the surface of the sensor structure body comprises applying adhesive to the optical fibre.

11. The method as claimed in claim 10, comprising applying adhesive to one or more of the strain sensors of the optical fibre.

12. A method of designing a multiple-axis force sensor comprising a sensor structure body and an optical fibre mounted on the sensor structure body, wherein the optical fibre comprises a plurality of strain sensors, the method comprising: receiving geometry of a sensor structure body; receiving parameters representative of: a target orientation of one or more strain sensors in an optical fibre when the optical fibre is mounted on a sensor structure body; a target location on a sensor structure body of one or more strain sensors in an optical fibre when the optical fibre is mounted on a sensor structure body; and a target minimum bend radius for the path of an optical fibre when the optical fibre is mounted on a sensor structure body; using the parameters to determine a set of apex locations on the sensor structure body such that, when an optical fibre is wrapped around the sensor structure body in a serpentine path such that each apex of the serpentine path is coincident with a respective apex location of the set of apex locations, the serpentine path has a minimum bend radius greater than the target minimum bend radius; and outputting the set of apex locations.

13. The method as claimed in claim 12, further comprising: receiving a target load capacity for the multiple-axis force sensor; and using the target load capacity to determine the geometry of the sensor structure body.

14. The method as claimed in claim 12 or 13, further comprising receiving a parameter representative of: a target number of strain sensors; and/or a separation distance between a respective pair of strain sensors in an optical fibre along the length of the optical fibre.

15. The method as claimed in claim 12, 13 or 14, comprising determining the target orientation and/or the target location of the one or more strain sensors using an unconstrained optimisation process.

16. The method as claimed in any one of claims 12 to 15, comprising determining the set of apex locations using a constrained optimisation process, wherein the constraints of the optimisation process comprise: the target minimum bend radius; the target location; and the target orientation.

17. The method as claimed in claim 16, wherein the constraints of the optimisation process further comprise a target distribution of strain sensors across a surface area of the sensor structure body.

18. The method as claimed in any one of claims 12 to 17, further comprising using the set of apex locations to determine a respective location and respective orientation for each of a plurality of strain sensors of an optical fibre when the optical fibre is mounted on the sensor structure body.

19. The method as claimed in claim 18, further comprising: generating a computer model of the force sensor comprising an optical fibre mounted on the sensor structure body of the force sensor such that the optical fibre follows the serpentine path; introducing a random change to the respective position and/or respective orientation of one or more of the plurality of strain sensors; and modelling the performance of the force sensor to generate a performance metric.

20. The method as claimed in claim 19, further comprising using the performance metric to adjust one or more of the apex locations.

21. The method as claimed in claim 18, 19 or 20, further comprising: calculating a value representative of the variation in the respective orientations of the plurality of strain sensors; comparing the value to a threshold variation; and in response to determining that the value is less than the threshold variation, adjusting one or more of the apex locations.

22. A multiple-axis force sensor comprising: a sensor structure body; and an optical fibre, comprising a plurality of strain sensors, extending around the sensor structure body in a serpentine path; wherein: the optical fibre is mounted such that the optical fibre comprises a plurality of apex portions and a plurality of straight portions, wherein each of the straight portions connects a respective pair of apex portions; and the straight portions of the optical fibre are fixedly mounted to the sensor structure body but the plurality of apex portions are moveable relative to the sensor structure body.

23. The multiple-axis force sensor as claimed in claim 22, wherein the strain sensors are Fibre Bragg Grating (FBG) sensors.

24. The multiple-axis force sensor as claimed in claim 22 or 23, wherein the surface of the sensor structure body adjacent each of the apex portions of the optical fibre is devoid of protrusions.

25. A method of using a multiple-axis force sensor to determine an estimate of one or more load components and/or a temperature change applied to the multipleaxis force sensor, wherein the multiple-axis force sensor comprises: a sensor structure body; and an optical fibre, comprising a plurality of strain sensors, extending around the sensor structure body in a serpentine path; wherein: the optical fibre is mounted such that the optical fibre comprises a plurality of apex portions and a plurality of straight portions, wherein each of the straight portions connects a respective pair of apex portions; and the straight portions of the optical fibre are fixedly mounted to the sensor structure body but the plurality of apex portions are moveable relative to the sensor structure body; and wherein the method comprises: transmitting light from a light source into the optical fibre of the multiple-axis force sensor; detecting a wavelength of light reflected by the strain sensors; and using the wavelength of the reflected light to determine an estimate of one or more load components and/or a temperature change applied to the multiple-axis force sensor.