Sensor arrangement for sensing a force and method for manufacturing a sensor arrangement

Abstract

The invention relates to a sensor arrangement for sensing a force, the sensor arrangement comprising a flexible circuit board, a plurality of air pressure sensors, a rigid core and a compliant layer covering the plurality of air pressure sensors and providing a measurement surface. The invention further relates to a method for manufacturing such a sensor arrangement.

Description

Technical Field

[0001] The present invention relates to a sensor arrangement for sensing force and a method for manufacturing a sensor arrangement for sensing force. Background Technology

[0002] When developing applications such as robotics, sensing forces applied to the robot's hand or another part of the robot (e.g., legs or manipulators) is crucial for giving the robot enhanced mobility and / or the ability to manipulate objects. Known implementations of sensor arrays that can be used in robotics applications to obtain feedback on applied forces are very expensive and lack sufficient resolution. Summary of the Invention

[0003] Therefore, one object of the present invention is to provide a sensor arrangement for sensing force that is different from or optimized from the prior art. Another object is to provide a method for manufacturing such a sensor arrangement.

[0004] These objectives are achieved through the subject matter of the main claims. Preferred embodiments may be derived, for example, from the dependent claims. The content of the claims becomes part of the specification through explicit reference.

[0005] This invention relates to a sensor arrangement for sensing force. The sensor arrangement includes a flexible circuit board. The sensor arrangement includes a plurality of barometric pressure sensors mounted on the flexible circuit board. The sensor arrangement includes a rigid core to which the flexible circuit board is wrapped and mounted, such that the flexible circuit board at least partially covers the rigid core, causing the plurality of barometric pressure sensors to protrude from the rigid core. The sensor arrangement also includes a compliant layer covering the plurality of barometric pressure sensors and providing a measuring surface.

[0006] This sensor arrangement can be manufactured at low cost and provides high resolution.

[0007] The flexible circuit board should be understood as being flexible when used alone, especially before being mounted onto the rigid core. This flexible circuit board is easy to manufacture and handle, thus reducing workload and cost. The pressure sensor can be a standard type used in many industrial or scientific applications. Therefore, they are very inexpensive. Pressure sensors typically provide an output signal (i.e., a pressure value) proportional to the force applied to the sensor. This can be considered the definition of a pressure sensor. Generally, any pressure sensor can be used.

[0008] The rigid core is typically made of a rigid material, such as plastic or metal. It provides stability to the sensor arrangement, especially when forces are applied. Thus, the compliant layer can deform in response to applied forces, while the rigid core absorbs the forces and provides an indeformable reference.

[0009] The characteristic that the flexible circuit board at least partially covers the rigid core generally means that at least a portion of the rigid core is covered by the flexible circuit board. Typically, the rigid core may have a surface designed to be covered by the flexible circuit board, and the flexible circuit board may partially or completely cover the surface. Therefore, the flexible circuit board may leave a portion of the surface of the rigid core uncovered.

[0010] The flexibility of the flexible circuit board typically means that it can be easily bent in its detached state, especially when it is not yet mounted on the rigid core. For example, the flexible circuit board can behave like a piece of cloth or rubber in its detached state.

[0011] The flexible circuit board can be mounted to the rigid core using adhesive or screws. However, other methods can also be used to mount the flexible circuit board to the rigid core.

[0012] Typically, the plurality of pressure sensors are mounted on the flexible circuit board before the flexible circuit board is mounted on the rigid core.

[0013] The flexible circuit board may include multiple conductor paths, such as wires connecting the multiple barometric pressure sensors, to power them and / or read data. Using the flexible circuit board is a highly efficient way to provide power and / or read capability to such barometric pressure sensors mounted on a rigid core of a separate shape, because the wires on the flexible circuit board automatically adapt to any desired shape.

[0014] The compliant layer is, in particular, a deformable layer that responds to forces applied to the measuring surface. This deformation is characteristic of the external force or other parameters, such as indenter shape or shear force. The compliant layer can be particularly flexible and / or elastic, allowing it to automatically return to its defined shape after the force application ceases. Forces applied to the measuring surface are typically transmitted by the compliant layer to the plurality of barometric pressure sensors, particularly for typical forces applied to the measuring surface, which affect the multiple pressure sensors. Therefore, very high-resolution force detection can be achieved even if the spacing between the plurality of pressure sensors is much wider than that of high-resolution sensors known in the prior art. This is especially true because more sophisticated force inference techniques can be used, such as those based on machine learning and / or artificial neural networks, as described in this application. In particular, a single compliant layer can cover all the pressure sensors.

[0015] According to one embodiment, the rigid core is dome-shaped. This is particularly suitable when the sensor arrangement is the tip of a robot or another manipulator. However, other shapes can also be used. For example, the sensor arrangement can be adapted to design foot sensors, calf sensors, thigh sensors, or breast sensors for a robot. The shape of the rigid core can be adjusted accordingly. For example, it can be planar, cylindrical, or any shape. Typically, the shape is designed so that contact force can simultaneously activate multiple barometric pressure sensors, thereby enabling force localization.

[0016] The rigid core may, in particular, have multiple facets. According to one embodiment, each pressure sensor, or at least a portion of the pressure sensors, is positioned on at least one of the multiple facets. Therefore, the orientation of each pressure sensor can be defined by the orientation of its corresponding facet. This does not preclude the arrangement of more than one pressure sensor on a given facet. It is also possible that some facets or other portions of the rigid core's surface may not have pressure sensors installed.

[0017] It should be noted that although the plurality of pressure sensors are placed on the flexible circuit board, the flexible circuit board is typically adapted to the shape of the facets of the rigid core. Therefore, the flexible circuit board itself forms facets.

[0018] In particular, the facets can have different orientations, thus allowing force measurements to be performed in different directions.

[0019] The compliance layer may comprise or be composed of plastic or rubber materials. Plastic materials may be, for example, thermoplastics, elastomers, thermoplastic elastomers, thermosetting materials, or similar materials. These materials have proven suitable for typical applications. However, other materials may also be used. In particular, they may be used in the manufacturing process of the aforementioned compliance layer.

[0020] The compliance layer can, in particular, transmit the force applied to the measuring surface to at least a portion of the plurality of pressure sensors. Specifically, it can be configured such that the force on at least a portion or most of the measuring surface is transmitted to more than one pressure sensor. This allows for improved resolution when measuring force using electronic force inference techniques.

[0021] The plurality of pressure sensors can be connected, in particular, via conductor paths on the flexible circuit board. These conductor paths can be flexible in that they automatically adapt to the surface of the rigid core when the flexible circuit board is wrapped. Such conductor paths allow for reliable and easy connection of the pressure sensors.

[0022] The plurality of pressure sensors can be connected, in particular, via conductor paths on the flexible circuit board. These conductor paths can be flexible in that they automatically adapt to the surface of the rigid core when the flexible circuit board is wrapped. Such conductor paths allow for reliable and easy connection of the pressure sensors.

[0023] The flexible circuit board can be, in particular, star-shaped. Specifically, it can include multiple arms or spokes connected at the central portion. This especially allows for the wrapping of a flexible circuit board on a dome-shaped rigid core, as can be seen, for example, in the accompanying drawings.

[0024] The plurality of barometric pressure sensors can be arranged at a distance of at least 1 mm, at least 2 mm, at least 3 mm, at least 4 mm, or at least 5 mm. They can also be arranged at a distance of at most 1 mm, at most 2 mm, at most 3 mm, at most 4 mm, or at most 5 mm. The distance can be the distance between the outer perimeters of the barometric pressure sensors. Every two different values ​​can be combined to form a suitable interval.

[0025] Specifically, the sensor arrangement can be a robot tip and / or a manipulator element of the robot. While this is a preferred application, it should be noted that the sensor arrangement can also be used in principle for several other applications, especially when force must be measured and / or when elements should be used for manipulation. Manipulation specifically refers to: for example, a manipulator element, which can be the same as the sensor arrangement, that can grasp or capture an object to be manipulated and manipulate the object, for example, according to its position or orientation. When doing so, the sensor arrangement can be used to measure force. In general, it can be said that the sensor arrangement disclosed herein can combine barometric pressure sensing technology with novel assembly methods. It can be further combined with machine learning techniques, such as those disclosed herein, to create high-resolution tactile sensors with high levels of robustness and, for example, a three-dimensional dome shape.

[0026] In one embodiment, the rigid core is a 3D-printed part. This allows for variable and efficient manufacturing. However, other manufacturing methods, such as drilling or molding, can also be used.

[0027] For example, at least 5, 10, 15, 19, 20, 25, 30, 35, or 37 barometric pressure sensors can be used. These can be encased in a dome-shaped central core. This assembly can be placed in a mold covered with a material (e.g., polyurethane) to provide protection for the flexible outer surface of the sensors while also enabling localized pressure measurements. Typically, the multiple barometric pressure sensors are individual elements, so each can be visually and / or physically distinguished from adjacent sensors.

[0028] The present invention also relates to a method for manufacturing a sensor arrangement for sensing force. The method includes the following steps:

[0029] - Provides a flexible circuit board on which multiple barometric pressure sensors are mounted.

[0030] - Provides a rigid core,

[0031] - The flexible circuit board is wrapped around and mounted on the rigid core such that the flexible circuit board at least partially covers the rigid core, and the plurality of pressure sensors protrude from the rigid core.

[0032] - Cover the plurality of barometric pressure sensors with a flexible layer to provide a measuring surface on the flexible layer.

[0033] This method is particularly applicable to the fabrication of sensor arrangements as described above. It should be noted that all statements given regarding sensor arrangements can also be applied to the methods for fabricating such arrangements, and vice versa, provided that such statements are technically appropriate.

[0034] The method provides for the inexpensive and efficient manufacture of sensor arrangements, especially the sensor arrangements according to the invention.

[0035] In one embodiment, the plurality of pressure sensors are already mounted on the flexible circuit board at the start of the method. However, in an alternative embodiment, mounting the plurality of pressure sensors on the flexible circuit board may be part of the method, as described further below, for example.

[0036] Mounting the flexible circuit board onto the rigid core can be performed, for example, by using glue, screws, or by clamping or otherwise securing the flexible circuit board to cover at least a portion of the surface of the rigid core.

[0037] Because the plurality of pressure sensors protrude from the rigid core, the force exerted on the pressure sensors by forces applied to the measuring surface can be improved. Specifically, this means that the rigid core contacts one surface of the flexible circuit board, and the plurality of pressure sensors are mounted on opposite surfaces of the flexible circuit board.

[0038] When the compliant layer covers the plurality of barometric pressure sensors, it typically also covers the flexible circuit board, particularly the portion of the flexible circuit board surrounding the plurality of barometric pressure sensors and at least a portion of the rigid core. Specifically, the compliant layer may directly contact the surface area of ​​the rigid core covered by the compliant layer but not by the flexible circuit board. Covering with the compliant layer can be performed in particular as further described below.

[0039] Preferably, covering the plurality of barometric pressure sensors with the flexible layer includes the following steps:

[0040] - Place the rigid core with the flexible circuit board into the mold.

[0041] - Fill the mold at least partially with material so that the plurality of pressure sensors are covered by the material.

[0042] - Transform the material into the aforementioned flexible layer.

[0043] This method of covering with the sturdiness layer provides a simple and cost-effective manufacturing process. The mold can specifically define the final shape of the measuring surface, particularly ensuring that the measuring surface of the sturdiness layer acquires a shape defined by the shape of the mold.

[0044] The mold can be partially or completely filled with material, depending on which portion of the rigid core or the flexible circuit board mounted on the rigid core the flexible layer should cover. Specifically, the mold can be filled with material to an amount sufficient to completely cover the flexible circuit board.

[0045] Transforming the material into the flexible layer means that an easy-to-handle material can be used, for example, because it is a fluid that can be easily filled into a mold.

[0046] The transformation may include, for example, the following steps:

[0047] - The material is degassed by placing the rigid core covered by the material and the flexible circuit board in a vacuum.

[0048] Therefore, materials that are fluid in their undegassed state but compliant in their degassed state can be used.

[0049] Degassing can be carried out or initiated, especially at room temperature, for example, in the temperature range between 15°C and 25°C. During vacuum conditions, the temperature may be higher compared to these values.

[0050] However, it should be noted that other techniques can also be used to form a compliance layer.

[0051] Providing the flexible circuit board may include one or both of the following steps:

[0052] - Cut at least a portion of the flexible circuit board from the sheet material.

[0053] - Multiple pressure sensors are arranged and installed on the flexible circuit board.

[0054] Therefore, fabricating the flexible circuit board having the plurality of pressure sensors can be part of the method. In an alternative embodiment, a flexible circuit board with multiple pressure sensors already mounted thereon can be used.

[0055] According to one embodiment, the rigid core is dome-shaped. This is particularly suitable when the sensor arrangement is the tip of a robot or another manipulator. However, other shapes can also be used. For example, the sensor arrangement can be adapted to design foot sensors, calf sensors, thigh sensors, or breast sensors for a robot. The shape of the rigid core can be adjusted accordingly. For example, it can be planar, cylindrical, or any shape. Typically, the shape is designed so that contact forces can simultaneously activate the multiple barometric pressure sensors, thereby enabling force localization.

[0056] The rigid core may, in particular, have multiple facets. According to one embodiment, each pressure sensor, or at least a portion thereof, is positioned on at least one of the multiple facets. Therefore, the orientation of each pressure sensor can be defined by the orientation of its corresponding facet. This does not preclude the arrangement of more than one pressure sensor on a given facet. It is also possible that no pressure sensors may be installed on any facet or other portion of the rigid core's surface.

[0057] It should be noted that although the plurality of pressure sensors are placed on the flexible circuit board, the flexible circuit board is typically adapted to the shape of the facets of the rigid core. Therefore, the flexible circuit board itself forms facets.

[0058] In particular, the facets can have different orientations, thus allowing force measurements to be performed in different directions.

[0059] The compliance layer may comprise or be composed of plastic or rubber materials. The plastic material may be a thermoplastic, elastomer, thermoplastic elastomer, thermosetting material, or similar material. These materials have proven suitable for typical applications; however, other materials may also be used. In particular, they may be used in the manufacturing process of the compliance layer described above.

[0060] The compliance layer can, in particular, transmit the force applied to the measuring surface to at least a portion of the pressure sensor. Specifically, it can be configured such that at least a portion or most of the force on the measuring surface is transmitted to more than one pressure sensor. This allows for improved resolution when measuring force using electronic force inference techniques.

[0061] The plurality of pressure sensors can be connected, in particular, via conductor paths on the flexible circuit board. These conductor paths can be flexible in particular, so as to automatically adapt to the surface of the rigid core when the flexible circuit board is wrapped. Such conductor paths allow for reliable and easy connection of the plurality of pressure sensors.

[0062] The flexible circuit board can be, in particular, star-shaped. Specifically, it can include multiple arms or spokes connected to the central portion. This allows the flexible circuit board to be wrapped around a dome-shaped rigid core, as can be seen, for example, in the accompanying drawings.

[0063] The barometric pressure sensors can be arranged at a distance of at least 1 mm, at least 2 mm, at least 3 mm, at least 4 mm, or at least 5 mm. They can also be arranged at a distance of at most 1 mm, at most 2 mm, at most 3 mm, at most 4 mm, or at most 5 mm. The distance can be the distance between the outer perimeters of the plurality of barometric pressure sensors. Every two different values ​​can be combined to form a suitable interval.

[0064] Specifically, the sensor arrangement can be a robot tip and / or a manipulator element of the robot. While this is a preferred application, it should be noted that the sensor arrangement can also be used in principle for several other applications, especially when force must be measured and / or when elements should be used for manipulation. Manipulation specifically refers to: for example, a manipulator element identical to the sensor arrangement can grasp or capture an object to be manipulated and manipulate the object, for example, based on its position or orientation. When doing so, the sensor arrangement can be used to measure force. In general, it can be said that the sensor arrangement disclosed herein can combine barometric pressure sensing technology with novel assembly methods. It can be further combined with machine learning techniques, such as those disclosed herein, to create high-resolution tactile sensors with high robustness levels and, for example, a three-dimensional dome shape.

[0065] In particular, the rigid core can be 3D printed. This may mean that providing the rigid core includes the step of 3D printing the rigid core. However, other manufacturing methods, such as drilling or molding, can also be used.

[0066] For example, at least 5, at least 10, at least 15, at least 19, at least 20, at least 25, at least 30, at least 35, or at least 37 barometric pressure sensors can be used. They can be encased in a dome-shaped central core. This assembly can be placed in a mold and covered with a material (such as polyurethane) to provide protection for the flexible outer surface of the sensors while still enabling localized pressure measurements.

[0067] Machine learning techniques can be used to provide super-resolution sensing of tactile interactions using sensor data. Therefore, the multiple barometric pressure sensors function as if there were actually more sensors. Machine learning algorithms can be integrated by first learning an intrinsic model of the finger pad using the finite element method, and then using transfer learning to correlate the real physical barometer with the intrinsic model. Force distribution maps (nodal forces with 3 degrees of freedom and a local coordinate system) can be predicted as a representation of the touch effect, and these can be classified into different operational scenarios, such as gripping, flipping detection, and twisting.

[0068] The method enables high-resolution sensing around the finger contour, making the system well-suited for applications where the location of object contact is unpredictable or can vary significantly. Furthermore, the hardware components used for the sensor arrangement are very inexpensive, especially compared to other sensors known in the prior art.

[0069] According to a preferred embodiment, the sensor arrangement according to the invention or the manufacturing method according to the invention further includes an electronic control module configured to perform a force inference method for the sensor arrangement. This allows the force inference function to be integrated into the sensor arrangement. The electronic control module may, for example, be located in or on the rigid core, or it may be located separately from the rigid core.

[0070] Specifically, the control module can be configured to execute a method for force inference to provide a force map of the measurement surface, the force map comprising multiple force vectors. Such a force map can provide relevant information about the applied force, such as that originating from an indenter pressing on the measurement surface or from the object to be operated.

[0071] The control module can be specifically configured to execute the force inference method and / or training method described further below.

[0072] According to a typical implementation, the force diagram may include at least 0.25 force vectors per square millimeter, at least 0.5 force vectors per square millimeter, at least 0.75 force vectors per square millimeter, at least 1 force vector per square millimeter, at least 1.5 force vectors per square millimeter, or at least 2 force vectors per square millimeter.

[0073] According to a typical implementation, the force diagram may include up to 0.25 force vectors per square millimeter, up to 0.5 force vectors per square millimeter, up to 0.75 force vectors per square millimeter, up to 1 force vector per square millimeter, up to 1.5 force vectors per square millimeter, or up to 2 force vectors per square millimeter.

[0074] These values ​​have proven to be suitable for typical application examples. However, other values ​​may also be used.

[0075] According to a typical implementation, the force diagram may include at least 500, at least 1000, or at least 2000 force vectors. According to a typical implementation, the force diagram may include up to 1000, up to 2000, up to 3000, or up to 4000 force vectors. Such values ​​have proven particularly suitable for applications where the sensor arrangement is at the tip of a robot. However, other values ​​may also be used.

[0076] Preferably, each force vector includes a normal force component, a first shear force component, and a second shear force component. This provides information not only about the normal force but also about the shear force, thus allowing for better adjustment in applications such as robot tip applications.

[0077] Specifically, the first shear force component may correspond to a first shear force, and the second shear force component may correspond to a second shear force. The first shear force may be perpendicular to the second shear force. In particular, the shear force components may be perpendicular to each other.

[0078] According to one embodiment, the control module can be configured to read temperature values ​​from the plurality of barometric pressure sensors and provide temperature information or a temperature map of the sensor arrangement based on the temperature values. This may provide additional information about the temperature distribution, for example, for control or monitoring applications. In particular, the plurality of barometric pressure sensors may have corresponding integrated temperature measurement functions that can be used for this purpose.

[0079] Further inventive aspects are described below. Such aspects may be combined, alone, or in combination with other features disclosed herein. They may also be considered as separate inventive aspects and may be the subject of the claims.

[0080] The present invention relates to a method for force inference in a sensor arrangement for sensing force.

[0081] Such sensor arrangements, especially those using the described method, may include multiple barometric pressure sensors. They may further include a compliance layer. This compliance layer may cover the multiple barometric pressure sensors and provide a measuring surface. For example, a sensor arrangement using the method of the present invention may be a sensor arrangement as described herein, or a sensor arrangement that can be manufactured according to the methods described herein. All disclosed embodiments and variations may be used with respect to the sensor arrangement or manufacturing method.

[0082] The force inference method includes the following steps:

[0083] - Read multiple pressure values ​​from the plurality of barometric pressure sensors, and

[0084] - A feedforward neural network is used to calculate a force map on the measurement surface based on the plurality of pressure values, the force map comprising a plurality of force vectors.

[0085] Using this method, force inference from the sensors can be performed by obtaining high-resolution and / or complex information from the multiple barometric pressure sensors. This is possible because feedforward neural networks have been found to provide force information with a finer resolution than the spacing between the multiple barometric pressure sensors. It can even provide more information. This capability is particularly evident if the feedforward neural network is trained correctly. Preferred implementations of the training are further given below.

[0086] The plurality of pressure sensors are particularly well-suited to generating output signals that depend on, for example, air pressure, such as being linearly correlated with the pressure applied to a particular pressure sensor. In particular, the pressure is transmitted from the measuring surface to the pressure sensors, where forces typically applied even with minimal extension to the measuring surface are also transmitted to the plurality of pressure sensors, making it possible to use techniques such as feedforward neural networks to achieve good resolution.

[0087] For details on sensor placement, please refer to the detailed description provided herein, including descriptions of embodiments and variations.

[0088] The force map can be, in particular, a map defined on a real measurement surface, which may include multiple points. At each point, some information may be defined, such as a force vector as described further below. The force map typically provides information about the forces applied to the measurement surface. For example, such forces may originate from one or more indenters pressing on the measurement surface, or from an object currently manipulated by the sensor arrangement, for example, when the sensor arrangement is a robot tip.

[0089] According to one embodiment, the feedforward neural network includes a transmission network and a reconstruction network. The transmission network maps the barometric pressure sensor to multiple virtual sensors of a finite element model of the sensor arrangement. The reconstruction network maps the virtual sensors of the finite element model to a force map. Each virtual sensor may include one or more virtual sensor points, and each virtual sensor point has a virtual sensor point value.

[0090] Therefore, the feedforward neural network is split in this implementation. This enhances functionality, especially the ability to train better, which will be described further below.

[0091] The transmission network can map actual barometric pressure sensors or output values ​​derived from barometric pressure sensors to a finite element model. The finite element model can be a virtual model of the actual sensor arrangement. It can be particularly used to enhance force inference capabilities. A finite element model can be built using the finite element method. It can include virtual representations of the real components and materials used. For example, Young's modulus and Poisson's ratio of the materials used can be used in the same way as the actual sensor arrangement. Furthermore, the distance and other geometric dimensions between the real sensor arrangement and the virtual finite element model can be the same. However, it should be noted that the finite element model is primarily a component used for training and does not necessarily have to be implemented solely for force inference after training. If training has been completed, the transmission network and the reconstruction network can be used separately from the complete finite element model, where force inference should be performed in each case, i.e., a force map should be obtained by first mapping the output values ​​read from the barometric pressure sensors to virtual sensor point values ​​through the transmission network, and then mapping the obtained virtual sensor point values ​​to the force map through the reconstruction network.

[0092] Typically, the transmission network and the reconstruction network are artificial neural networks. Mapping may specifically mean feeding input values ​​into the network, and the network generating output values ​​based on its training. The training can be adapted to multiple values ​​that define the network's behavior. For example, approximately one million values ​​can be used to define the network's behavior. In the case of the transmission network, it may receive data from the multiple barometers, and it may generate multiple dummy sensor point values. In the case of the reconstruction network, it can be fed with dummy sensor point values, and it can generate a force map. The entire feedforward neural network, whether split or not, can accept data from the multiple barometers, and it can generate a force map.

[0093] A virtual sensor can be thought of as dividing a real sensor into sensor points. While a real sensor, such as a barometer, converts a force applied to it into an output signal, a virtual sensor converts this force into multiple sensor point values. Typically, the sensor point values ​​are located in a region of the finite element model, which corresponds to the barometer in an actual sensor arrangement. This region in the finite element model can also be smaller or larger, for example, 10% or 50% smaller or larger. The virtual sensor concept also considers that barometers are not typically located at a point known to be accurate enough to use a precise location in force inference. Despite these variations, reliable force inference can be performed using a finite element model with virtual sensor points.

[0094] The training aspects of the network will be described next. The training steps mentioned in this paragraph should specifically be considered as steps performed before force inference of the actual force measurement. Therefore, a force inference method can be considered as a combination of training steps performed before force inference, and force inference using one or more trained networks. A force inference method can also be considered as force inference itself, using one or more appropriately trained networks. Individual training methods are further described below. They can be performed independently of any force inference. Typically, force inference, which reads the barometric pressure sensor and generates a force map, is considered as the action to be performed in the use case, i.e., when the sensor arrangement will be used to measure or evaluate the force applied to the measurement surface, for example, because the sensor arrangement is currently manipulating an object or otherwise in contact with an object applying pressure to the measurement surface.

[0095] According to one implementation, the reconstruction network may have been trained by performing the following steps prior to force inference:

[0096] - Multiple simulations are performed in the finite element model. Each simulation involves simultaneously applying one or more simulated forces to the simulated measurement surface of the finite element model, thereby calculating a simulated force map on the simulated measurement surface. The simulated force map includes multiple simulated force vectors, and the corresponding virtual sensor point values ​​are calculated using the finite element model.

[0097] - Train and reconstruct the network using the calculated simulated force map and the corresponding calculated virtual sensor point values.

[0098] Such training steps of the reconstructed network can be used to properly train the network so that it can generate accurate and detailed force maps from virtual sensor point values, which are typically the expected outputs of the sensor arrangement. These virtual sensor point values ​​can be obtained, in particular, through a transmission network.

[0099] It has been demonstrated that training the reconstruction network using only simulation is suitable. In particular, simulations can be used to train the reconstruction network in a manner that allows it to detect not only one force applied to the measurement surface but also multiple forces. This is far easier than training the network using real force tests, where applying two or more forces simultaneously is complex due to the need to avoid collision problems and the complexity of experimental setups. It has been shown that high reliability of the reconstructed force map can be obtained by training the reconstruction network solely through simulation. Simulations can be performed, in particular, in a computer or another programmable and / or automated data processing entity.

[0100] Training in the finite element model can be performed, in particular, through pure computer simulation. Therefore, the simulated forces are applied only in such computer simulations. The simulated measurement surface is typically the surface of the finite element model, such as a compliant layer of the model. Thus, the simulated measurement surface exists only in the simulation, while the actual measurement surface is the surface where the sensors are arranged.

[0101] Simulated forces are applied to a simulated measurement surface. This results in the formation of a simulated force map. The simulated force map comprises multiple simulated force vectors, each of which gives a local value of the simulated force map. The simulated force map can be represented and / or calculated as the deformation of the simulated measurement surface. It can be calculated, in particular, using the finite element method.

[0102] Virtual sensor point values ​​can also be calculated using the finite element method. Specifically, the simulated forces and structural and material properties of the finite element model representing the actual sensor arrangement determine the simulated force map and the virtual sensor point values. This illustrates the relationship between the simulated force map and the virtual sensor point values.

[0103] In force inference, virtual sensor point values ​​can be generated based on data from barometric pressure sensors that indirectly measure real forces. Using the relationship between the simulated virtual sensor point values ​​and the simulated force map, the force map can be reconstructed through the aforementioned reconstruction network.

[0104] It should be noted that generating a force map representation from virtual sensor point values ​​is called reconstruction. Therefore, the network that performs this reconstruction is referred to as a reconstruction network.

[0105] When training the reconstructed network, data from the simulation can be used. Such data may in particular include simulated force maps and corresponding virtual sensor point values.

[0106] According to one embodiment, the simulated force applied to the simulated measurement surface is generated based on individual simulated indenters having simulated indenter shapes. The shape may specifically relate to the portion of the simulated indenter that contacts the simulated measurement surface in the simulation. Therefore, the simulated indenter is the object used to define the simulated force in the simulation.

[0107] According to one embodiment, the simulated indenter shape is selected from a group comprising at least a pointed, circular, triangular, square, hemispherical, cubic, and cylindrical shape. Such simulated indenter shapes have proven suitable because they correspond to typical shapes of real objects that contact the measurement surface in applications. Using these different indenter shapes can significantly improve the training of the reconstruction network to reconstruct corresponding or similar shapes applied to real measurement surfaces. It should be noted that each of the mentioned shapes can be used, or only one of the mentioned shapes can be used, or a selection of the mentioned shapes can be used. Alternatively or additionally, other shapes can be used. When multiple indenters are used in the simulation, these indenters may have the same or different shapes.

[0108] According to one implementation, the reconstruction network is trained using multiple different simulated indenter shapes. This allows the reconstruction network to be trained to distinguish the forces generated by different indenter shapes. In particular, one or more simulations can be performed for each used indenter shape or each combination of used indenter shapes. Such simulations can differ, for example, in the number of indenters and / or the location where one or more indenters are applied.

[0109] According to one embodiment, the reconstruction network is trained using simulated indenters of various sizes. Alternatively, or as an alternative to using different shapes, this allows the reconstruction network to be trained to distinguish between indenters or other objects applying different magnitudes of force. For example, contact portions of different sizes corresponding to the simulated measurement surface can be used. Statements relating to simulation performance given by using different indenter shapes apply accordingly. Furthermore, different indenter shapes and sizes can be combined.

[0110] According to one implementation, the reconstructed network is trained using at least a portion of simulation, including the simultaneous application of simulated forces generated based on two or more simulated indenters. This allows the reconstructed network to distinguish between forces applied by only one indenter and forces applied by two or more indenters. This can be done, in particular, in simulation, which is much easier than preparing experimental setups for such applications involving two or more indenters.

[0111] According to one implementation, the reconstruction network is trained with at least a portion of the simulation, including applying simulated forces generated based on only one simulated indenter. This allows for specific training for force-oriented reconstruction when only one indenter is applied.

[0112] For example, the following number of simulations can be performed in a typical training session.

[0113] When training a single-contact reconstruction network, 10,000 to 50,000 or 30,000 simulations can be performed.

[0114] When training a dual-contact reconstruction network, 5,000 to 20,000 or 10,000 simulations can be performed.

[0115] When training a three-contact reconstructed network, 5,000 to 20,000 or 10,000 simulations can be performed.

[0116] When training a four-contact reconstructed network, 5,000 to 20,000 or 10,000 simulations can be performed.

[0117] When reconstructing a network using five-contact training, 5,000 to 20,000 or 10,000 simulations can be performed.

[0118] However, these are just typical or preferred numbers. Generally, any number of simulations can be performed. For example, two-contact represents the simultaneous application of two forces, three-contact represents the simultaneous application of three forces, four-contact represents the simultaneous application of four forces, and five-contact represents the simultaneous application of five forces. Such simulations can be combined during training.

[0119] According to one embodiment, each simulated force vector includes a normal force component, a first shear force component, and a second shear force component. Therefore, the force map provides information about these components. It should be noted that in a typical implementation according to the prior art, the shear force cannot be reconstructed. However, it has been shown that when such simulated force vectors with the aforementioned components are used to train the reconstruction network via simulation, the shear force can be reconstructed in addition to the normal force. This provides additional information that is valuable in several applications, such as in robotic applications used to manipulate objects.

[0120] According to one embodiment, in the simulated force vector, the first shear force component corresponds to a first shear force, and the second shear force component corresponds to a second shear force. Specifically, the first shear force is perpendicular to the second shear force. This vertical orientation of the shear forces provides easily accessible information.

[0121] It should be noted that the force vector may alternatively have more or fewer than three components.

[0122] According to one implementation, the reconstruction network is trained using multiple simulated forces with different shear force components. This enables the reconstruction network to distinguish between different shear forces applied to the measurement surface. The shear forces between different force components used in a simulation and / or between different simulations may differ.

[0123] According to one implementation, the reconstruction network is trained using multiple simulated forces with different normal force components. This allows the reconstruction network to distinguish between different normal forces applied to the measurement surface. The normal forces between different forces used in a single simulation and / or between different simulations may differ.

[0124] It should be noted that the concepts of using different simulated indenters, different simulated indenter shapes, different simulated indenter sizes, and / or different simulated shear forces or different simulated shear force components can also be applied to other purposes when training neural networks for force inference. This is independent of the sensor arrangement given in this paper. The same applies to real indenters and / or forces.

[0125] According to one implementation, the transmission network may have been trained by performing the following steps prior to force inference:

[0126] - Perform multiple force tests on the sensor arrangement, each force test involving applying a force at a location on the measuring surface of the sensor arrangement via an indenter, simultaneously measuring the force applied by the indenter and simultaneously measuring the pressure value using a barometric pressure sensor.

[0127] For each force test, a corresponding simulation is performed using the finite element model. Each simulation involves applying a simulated force to the simulated measurement surface of the finite element model, thereby calculating a simulated force map on the simulated measurement surface. The simulated force map includes multiple simulated force vectors, where the simulated force corresponds to the measured force and is applied at positions on the simulated measurement surface corresponding to the positions on the measurement surface. The corresponding virtual sensor point values ​​are then calculated using the finite element model.

[0128] - Train the transmission network using the measured pressure values ​​and the corresponding calculated virtual sensor point values.

[0129] Compared to simulation, force testing is a test performed using a real physical sensor arrangement. The indenter can be an object specifically designed to contact the measurement surface. Force testing can be performed by moving the sensor arrangement relative to a fixed indenter, or by moving the indenter relative to a fixed sensor arrangement. Furthermore, the movement of the sensor arrangement and the indenter can be applied. Force can be measured during the application of the indenter and can form the basis for the simulated force applied in the simulation. Measuring force has proven appropriate rather than attempting to apply a specifically defined force, as the latter approach is more complex, even if it is possible. The pressure value is typically the output signal of the multiple barometric pressure sensors.

[0130] It should be noted that it has been found that it is unnecessary to apply multiple pressure heads simultaneously for force testing to prepare the feedforward neural network for correctly evaluating multiple forces. As mentioned above, this can be accomplished through simulations used to train the transmission network.

[0131] The simulation used to train the transmission network can be performed using the same finite element model as the simulation used to train the reconstructed network.

[0132] In simulations, simulated forces, along with the structural and material properties of the finite element model, are typically the basis for calculations performed using the finite element model. Specifically, the simulated forces result in calculated simulated force maps and calculated virtual sensor point values. Therefore, the finite element model is used to calculate the virtual sensor point values ​​corresponding to the forces actually applied to the measurement surface.

[0133] The simulated force can correspond in particular to a real measured force, for example, it can have the same components, the same absolute value, and / or the same direction. Specifically, the simulated force can have an integral over the contact area of ​​the simulated indenter, which applies the simulated force to the simulated measurement surface. The simulated force applied to the simulated measurement surface is equal to or has a predetermined relationship with the measured force, or is the integral of the real and / or measured forces over the real contact area. For example, this can be related to the magnitude and / or direction of the force. A predefined variation between the measured force and the simulated force can also be used, which can also be considered as a corresponding force.

[0134] For example, a camera can be used to measure position through image recognition, or the position can be calculated based on machine parameters during force testing. Simulated forces can be applied to the same locations on the simulated measurement surface as the locations on the real measurement surface where real forces are applied. This provides a good correspondence between experiment and simulation.

[0135] When training the transmission network, experimental data and simulation data can be used. Such simulation data may in particular include pressure values ​​from barometric sensors and values ​​from virtual sensor points in the corresponding simulations.

[0136] According to one embodiment, force tests for training the transmission network are performed using multiple indenters, each with its own indenter shape. The shape may specifically relate to the portion of the indenter that contacts the measurement surface during the force test. Thus, the indenter is the object used in the force test to define the force applied to the measurement surface. In particular, multiple force tests can be performed, with one of a set of indenter shapes used in each force test. Typically, only one indenter is used per force test.

[0137] According to one embodiment, the indenter shape is selected from a group comprising at least a pointed tip, a circle, a triangular cross-section, a square cross-section, a hemisphere, a cube, and a cylinder. Such indenter shapes have proven suitable because they correspond to the typical shape of the object contacting the measurement surface in the application. Using these different indenter shapes can significantly improve the training of the transmission network to reconstruct corresponding or similar shapes applied to the measurement surface. It should be noted that each of the mentioned shapes can be used, only one of the mentioned shapes can be used, or a selection of the mentioned shapes can be used. Alternatively or additionally, other shapes can be used.

[0138] According to one implementation, simulations are performed using simulated forces based on simulated indenters, where the shapes of the simulated indenters correspond to the shapes of the actual indenters used in the respective force tests. This ensures optimal correspondence between the force tests and the simulations, thereby enabling ideal training of the transmission network.

[0139] According to one embodiment, the transmission network is trained using a variety of different indenter shapes. This enables the transmission network to be trained to distinguish the forces generated by different indenter shapes. Typically, the different indenter shapes are distributed across multiple force tests, since only one indenter is applied in each force test.

[0140] According to one embodiment, the transmission network is trained using multiple indenters of different sizes. Alternatively, or as an alternative, using different shapes, this can train the transmission network to distinguish between indenters or other objects applying different magnitudes of force. For example, contact portions of different sizes with the measuring surface can be used.

[0141] According to one embodiment, the transmission network is trained with multiple indenters, applying corresponding shear forces for at least a portion of the force tests used to train the transmission network. This trains the transmission network to distinguish between different shear forces applied to the measurement surface. In particular, multiple force tests can be performed using different shear forces or shear force components.

[0142] According to one embodiment, the measured forces each include a normal force component, a first shear force component, and a second shear force component. Therefore, the measured forces can provide information about these components. It should be noted that in typical embodiments according to the prior art, shear force cannot be measured. However, it has been shown that when such simulated force vectors with the aforementioned components are used to train a transmission network via force testing, shear force can be reconstructed in addition to the normal force. This provides additional information that is valuable in several applications, such as in robotic applications used to control robot tips.

[0143] According to one embodiment, in the measured force, the first shear force component corresponds to a first shear force, and the second shear force component corresponds to a second shear force. Specifically, the first shear force is perpendicular to the second shear force. This provides easily accessible information due to the perpendicular direction of the shear force.

[0144] It should be noted that the measured force may alternatively have more or fewer than three components.

[0145] The measured forces can be represented in a global coordinate system. They can also be represented by the local normal components at the point of contact, perpendicular to the measuring surface, with shear forces perpendicular to and / or mutually perpendicular to the normal components. This can be considered equivalent because coordinate transformations can be used to calculate the components in another coordinate system.

[0146] According to one embodiment, the transmission network is trained using multiple forces with different shear force components. This trains the transmission network to distinguish between different shear forces applied to the measurement surface. The shear force can vary between different components of the force used in a single force test and / or between different force tests.

[0147] According to one embodiment, the transmission network is trained using multiple forces with different normal force components. This allows the transmission network to be trained to distinguish between different normal forces applied to the measurement surface. The normal forces, in particular, can vary between different force tests and corresponding simulations.

[0148] According to one embodiment, the force applied by the indenter is measured using a force sensor located in or near the indenter. This force sensor can measure the force applied by the indenter to a measuring surface. In particular, it can measure the three components of the force, as described above. Positioning the force sensor near the indenter can particularly include: positioning the force sensor so that it contacts the indenter and / or positioning it between the indenter and the object on which the indenter is mounted.

[0149] According to one implementation, each simulated force vector includes a normal force component, a first shear force component, and a second shear force component. This can specifically correspond to a measured force. Therefore, in force testing, simulated forces can be used to simulate forces corresponding to those actually applied.

[0150] According to one embodiment, the feedforward neural network directly maps the plurality of barometric pressure sensors to a force map. This can be viewed as an alternative implementation that splits the feedforward neural network into a transmission network and a reconstruction network. In particular, in this embodiment, no pressure values ​​are mapped to virtual sensor point values. Instead, only one neural network is trained and maps the pressure values ​​directly to the force map.

[0151] For example, 20 to 100 force tests and corresponding simulations, or 50 force tests and corresponding simulations, can be performed to properly train the transmission network.

[0152] As a further example, at least 20 force tests, at least 50 force tests, at least 100 force tests, at least 500 force tests, at least 1,000 force tests, at least 2,000 force tests, or at least 10,000 force tests and / or up to 500 force tests, up to 1,000 force tests, up to 2,000 force tests, up to 10,000 force tests, or up to 50,000 force tests may be performed. However, other numbers of tests may also be used.

[0153] Force testing can be conducted in particular where the force is not predetermined but measured under each condition. Different apparatus parameters can be used to obtain different forces.

[0154] According to one implementation, the feedforward neural network may have been trained by performing the following steps prior to force inference:

[0155] - Perform multiple force tests on the sensor arrangement, each force test involving applying a force at a location on the measuring surface of the sensor arrangement via an indenter, simultaneously measuring the force applied by the indenter and simultaneously measuring the pressure value using a barometric pressure sensor.

[0156] For each force test, a corresponding simulation is performed using a finite element model arranged with the sensors. Each simulation involves applying a simulated force to a simulated measurement surface of the finite element model, thereby calculating a simulated force map on the simulated measurement surface. The simulated force map includes multiple simulated force vectors, where the simulated force corresponds to the measured force, and the position applied to the simulated measurement surface corresponds to the position on the measurement surface.

[0157] - The feedforward neural network is trained using the measured pressure values ​​and the corresponding computational simulation force diagram.

[0158] Such training can also be performed without using virtual sensor point values. For example, it can be used in implementations where the feedforward neural network directly maps pressure values ​​to a force map, as described above. However, it can also be implemented by splitting the feedforward neural network into a transmission network and a reconstruction network as discussed above, specifically by training the transmission network and the reconstruction network separately.

[0159] For details regarding force testing and simulation, refer to the statements above regarding the training of the transmission network and the training of the reconstructed network.

[0160] According to one embodiment, the force test used to train the feedforward neural network is performed using multiple indenters, each with a corresponding indenter shape. The shape may be specifically related to the portion of the indenter that contacts the measurement surface in the force test. Thus, an indenter is an object used to apply force to the measurement surface in a force test.

[0161] According to one embodiment, the indenter shape is selected from a group comprising at least a pointed tip, a circle, a triangular cross-section, a square cross-section, a hemisphere, a cube, and a cylinder. Such indenter shapes have proven suitable because they correspond to the typical shape of the object contacting the measurement surface in the application. Using these different indenter shapes can significantly improve the training of the feedforward neural network to reconstruct the corresponding or similar shape applied to the measurement surface. It should be noted that each of the mentioned shapes can be used, only one of the mentioned shapes can be used, or a selection of the mentioned shapes can be used. Alternatively or additionally, other shapes can be used.

[0162] According to one implementation, simulations are performed using simulated forces based on simulated indenters, where the shapes of the simulated indenters correspond to the shapes of the actual indenters used in the respective force tests. This ensures optimal correspondence between the force tests and the simulations, allowing the feedforward neural network to be trained ideally.

[0163] According to one embodiment, the feedforward neural network is trained using multiple different indenter shapes. This trains the feedforward neural network to distinguish the forces generated by different indenter shapes.

[0164] According to one embodiment, the feedforward neural network is trained using multiple indenters of different sizes. Alternatively, or as an alternative, using different shapes, this can train the feedforward neural network to distinguish between indenters or other objects applying different magnitudes of force. For example, contact portions of different sizes relative to the measuring surface can be used.

[0165] According to one embodiment, the feedforward neural network is trained with the indenter, applying a corresponding shear force at least for a portion of the force test used to train the feedforward neural network. This trains the feedforward neural network to distinguish between different shear forces applied to the measurement surface. In particular, multiple force tests can be performed using different shear forces or shear force components.

[0166] According to one embodiment, the measured forces each include a normal force component, a first shear force component, and a second shear force component. Therefore, the measured forces can provide information about these components. It should be noted that in typical embodiments according to the prior art, shear force cannot be measured. However, it has been shown that when a feedforward neural network with force testing is trained using such a simulated force vector having the aforementioned components, shear force can be reconstructed in addition to the normal force. This provides additional information that is valuable in several applications, such as in robotic applications used to control robot tips.

[0167] According to one embodiment, in the measured force, the first shear force component corresponds to a first shear force, and the second shear force component corresponds to a second shear force. Specifically, the first shear force is perpendicular to the second shear force. This provides easily accessible information due to the perpendicular direction of the shear force.

[0168] It should be noted that the measured force may alternatively have more or fewer than three components.

[0169] According to one embodiment, the feedforward neural network is trained using multiple forces with different shear force components. This trains the feedforward neural network to distinguish between different shear forces applied to the measurement surface. The shear force may vary between different components of the force used in a single force test and / or between different force tests.

[0170] According to one embodiment, the feedforward neural network is trained using multiple forces with different normal force components. This trains the feedforward neural network to distinguish between different normal forces applied to the measurement surface. The normal force can vary particularly between different force tests.

[0171] According to one embodiment, the force applied by the indenter is measured using a force sensor located in or near the indenter. This force sensor measures the force applied by the indenter to a measuring surface. In particular, it can measure the three components of the force, as described above. Positioning the force sensor near the indenter may include: positioning the force sensor such that it contacts the indenter and / or positioning it between the indenter and the object on which the indenter is mounted.

[0172] According to one implementation, each simulated force vector includes a normal force component, a first shear force component, and a second shear force component. This can specifically correspond to a measured force. Therefore, in force testing, simulated forces can be used to simulate forces corresponding to those actually applied.

[0173] The following text describes aspects relevant to the actual process of force inference, rather than primarily aspects relevant to training.

[0174] According to one implementation, the pressure values ​​on which the calculated force map is based are read out simultaneously or over a predefined time period. This ensures that all pressure values ​​relate to the same force application.

[0175] According to a typical implementation, the force diagram includes at least 0.25 force vectors per square millimeter, at least 0.5 force vectors per square millimeter, at least 0.75 force vectors per square millimeter, at least 1 force vector per square millimeter, at least 1.5 force vectors per square millimeter, or at least 2 force vectors per square millimeter.

[0176] According to a typical implementation, the force diagram includes a maximum of 0.25 force vectors per square millimeter, a maximum of 0.5 force vectors per square millimeter, a maximum of 0.75 force vectors per square millimeter, a maximum of 1 force vector per square millimeter, a maximum of 1.5 force vectors per square millimeter, or a maximum of 2 force vectors per square millimeter.

[0177] These force vector densities have proven suitable for typical applications because they provide sufficient resolution and can be obtained with widely available computational power. Each lower value can be combined with each higher value to form a suitable range. Furthermore, other force vector densities can be used.

[0178] According to a typical implementation, the force diagram includes at least 500, at least 1000, or at least 2000 force vectors. According to a typical implementation, the force diagram includes at most 1000, at most 2000, at most 3000, or at most 4000 force vectors. Such an implementation can be used, for example, in situations where the sensor arrangement is at the tip of a robot approximately the size of a human.

[0179] According to a preferred embodiment, each force vector includes a normal force component, a first shear force component, and a second shear force component. This allows the force map to provide suitable three-dimensional information.

[0180] Specifically, the first shear force component may correspond to a first shear force, and the second shear force component may correspond to a second shear force. The first shear force may, in particular, be perpendicular to the second shear force. This allows the force diagram to provide appropriate shear force information for the applied force.

[0181] According to one embodiment, the method for force inference further includes: reading temperature values ​​from the plurality of barometric pressure sensors and providing temperature information or a temperature map of the sensor arrangement based on the temperature values. This can provide additional temperature information that can be used, for example, in robot control applications. The temperature measurement functionality present in the plurality of barometric pressure sensors can be used for this purpose.

[0182] It should be noted that when the method includes both force maps and simulated force maps, the force maps are typically related to the sensor arrangement, while the simulated force maps are related to the finite element model. A statement used for one force map can generally be applied to both force maps.

[0183] The following sections describe different methods used to train the network. These methods are not part of the force inference method but are performed separately to train the network. For each feature, refer to the statements already given above regarding the methods for training the network and force inference to avoid repetition.

[0184] This invention relates to a method for training a reconstructed network.

[0185] The reconstruction network maps virtual sensors from a finite element model of a sensor arrangement to a force map. The sensor arrangement includes multiple pressure sensors and a compliant layer covering the multiple pressure sensors and providing a measurement surface. The force map includes multiple force vectors.

[0186] - Each virtual sensor includes one or more virtual sensor points, and each virtual sensor point has a virtual sensor point value.

[0187] -The reconstructed network is trained through the following steps:

[0188] - Multiple simulations are performed in the finite element model. Each simulation involves simultaneously applying one or more simulated forces to the simulated measurement surface of the finite element model, thereby calculating a simulated force map on the simulated measurement surface. The simulated force map includes multiple simulated force vectors, and the corresponding virtual sensor point values ​​are calculated using the finite element model.

[0189] - Train the reconstructed network using the calculated simulated force map and the corresponding calculated virtual sensor point values.

[0190] According to one embodiment, the simulated force applied to the simulated measurement surface is generated based on individual simulated indenters having simulated indenter shapes.

[0191] According to one embodiment, the simulated indenter shape is selected from the group consisting of at least a pointed tip, a circle, a triangular cross-section, a square cross-section, a hemisphere, a cube, and a cylinder.

[0192] According to one implementation, the reconstruction network is trained using multiple different simulated indenter shapes.

[0193] According to one implementation, the reconstructed network is trained using simulated pressure heads of various sizes.

[0194] According to one implementation, the reconstructed network is trained with at least a portion of simulation, which includes simultaneously applying simulated forces based on two or more simulated pressure heads.

[0195] According to one implementation, the reconstructed network is trained with at least a portion of simulation, which includes applying a simulated force generated based solely on a simulated pressure head.

[0196] According to one implementation, each simulated force vector includes a normal force component, a first shear force component, and a second shear force component.

[0197] According to one embodiment, in the simulated force vector, the first shear force component corresponds to a first shear force, and the second shear force component corresponds to a second shear force, wherein the first shear force is perpendicular to the second shear force.

[0198] According to one implementation, the reconstructed network is trained using multiple simulated forces with different shear force components.

[0199] According to one implementation, the reconstructed network is trained using multiple simulated forces with different normal force components.

[0200] According to one implementation, the reconstructed network is used in the method described above with respect to the use of a transport network and a reconstructed network.

[0201] According to their respective implementation methods

[0202] The force diagram contains at least 0.25 force vectors per square millimeter, at least 0.5 force vectors per square millimeter, at least 0.75 force vectors per square millimeter, at least 1 force vector per square millimeter, at least 1.5 force vectors per square millimeter, or at least 2 force vectors per square millimeter.

[0203] and / or

[0204] - The force diagram contains up to 0.25 force vectors per square millimeter, up to 0.5 force vectors per square millimeter, up to 0.75 force vectors per square millimeter, up to 1 force vector per square millimeter, up to 1.5 force vectors per square millimeter, or up to 2 force vectors per square millimeter.

[0205] According to one embodiment, each force vector includes a normal force component, a first shear force component, and a second shear force component.

[0206] According to one embodiment,

[0207] - The first shear force component corresponds to a first shear force, and the second shear force component corresponds to a second shear force, and

[0208] - The first shear force is perpendicular to the second shear force.

[0209] This also applies to simulated force diagrams and their simulated force vectors.

[0210] This invention relates to a method for training a transmission network.

[0211] -The transmission network maps multiple barometric pressure sensors of the sensor arrangement to multiple virtual sensors in the finite element model of the sensor arrangement, the sensor arrangement including multiple barometric pressure sensors and a compliant layer covering the multiple barometric pressure sensors and providing a measuring surface.

[0212] - Each virtual sensor includes one or more virtual sensor points, and each virtual sensor point has a virtual sensor point value.

[0213] -The transmission network is trained through the following steps:

[0214] - Perform multiple force tests on the sensor arrangement. Each force test involves applying a force at a location on the measuring surface of the sensor arrangement via an indenter, simultaneously measuring the force applied by the indenter and simultaneously measuring the pressure value using the multiple barometric pressure sensors.

[0215] For each force test, a corresponding simulation is performed using the finite element model. Each simulation involves applying a simulated force to the simulated measurement surface of the finite element model, thereby calculating a simulated force map on the simulated measurement surface. The simulated force map includes multiple simulated force vectors, where the simulated forces correspond to the measured forces, and the positions applied to the simulated measurement surface correspond to the positions on the measured surface. The corresponding virtual sensor point values ​​are then calculated using the finite element model.

[0216] - The transmission network is trained using the measured pressure values ​​and the corresponding calculated virtual sensor point values.

[0217] According to one embodiment, the force test for training the transmission network is performed using multiple indenters, each having its own indenter shape.

[0218] According to one embodiment, the shape of the indenter is selected from the group consisting of at least a pointed tip, a circle, a triangular cross-section, a square cross-section, a hemispherical shape, a cube, and a cylinder.

[0219] According to one embodiment, simulations are performed using simulated forces based on the simulated indenter, wherein the shapes of the simulated indenters correspond to the shapes of the actual indenters used in the corresponding force tests.

[0220] According to one embodiment, the transmission network is trained using multiple different pressure head shapes.

[0221] According to one embodiment, the transmission network is trained using multiple pressure heads of different sizes.

[0222] According to one embodiment, the transmission network is trained by applying a corresponding shear force to an indenter used for at least a portion of force tests for training the transmission network.

[0223] According to one embodiment, the measured forces each include a normal force component, a first shear force component, and a second shear force component.

[0224] According to one embodiment, in the measured force, the first shear force component corresponds to a first shear force, and the second shear force component corresponds to a second shear force, wherein the first shear force is perpendicular to the second shear force.

[0225] According to one embodiment, the transmission network is trained using multiple forces with different shear force components.

[0226] According to one implementation, the transmission network is trained using multiple forces with different normal force components.

[0227] According to one embodiment, the force applied by the indenter is measured using a force sensor located in or near the indenter.

[0228] According to one implementation, each simulated force vector includes a normal force component, a first shear force component, and a second shear force component.

[0229] According to one embodiment, the transmission network is used in the above-described methods for using and reconstructing the transmission network.

[0230] This invention relates to a method for training a feedforward neural network.

[0231] -The feedforward neural network calculates a force map on the measurement surface of a sensor arrangement based on pressure values ​​from barometric pressure sensors. The sensor arrangement includes multiple barometric pressure sensors and a compliant layer covering the multiple barometric pressure sensors and providing the measurement surface. The force map includes multiple force vectors.

[0232] -The feedforward neural network is trained through the following steps:

[0233] - Perform multiple force tests on the sensor arrangement. Each force test involves applying a force at a location on the measuring surface of the sensor arrangement via a pressure head, simultaneously measuring the force applied by the pressure head and simultaneously measuring the pressure value using the multiple barometric pressure sensors.

[0234] For each force test, a corresponding simulation is performed using a finite element model arranged with the sensors. Each simulation involves applying a simulated force to a simulated measurement surface of the finite element model, thereby calculating a simulated force map on the simulated measurement surface. The simulated force map includes multiple simulated force vectors, where the simulated forces correspond to the measured forces, and the positions where the simulated forces are applied on the simulated measurement surface correspond to the positions on the measurement surface.

[0235] - The feedforward neural network is trained using the measured pressure values ​​and the corresponding computational simulation force diagram.

[0236] According to one embodiment, the force test used to train the feedforward neural network is performed using multiple indenters, each having a corresponding indenter shape.

[0237] According to one embodiment, the indenter shape is selected from the group consisting of at least a pointed tip, a circle, a triangular cross-section, a square cross-section, a hemispherical shape, a cube, and a cylinder.

[0238] According to one embodiment, the simulation is performed using the simulated force based on the simulated indenter, wherein the shape of each simulated indenter corresponds to the shape of the actual indenter used in the corresponding force test.

[0239] According to one implementation, the feedforward neural network is trained using multiple different indenter shapes.

[0240] According to one implementation, the feedforward neural network is trained using multiple pressure heads of different sizes.

[0241] According to one embodiment, the feedforward neural network is trained with an indenter by applying a corresponding shear force, at least for a portion of the force test used to train the feedforward neural network.

[0242] According to one embodiment, the measured forces each include a normal force component, a first shear force component, and a second shear force component.

[0243] According to one embodiment, in the measured force, the first shear force component corresponds to a first shear force, and the second shear force component corresponds to a second shear force, wherein the first shear force is perpendicular to the second shear force.

[0244] According to one embodiment, the feedforward neural network is trained using multiple forces with different shear force components.

[0245] According to one implementation, the feedforward neural network is trained using multiple forces with different normal force components.

[0246] According to one embodiment, a force sensor located in or near the indenter is used to measure the force.

[0247] According to one implementation, each simulated force vector includes a normal force component, a first shear force component, and a second shear force component.

[0248] According to one embodiment, in the simulated force vector, the first shear force component corresponds to a first shear force, and the second shear force component corresponds to a second shear force, wherein the first shear force is perpendicular to the second shear force.

[0249] According to one embodiment, the feedforward neural network is used in the force inference method described above.

[0250] According to their respective implementation methods

[0251] The force diagram contains at least 0.25 force vectors per square millimeter, at least 0.5 force vectors per square millimeter, at least 0.75 force vectors per square millimeter, at least 1 force vector per square millimeter, at least 1.5 force vectors per square millimeter, or at least 2 force vectors per square millimeter.

[0252] and / or

[0253] - The force diagram contains up to 0.25 force vectors per square millimeter, up to 0.5 force vectors per square millimeter, up to 0.75 force vectors per square millimeter, up to 1 force vector per square millimeter, up to 1.5 force vectors per square millimeter, or up to 2 force vectors per square millimeter.

[0254] According to one embodiment, each force vector includes a normal force component, a first shear force component, and a second shear force component.

[0255] According to one embodiment, the first shear force component corresponds to a first shear force, and the second shear force component corresponds to a second shear force, wherein the first shear force is perpendicular to the second shear force.

[0256] The following describes in detail a sensor arrangement to which the methods disclosed herein can be applied. Further reference can be made to the details or explanations of this sensor arrangement given herein, which can then be applied accordingly.

[0257] In particular, in the methods disclosed herein, the sensor arrangement may be a sensor arrangement for sensing force, said sensor arrangement comprising:

[0258] - Flexible circuit board

[0259] - Multiple barometric pressure sensors are mounted on a flexible circuit board.

[0260] - A rigid core, the flexible circuit board being wrapped around and mounted onto the rigid core such that the flexible circuit board at least partially covers the rigid core, the plurality of pressure sensors protruding from the rigid core, and

[0261] - A flexible layer covers the multiple pressure sensors and provides a measurement surface.

[0262] However, it should be noted that the concepts of force inference and training disclosed herein can also be applied to other sensor arrangements. This particularly involves using different indenter sizes, different indenter shapes, different shear forces, and / or different shear force components. These concepts can be generalized.

[0263] According to one embodiment, the rigid core is dome-shaped.

[0264] According to one embodiment, the rigid core has a plurality of facets, wherein each pressure sensor is positioned on one of the plurality of facets.

[0265] According to one embodiment, the flexible layer comprises or is composed of a plastic material or rubber.

[0266] According to one embodiment, the plastic material is a thermoplastic, an elastomer, a thermoplastic elastomer, or a thermosetting plastic.

[0267] According to one embodiment, the compliant layer transmits the force applied to the measuring surface to at least a portion of the plurality of barometric pressure sensors.

[0268] According to one embodiment, the plurality of pressure sensors are connected via conductor paths on the flexible circuit board.

[0269] According to one embodiment, the flexible circuit board is asterisk-shaped.

[0270] According to one embodiment, the flexible circuit board includes a plurality of arms connected to a central portion.

[0271] According to their respective implementations, the plurality of barometric pressure sensors are arranged at a distance of at least 1 mm, at least 2 mm, at least 3 mm, at least 4 mm, or at least 5 mm and / or at a distance of at most 1 mm, at most 2 mm, at most 3 mm, at most 4 mm, or at most 5 mm.

[0272] According to one embodiment, the sensor arrangement is the robot tip and / or the robot's manipulation element.

[0273] According to one embodiment, the rigid core is a 3D printed part.

[0274] The present invention also relates to a force inference module for force inference of a sensor arrangement for sensing force, the force inference module being configured to perform the methods as disclosed herein. All embodiments and variations of the methods can be applied.

[0275] This invention relates to a sensor arrangement for sensing force, said sensor arrangement comprising one, some, or all of the following:

[0276] - Flexible circuit board

[0277] Multiple barometric pressure sensors are mounted on the flexible circuit board.

[0278] - A rigid core, the flexible circuit board being wrapped around and mounted onto the rigid core such that the flexible circuit board at least partially covers the rigid core, the plurality of pressure sensors protruding from the rigid core.

[0279] - A flexible layer covers the multiple barometric pressure sensors and provides a measuring surface, and

[0280] -The force inference module according to the present invention.

[0281] Regarding sensor arrangements that include force inference modules, all embodiments and variations of force inference modules and sensor arrangements and their components can be applied, especially those described herein. Attached Figure Description

[0282] Further aspects and advantages will be apparent to those skilled in the art from the following description based on the accompanying drawings. These show:

[0283] Figure 1 Sensor placement,

[0284] Figure 2 Rigid core

[0285] Figure 3 Flexible circuit board

[0286] Figure 4 A rigid core with a flexible circuit board.

[0287] Figure 5 : Exploded view of the mold

[0288] Figure 6 The assembled mold,

[0289] Figure 7 : An exploded view of a mold having a rigid core covered by a flexible circuit board.

[0290] Figure 8 The mold and the rigid core with a flexible circuit board covering the pressure sensor.

[0291] Figure 9 Schematic diagram of force inference.

[0292] Figure 10 Finite element model

[0293] Figure 11 Several different pressure heads,

[0294] Figure 12 Arrangements for conducting force tests.

[0295] Figure 13 : A flowchart of the transmission network training process

[0296] Figure 14 : Reconstruct the flowchart of the network training process,

[0297] Figure 15 : A flowchart of the feedforward neural network training process, and

[0298] Figure 16 : A diagram of an attempt. Detailed Implementation

[0299] Figure 1 A sensor arrangement 10 according to an embodiment of the present invention is shown.

[0300] The sensor arrangement 10 includes a dome-shaped rigid core 100. The rigid core 100 is partially covered by a flexible circuit board 300, which is fixedly mounted on the rigid core 100. The flexible circuit board 300 is covered by a compliance layer 200.

[0301] Multiple barometric pressure sensors 400 are applied to the flexible circuit board 300. They protrude from the rigid core 100. The compliant layer 200 provides a measuring surface 210 on which forces can be applied. The compliant layer 200 is flexible and resilient, such that forces applied to the measuring surface 210 cause localized deformation of the measuring surface 210, wherein the compliant layer 200 transmits these forces to at least a portion of the multiple barometric pressure sensors 400. Therefore, the multiple barometric pressure sensors 400 can be used to evaluate forces or applied forces.

[0302] The flexible circuit board 300 includes a plurality of facets. These facets correspond to facets constructed on the rigid core 100, as detailed below. Figure 2 As shown.

[0303] The flexible circuit board 300 has a central portion 305, from which, in the current embodiment, six arms extend in a manner. The central portion 305 can be considered a facet. All the arms are as follows: Figure 3 As shown. In Figure 1 In the figure, only three arms, namely the first arm 310, the second arm 320 and the third arm 330, are visible and are indicated by the reference numerals in the figure.

[0304] Each arm is divided into three facets; for example, the first arm 310 is divided into a first facet 311, a second facet 312, and a third facet 313. The other arms are correspondingly divided, wherein the facets 321, 322, 323, 331, 332, and 333 of the flexible circuit board 300 are... Figure 1 As can be seen in the text.

[0305] In the current embodiment, each facet holds a pressure sensor 400. Furthermore, the central portion 305 holds a pressure sensor 400. It should be noted that other configurations are also possible; for example, the facets may include more than one pressure sensor 400, or none at all.

[0306] It should be noted that the pressure sensors 400 are spaced apart on the flexible circuit board 300. However, finer resolution regarding the applied force can be achieved using the techniques described below.

[0307] Figure 2 The rigid core 100 is shown separately. The rigid core 100 comprises a total of six surface regions, of which the first surface region 110, the second surface region 120, and the third surface region 130 are visible and... Figure 2 The diagram shows that each surface region 110, 120, and 130 is divided into three facets. For example, the first surface region 110 is divided into a first facet 111, a second facet 112, and a third facet 113. Other surface regions are correspondingly divided, with facets 121, 122, 123, 131, 132, and 133... Figure 2 As can be seen, the central portion 105 connects to the surface area at the top of the rigid core 100.

[0308] The facets of the rigid core 100 define the facets of the flexible circuit board 300. Specifically, the facets have different orientations, and the flexible circuit board 300 is adapted to each orientation of the facets.

[0309] exist Figure 2 The image also clearly shows that the rigid core 100 is dome-shaped, for example, it could be used for the fingertips of a robot.

[0310] Figure 3The flexible circuit board 300, on which multiple of the aforementioned pressure sensors 400 are mounted, is shown separately. As previously described, the flexible circuit board 300 has six arms 310, 320, 330, 340, 350, and 360, which are connected together at the central portion 305. In the current embodiment, a total of 19 pressure sensors 400 are mounted on the flexible circuit board 300. In other embodiments, more or fewer pressure sensors may be used.

[0311] It should be noted that, Figure 3 The facets are not shown because they are not an inherent feature of the flexible circuit board 300. Figure 1 The small side of the flexible circuit board 300 shown is where the flexible circuit board 300 is mounted. Figure 2 The results on the rigid core 100 shown.

[0312] It should be noted that corresponding holes 315, 325, 335, 345, 355, and 365 are provided in each arm 310, 320, 330, 340, 350, and 360, which can be used, for example, to fasten the flexible circuit board 300 to the rigid core 100 during manufacturing.

[0313] Figure 4 Showing Figure 3 The flexible circuit board 300 is mounted on Figure 2 The rigid core 100 is thus formed on the flexible circuit board 300, which has acquired the structure of the rigid core 100. Figure 4 The layout shown is not yet complete. Figure 1 The softening layer 200 is shown below. The following description, in conjunction with the accompanying drawings, explains how the softening layer 200 and its measuring surface 210 are formed.

[0314] Figure 5 An exploded view of mold 500 is shown. Mold 500 includes a first component 510 and a second component 520. (As shown...) Figure 5 As shown, a hollow interior 530 is formed inside components 510 and 520, such that when components 510 and 520 are assembled, the hollow interior 530 is only open at the top of the mold 500. Furthermore, the mold 500 includes a top 540 to secure the arrangement of the rigid core and the flexible circuit board mounted thereon, as shown. Figure 4 As shown.

[0315] Figure 6 The mold 500 is shown in an assembled state. Therefore, the hollow interior 530 is only open at the top of the mold 500, and the top 540 extends across the hollow interior 530.

[0316] Figure 7 The diagram shows the arrangement of the already explained mold 500, the rigid core, the flexible circuit board, and the plurality of pressure sensors mounted thereon. Figure 7 An exploded view is shown, while Figure 8 The same parts are shown in their assembled state. Figure 8 In the indicated state, the rigid core 100 is mounted to the top 540 of the mold, and the rigid core 100 protrudes downward from the top 540 into the hollow interior 530.

[0317] exist Figure 8 In the indicated state, material, such as plastic, can be fluidized into the hollow interior 530. This is easily handled due to the fluid properties. The material can be filled into the hollow interior 530 such that the flexible circuit board 300 and the rigid core 100 are covered by the material at a level corresponding to the position where the compliant layer 200 should cover the flexible circuit board 300 and the rigid core 100. The surface of the hollow interior 530 defines the measurement surface 210 in the final state.

[0318] After filling with material, the mold 500, rigid core 100, flexible circuit board 300 mounted thereon, and the filled material are placed in a vacuum chamber. The vacuum chamber is evacuated, and the material is degassed. Through degassed material, the material is transformed into the flexible layer 200, thereby manufacturing... Figure 1 The sensor arrangement 10 shown is illustrated.

[0319] The process illustrated in these figures is the manufacturing process for sensor arrangement 10, which requires only a few specific components and is easy to perform. Therefore, costs can be significantly reduced compared to more expensive embodiments known in the prior art.

[0320] Figure 9 A schematic diagram of a method for force inference for a sensor arrangement 10 (e.g., the sensor arrangement 10 described above) is shown. As already mentioned, the sensor arrangement 10 includes a plurality of pressure sensors 400. These pressure sensors 400 generate respective pressure values ​​R1, R2, ..., Rx as their respective output values, indicating the pressure sensed by the respective pressure sensor 400 at a location below the fabrication layer 200.

[0321] This pressure value R forms the input to the transfer network TN, which is a neural network that maps the pressure sensor 400 to multiple virtual sensors in the finite element model 10a of the sensor arrangement 10. (Refer to the following...) Figure 10The virtual sensors are further described below. Each virtual sensor includes one or more virtual sensor points, each virtual sensor point having virtual sensor point values ​​S1, S2, ..., Sx. This will be further referenced below. Figure 10 Further detailed description.

[0322] The transmission network TN maps pressure values ​​R to virtual sensor point values ​​S, meaning that the transmission network TN provides a set of virtual sensor point values ​​S as output for each combination of pressure values ​​R that it takes as input. This requires training the transmission network TN, which can be done in particular as described herein.

[0323] The virtual sensor point values ​​S1, S2, ..., Sx form the input to the reconstruction network RN, which is a neural network that maps the virtual sensors of the finite element model 10a to the force map FM. The force map FM includes multiple force vectors F1, F2, ..., Fx, each of which has three components: a normal force component and two perpendicular shear force components. Therefore, each force vector F gives the value and direction of the force applied to a specific point on the measurement surface 210. The force map FM further refers to... Figure 16 explain.

[0324] The reconstruction network RN maps virtual sensor point values ​​S to the force map FM, meaning that the reconstruction network RN provides a set of force vectors F as outputs for each combination of virtual sensor point values ​​S that it takes as input. This requires training the reconstruction network RN, which can be done in particular as described herein.

[0325] The transmission network TN and the reconstruction network RN together form a feed-forward neural network (FFNN), which can be viewed as a neural network that maps the barometric pressure sensor 400 to the force FM. It is divided into two parts, as shown and explained.

[0326] To train the transmission network TN, method T1 can be used. To train the reconstruction network RN, method T2 can be used. To train the entire feedforward neural network FFNN, method T3 can be used. These methods are further described below.

[0327] Using neural networks or artificial intelligence as a generalization, more information can be extracted from barometric pressure sensors than from direct force inference without AI. Specifically, the applied force can be assessed with a resolution much larger than the 400-pixel spacing of a barometric pressure sensor. Furthermore, additional information, such as shear force and / or the amount and location of the applied pressure head, can be extracted. This information is contained in the force map FM calculated based on the pressure value R.

[0328] Figure 10 A finite element model 10a of the sensor arrangement 10 is shown. The finite element model 10a is used for... Figure 9 In the process of force deduction described, it should be noted that... Figure 10 The diagram illustrates structural details of the sensor arrangement 10, but does not provide specific details of the implementation of the finite element calculations, as such a finite element concept relies on known techniques. In principle, the finite element model 10a is an electronic representation of the actual sensor arrangement 10, and thus the behavior of the sensor arrangement 10 can be simulated using the finite element model 10a.

[0329] All components of the sensor arrangement 10 have corresponding components in the finite element model 10a, where the components in the finite element model 10a are represented by the letter "a". The structural difference between the sensor arrangement 10 and the finite element model 10a is that the virtual sensor 400a in the finite element model 10a replaces the barometric pressure sensor 400 in the sensor arrangement 10. Each virtual sensor 400a includes one or more sensor points 410a, and an embodiment in which each virtual sensor 400a includes 12 virtual sensor points 410a is shown. Each virtual sensor point 410a has a corresponding virtual sensor point value S, as already specified for... Figure 9 The above is discussed. However, it is also possible to use a different number of virtual sensor points 410a for each virtual sensor 400a.

[0330] Therefore, the simulated force 605a applied to the simulated measurement surface 210a of the finite element model 10a is transmitted to the virtual sensor 400a and its virtual sensor point 410a through the finite element representation of the compliance layer 200 (i.e., the simulated compliance layer 200a). This transmitted force produces a corresponding virtual sensor point value S. This can be used to perform simulations, giving a corresponding virtual sensor point value S for each applied simulated force 605a or combination of simulated forces 605a.

[0331] This simulated force 605a can be applied by the simulated pressure head 600a, wherein Figure 10 Two such simulated pressure heads 600a are shown as examples. Using these simulated pressure heads 600a, simulated forces can be applied to the simulated measurement surface 210a, and the virtual sensor point value S can be calculated using standard finite element model methods.

[0332] Data obtained from such simulations can be used to train the reconstructed network RN, typically using multiple such simulations, such as 1,000 or approximately 10,000 simulations. These simulations usually employ different types of simulation indenters 600a, particularly those with different shapes and / or sizes, and varying numbers of simulation indenters 600a, such as one, two, and / or three indenters 600a. Such simulations can be performed purely by computer simulation, requiring no complex experimental setup. This allows for highly efficient and reliable training of the reconstructed network RN, thus enabling it to achieve greater reconstruction efforts and FM capabilities even with limited experimental capabilities.

[0333] Figure 11 The shapes of four different indenters 600 are schematically shown. These can be physical indenters 600 used in experimental setups, as described below. Figure 12 Further description, or perhaps analogous to pressure head 600a.

[0334] Figure 11 a shows an indenter 600 with a flat shape in contact with the measuring surface 210. Figure 11 b shows the indenter 600 with a pointed contact portion. Figure 11 c shows an indenter 600 with a contact portion shaped like a hemisphere. Figure 11 d shows that it has the same as Figure 11 The indenter 600 shown in c is of the same type but has a smaller contact portion. Using this different indenter 600 allows for optimization of neural network training for this different shape, meaning that the neural network trained with this different indenter 600 is better able to reconstruct the force applied by an indenter 600 with a different shape compared to a similar indenter 600. In other words, the force map FM reconstructed after applying an indenter 600 with a flat shape will differ from the force map FM reconstructed after applying an indenter 600 with a hemispherical shape.

[0335] Figure 12 An experimental apparatus 700 for performing force tests is shown. The apparatus 700 includes a base 710, on which a first robotic arm 720 is mounted. A positioning joint 730 is located on the first robotic arm 720. A second robotic arm 740 is fixed to the joint 730. The joint 730 is usable for actively moving the second robotic arm 740, wherein an electric drive (not shown) is used for this movement.

[0336] At the other end of the second robotic arm 740, a sensor arrangement 10 as described above is positioned. This is shown schematically only, wherein the outer surface of the sensor arrangement 10 is the measurement surface 210 already described.

[0337] The experimental apparatus 700 also includes a top 750 at which a force sensor 610 is mounted. A pressure head 600 is mounted at the force sensor 610. An articulation 730 can be used to press the sensor arrangement 10 onto the pressure head 600, wherein during such a force test, a pressure value R is read from a pressure sensor 400, and a force 605 applied by the pressure head 600 to the measuring surface 210 is measured using the force sensor 610. The force sensor 610 measures three-dimensional forces, thereby measuring the normal force component and the shear force component. The three-dimensional forces can be represented in a global coordinate system, or they can be represented by a normal component perpendicular to a point on the measuring surface 210 and two shear force components that are generally perpendicular to the normal component and generally perpendicular to each other. If the components are known in another coordinate system, coordinate transformations can be used to calculate the components in that coordinate system.

[0338] The position of the indenter 600 contacting the measuring surface 210 is observed by the camera 620. This allows the coordinates of the position on the measuring surface 210 to be calculated via image recognition. Alternatively, machine parameters can be used, for example, to calculate such a position.

[0339] The indenter 600 is fixed, while the sensor arrangement 10 is movable within the experimental apparatus 700, allowing the use of known articulation devices (e.g., from a 3D printer). However, it should be noted that force testing can be performed alternatively in different ways, such as by moving the indenter 600 with the fixed sensor arrangement 10, or by moving both the sensor arrangement 10 and the indenter 600.

[0340] Data from such force tests can be used for training. Figure 9 The neural network shown below will be described further below.

[0341] Figure 13 A schematic diagram of method T1 for training the transmission network TN is shown.

[0342] In the first step T1_1, as about Figure 12 Multiple force tests are performed as described. For such force tests, different indenters 600 with different shapes and / or sizes are preferably used, wherein only one indenter 600 is used in each force test in the described embodiment.

[0343] In step T1_2, multiple simulations are performed using the finite element model 10a, with one simulation performed for each force test. The force 605 measured by the force sensor 610 in the force test is used in the corresponding simulation to apply the simulated force 605a. The position on the simulated measurement surface 210a is the same as the position on the measurement surface 210 in the force test, where such a position can be calculated, for example, from machine parameters or derived from image recognition, as already referenced. Figure 12 As described, the shape of the simulated indenter 600a is identical to that of the actual indenter 600. In each force test, the virtual sensor point value S is calculated based on the applied simulated force 605a using standard finite element simulation.

[0344] In step T1_3, the transmission network TN is trained with data obtained through force testing and simulation, particularly using pressure values ​​R from the barometric pressure sensor 400 from the force test and calculated virtual sensor point values ​​S from the corresponding simulation.

[0345] Figure 14 Method T2 for training the reconstructed network RN is shown.

[0346] In the first step T2_1, multiple simulations are performed using the finite element model 10a, preferably using multiple different numbers of indenters, and more preferably using multiple different indenter shapes and sizes. In each simulation, the simulated force diagram FMa is calculated on the simulated measurement surface 210a, and the corresponding virtual sensor point value S is calculated.

[0347] Using such a simulated force map FMa and virtual sensor point values ​​S, the transmission network TN is trained in step T2_2 so that it can reconstruct the force map from the simulated sensor point values ​​S.

[0348] Figure 15 Method T3 for training the entire feedforward neural network FFNN is shown.

[0349] In the first step T3_1, multiple force tests are performed, such as those related to... Figure 12 As explained. These force tests transmit the applied force 605 (as measured by the force sensor 610), the corresponding position, and the measured pressure value R of the pressure sensor 400.

[0350] In the second step T3_2, multiple corresponding simulations are performed using the finite element model 10a of the sensor arrangement 10. Each simulation involves applying a simulated force 605a to the simulated measurement surface 210a of the finite element model 10, at the same position as the actual position on the measurement surface 210, and the simulated indenter 600a has the same indenter shape as the real indenter 600. Thus, a simulated force diagram FMa is calculated on the simulated measurement surface 210a.

[0351] In a further step T3_3, the measured pressure value R from the force test and the corresponding simulated force map FMa derived from the simulation are used to train the entire feedforward neural network (FFNN), wherein, in the implementation shown, both the transmission network TN and the reconstruction network RN are trained.

[0352] It should be noted that, even when using only one neural network, it is also possible to use information about... Figure 15 The described process refers to the absence of a split between the transport network TN and the reconfiguration network RN. Figure 9 In the implementation scenario shown, except for the one regarding Figure 13 and 14 In addition to the methods described, it is also possible to perform actions related to... Figure 15 The described method optimizes the transport network TN and the reconfigured network RN.

[0353] Figure 16 The sensor arrangement 10 is shown in schematic diagram with a force map FM. The force map FM includes multiple force vectors F located around the measuring surface 210. Although Figure 16 The diagram shows two force vectors F, but in typical implementations, more force vectors F can be used. For example, in an exemplary embodiment, one force vector F per square millimeter can be used.

[0354] Each force vector F has a normal force component F0. N First shear force component F S1 Second shear force component F S2 The normal force component F N The normal force component of the applied force is given, which is the component perpendicular to the local orientation of the measuring surface 210. The shear force component F... S1 F S2 The shear force values ​​applied to the measuring surface 210 at the corresponding points are given. The shear forces are generally parallel to the local orientation of the measuring surface 210 and are generally perpendicular to each other and perpendicular to the normal force. This may particularly concern the non-deformed orientation of the measuring surface, which can define the orientation of the force vector F, especially the orientation of its normal component.

[0355] Therefore, each force vector F gives the strength and direction of the force applied at a specific point on the measuring surface 210. Such a force may, for example, originate from the indenter 600.

[0356] It should be noted that other definitions of the force vector F can also be used; for example, only the normal force component can be evaluated or the shear force can have an alternative definition.

[0357] In the case of simulating the force diagram FMa, the simulated force vector Fa of this simulated force diagram FMa on the simulated measurement surface 210a can have corresponding simulated components, such as the normal force component F. N a. First shear force component F S1 a and the second shear force component F S2 a. This simulation's attempt at FMa is aimed at... Figure 10 Specifically calculated in the simulation performed on the described finite element model.

[0358] The steps of the method of the present invention can be performed in the given order. However, they can also be performed in other orders, provided that this is technically reasonable. In one embodiment, the method of the present invention can be performed, for example, by a specific combination of steps, without performing further steps. However, other steps, including those not mentioned, can also be performed.

[0359] It should be noted that features may be described in combination in the claims and specification, for example, to provide better understanding, although in fact these features may be used or implemented independently of each other. Those skilled in the art will recognize that such features may be combined with other features or combined with features that are independent of each other.

[0360] References in dependent claims may indicate preferred combinations of features, but do not exclude other combinations of features.

[0361] Reference symbol list:

[0362] 10: Sensor Placement

[0363] 100: Rigid Core

[0364] 105: Central Part

[0365] 110: First surface region

[0366] 111: Small Noodles

[0367] 112: Small noodles

[0368] 113: Small noodles

[0369] 120: Second surface region

[0370] 121: Small Noodles

[0371] 122: Small Noodles

[0372] 123: Small noodles

[0373] 130: Third surface region

[0374] 131: Small Noodles

[0375] 132: Small Noodles

[0376] 133: Small Noodles

[0377] 200: Softening layer

[0378] 210: Measuring surface

[0379] 300: Flexible Circuit Board

[0380] 305: Central Section

[0381] 310: First Arm

[0382] 311: Small Noodles

[0383] 312: Small noodles

[0384] 313: Small Noodles

[0385] 315: Kong

[0386] 320: Second Arm

[0387] 321: Small Noodles

[0388] 322: Small Noodles

[0389] 323: Small Noodles

[0390] 325: Hole

[0391] 330: Third Arm

[0392] 331: Small Noodles

[0393] 332: Small Noodles

[0394] 333: Small Noodles

[0395] 335: Kong

[0396] 340: The Fourth Arm

[0397] 345: Kong

[0398] 350: The Fifth Arm

[0399] 355: Hole

[0400] 360: The Sixth Arm

[0401] 365: Kong

[0402] 400: Barometric Pressure Sensor

[0403] 500: Mold

[0404] 510: First component

[0405] 520: Second component

[0406] 530: Hollow interior

[0407] 540: Top

[0408] 600: Pressure Head

[0409] 605: Strength

[0410] 610: Force Sensor

[0411] 620: Camera

[0412] 700: Experimental Apparatus

[0413] 710: Bottom

[0414] 720: The First Robotic Arm

[0415] 730: Joint

[0416] 740: Second robotic arm

[0417] 750: Top

[0418] 10a: Finite element model

[0419] 210a: Simulated measurement surface

[0420] 400a: Virtual Sensor

[0421] 410a: Virtual sensing point

[0422] 600a: Simulated pressure head

[0423] 605a: Simulated Force

[0424] Other reference numerals with the letter 'a': Components of finite element model 10a; TN: Transmission network.

[0425] RN: Reconstructing Networks

[0426] FFNN: Feedforward Neural Network

[0427] T1: Methods for training transmission networks

[0428] T2: Methods for training and reconstructing networks

[0429] T3: Methods for training feedforward neural networks

[0430] R: Pressure value

[0431] S: Virtual sensor point value

[0432] FM: Striving

[0433] F: Force vector

[0434] FMa: Simulated Force Map

[0435] Fa: Simulated force vector

[0436] F N (Normal force component of a force vector)

[0437] F S1 The first shear force component (of the force vector)

[0438] F S2 The second shear force component (of the force vector)

[0439] F N a: (Normal force component of the simulated force vector)

[0440] F S1 a: (The first shear force component of the simulated force vector)

[0441] F S2 a: (The second shear force component of the simulated force vector)

Claims

1. A sensor arrangement (10) for sensing force, characterized in that: The sensor arrangement (10) includes: A flexible circuit board (300), wherein the flexible circuit board (300) includes a plurality of arms (310, 320, 330, 340, 350, 360) connected at a central portion (305). Multiple barometric pressure sensors (400) are mounted on the flexible circuit board (300). A rigid core (100), wherein the rigid core (100) is dome-shaped, and a flexible circuit board (300) is wrapped around and mounted on the rigid core (100) such that the flexible circuit board (300) at least partially covers the rigid core (100), and the plurality of pressure sensors (400) protrude from the rigid core (100), and A flexible layer (200) covers the plurality of pressure sensors (400) and provides a measuring surface (210).

2. The sensor arrangement (10) according to claim 1, characterized in that: The rigid core (100) has a plurality of facets (111, 112, 113, 121, 122, 123, 131, 132, 133), wherein each pressure sensor (400) is positioned on one of the plurality of facets (111, 112, 113, 121, 122, 123, 131, 132, 133).

3. The sensor arrangement (10) according to claim 1, characterized in that: The flexible layer (200) comprises or is composed of plastic material or rubber.

4. The sensor arrangement (10) according to claim 3, characterized in that: The plastic material is a thermoplastic, elastomer, thermoplastic elastomer, or thermosetting plastic.

5. The sensor arrangement (10) according to claim 1, characterized in that: The compliant layer (200) transmits the force applied to the measuring surface (210) to at least a portion of the plurality of barometric pressure sensors (400).

6. The sensor arrangement (10) according to claim 1, characterized in that: The plurality of pressure sensors (400) are connected via conductor paths on the flexible circuit board (300).

7. The sensor arrangement (10) according to claim 1, characterized in that: The flexible circuit board (300) is asterisk-shaped.

8. The sensor arrangement (10) according to claim 1, characterized in that: The plurality of barometric pressure sensors (400) are arranged at a distance of at least 1 mm, at least 2 mm, at least 3 mm, at least 4 mm or at least 5 mm and / or at a distance of at most 1 mm, at most 2 mm, at most 3 mm, at most 4 mm or at most 5 mm.

9. The sensor arrangement (10) according to claim 1, characterized in that: The sensor arrangement (10) is the robot tip and / or the robot's manipulation element.

10. The sensor arrangement (10) according to claim 1, characterized in that: The rigid core (100) is a 3D printed part.

11. A method for manufacturing a sensor arrangement (10) for sensing force, characterized in that: The method includes the following steps: A flexible circuit board (300) is provided, on which multiple barometric pressure sensors (400) are mounted, wherein the flexible circuit board (300) includes multiple arms (310, 320, 330, 340, 350, 360) connected at a central portion (305). A rigid core (100) is provided, wherein the rigid core (100) is dome-shaped. The flexible circuit board (300) is wrapped around and mounted on the rigid core (100) such that the flexible circuit board (300) at least partially covers the rigid core (100), and the plurality of pressure sensors (400) protrude from the rigid core (100). The plurality of barometric pressure sensors (400) are covered with a flexible layer (200) to provide a measuring surface (210) on the flexible layer (200).

12. The method according to claim 11, characterized in that: Covering the plurality of barometric pressure sensors (400) with the flexible layer (200) includes the following steps: The rigid core (100) having the flexible circuit board (300) is placed in the mold (500). The mold (500) is at least partially filled with material such that the plurality of pressure sensors (400) are covered by the material. The material is then transformed into the flexible layer (200).

13. The method according to claim 12, characterized in that: The conversion includes the following steps: The material is degassed by placing the rigid core (100) with the flexible circuit board (300) covered by the material in a vacuum.

14. The method according to claim 11, characterized in that: Providing the flexible circuit board (300) includes one or two of the following steps: At least a portion of the flexible circuit board (300) is cut from the sheet. The plurality of pressure sensors (400) are arranged and mounted on the flexible circuit board (300).

15. The method according to claim 11, characterized in that: The rigid core (100) has multiple facets (111, 112, 113, 121, 122, 123, 131, 132, 133), and The flexible circuit board (300) is wrapped around and mounted to the rigid core (100) such that each pressure sensor (400) is positioned on one of the plurality of facets (111, 112, 113, 121, 122, 123, 131, 132, 133).

16. The method according to claim 11, characterized in that: The flexible layer (200) comprises or is composed of plastic material or rubber.

17. The method according to claim 16, characterized in that: The plastic material is a thermoplastic, elastomer, thermoplastic elastomer, or thermosetting material.

18. The method according to claim 11, characterized in that: The compliant layer (200) transmits the force applied to the measuring surface (210) to at least a portion of the plurality of barometric pressure sensors (400).

19. The method according to claim 11, characterized in that: The plurality of pressure sensors (400) are connected via conductor paths on the flexible circuit board (300).

20. The method according to claim 11, characterized in that: The flexible circuit board (300) is asterisk-shaped.

21. The method according to claim 11, characterized in that: The plurality of barometric pressure sensors (400) are arranged at a distance of at least 1 mm, at least 2 mm, at least 3 mm, at least 4 mm or at least 5 mm and / or at a distance of at most 1 mm, at most 2 mm, at most 3 mm, at most 4 mm or at most 5 mm.

22. The method according to claim 11, characterized in that: The sensor arrangement (10) is the robot tip and / or the robot's manipulation element.

23. The method according to claim 11, characterized in that: Providing the rigid core (100) includes the following steps: 3D printing of the rigid core (100).

24. The sensor arrangement (10) according to claim 1 or the method according to claim 11, characterized in that: It further includes an electronic control module configured to perform a method for force inference of the sensor arrangement (10).

25. The sensor arrangement (10) according to claim 24, characterized in that: The electronic control module is configured to perform a method for force inference to provide a force map (FM) of the measurement surface (210), the force map (FM) including a plurality of force vectors (F).

26. The sensor arrangement (10) according to claim 25, characterized in that: The force map (FM) includes at least 0.25 force vectors per square millimeter, at least 0.5 force vectors per square millimeter, at least 0.75 force vectors per square millimeter, at least 1 force vector per square millimeter, at least 1.5 force vectors per square millimeter, or at least 2 force vectors per square millimeter. and / or The force diagram includes at most 0.25 force vectors per square millimeter, at most 0.5 force vectors per square millimeter, at most 0.75 force vectors per square millimeter, at most 1 force vector per square millimeter, at most 1.5 force vectors per square millimeter, or at most 2 force vectors per square millimeter.

27. The sensor arrangement (10) according to claim 25, characterized in that: Each force vector includes a normal force component, a first shear force component, and a second shear force component.

28. The sensor arrangement (10) according to claim 27, characterized in that: The first shear force component corresponds to the first shear force, and the second shear force component corresponds to the second shear force. The first shear force is perpendicular to the second shear force.

29. The sensor arrangement (10) according to claim 24, characterized in that: The electronic control module is configured to read temperature values ​​from the plurality of barometric pressure sensors (400) and provide temperature information or a temperature map of the sensor arrangement (10) based on the temperature values.

Images