Six-dimensional force sensor decoupling method and device

Abstract

The invention belongs to the technical field of force sensors, and particularly discloses a six-dimensional force sensor decoupling method and device, and the method comprises the steps: obtaining a system state vector set of carrying equipment where a six-dimensional force sensor is located in a movement process, and original sensor data generated by an output channel in the six-dimensional force sensor; respectively setting a working condition mapping function for compensating the six-dimensional force sensor and a difference compensation function for compensating the six-dimensional force sensor through the system state vector set and the original sensor data; decoupling a linear part and a nonlinear part of a true value corresponding to the original sensor data according to the working condition mapping function and the difference compensation function; the nonlinear part is used as increment correction to be superposed with the linear part, and a superposition result is decoupled to obtain output data of each output channel; the method has the following advantages that layered precise compensation of high dynamic and slowly varying errors is realized, and the measurement precision and the real-time performance are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of force sensors, and in particular, to a six-dimensional force sensor decoupling method and device. Background Art

[0002] Six-dimensional force sensors have been widely used in robots, drones, automated equipment, and other transportation equipment to simultaneously measure force information in three directions and torque information in three directions. Their typical applications include:

[0003] Industrial robots use the multi-axis information of sensors for flexible control in assembly, grinding, grasping and other processes; when drones or other aircraft are hanging and transporting goods or performing aerial cooperation tasks, force sensors are used to obtain the multi-dimensional effects of the load on the airframe; surgical robots, rehabilitation robots, etc. require precise force feedback to improve safety and control accuracy. However, for the high-frequency coupling factors of equipment in real-time motion (including attitude changes, wind field disturbances, load offsets, etc.) and the low-frequency bias factors of equipment in slow environmental changes (temperature drift, mechanism looseness, load gradual change, etc.), it is impossible to capture high-order coupling and non-linear residuals in strong motion conditions, resulting in a significant reduction in measurement accuracy in complex tasks.

[0004] Therefore, a six-dimensional force sensor decoupling method and device are proposed to solve the above-mentioned problems. Summary of the Invention

[0005] The present invention aims to provide a six-dimensional force sensor decoupling method and device to solve or improve the problem that the six-dimensional force sensor decoupling method is difficult to accurately compensate high-frequency dynamic coupling and low-frequency slow-varying bias simultaneously, resulting in a reduction in measurement accuracy under complex motion conditions.

[0006] In view of this, the first aspect of the present invention is to provide a six-dimensional force sensor decoupling method.

[0007] The second aspect of the present invention is to provide a device.

[0008] The first aspect of the present invention provides a six - dimensional force sensor decoupling method and device, including the following steps: obtaining a set of system state vectors of the carrier device where the six - dimensional force sensor is located during movement, and original sensor data generated by output channels in the six - dimensional force sensor that are affected by the coupling of the set of system state vectors; respectively setting a working condition mapping function for compensating the influence of the real - time high - frequency changes of the carrier device on the six - dimensional force sensor, and a difference compensation function for compensating the influence of the stable and continuous errors of the carrier device on the six - dimensional force sensor through the set of system state vectors and the original sensor data; decoupling the linear part and the non - linear part of the corresponding true value of the original sensor data according to the working condition mapping function and the difference compensation function; taking the non - linear part as an incremental correction and superimposing it on the linear part, and decoupling the superimposed result to obtain the output data of each output channel.

[0009] In any of the above - mentioned technical solutions, the step of obtaining the linear part includes: obtaining a linear synthesis matrix of the output channel under the current set of system state vectors through the working condition mapping function, and taking the linear synthesis matrix as the mapping relationship between the original sensor data and the true value; obtaining a comprehensive output bias of the influence of the carrier device on the measurement of the six - dimensional force sensor under the current set of system state vectors, and taking the comprehensive output bias as the error compensation between the original sensor data and the true value; generating the linear part of the true value through the mapping relationship and the dynamic compensation.

[0010] In any of the above - mentioned technical solutions, the linear synthesis matrix is generated by linearly superimposing an initial calibration matrix of the carrier device in a stationary state and a dynamic correction matrix in a moving state.

[0011] In any of the above - mentioned technical solutions, the comprehensive output bias is generated by combining a static bias and a dynamic bias of the carrier device in an unloaded state.

[0012] In any of the above - mentioned technical solutions, the non - linear part is used to characterize the error caused by high - order non - linear factors between the linear part and the true value.

[0013] In any of the above - mentioned technical solutions, the non - linear part is solved and calculated under each set of system state vectors through polynomial basis functions, and the construction steps of the polynomial basis functions include: setting cross - terms for dividing the state vectors according to the dominant degree of each state vector in the set of system state vectors affecting the measurement of the true value by the six - dimensional force sensor; establishing a polynomial basis function for characterizing the non - linear interaction relationship between the original sensor data and each state vector through all the cross - terms.

[0014] In any of the above technical solutions, when the polynomial basis function is established through the cross terms, a weight matrix is ​​set for each cross term according to the degree to which each output channel is affected by the change of the system state vector set.

[0015] In any of the above technical solutions, the step of obtaining the nonlinear part includes: forming an extended input vector through the original sensor data and the system state vector set; inputting the extended input vector into a polynomial basis function to obtain a six-dimensional nonlinear compensation amount; and using the six-dimensional nonlinear compensation amount as the nonlinear part.

[0016] In any of the above technical solutions, at each unit time, the current system state vector set is updated according to the current operating condition of the transport equipment.

[0017] The second aspect of the present invention provides a device, including: a state and data acquisition module, which is used to collect the operating state information of the carrier and form the system state vector set; it is also connected to each output channel of the six-dimensional force sensor to obtain the original sensor data; a function and setting module, which is used to select or interpolate the operating condition mapping function to compensate for the measurement error caused by high-frequency changes in real time, and at the same time set the difference compensation function according to slow or stable factors to correct the stable continuous error; a linear and nonlinear decoupling module, which is used to separate the linear part decoupling result and the nonlinear part decoupling result according to the operating condition mapping function and the difference compensation function; an incremental correction and output module, which is used to combine the linear part and the nonlinear part, and decouple to obtain the output data of each output channel.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] By obtaining the real-time motion status of the carrier and setting the working condition mapping function according to the real-time high-frequency changes, the high-frequency dynamic coupling problems caused by rapid posture changes, transient acceleration fluctuations and load disturbances are effectively solved, so that the six-dimensional force sensor can still maintain high measurement accuracy and stability in fast and intense motion scenarios.

[0020] The difference compensation function is used to compensate for stable and continuous errors such as temperature drift, slow load changes, and gradual loosening of mechanical structures that may occur during the long-term operation of the carrier equipment. This effectively improves the measurement accuracy of the six-dimensional force sensor under long-term or slowly changing environmental conditions, and ensures the reliability and consistency of long-term measurements.

[0021] Through the hierarchical decoupling strategy of the linear part and the non-linear part, this algorithm significantly enhances the compensation ability for non-linear residuals and high-order coupling factors in complex motion conditions, effectively reduces the errors that cannot be captured by the linear model, and improves the overall decoupling accuracy of the sensor in complex task scenarios; the design idea of taking the non-linear part as an incremental correction and superimposing it on the linear part enables the entire algorithm to have both good real-time performance and high computational efficiency, and can effectively improve the credibility and stability of the output data of the force sensor without significantly increasing the computational burden of the system.

[0022] Additional aspects and advantages of embodiments according to the present invention will become apparent in the following description section, or will be learned by practice of embodiments according to the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The above and / or additional aspects and advantages of the present invention will become apparent and be readily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0024] Figure 1 is a flowchart of the method steps of the present invention;

[0025] Figure 2 is a data output diagram of the present invention with a unit time interval of 50 milliseconds;

[0026] Figure 3 is an example mapping relationship table of the state vector and the output channels of the present invention;

[0027] Figure 4 is a logical block diagram of the device structure of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0028] In order to more clearly understand the above objects, features and advantages of the present invention, the present invention will be further described in detail below with reference to the drawings and specific embodiments. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments may be combined with each other.

[0029] In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention may be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0030] Please refer to Figures 1-4 , and a six-dimensional force sensor decoupling method and device according to some embodiments of the present invention will be described below.

[0031] An embodiment of the first aspect of the present invention provides a six-dimensional force sensor decoupling method. In some embodiments of the present invention, as Figures 1-4 shown, the method includes:

[0032] S101. Obtain the system state vector set of the carrier device where the six - axis force sensor is located during movement, and the original sensor data generated by the output channels in the six - axis force sensor that are affected by the coupling of the system state vector set.

[0033] Here, the system state vector set includes various attitude and motion information of the carrier device (such as unmanned aerial vehicle, robotic arm, automated guided vehicle, manned platform, etc.) during actual movement, including but not limited to: attitude angles (such as pitch angle θ, roll angle φ, yaw angle ψ), linear acceleration, angular velocity, position information, velocity information, and load - related parameters (such as load mass, load center - of - gravity position, swing amplitude, wind interference, etc.). These state variables together constitute a multi - dimensional state vector to characterize the overall operation of the carrier device at any moment or any motion stage. The six - axis force sensor includes several force channels and torque channels, generally at least six main output channels (corresponding to Fx, Fy, Fz, Mx, My, Mz respectively), and in some high - precision or redundant designs, multiple measurement channels or temperature compensation channels can also be expanded. When the carrier device is in motion, the measurement results of these channels often have coupling interference with each other - that is, the output value of a certain path of the sensor not only depends on the force applied in a single direction, but also is affected by forces and torques in other directions and changes in the external environment. If this coupling error is not eliminated, it will directly reduce the measurement accuracy and control performance.

[0034] As can be seen from the above, when the carrier device is an unmanned aerial vehicle, its built - in or external IMU updates the attitude angles in real - time at a high frequency (such as 200Hz, 500Hz or even higher sampling rate); in a ground robot or robotic arm, the corresponding inclination information or joint rotation angle information can usually be obtained from its joint sensors or attitude sensors. Taking the unmanned aerial vehicle scenario as an example: when the unmanned aerial vehicle performs a maneuvering flight, the pitch and roll movements will cause continuous changes in linear acceleration and angular velocity, and these quantities can be directly obtained through the accelerometer and gyroscope of the IMU. For a robotic arm, the linear acceleration and angular velocity values of the end - effector can be calculated from the joint angular velocity and kinematic model. If the carrier device is flying or operating outdoors (for example, an unmanned aerial vehicle is affected by a wind field outdoors), the wind speed and direction of the current environmental wind field can also be obtained through a barometer, anemometer or preset meteorological data, and incorporated into the system state vector. The six - axis force sensor usually consists of strain gauges, piezoelectric sensing elements, fiber optic sensing elements or other multi - axis detection elements. Each detection direction has different permutations and combinations and bridge circuit designs with respect to force or torque, and provides several analog or digital outputs. These output channels may not be limited to the six most basic channels, but may also include temperature compensation channels and redundant channels to further improve the accuracy or fault tolerance rate.

[0035] Specifically, an exemplary formula for the system state vector set is:

[0036]

[0037] Wherein, X state (t) is the system state vector set, and t is the unit time; θ, φ, ψ are the pitch, roll, and yaw angles of the unmanned aerial vehicle; are the angular velocities on different axes; a x , a y , α z are the linear accelerations on different axes; v w , α w are the wind speed and wind direction; m c is the payload mass; Δx, Δy, Δz are the centroid offsets on different axes. The above quantities can be continuously expanded as needed to include more parameters related to sensor coupling.

[0038] Specifically, multiple channels (including redundant or temperature compensation channels) of the six-axis force sensor are uniformly denoted by the following formula:

[0039]

[0040] Wherein, S raw (t) is the original sensor data; n is the current number of output channels; s n (t) is the nth original data; T is the transpose of the matrix; is a real column vector of dimension n.

[0041] S102, respectively set a working condition mapping function for compensating the influence of the real-time high-frequency change of the carrier equipment on the six-axis force sensor through the system state vector set and the original sensor data, and a difference compensation function for compensating the influence of the stable and continuous error of the carrier equipment on the six-axis force sensor.

[0042] Here, such as inertial forces and inertial torques generated by rapid changes in attitude angles, rapid acceleration or deceleration, short-period wind field pulsations, high-frequency components of load swings, and high-acceleration impacts at the end caused by rapid joint movements of the robotic arm. These factors will significantly change the output coupling relationship of the six-axis force sensor in a short time and require a working condition mapping function to achieve real-time compensation for rapid changes. For example, the payload mass changes slowly over time (such as gradual reduction of the payload during volatilization or liquid transportation), slow zero-point changes of the sensor caused by temperature drift, slow deformation or creep of the mechanical structure, steady wind speed and wind direction maintained in the environment for a long time, and deformation offsets caused by fatigue loosening after long-term operation. These slow-changing factors often do not trigger large measurement offsets instantaneously in a very short time, but will accumulate non-negligible offsets or zero drifts over a long period, and a difference compensation function is needed to correct these persistent quantization errors.

[0043] As described above, when the drone performs drastic maneuvers, such as instantaneously pulling up the pitch angle from 0° to 30° while accompanied by a change in the roll angle, the fuselage generates large linear and angular accelerations; all the state vectors in the system state vector change significantly within a short period of time; the working condition mapping function needs to find a set of compensation coefficients or interpolation matrices that best match the current attitude combination based on the rapidly changing attitude and acceleration values, and output in real time the correction of the first-order coupling deviation and part of the high-order coupling effect generated by the measurement of the six-axis force sensor. If the six-axis force sensor is installed at the end of the robotic arm, when a certain joint rotates at a high speed with a large angle instantaneously, high acceleration and high angular velocity will appear at the end; the working condition mapping function can use the joint sensors of the robotic arm or inverse kinematics to calculate the attitude and acceleration information of the end, and then combine the function obtained by prior training or interpolation to compensate for the instantaneous coupling error generated by this high-speed movement; for example, for certain specific attitudes of the robotic arm, there may be structural coupling (such as the linkage between joint 2 and joint 4) resulting in a temporary overshoot error of the force sensor, and the working condition mapping function will correct this specific linkage attitude. When the drone or the ground vehicle hangs a load and maneuvers around obstacles in a narrow space, the swing frequency of the load increases, and the swing amplitude periodically affects the output of the six-axis force sensor; the working condition mapping function can estimate the model in real time through the swing angle or swing frequency of the load, extract the periodic additional torque brought by the swing, and obtain the corresponding compensation result.

[0044] S103, decouple the linear part and the nonlinear part of the original sensor data corresponding to the true value according to the working condition mapping function and the difference compensation function.

[0045] Here, the linear part decoupling is used to quickly strip out the coupling relationships that can be described by a linear model, such as most initial calibration matrices, linear tilt compensation caused by the attitude, low-order linear correlations between inertial forces and sensor outputs, etc.; the nonlinear part decoupling is used to further analyze those complex coupling phenomena that are no longer sensitive to the linear model or need to introduce high-order cross terms, neural network mapping, etc. to describe, such as the combined action of load swing and fuselage acceleration, high-frequency flutter, compliant deformation of the robotic arm, coupled nonlinear strain, etc.

[0046] As can be seen from the above, by integrating the high-frequency correction amount corresponding to the working condition mapping function and the slow-varying bias amount corresponding to the difference compensation function into the linear decoupling model and the corresponding extended correction process respectively, a clearer "deviation correction path" can be obtained for the original sensor signal. For example, in the conventional motion range of the robotic arm joint, the error characteristics of the force sensor can often be compensated for most parts by a relatively direct linear method, and the working condition mapping function is used to achieve real-time correction of the instantaneous changes in attitude or load. At the same time, if the robotic arm is in a working environment from room temperature to high temperature for a long time, the continuous zero-point offset that occurs during this process will be hedged or offset by the difference compensation function to form a correction of a low-frequency slow variable. After the linear part is initially completed, the remaining deeper measurement errors are usually closely related to coupled non-linearity. For example, when mechanical component deformation, load center of gravity swing, and inertial force are coupled with each other, it is necessary to introduce non-linear thinking to distinguish them. Once such non-linear characteristics are found, they can often be corrected by the working condition mapping function with a certain degree of complexity within a short time window or specifically fitted by other higher-order models, so that the final output of the sensor gradually approaches the true value.

[0047] For example, when a drone makes a large-angle dive after a rapid climb, the body attitude and airflow environment will change significantly in a short time, thus causing mutual interference among multiple axes in the output of the force sensor. At this time, the working condition mapping function can immediately give a set of rapid correction parameters according to the instantaneous attitude angle and acceleration conditions to correct the most critical and fastest-responding dynamic interference, so that the linear decoupling remains within a relatively accurate range. If the external temperature changes significantly after the drone has flown for a period of time or the suspended load is slowly released, the difference compensation function is required to make periodic adjustments to these "slow-heating" or "gradually increasing" offsets to ensure that the final measurement result does not accumulate excessive errors over time.

[0048] Specifically, the steps to obtain the linear part include:

[0049] Obtain the linear synthesis matrix of the output channel under the current system state vector set through the working condition mapping function, and use the linear synthesis matrix as the mapping relationship between the original sensor data and the true value.

[0050] Obtain the comprehensive output bias of the influence of the carrier equipment on the six-axis force sensor measurement under the current system state vector set, and use the comprehensive output bias as the error compensation between the original sensor data and the true value.

[0051] Generate the linear part of the true value through the mapping relationship and dynamic compensation.

[0052] For the above specific description, first, it is necessary to determine the linear synthesis matrix corresponding to each output channel under the current system state vector set according to the working condition mapping function. Since the working condition mapping function often accumulates a large amount of correction data or rules regarding different postures, different external environments, and different load states in advance, during the specific implementation, the current system state (such as attitude angle, angular velocity, linear acceleration, the information of the mounted mass and center of gravity, etc.) can be input into the working condition mapping function to index or interpolate the linear synthesis matrix specifically for the current working condition. The linear synthesis matrix obtained in this way can be regarded as the direct mapping relationship between the original sensor data and its linear part true value under the current working condition, encompassing the impacts generated by various attitude couplings, inertial effects, and general structural errors at this moment, enabling the subsequent calculations to more accurately extract the relatively simple linear components.

[0053] Next, it is necessary to obtain the comprehensive output bias of the vehicle equipment on the six-axis force sensor measurement under the current system state vector set. This step is usually based on the induction of slow-varying factors or zero-offset information, or the bias estimation obtained from the combined effects of various factors such as the installation method of the vehicle equipment, long-term temperature drift, and slow load deformation. For example, in the application scenario of an unmanned aerial vehicle, if it has been determined that a certain degree of sensor zero-point change may occur with a temperature increase of two or three degrees, or in the scenario of a robotic arm, it is confirmed that a fixed offset will continuously occur in the measurement data after the thermal deformation of a specific joint, then after obtaining the current system state vector, the difference compensation module can find the set of bias compensation values most suitable for the current working condition. Through this method, such slow bias amounts that do not fluctuate significantly with small-range dynamic changes can be stably superimposed into the application of the linear synthesis matrix, thereby continuously correcting the measurement impact on the original sensor data.

[0054] After the linear synthesis matrix and the comprehensive output bias are both determined, they can be combined into a unified mapping framework to generate the linear part of the true value of each output channel. Specifically, first, perform the corresponding linear operation on the original sensor data according to the linear synthesis matrix to obtain the most basic linearly decoupled output. At the same time, use the comprehensive output bias as the correction amount added or subtracted from this result to ensure that those changes in the vehicle equipment with long periods or relatively stable states (such as fixed postures, small but persistent load distribution changes, etc.) are also included in the correction scope. Doing so can eliminate most of the predictable linear couplings and basic offsets in a relatively short operation process, making the subsequent consideration of non-linear errors or other incremental compensations a simpler and clearer process.

[0055] As described above, taking an industrial robot as an example, there may be a section of joint motion trajectory that is relatively stable and repetitive. Each time the robot moves to this trajectory section, the attitude angle of the robotic arm and the load states of each joint are relatively fixed. At this time, the linear synthesis matrix obtained through the working condition mapping function can highly accurately match the relationship between the sensor output and the actual force under this trajectory, and the comprehensive output offset calibrated in advance by the operation and maintenance personnel can also quickly offset the measurement inaccuracy that is prone to occur during repeated operations. As a result, such linear partial decoupling can better eliminate redundant interference, enable the robotic arm to maintain high-precision alignment during multiple repeated processes, and reduce rework or part damage caused by unstable force control.

[0056] When a drone is transporting a suspended load, if it has been measured that under the current light wind environment and with a slightly swinging load, the linear sensitivity of the sensor in some axes will change significantly, then by invoking the corresponding linear synthesis matrix through the working condition mapping function, a refined operation can be performed on the current raw data. At the same time, if slow-changing parameters such as the ambient temperature or the depletion of the load itself are also observed, the difference compensation will output a set of offset values for the zero point or the most basic quantity, so that each linear mapping result can return to an approximately true initial reference. This process ultimately enables the drone to minimize the coupling interference caused by the load to the measurement signal during flight in mid-air and continuously provide a more reliable measurement feedback to the flight control system.

[0057] Specifically, the linear part is the following formula:

[0058] F dyn (t) = C L (X state (t))S raw (t) + b L (X state (t))

[0059] In the formula, F dyn (t) is the linear part; C L (X state (t)) is the linear synthesis matrix; b L (X state (t)) is the comprehensive output offset.

[0060] In any of the above embodiments, the linear synthesis matrix is generated by linearly superimposing the initial calibration matrix of the carrier device in the stationary state and the dynamic correction matrix in the moving state.

[0061] In this embodiment, the generation of the linear synthesis matrix can be regarded as a key link that fuses static calibration and dynamic compensation information. The initial calibration matrix obtained in the stationary state of the device contains the reference mapping relationships for factors such as sensor-to-body mounting errors, basic zero offsets, and conventional torque couplings. When the carrier device enters the actual motion state, the continuous changes in attitude, the adjustment of load eccentricity with motion, and the interference of the external environment (such as air flow or mechanical vibration) will significantly change the measurement environment of the sensor instantaneously or in a short period, resulting in additional coupling quantities that were not included in the original static calibration. At this time, applying a mechanism that can dynamically evaluate and output a correction matrix becomes the key to compensating for motion interference and high-dynamic coupling in a timely manner.

[0062] Linearly superimposing these two matrices can bring a more comprehensive and flexible compensation ability than using a single matrix alone. The initial calibration matrix often covers most of the quantization errors in the stationary or basic calibration environment of the device, ensuring that the readings of the sensor can maintain a reliable reference accuracy without significant motion. During the actual operation of the real machine, once the system detects significant fluctuations in the device attitude or external interference, it can retrieve or calculate the dynamic correction matrix and add it numerically to the initial calibration matrix to obtain the linear synthesis matrix that best fits the current instantaneous working conditions. This incremental calibration idea can not only continue the experience and accuracy accumulated during the initial calibration but also quickly complete the correction of various newly emerging couplings in a highly dynamic scenario.

[0063] Specifically, the linear synthesis matrix is the following formula:

[0064] C L (X state (t))=C0(X state (t))+C d (X state (t))

[0065] In the formula, is the initial calibration matrix; is the dynamic correction matrix.

[0066] In any of the above embodiments, the comprehensive output bias is generated by combining the static bias and the dynamic bias of the carrier device in the unloaded state.

[0067] In this embodiment, when obtaining the comprehensive output bias, the idea is to first combine the static bias measured when the carrier device is in the no-load state with the dynamic bias generated during the actual operation of the device (including different postures, accelerations, temperature environments, etc.) to form a bias compensation amount that is most applicable in the current situation. The result of this combination can not only retain the high-precision zero-point reference obtained during no-load static calibration but also introduce additional corrections for variable working conditions, thereby ensuring continuous calibration of the sensor output zero point or reference in the real motion scenario. Another part is the dynamic bias during the actual operation of the device. It can stem from inertial forces brought about by device movement, wind field disturbances, rapid posture changes, or gradual changes in external environmental conditions such as temperature and humidity, and may also be related to small-scale creep of mechanical contacts or structural loads after long-term operation. In practice, the continuous wind force and occasional rotor airflow changes encountered during the high-altitude flight of an unmanned aerial vehicle will cause slight offsets in the measurement zero points of certain axes; or, when some ground transportation vehicles are traveling at high speeds, road vibrations and tire dynamic loads will cause certain deformations in the sensor mounting base, resulting in drift for a period of time. These additional biases all fall within the category of dynamic biases. If only the calibration in the static stage is relied on for correction, it may lead to a rapid decline in measurement accuracy and consistency in high-dynamic scenarios.

[0068] Combining the static bias and the dynamic bias can be understood as a multi-source information fusion process, that is, when the system performs real-time calculation or interpolation and looks up tables, it will first call the static bias value to provide a general reference, and at the same time use the dynamic bias information for the current state to append or correct this reference. If there are obvious changes in the external environment or device state, such as the ambient temperature continuously rising above a certain threshold or the unmanned aerial vehicle encountering stronger instantaneous airflow in the air, then the program will calculate a new dynamic bias amount accordingly, and then update the finally combined comprehensive output bias, keeping the error between the sensor and the real state at a low level continuously.

[0069] Specifically, the comprehensive output bias is the following formula:

[0070] b L (X state (t))=b0(X stat (t))+b d (X state (t))

[0071] In the formula, b0(X state (t)) is the static bias; b d (X state (t)) is the dynamic bias.

[0072] In any of the above embodiments, the non-linear part is used to characterize the error caused by high-order non-linear factors between the linear part and the true value.

[0073] In this embodiment, in the measurement and control scenarios of various actual carrying devices, simply describing the data of the six-axis force sensor linearly is often insufficient to cover all interferences and couplings. When the system is in a high-dynamic operating state, or there are relatively complex external environmental influences, the force and torque measurements often exhibit more coupled "nonlinear" components. If only relying on the calculation results of the linear part, there may still be a part of the unexplained error left. That is to say, although linear mapping can remove most of the conventional offsets and couplings, under some load and motion modes, multiple factors such as local structural deformation, vibration coupling, combined action of load swing and inertial force, material elastic nonlinearity, and air flow will be superimposed during the sensor measurement process, presenting a higher-level interaction characteristic. The nonlinear part is established for these higher-order factors, and the gap between the linear output and the true value is reduced through additional complex mapping or compensation means.

[0074] After the decoupling at the linear level has been completed, the remaining error that still needs to be observed is often those nonlinear factors that are difficult to explain by simple mathematical relationships. Through the nonlinear compensation link, these elusive interactions can be further analyzed and fitted. For example, in the application of a robotic arm, once the robotic arm rotates at high speed, the load swing at the end or the slight deformation of the arm body will trigger more couplings in more directions; when the linear compensation cannot completely remove this influence, this remaining significant influence will be mapped into the nonlinear model, and then a more flexible higher-order or multi-term cross method will be introduced for correction. After this layer of operation, the difference between the linear output and the true value is greatly reduced, and the robotic arm can maintain high-precision force control and stable operation.

[0075] Specifically, the nonlinear part performs the solution calculation for each system state vector set through polynomial basis functions, and the construction steps of the polynomial basis functions include:

[0076] According to the dominant degree of each state vector in the system state vector set affecting the true value of the six-axis force sensor measurement, cross terms for dividing the state vectors are set.

[0077] A polynomial basis function for characterizing the nonlinear interaction relationship between the original sensor data and each state vector is established through all the cross terms.

[0078] For the above specific description, based on the influence degree of each state vector in the system state vector set on the measured true value of the six-axis force sensor, first determine which "cross terms" may need to be introduced. During the actual operation of the carrier equipment, various state parameters (such as attitude angles, linear and angular velocities, load swing amplitudes, external wind fields, etc.) often do not affect the measured values independently of each other, but may form correlation effects in some cases. For example, when a robotic arm is running at high speed, the superposition of the high-speed movement of a certain joint and the inertial force of the end tooling is likely to generate higher-order strains or non-linear deformations; similarly, for an unmanned aerial vehicle, the multiplication of the angle of attack and the wind speed may bring more severe jitter interference. Through empirical observations and data statistics of these real scenarios, it is possible to initially judge which state variables have the most significant mutual influence, and then reserve positions for the cross terms of these variable pairs in the polynomial basis.

[0079] Establish a complete set of polynomial basis functions through all cross terms to express the non-linear mapping relationship between the original sensor data and each state vector. In this step, the key lies in how to make these basis functions "clear and operable". In relatively simple cases, perhaps only the first-order, second-order, or third-order cross terms need to be combined. However, if the operating boundaries of some carrier equipment are very wide and the coupling order between variables is higher, then more orders or more complex terms may need to be included. In this way, each polynomial basis function is equivalent to a "slot", waiting to be filled with various multiplied or combined state variables. For example, a ground high-speed vehicle is equipped with a six-axis force sensor, and the vehicle has lateral acceleration, longitudinal acceleration, and an observable wind direction and speed. When the actual data shows that the vehicle is more likely to have a certain steep additional coupling when encountering a large crosswind, then the basis function needs to include the product between "lateral acceleration" and "wind speed", and perhaps also some combinations of the wind direction and the vehicle's roll angle, so that this important coupling effect will not be missed when calculating the non-linear compensation later. In this way, a function library containing a large number of cross terms can be finally formed, and each function specifically corresponds to a certain specific high-order or multi-variable interaction. When the algorithm detects that such an interaction starts to be significant during actual operation, it can compensate for a large number of errors that cannot be solved linearly by calling and calculating the corresponding basis terms.

[0080] After building such a large polynomial basis based on all cross terms, the corresponding calculations can be performed on each current system state vector during execution; in other words, the algorithm will analyze the collected dynamic parameters, fill these cross templates with them, and then combine all the results to calculate the compensation amount of the non-linear part at this time. Thus, even if the sensor data shows higher-order couplings different from before, enough fitting space can be captured in this polynomial framework for correction.

[0081] As can be seen from the above, the originally extremely complex and difficult-to-estimate high-dynamic interference is disassembled into multiple mutually independent and crossable polynomial basis functions for management. By examining the explicit influence of each state quantity in the system state vector set, the most critical interaction terms can be effectively collected; and the polynomial basis functions constructed based on all the cross terms ultimately become the mathematical carriers relied on in the calculation of the non-linear part. In this way, there is no longer a need to rely solely on simple or single empirical formulas to handle the extremely cumbersome real environment interference. Instead, on the basis of hierarchical and variable cross-combinations, large-scale data can be fully utilized for training, updating, and online calling, gradually enabling more accurate compensation and correction of the high-order coupled measurement errors.

[0082] In any of the above embodiments, when establishing the polynomial basis function through the cross terms, a weight matrix is set for each cross term according to the degree of influence of each output channel on the change of the system state vector set.

[0083] In this embodiment, during the process of constructing the polynomial basis function, not only all possible cross terms need to be clearly listed, but also a matching weight matrix needs to be set for each cross term, so as to distinguish the influence differences shown by different channels due to system state changes during specific calculations. That is to say, even if two cross terms both seem to be in the form of "second-order" or "multivariable" multiplication in the state vector, the measurement offsets they bring to each output channel of the six-axis force sensor are not necessarily the same. By assigning respective weight matrices to each cross term, the error increment or bias correction that a certain output channel will generate due to this cross term under the current working conditions can be captured more precisely.

[0084] The significance of setting these weight matrices lies in that each channel output by the six-axis force sensor does not respond equivalently to the system state quantities. For example, a certain channel mainly measures the force in the horizontal direction and is not very sensitive to vertical loads or some changes associated with the pitch angle; however, when the wind speed or lateral load changes are combined with the pitch angle to a certain extent, this channel may still be involved. Therefore, in the polynomial basis, if "wind speed × pitch angle" is a relatively minor interference to this channel, offline calibration or training will give a smaller weight, while for another channel - such as a more lateral force measurement channel, this cross term may cause more significant coupling, and then a relatively large weight coefficient will be given. These weights in different dimensions can ultimately be combined into a table or a matrix, and some systems will also encapsulate them in a programming manner for subsequent flexible updating.

[0085] Specifically, the steps for obtaining the non-linear part include:

[0086] An extended input vector is jointly composed of the original sensor data and the system state vector set.

[0087] Input the extended input vector into the polynomial basis function to obtain a six-dimensional non-linear compensation quantity.

[0088] Take the six-dimensional non-linear compensation quantity as the non-linear part.

[0089] For the above specific description, synchronously read the original output of the sensor at the current moment, such as the real-time measurement values of the six-axis force sensor on each channel; at the same time, the system state vector of the carrier device will also be obtained, including various parameters that may affect the sensor measurement, such as attitude angle, acceleration, payload position, temperature change, wind field interference, etc. Looking at only one channel or a single variable, it is often difficult to reflect the complex dynamic interaction. However, when all relevant information is put together, it can form an "extended input vector" that covers a more comprehensive and multi-dimensional range. Specifically, this vector not only preserves the sensor output itself but also carries various detailed features of the environment and the operating conditions of the device, providing sufficient computational basis for subsequent non-linear modeling. After having the above extended input vector, put it into the pre-constructed polynomial basis function model. The polynomial basis function combines all possible high-order interaction terms, the products of state variables that are estimated to be necessary to include, and the non-linear correlations between various variables into a unified mapping structure. Then, the algorithm can calculate the corresponding non-linear compensation contribution degree for each item according to the weights or coefficients of each item in these basis functions. Summing up these contributions, the so-called "six-dimensional non-linear compensation quantity" can be obtained, that is, for each force / torque channel, there is a more refined correction quantity to correct the complex coupling at the current moment. When the polynomial basis function completes the mapping and weighting process of the extended input vector, the six output components become the core basis for high-order correction of the sensor at that moment, that is, the non-linear part. It usually appears in the form of a six-dimensional vector and corresponds to the additional correction quantities of the three-axis force and the three-axis torque respectively. Next, according to the need, this non-linear part can be directly incrementally superimposed on the "already linearly compensated" data. In this way, the entire measurement process completes the adaptation to high-order non-linear coupling: if the environment and the motion state change more extremely, this non-linear part will naturally undergo new calculations and updates during the next processing, so as to maintain timely response to the new operating conditions.

[0090] As described above, the original data and the system state are first combined to form an extended input, enabling the model to see a sufficient number of variables. Then, this set of inputs is brought into the polynomial basis function, and through the pre-set cross relationships and weights, the non-linear compensation amplitude for each force / torque channel at a certain moment is calculated. Finally, this compensation value is established as the non-linear part at the current moment to achieve a deeper correction of the high-order coupling error remaining in the system. Whether in high-speed moving vehicles, precision robots, medical surgical robotic arms, or ground vehicle scenarios, as long as there are complex disturbances that are difficult to be directly corrected by traditional linear theories, this extended input + polynomial basis strategy can provide a more flexible and adaptive solution, ensuring that the measurement results of the six-axis force sensor always maintain a high level of accuracy and reliability.

[0091] Specifically, the polynomial basis function is given by the following formula:

[0092]

[0093] In the formula, \(z(t)\) is the combined extended input vector, formed by splicing the original sensor measurement data and the UAV state parameters; \(P\) l (z(t)) is the polynomial basis function composed of the components of the input vector \(z(t)\), usually including the following specific forms: linear term (first-order term): the \(i\)-th component of \(z\) i (t) itself, such as the original sensor output of a certain channel, or a certain attitude angle; quadratic cross term: \(z\) i (t) \(z\) j (t) represents the product of any two components of \(z(t)\), such as the cross product term of a certain sensor output and a certain state quantity (such as attitude angle or acceleration); cubic or higher-order terms: \(z\) i (t) \(z\) j (t) \(z\) k (t) represents the three-dimensional or higher-order cross product term to capture more complex non-linear interaction relationships, such as the three-dimensional interaction effects of sensor output, payload eccentricity, and acceleration, etc.; \(W\) l is the weight coefficient matrix corresponding to the above polynomial basis function; \(Q\) is the total number of basis functions included in the polynomial basis function set, that is, the total number of all linear terms, quadratic terms, and cubic cross terms, etc.; \(l\) is an index variable that traverses all polynomial basis functions, used to represent the serial number of a specific basis function currently in use; \(k\) is a component, and \(k\in\{1,...,6\}\), respectively representing \(F\) x , \(F\) y , \(F\) z , \(M\) x , \(M\) y , \(M\) z ; \(f\) nl(z(t)) is a non - linear compensation function, implemented through polynomial fitting or neural network models, used to compensate for the non - linear coupling errors remaining after the above - mentioned linear decoupling.

[0094] S104, superimpose the non - linear part as an incremental correction on the linear part, and decouple the superimposed result to obtain the output data of each output channel.

[0095] Here, since the linear part can usually be quickly completed with relatively limited computing resources, while the non - linear part often depends on more time - consuming algorithms (such as high - order data mapping, neural networks, or more complex interpolation models), during execution, people often first use an easily implemented linear mechanism to eliminate a large number of conventional coupling errors, and then incrementally add the non - linear compensation as a refinement means to the final result. This not only takes into account real - time performance but also enables the system to flexibly adjust the degree of non - linear compensation according to the current computing power and accuracy requirements when dealing with various harsh or complex environments.

[0096] As can be seen from the above, in some high - requirement scenarios, to ensure both accuracy and speed, the system first completes linear decoupling in a very short cycle, and then periodically (or under appropriate trigger conditions) calculates the non - linear correction amount again. Then, all corrections are completed through incremental superposition. In terms of results, once the non - linear compensation value is updated, it can be immediately added back to the decoupled result, and the subsequent output data will be instantly more accurate. And if the external interference conditions do not change significantly within a period of time, the non - linear compensation value can also remain relatively stable and does not need to be changed frequently. For example, when a robot uses a six - dimensional force sensor for high - precision assembly or polishing operations, if only relying on the compensation of the linear part, most of the simple interferences caused by joint movement may be removed, but in some highly delicate or complex spatial movements, the remaining errors will still cause inaccurate positioning and even damage to the processed parts. At this time, if the non - linear correction amount obtained through higher - order modeling in advance can be incorporated, then the very small coupling deviations that still exist originally will be further compensated, so that the robotic arm will be closer to being safe and smooth during flexible assembly, precision polishing, or grinding, and there will be no phenomenon of sudden changes in force output.

[0097] Forming a clear hierarchical mode of non - linear correction and linear decoupling not only makes the basic coupling processing lighter, faster, and feasible, but also retains sufficient degrees of freedom to accurately handle complex and non - linear on - site interferences. In this way, not only the strict requirements for real - time performance are met, but also a deeper improvement is achieved in terms of accuracy, thus supporting a variety of high - precision force feedback requirements in broad fields such as industrial assembly, medical robot surgical assistance, and high - dynamic motion vehicle control.

[0098] Specifically, the formula for the output data is:

[0099]

[0100] where s u (t) is the filtered output of the u-th sensor channel; [F ral (t)] k is the decoupled final result of six-dimensional force / torque at the current moment of a specific output channel in the output data of the six-dimensional force sensor, that is, the output data.

[0101] In any of the above embodiments, at each unit time, the current system state vector set is updated according to the current working condition of the carrying device.

[0102] In this embodiment, only by updating the system state vector set at each unit time can the control and measurement of the carrying device maintain the real-time perception ability of external changes. Otherwise, if the state quantity lags or is too rough, then all force sensor decoupling and control strategies will be based on "outdated information", resulting in the analysis and control of force and torque not being able to keep up with the actual needs. The usage environments of different carrying devices are often different, and it may change from high-speed turning to low-speed translation, from gentle breeze to strong wind, from normal temperature to high heat state, etc. After obtaining a new state vector in each control cycle, the compensation mode and parameters can be flexibly switched or updated, so that both linear and nonlinear compensation can be corrected following the change of the working condition, ensuring that the measurement always remains in a better state. Some devices also use the data in the state vector for a certain degree of fault detection or health monitoring. Once it is found that some variables show abnormal transitions, preventive measures such as reducing frequency or emergency load reduction can be taken in a timely manner, which can not only protect the carrying device, but also protect the target being transported or processed.

[0103] A six-dimensional force sensor decoupling method provided by the present invention enables the sensor to accurately compensate for high-frequency dynamic coupling in real time when facing rapid attitude changes, transient interferences, and high-frequency load fluctuations by specifically setting a working condition mapping function for real-time high-frequency changes, ensuring a significant improvement in the measurement accuracy in high-dynamic motion scenarios.

[0104] Using the difference compensation function to compensate for the stable and continuous errors caused by temperature drift, slow load change, and gradual loosening of the mechanical structure effectively improves the zero-point stability and measurement consistency of the six-dimensional force sensor during long-term operation, making the long-term monitoring data more reliable.

[0105] By performing a refined separation process on the linear part and the nonlinear part, the nonlinear errors that cannot be eliminated by the linear model are effectively captured and compensated, further improving the overall measurement accuracy in complex motion working conditions (such as violent motion, strong interference environment).

[0106] The non-linear part compensation is quickly and effectively applied to the measurement result of the linear part by using the incremental superposition method, taking into account both computational efficiency and accuracy requirements, and can simultaneously meet the dual requirements of real-time performance and high accuracy in practical applications.

[0107] It enables the six-axis force sensor to more accurately reflect the actual force conditions in various high-precision application fields such as robot flexible operation, complex hanging transportation of unmanned aerial vehicles, and medical precision control, greatly expanding the applicability of the sensor in multiple application fields.

[0108] An embodiment of the second aspect of the present invention proposes a device 2. In some embodiments of the present invention, as Figure 4 shown, the device 2 includes:

[0109] A state and data acquisition module 201, configured to collect the operation state information of the carrier device and form a system state vector set; and is also connected to each output channel of the six-axis force sensor to obtain the original sensor data.

[0110] A function and setting module 202, configured to select or interpolate a working condition mapping function to perform real-time compensation for the measurement error caused by high-frequency changes, and at the same time set a difference compensation function according to slow or stable factors to correct the stable continuous error.

[0111] A linear and non-linear decoupling module 203, configured to separate the linear part decoupling result and the non-linear part decoupling result according to the working condition mapping function and the difference compensation function.

[0112] An incremental correction and output module 204, configured to combine the linear part and the non-linear part and perform decoupling to obtain the output data of each output channel.

[0113] A device provided by the present invention can, by setting a status and data acquisition module, collect the system status information and original sensor data during the operation of a carrier device in real time, ensuring that the decoupling process is always based on accurate and comprehensive data; meanwhile, the design of the function and setting module enables the device to set a working condition mapping function and a difference compensation function respectively for interferences at different time scales, compensate the dynamic coupling error caused by high-frequency changes in real time, and effectively correct the stable and continuous error generated during long-term operation, significantly improving the stability and accuracy of measurement. In addition, the linear and non-linear decoupling module further enhances the device's ability to handle complex working conditions, enabling the linear error and non-linear error to be processed hierarchically, fully capturing and correcting the high-order non-linear error that cannot be described by a simple linear model, and significantly improving the accuracy and reliability of decoupling; while the incremental correction and output module ensures both the real-time nature and high accuracy of the final output data, and can quickly and accurately obtain the true measurement results of each output channel. Overall, the device has a clear structure and a modular function design, can efficiently meet the accurate force measurement requirements of various high-dynamic and multi-disturbance application scenarios such as robots, drones, and automation equipment, and has obvious practical value and technical advantages.

[0114] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "longitudinal", "transverse", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation to the present invention.

[0115] The embodiments described above are only used to describe the preferred mode of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solution of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A decoupling method for a six-axis force sensor, characterized in that It includes the following steps: Obtain the system state vector set of the carrier equipment where the six - axis force sensor is located during the movement process, and the original sensor data generated by the output channels in the six - axis force sensor that are affected by the coupling of the system state vector set; Respectively set a working condition mapping function for compensating the influence of the real - time high - frequency change of the carrier equipment on the six - axis force sensor, and a difference compensation function for compensating the influence of the stable and continuous error of the carrier equipment on the six - axis force sensor through the system state vector set and the original sensor data; Decouple the linear part and the nonlinear part of the corresponding true value of the original sensor data according to the working condition mapping function and the difference compensation function; Superimpose the nonlinear part as an incremental correction with the linear part, and decouple the superimposed result to obtain the output data of each output channel.

2. The decoupling method of the six-axis force sensor according to claim 1, characterized in that The steps for obtaining the linear part include: Obtain the linear synthesis matrix of the output channel under the current system state vector set through the working condition mapping function, and use the linear synthesis matrix as the mapping relationship between the original sensor data and the true value; Obtain the comprehensive output bias of the influence of the carrier equipment on the measurement of the six - axis force sensor under the current system state vector set, and use the comprehensive output bias as the error compensation between the original sensor data and the true value; Generate the linear part of the true value through the mapping relationship and the dynamic compensation.

3. The six-axis force sensor decoupling method according to claim 2, characterized in that The linear synthesis matrix is generated by linearly superimposing the initial calibration matrix of the carrier equipment in the static state and the dynamic correction matrix in the moving state.

4. The decoupling method of the six-axis force sensor according to claim 2, characterized in that The comprehensive output bias is generated by combining the static bias and the dynamic bias of the carrier equipment in the no - load state.

5. The decoupling method of the six-axis force sensor according to claim 1, wherein The nonlinear part is used to characterize the error caused by high - order nonlinear factors between the linear part and the true value.

6. The decoupling method of the six-axis force sensor according to claim 5, wherein The nonlinear part is solved and calculated under each system state vector set through polynomial basis functions, and the construction steps of the polynomial basis functions include: Set the cross - terms for dividing the state vectors according to the dominant degree of each state vector in the system state vector set affecting the measurement of the true value by the six - axis force sensor; Establish polynomial basis functions for characterizing the nonlinear interaction relationship between the original sensor data and each state vector through all the cross - terms.

7. The six-axis force sensor decoupling method according to claim 6, wherein, When establishing the polynomial basis functions through the cross - terms, set a weight matrix for each cross - term according to the influence degree of each output channel on the change of the system state vector set.

8. The decoupling method of the six-axis force sensor according to claim 6, wherein The steps for obtaining the nonlinear part include: Jointly form an extended input vector through the original sensor data and the system state vector set; Input the extended input vector into the polynomial basis function to obtain a six - axis nonlinear compensation amount; Use the six - axis nonlinear compensation amount as the nonlinear part.

9. The six-axis force sensor decoupling method according to any one of claims 1-8, characterized in that At each unit time, the current system state vector set is updated according to the current working condition of the carrier equipment.

10. An apparatus for implementing the decoupling method of the six-axis force sensor according to any one of claims 1-9, characterized in that, It includes: A state and data acquisition module, which is used to collect the operation state information of the carrier equipment and form the system state vector set; It is also connected to each output channel of the six-axis force sensor to obtain raw sensor data; A function and setting module is used to select or interpolate the working condition mapping function to perform real-time compensation for measurement errors caused by high-frequency changes, and at the same time set the difference compensation function according to slow or stable factors to correct stable continuous errors; A linear and non-linear decoupling module is used to separate the linear part decoupling result and the non-linear part decoupling result according to the working condition mapping function and the difference compensation function; An incremental correction and output module is used to combine the linear part and the non-linear part and perform decoupling to obtain the output data of each output channel.

Images