Block CMA-ES-based full-dimensional load decoupling system and method for six-dimensional force sensor

Abstract

The invention discloses a block CMA-ES-based full-dimensional load decoupling system and method for a six-dimensional force sensor, and the system employs a data collection and preprocessing unit to collect a voltage signal of the six-dimensional force / torque sensor, generates a model training data set through preprocessing, employs a parameter identification method based on a block adaptive covariance matrix evolution strategy, and achieves the full-dimensional load decoupling of the six-dimensional force / torque sensor. And identifying the segmented second-order polynomial multi-input multi-output decoupling model to obtain an optimal MIMO model parameter, and finally inputting a voltage signal under a full-dimensional loading condition to realize real-time output of six-dimensional force / torque information. According to the method, a sectional second-order polynomial MIMO decoupling model is adopted, parameter optimization identification is carried out based on block CMA-ES, the method belongs to a full-dimensional load decoupling method based on a single-dimensional modeling and multi-dimensional verification strategy, and the measurement precision of the sensor under the multi-dimensional load is remarkably improved while high decoupling efficiency is guaranteed.

Description

Technical Field

[0001] This invention belongs to the technical field of six-dimensional force / torque sensor decoupling, and mainly relates to a six-dimensional force sensor full-dimensional load decoupling system and method based on block CMA-ES. Background Technology

[0002] Six-dimensional force / torque sensors can simultaneously sense three-dimensional force and torque components in space, making them key components for precise force control in fields such as robotics, aerospace, and precision medical devices. However, inter-dimensional coupling interference has always been a core challenge limiting their measurement accuracy, while static decoupling algorithms are an effective way to improve accuracy.

[0003] Currently, due to the sheer number and difficulty in traversing multidimensional full-space load points, mainstream decoupling algorithms generally employ single-dimensional load data for modeling. However, existing research often focuses on improving single-dimensional accuracy, neglecting the verification and evaluation of multidimensional composite load conditions. Since sensors are frequently subjected to complex composite stresses in real-world applications, algorithms lacking multidimensional verification struggle to accurately reflect the sensor's overall performance, resulting in insufficient accuracy in practical use. Therefore, researching a decoupling algorithm that balances the convenience of single-dimensional modeling with the rigor of multidimensional verification, ensuring high-precision output under full-dimensional load conditions, is of profound significance for improving the practical application of sensors. Summary of the Invention

[0004] This invention addresses the problem that most existing multidimensional force sensor decoupling algorithms rely solely on single-dimensional load data for modeling, lacking effective verification for multidimensional composite load conditions. This results in insufficient utilization of the full-dimensional load coupling characteristics and limited measurement accuracy under complex real-world conditions. The invention provides a six-dimensional force sensor full-dimensional load decoupling system and method based on block-based CMA-ES. It utilizes a data acquisition and preprocessing unit to collect voltage signals from a six-dimensional force / torque sensor. After preprocessing, a model training dataset is generated. A parameter identification method based on a block-based adaptive covariance matrix evolution strategy is employed to identify the piecewise second-order polynomial multi-input multi-output decoupling model, obtaining the optimal MIMO model parameters. Finally, the voltage signal under full-dimensional loading conditions is input to achieve real-time output of six-dimensional force / torque information. This invention employs a piecewise second-order polynomial MIMO decoupling model and uses block-based CMA-ES for parameter optimization and identification. It belongs to the full-dimensional load decoupling method based on a "single-dimensional modeling, multi-dimensional verification" strategy, significantly improving the sensor's measurement accuracy under multidimensional loads while ensuring high decoupling efficiency.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is: a six-dimensional force sensor full-dimensional load decoupling system based on block CMA-ES, which includes at least a data acquisition and preprocessing unit, a parameter identification unit based on block CMA-ES, and a piecewise second-order polynomial MIMO decoupling unit.

[0006] The data acquisition and preprocessing unit is used to calibrate the six-dimensional force / torque sensor, acquire data from the six-dimensional force / torque sensor, and generate a dataset for model training after data preprocessing; the data preprocessing includes at least zero-point calibration.

[0007] The block-based CMA-ES parameter identification unit: Based on the preprocessed training dataset, a full parameter set containing first-order and second-order coefficient matrices in both positive and negative directions is constructed. A block strategy is adopted to divide the high-dimensional full parameter set into multiple low-dimensional parameter sub-blocks. For each low-dimensional parameter sub-block, the CMA-ES algorithm is used to independently optimize it. The root mean square error between the decoupled model output and the calibration true value is taken as the objective function to obtain the optimal parameters of each sub-block. Finally, the optimal parameters of each sub-block after optimization are recombined to obtain the globally optimal decoupled model parameters.

[0008] The piecewise second-order polynomial MIMO decoupling unit loads the optimal decoupling model parameters based on the output of the block CMA-ES parameter identification unit, and calculates and outputs the decoupled six-dimensional force / torque information based on the real-time sensor voltage signal.

[0009] As an improvement of the present invention, the six-dimensional force / torque sensor data acquired in the data acquisition and preprocessing unit is full-dimensional training data, including single-dimensional data or various composite data from two-dimensional to six-dimensional.

[0010] To achieve the above objectives, the present invention also adopts the following technical solution: a full-dimensional load decoupling method for a six-dimensional force sensor based on block-based CMA-ES, comprising the following steps:

[0011] S1. Data Acquisition: The six-dimensional force / torque sensor is installed on the weight-pump calibration platform. Specific single-dimensional and multi-dimensional force / torque loads are applied by loading and unloading the calibration weights. The six-channel voltage signal output by the data acquisition card is acquired in real time.

[0012] S2. Data preprocessing: Perform data preprocessing on the voltage signal acquired in step S1. The data preprocessing includes at least zero-point correction. The zero-point correction specifically includes: acquiring the six-channel voltage output of the sensor under no-load conditions as the zero-point reference value, and calculating the difference between the measured voltage under each loading condition and the corresponding channel zero-point reference value as the effective voltage signal.

[0013] S3. Parameter identification based on block-based CMA-ES: The positive and negative coefficient matrices of the piecewise second-order polynomial MIMO decoupling model are reduced in dimension and grouped by a block-based strategy, and CMA-ES is used to iteratively optimize each group of parameter sub-blocks to obtain the optimal decoupling model parameters.

[0014] S4. Piecewise second-order polynomial MIMO decoupling: Based on the optimal decoupling model obtained in step S3, the preprocessed voltage signal is input, and the loading direction of the input signal in each dimension is identified by the sign judgment logic. Accordingly, the positive parameter group and the negative parameter group are dynamically switched. The decoupling value of each dimension is calculated using the second-order polynomial model, and the decoupled six-dimensional force / torque information is output.

[0015] Compared with the prior art, the present invention has the following beneficial effects:

[0016] (1) The six-dimensional force / torque sensor full-dimensional load decoupling system and method based on block CMA-ES proposed in this invention improves the robustness under complex working conditions. It adopts the strategy of "single-dimensional modeling and multi-dimensional verification" and uses multi-dimensional composite load data to constrain the model, which solves the problem that the traditional algorithm loses accuracy in actual complex stress environment due to lack of multi-dimensional verification.

[0017] (2) The six-dimensional force / torque sensor full-dimensional load decoupling system and method based on block CMA-ES proposed in this invention takes into account both high decoupling accuracy and real-time performance. It adopts a piecewise second-order polynomial MIMO model, which can accurately fit the nonlinear characteristics of the sensor and has a small computational load, meeting the stringent requirements for real-time force control in fields such as robotics.

[0018] (3) The six-dimensional force / torque sensor full-dimensional load decoupling system and method based on block CMA-ES proposed in this invention reduces the dimension of the search space by dividing the parameters to be identified into blocks, which effectively improves the convergence speed of the CMA-ES algorithm. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the workflow of the block-based CMA-ES parameter identification unit in the system of the present invention;

[0020] Figure 2 This is a schematic diagram of the workflow of the piecewise second-order polynomial MIMO decoupling unit in the system of the present invention;

[0021] Figure 3 This is a schematic diagram of the structure of the piecewise second-order polynomial MIMO decoupling model of the system of the present invention. Detailed Implementation

[0022] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are for illustrative purposes only and are not intended to limit the scope of the invention.

[0023] Example 1

[0024] The six-dimensional force / torque sensor full-dimensional load decoupling system based on block CMA-ES parameter identification includes at least a data acquisition and preprocessing unit, a block CMA-ES parameter identification unit, and a piecewise second-order polynomial MIMO decoupling unit.

[0025] The data acquisition and preprocessing unit includes a weight-based calibration platform, a data acquisition card, and a signal processing module. The weight-based calibration platform is used to calibrate the six-dimensional force / torque sensor by applying specific single-dimensional and multi-dimensional force / torque loads to the sensor using suspended weights. The data acquisition card is used to acquire the sensor's six-channel voltage signals in real time and transmit them to the signal processing module, collecting full-dimensional training data, including not only single-dimensional loading data but also various composite loading data from two to six dimensions. The signal processing module runs on a host computer and receives the voltage signals output from the data acquisition card, performing data preprocessing. This data preprocessing includes at least zero-point correction and generating a model training dataset.

[0026] The block-based CMA-ES parameter identification unit is used to identify the parameters of the piecewise second-order polynomial MIMO decoupling unit, specifically as follows: Figure 1 As shown: First, the collected data is preprocessed. Second, a full parameter set containing first-order and second-order coefficient matrices in both positive and negative directions is constructed and initialized. Then, a block-based strategy is adopted to divide the high-dimensional full parameter set into multiple low-dimensional parameter sub-blocks. For each parameter sub-block, the CMA-ES algorithm is used for independent optimization, with the objective function being to minimize the root mean square error between the MIMO decoupling model output and the calibration true value. Finally, the optimized parameters of each sub-block are recombined to obtain the globally optimal decoupling model parameters.

[0027] The piecewise second-order polynomial MIMO decoupling unit loads the optimal decoupling model parameters, calculates and outputs the decoupled six-dimensional force / torque information based on the real-time sensor voltage signal preprocessed with zero-point correction, combined with positive and negative direction determination logic. Figure 2 As shown, the real-time sensor voltage signal, which has been preprocessed by zero-point correction, is input into the piecewise second-order polynomial MIMO decoupling model, and the decoupled six-dimensional force / torque information is calculated and output.

[0028] A method for full-dimensional load decoupling of a six-dimensional force / torque sensor based on block-based CMA-ES, using the above system, specifically includes the following steps:

[0029] Step S1, Data Acquisition: The six-dimensional force / torque sensor is installed on the weight-type calibration platform. Specific single-dimensional and multi-dimensional force / torque loads are applied by loading and unloading the calibration weights. The six-channel voltage signal output by the data acquisition card is acquired in real time.

[0030] Step S2, Data Preprocessing: Perform data preprocessing on the voltage signal acquired in step S1. The data preprocessing includes at least zero-point correction.

[0031] Zero-point calibration specifically includes: acquiring the six-channel voltage output of the sensor under no-load conditions as the zero-point reference value, and subtracting the corresponding channel's zero-point reference value from the original measured voltage under each loaded condition to obtain the effective voltage signal after zero-point calibration. For the first The sampling point The effective voltage signal after channel zero-point correction. For the first The sampling point The raw voltage signal acquired by the channel, For the first Zero-point reference value under no-load conditions:

[0032] , .

[0033] Step S3: Parameter identification based on block-based CMA-ES: The positive and negative coefficient matrices of the piecewise second-order polynomial MIMO decoupling model are reduced in dimension and grouped using a block-based strategy. CMA-ES is then used to iteratively optimize each group of parameter sub-blocks to obtain the optimal decoupling model parameters.

[0034] S31: Parameter vectorization and block partitioning. The parameter matrix of the piecewise second-order polynomial MIMO model to be identified is flattened into a one-dimensional full parameter vector. The block size is set, and the full parameter vector is divided into... Each parameter sub-block; the parameter matrix includes a positive first-order coefficient matrix. Negative first-order coefficient matrix positive quadratic coefficient matrix Negative quadratic coefficient matrix ;

[0035] S32: Block-based iterative optimization, using a serial or parallel approach, sequentially selects one parameter sub-block as the current optimization object, while keeping the values ​​of the remaining parameter sub-blocks fixed.

[0036] S33: CMA-ES evolutionary computation, for the current parameter sub-block, initializes the population mean, step size, and population size, generates a candidate solution population in the parameter space, substitutes the candidate solutions into the decoupled model, calculates the root mean square error (RMSE) between the model output value and the calibration reference true value as the fitness function, updates the covariance matrix and evolutionary path based on the fitness ranking results, generates the next generation population, until the preset number of iterations or convergence threshold is reached. The total number of samples in the training dataset. The first calculated for the model The first sample The estimating force of the dimension, For the first The first sample The true strength value of the dimension is used to calculate the fitness function. :

[0037]

[0038] Among them, among them, The total number of samples in the training dataset. The first calculated for the model The first sample The estimating force of the dimension, For the first The first sample The true power value of a dimension;

[0039] S34: Global parameter update and reorganization, updating the optimized parameter sub-blocks back to the full parameter vector, completing a round of full-dimensional parameter identification, and finally reorganizing the full parameter vector into the optimal parameter matrix for output.

[0040] Step S4, Piecewise Second-Order Polynomial MIMO Decoupling: Based on the optimal decoupling model obtained in Step S3, the preprocessed voltage signal is input, and the loading direction of the input signal in each dimension is identified by the sign judgment logic. Accordingly, the positive parameter group and the negative parameter group are dynamically switched. Then, the decoupling value of each dimension is calculated using the second-order polynomial model to eliminate inter-dimensional coupling interference, and the decoupled six-dimensional force / torque information is calculated and output.

[0041] S41: Direction determination and parameter selection. For each dimension of the input signal, determine its value (positive or negative). If the input signal is greater than 0, call the positive parameter set. The positive parameter set is... or The part corresponding to the positive characteristic; if the input signal is less than or equal to 0, the negative parameter set is called; this step achieves an accurate description of the sensor's tensile / compressive characteristics;

[0042] S42: Polynomial decoupled computation, based on the MIMO model, such as Figure 3As shown, the decoupling values ​​for each dimension are calculated. The calculation formula includes first-order linear terms and second-order nonlinear terms:

[0043]

[0044] in, , Representing force , Representing force , Representing force , Indicates torque , Indicates torque , Indicates torque ; The actual output voltage values ​​in various dimensions , Indicates voltage , Representing force , Representing force , Indicates torque , Indicates torque , Indicates torque Coupling interference terms in different dimensions , The definition is as follows:

[0045] ,

[0046] in, These are the secondary and primary decoupling coefficients of the sensor, respectively.

[0047] Test case

[0048] This test case selected several existing six-dimensional force decoupling algorithms for comparative performance evaluation, including least squares (LS), support vector regression (SVR), backpropagation neural network (BPNN), and the piecewise second-order polynomial MIMO decoupling method based on block CMA-ES proposed in this invention. Decoupling tests were conducted on all six-dimensional force / torque sensor calibration data. The main evaluation metric was decoupling accuracy, specifically including: single-dimensional error for verifying basic modeling capabilities, and multi-dimensional error for verifying decoupling performance under complex working conditions.

[0049] One-dimensional error is used to characterize the deviation of the measured value from the actual applied value under single-axis independent loading conditions. Multi-dimensional error is used to characterize the deviation of the measured value from the actual applied value under multi-axis compound loading conditions. The accuracy calculation formula is defined as follows:

[0050]

[0051] in, express Full-scale value of directional load; express The maximum difference between the actual load value and the true measured load value in the direction.

[0052] The above methods were used to conduct a comparative experiment on indicators, and the test results are shown in Table 1 below:

[0053] Table 1

[0054]

[0055] Table 1 above shows a comparison of the decoupling accuracy test results of different decoupling algorithms. As can be seen from Table 1, in terms of single-dimensional measurement accuracy, the method of this invention is significantly better than the LS method and comparable to the SVR and BPNN algorithms; while in terms of multi-dimensional measurement accuracy, the method of this invention is generally better than other methods.

[0056] In summary, this invention proposes a six-dimensional force / torque sensor full-dimensional load decoupling system and method based on block-based CMA-ES. It employs a "single-dimensional modeling, multi-dimensional verification" strategy to improve robustness under complex working conditions, while simultaneously ensuring high decoupling accuracy and real-time performance. Furthermore, by segmenting the parameters to be identified, the dimensionality of the search space is reduced, effectively improving the convergence speed of the CMA-ES algorithm. Through the system and method of this invention, accurate decoupling under full-dimensional loads from a six-dimensional force / torque sensor is achieved.

[0057] It should be noted that the above content merely illustrates the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. For those skilled in the art, various improvements and modifications can be made without departing from the principle of the present invention, and all such improvements and modifications fall within the scope of protection of the claims of the present invention.

Claims

1. A six-dimensional force sensor full-dimensional load decoupling system based on block-based CMA-ES, characterized in that... It includes at least a data acquisition and preprocessing unit, a block-based CMA-ES parameter identification unit, and a piecewise second-order polynomial MIMO decoupling unit; The data acquisition and preprocessing unit is used to calibrate the six-dimensional force / torque sensor, acquire data from the six-dimensional force / torque sensor, and generate a dataset for model training after data preprocessing; the data preprocessing includes at least zero-point calibration. The block-based CMA-ES parameter identification unit: Based on the preprocessed training dataset, a full parameter set containing first-order and second-order coefficient matrices in both positive and negative directions is constructed. A block strategy is adopted to divide the high-dimensional full parameter set into multiple low-dimensional parameter sub-blocks. For each low-dimensional parameter sub-block, the CMA-ES algorithm is used to independently optimize it. The root mean square error between the decoupled model output and the calibration true value is taken as the objective function to obtain the optimal parameters of each sub-block. Finally, the optimal parameters of each sub-block after optimization are recombined to obtain the globally optimal decoupled model parameters. The piecewise second-order polynomial MIMO decoupling unit loads the optimal decoupling model parameters based on the output of the block CMA-ES parameter identification unit, and calculates and outputs the decoupled six-dimensional force / torque information based on the real-time sensor voltage signal.

2. The six-dimensional force sensor full-dimensional load decoupling system based on block CMA-ES as described in claim 1, characterized in that: The six-dimensional force / torque sensor data acquired in the data acquisition and preprocessing unit is full-dimensional training data, including single-dimensional data or various composite data from two to six dimensions.

3. The six-dimensional force sensor full-dimensional load decoupling system based on block CMA-ES as described in claim 2, characterized in that: The data acquisition and preprocessing unit includes at least a weight-based calibration platform, a data acquisition card, and a signal processing module; wherein... The weight-pump calibration platform applies a load to the six-dimensional force / torque sensor by suspending weights, wherein the load is a single-dimensional or multi-dimensional force / torque load. The data acquisition card: acquires the six-channel voltage signals of the sensor in real time and transmits them to the signal processing module; The signal processing module runs on the host computer and is used to receive the voltage signal output by the data acquisition card, preprocess the data, and generate a dataset for model training.

4. A method for full-dimensional load decoupling of a six-dimensional force sensor based on block-based CMA-ES, using the system as described in claim 1, characterized in that... Includes the following steps: S1. Data Acquisition: The six-dimensional force / torque sensor is installed on the weight-pump calibration platform. Specific single-dimensional and multi-dimensional force / torque loads are applied by loading and unloading the calibration weights. The six-channel voltage signal output by the data acquisition card is acquired in real time. S2. Data preprocessing: Perform data preprocessing on the voltage signal acquired in step S1. The data preprocessing includes at least zero-point correction. The zero-point correction specifically includes: acquiring the six-channel voltage output of the sensor under no-load conditions as the zero-point reference value, and calculating the difference between the measured voltage under each loading condition and the corresponding channel zero-point reference value as the effective voltage signal. S3. Parameter identification based on block-based CMA-ES: The positive and negative coefficient matrices of the piecewise second-order polynomial MIMO decoupling model are reduced in dimension and grouped by a block-based strategy, and CMA-ES is used to iteratively optimize each group of parameter sub-blocks to obtain the optimal decoupling model parameters. S4. Piecewise second-order polynomial MIMO decoupling: Based on the optimal decoupling model obtained in step S3, the preprocessed voltage signal is input, and the loading direction of the input signal in each dimension is identified by the sign judgment logic. Accordingly, the positive parameter group and the negative parameter group are dynamically switched. The decoupling value of each dimension is calculated using the second-order polynomial model, and the decoupled six-dimensional force / torque information is output.

5. The method for full-dimensional load decoupling of a six-dimensional force sensor based on block-based CMA-ES as described in claim 4, characterized in that: Step S3, based on block-based CMA-ES parameter identification, specifically includes the following steps: S31. Parameter Vectorization and Blocking: Flatten the parameter matrix of the piecewise second-order polynomial MIMO model to be identified into a one-dimensional full parameter vector, set the block size, and divide the full parameter vector into... Each parameter sub-block; the parameter matrix includes at least a positive first-order coefficient matrix. Negative first-order coefficient matrix positive quadratic coefficient matrix Negative quadratic coefficient matrix ; S32. Block-based iterative optimization: Using a serial or parallel method, select one parameter sub-block as the current optimization object in turn, while keeping the values ​​of the remaining parameter sub-blocks fixed. S33, CMA-ES Evolutionary Calculation: For the current parameter sub-block selected in step S32, initialize the population mean, step size and population size, generate a candidate solution population in the parameter space, substitute the candidate solutions into the decoupled model, calculate the root mean square error between the model output value and the calibration reference true value as the fitness function, update the covariance matrix and evolutionary path according to the fitness ranking results, generate the next generation population, until the preset number of iterations or convergence threshold is reached; S34. Global Parameter Update and Reorganization: Update the parameter sub-blocks optimized in step S33 back into the full parameter vector to complete one round of full-dimensional parameter identification. After the parameter identification of all sub-blocks is completed, the final full parameter vector is reorganized into the optimal parameter matrix output.

6. The method for full-dimensional load decoupling of a six-dimensional force sensor based on block-based CMA-ES as described in claim 5, characterized in that: The fitness function in step S33 Specifically: ； in, The total number of samples in the training dataset. The first calculated for the model The first sample The estimating force of the dimension, For the first The first sample The true power of a dimension.

7. The method for full-dimensional load decoupling of a six-dimensional force sensor based on block-based CMA-ES as described in claim 5 or 6, characterized in that: The piecewise second-order polynomial MIMO decoupling step S4 specifically includes the following steps: S41. Direction Determination and Parameter Selection: For each dimension of the input signal, determine its value (positive or negative). If the input signal is greater than 0, call the positive parameter set; if the input signal is less than or equal to 0, call the negative parameter set. The positive parameter set is the part of the parameter matrix corresponding to the positive characteristic, and the negative parameter set is the part of the parameter matrix corresponding to the negative characteristic. S42. Polynomial Decoupling Calculation: Based on the MIMO model, calculate the decoupling values ​​for each dimension, specifically including first-order linear terms and second-order nonlinear terms. ； in, , Representing force , Representing force , Representing force , Indicates torque , Indicates torque , Indicates torque diagonal matrix , This represents the actual output voltage value in each dimension. Indicates voltage , Representing force , Representing force , Indicates torque , Indicates torque , Indicates torque , These represent coupling interference terms in different dimensions. Indicates the effect of other dimensions on force Coupling interference, Indicates the effect of other dimensions on force Coupling interference, Indicates the effect of other dimensions on force Coupling interference, Indicates the effect of other dimensions on force Coupling interference, Indicates the effect of other dimensions on force Coupling interference, Indicates the effect of other dimensions on force Coupling interference.

8. The method for full-dimensional load decoupling of a six-dimensional force sensor based on block-based CMA-ES as described in claim 7, characterized in that: In step S42, coupling interference terms of different dimensions Specifically: , ； in, These are the secondary and primary decoupling coefficients of the sensor, respectively.

Images