Performance and life prediction method of soft robots based on artificial intelligence

Abstract

The present invention provides a method for predicting the performance and life of a soft robot based on artificial intelligence, which relates to the technical field of soft robots, including combining the key machine data obtained by preprocessing the collected test data under different test conditions and finite element simulation data of the soft robot with a deep learning algorithm, constructing a life and performance prediction model to predict the life and performance of the soft robot to obtain key prediction results; optimizing the design of the soft robot according to the key prediction results. By combining the key machine data obtained by preprocessing the collected test data and finite element simulation data of the soft robot with a deep learning algorithm, constructing a life and performance prediction model to predict the life and performance of the soft robot to obtain key prediction results; optimizing the design of the soft robot according to the key prediction results, it is possible to achieve efficient prediction of the performance and life of the soft robot, provide data support for design optimization, and reduce R&D costs.

Description

Technical Field

[0001] The present invention relates to the technical field of soft robots, and in particular to a method for predicting the performance and life of soft robots based on artificial intelligence. Background Art

[0002] In recent years, soft robots, as a type of robot with high flexibility and adaptability, have been widely used in medical, agricultural, industrial and other fields. Traditional soft robot design mainly relies on experiments and finite element simulation analysis. These methods not only consume a lot of time and resources, but also have problems such as high design difficulty and high experimental costs, resulting in low design and development efficiency. Therefore, how to efficiently predict the performance and life of soft robots, provide data support for design optimization, reduce R&D costs and improve R&D efficiency has become one of the current research focuses.

[0003] Therefore, an artificial intelligence-based method for predicting the performance and lifespan of soft robots is proposed. Summary of the invention

[0004] The present invention provides a method for predicting the performance and life of a soft robot based on artificial intelligence, which is used to construct a life and performance prediction model by combining the key machine data obtained after data preprocessing of the collected soft robot test data and finite element simulation data with a deep learning algorithm to obtain key prediction results for the life and performance of the soft robot; the soft robot design is optimized according to the key prediction results, which can achieve efficient prediction of the performance and life of the soft robot, provide data support for design optimization, and reduce research and development costs.

[0005] The present invention provides a method for predicting the performance and life of a soft robot based on artificial intelligence, comprising:

[0006] Step 1: collect test data and finite element simulation data of the current soft robot under different test conditions in real time, and output them as the first experimental data;

[0007] Step 2: preprocessing the first experimental data to obtain key machine data;

[0008] Step 3: Using the key machine data, a life and performance prediction model is constructed based on deep learning to predict the performance and life of the current soft robot and obtain key prediction results;

[0009] Step 4: Optimize the current soft robot design based on the key prediction results.

[0010] Preferably, the test data and finite element simulation data of the current soft robot under different test conditions are collected in real time and output as the first experimental data, including:

[0011] According to the design characteristics and expected application scenarios of the soft robot, different expected test conditions are set;

[0012] Using the set test equipment, under each type of expected test condition, a mechanical property test is performed on the current soft robot to obtain first key test data, and the first key test data is marked with the corresponding expected test condition;

[0013] The initial machine model of the current soft robot is established by using the set finite element analysis software, and the model boundary conditions and loading conditions are set for the initial machine model according to the different material properties and geometric parameter groups that can be used by the current soft robot;

[0014] Running a finite element simulation on the initial machine model to obtain first simulation data of the current soft robot under different material properties and geometric parameter groups;

[0015] Performing sensitivity assessment and key data screening on the first simulation data to obtain first key simulation data, and marking corresponding material properties and geometric parameter groups on the first key simulation data;

[0016] The acquired first key simulation data and first key test data are output together as first experimental data.

[0017] Preferably, performing sensitivity assessment and key data screening on the first simulation data to obtain first key simulation data includes:

[0018] Expanding the real-time parameter value of the characteristic parameter of the material corresponding to the current first simulation data to obtain a characteristic parameter threshold range of the current material characteristic parameter;

[0019] For the initial machine model, sequentially changing the characteristic parameter values ​​in the finite element simulation according to the characteristic parameter threshold range, to obtain characteristic-simulation data corresponding to the current first simulation data;

[0020] Taking each characteristic parameter value in the characteristic parameter threshold interval as an independent variable and the corresponding acquired characteristic-simulation data as a dependent variable, a characteristic-simulation data change curve is established;

[0021] Performing a curve trend analysis on the characteristic-simulation data change curve to obtain a first characteristic influence coefficient of the current characteristic parameter on the first simulation data;

[0022] After performing weighted average calculation on the first characteristic influence coefficient of all acquired characteristic parameters on the current first simulation data, a first evaluation coefficient is obtained;

[0023] The geometric parameters in the corresponding geometric parameter group of the current first simulation data are enlarged or reduced as a whole according to a number of set ratios to obtain a number of first geometric parameter groups;

[0024] For the initial machine model, according to the obtained first geometric parameter group, the geometric parameter group value in the finite element simulation is changed to obtain geometric-simulation data corresponding to the current first simulation data;

[0025] Taking the set ratio as an independent variable and the corresponding geometry-simulation data of the first geometry parameter group obtained according to the set ratio as a dependent variable, a geometry-simulation data variation curve is established;

[0026] Performing a curve trend analysis on the characteristic-simulation data variation curve to obtain a second characteristic influence coefficient of the geometric parameter group on the first simulation data, and outputting the coefficient as a second evaluation coefficient;

[0027] Calculate the average value of the first evaluation coefficient and the second evaluation coefficient to obtain a sensitive evaluation coefficient;

[0028] The corresponding first simulation data whose sensitivity evaluation coefficient is greater than the set sensitivity coefficient threshold is output as the first key simulation data.

[0029] Preferably, the calculation formula of the first characteristic influence coefficient is as follows:

[0030] ; In the formula, It is represented as the first characteristic influence coefficient of the current characteristic parameter on the first simulation data; It is expressed as the slope of the curve of the current characteristic-simulation data change curve; It is represented as the maximum value of the characteristic-simulation data in the characteristic-simulation data variation curve corresponding to the current characteristic parameter; It is represented by the minimum value of the characteristic-simulation data in the characteristic-simulation data change curve corresponding to the current characteristic parameter; It is expressed as the absolute average value of the difference between the maximum and minimum values ​​of the characteristic-simulation data change curve corresponding to all characteristic parameters; It is expressed as the total number of peaks of the characteristic-simulation data change curve corresponding to the current characteristic parameter; It is expressed as the average number of times the peak value of the characteristic-simulation data change curve corresponding to all characteristic parameters appears; e is expressed as a constant, and its value is 2.7.

[0031] Preferably, the key machine data is used to construct a life and performance prediction model based on deep learning to predict the performance and life of the current soft robot, and key prediction results are obtained, including:

[0032] Dividing the key machine data according to a set division ratio to obtain target training data and target test data;

[0033] Using the target training data to train a neural network to obtain an initial prediction model;

[0034] Testing the initial prediction model based on the target test data to obtain a first test result;

[0035] By using the set model performance evaluation index to evaluate the model performance of the initial prediction model based on the first test result, a prediction performance coefficient is obtained;

[0036] When the predicted performance coefficient is greater than a set performance index threshold, the current initial prediction model is used as a life and performance prediction model;

[0037] When the predicted performance coefficient is not greater than the set performance index threshold, a set optimization algorithm is used to optimize the model performance parameters in the current initial prediction model to generate a life and performance prediction model;

[0038] The real-time basic design characteristic data of the current soft robot is obtained and input into the life and performance prediction model to obtain key prediction results.

[0039] Preferably, the basic design characteristic data refers to the geometric parameter set, material properties and working conditions of the soft robot.

[0040] Preferably, it also includes:

[0041] Regularly conduct a comprehensive analysis of the usage time and historical prediction performance of the life and performance prediction model to obtain a first dynamic adjustment coefficient;

[0042] The calculation formula of the first dynamic adjustment coefficient is as follows:

[0043] ; In the formula, It is represented as the first dynamic adjustment coefficient of the life and performance prediction model at the current moment; It is represented as the weight of the influence of the model usage time on the calculation of the first dynamic adjustment coefficient; It is represented by the usage time of the life and performance prediction model; It is represented by the i-th historical prediction performance coefficient of the current life and performance prediction model that is greater than the set performance index threshold, where i=1, 2, 3, , ; It is expressed as the total number of historical prediction performance coefficients greater than the set performance index threshold; It is expressed as the average value of the time interval between the acquisition time and the current time of all historical prediction performance coefficients greater than the set performance index threshold; It is represented by the jth historical prediction performance coefficient of the current life and performance prediction model that is not greater than the set performance index threshold, where j=1, 2, 3, , ; It is expressed as the total number of historical prediction performance coefficients that are not greater than the set performance index threshold; It is expressed as the average value of the time interval between the acquisition time and the current time of all historical prediction performance coefficients that are not greater than the set performance index threshold; It is represented as a set performance index threshold of the life and performance prediction model; It is expressed as the weight of the influence of the model prediction performance on the calculation of the first dynamic adjustment coefficient;

[0044] The first dynamic adjustment coefficient is used to adjust the model optimization cycle used by the current life and performance prediction model, and the model optimization data volume used at the last model optimization moment, to obtain a new model optimization cycle and a new model optimization data volume;

[0045] Determining the next model optimization time of the current life and performance prediction model according to the new model optimization cycle and the previous model optimization time;

[0046] Collecting new design data and new test data of the soft robot with a new model optimization data volume as training data, and inputting them into the life and performance prediction model at the next model optimization moment to perform model retraining and testing to obtain an optimized test result;

[0047] The optimization test results are combined with set model performance evaluation indicators for analysis, and the model performance parameters of the life and performance prediction model are iteratively optimized.

[0048] Preferably, according to the key prediction results, the current soft robot design is optimized, including:

[0049] Based on the key prediction results, the current soft robot design is optimized, including:

[0050] Comparing the predicted machine life and predicted machine performance in the key prediction results with the corresponding expected machine life requirements and expected machine performance requirements, respectively;

[0051] If the predicted machine life meets the expected machine life requirements, and the predicted machine performance meets the expected machine performance requirements, then there is no need to optimize the current soft robot design;

[0052] If the predicted machine life does not meet the expected machine life requirement, or the predicted machine performance does not meet the expected machine performance requirement, the design parameters in the current soft robot design are optimized using a preset optimization design scheme to obtain optimized design parameters;

[0053] The optimized design parameters are used to optimize the current soft robot design.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] By combining the key machine data obtained after data preprocessing of the collected soft robot test data and finite element simulation data with the deep learning algorithm, a life and performance prediction model is constructed to obtain key prediction results for the life and performance of the soft robot; the soft robot design is optimized based on the key prediction results, which can achieve efficient prediction of the soft robot's performance and life, provide data support for design optimization, and reduce R&D costs.

[0056] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0057] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0059] Figure 1 The present invention is a flowchart of a method for predicting the performance and life of a soft robot based on artificial intelligence. DETAILED DESCRIPTION

[0060] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0061] The embodiment of the present invention provides a method for predicting the performance and life of a soft robot based on artificial intelligence, such as Figure 1 As shown, including:

[0062] Step 1: collect test data and finite element simulation data of the current soft robot under different test conditions in real time, and output them as the first experimental data;

[0063] Step 2: preprocessing the first experimental data to obtain key machine data;

[0064] Step 3: Using the key machine data, a life and performance prediction model is constructed based on deep learning to predict the performance and life of the current soft robot and obtain key prediction results;

[0065] Step 4: Optimize the current soft robot design based on the key prediction results.

[0066] In this embodiment, the soft robot is a type of robot with high flexibility and adaptability, which is widely used in medical, agricultural, industrial and other fields; the test conditions refer to different working conditions, such as load, speed, temperature, etc.; the test data refers to the mechanical properties test data of the soft robot, including maximum stress, deformation, fatigue life, etc.; the finite element simulation data includes stress, deformation, displacement and other data.

[0067] In this embodiment, the first experimental data refers to the test data under different test conditions of the soft robot and the finite element simulation data; data preprocessing refers to data cleaning, feature extraction and standardization processing, wherein data cleaning refers to removing invalid, repeated or abnormal collected data to ensure the quality of the data; feature extraction refers to extracting the influence characteristics of the first experimental data and the basic design characteristic data associated with the current soft robot on the mechanical properties and life; the purpose of standardization is to avoid the influence of the dimensions of different features on the model training; key machine data refers to the data obtained after data preprocessing of the first experimental data and the basic design characteristic data associated with the current soft robot.

[0068] The beneficial effects of the above technical solution are: by combining the key machine data obtained after data preprocessing of the collected soft robot test data and finite element simulation data with the deep learning algorithm, a life and performance prediction model is constructed to obtain key prediction results for the life and performance of the soft robot; the soft robot design is optimized according to the key prediction results, which can achieve efficient prediction of the performance and life of the soft robot, provide data support for design optimization, and reduce R&D costs.

[0069] The embodiment of the present invention provides a method for predicting the performance and life of a soft robot based on artificial intelligence, which collects test data and finite element simulation data of the current soft robot under different test conditions in real time and outputs them as first experimental data, including:

[0070] According to the design characteristics and expected application scenarios of the soft robot, different expected test conditions are set;

[0071] Using the set test equipment, under each type of expected test condition, a mechanical property test is performed on the current soft robot to obtain first key test data, and the first key test data is marked with the corresponding expected test condition;

[0072] The initial machine model of the current soft robot is established by using the set finite element analysis software, and the model boundary conditions and loading conditions are set for the initial machine model according to the different material properties and geometric parameter groups that can be used by the current soft robot;

[0073] Running a finite element simulation on the initial machine model to obtain first simulation data of the current soft robot under different material properties and geometric parameter groups;

[0074] Performing sensitivity assessment and key data screening on the first simulation data to obtain first key simulation data, and marking corresponding material properties and geometric parameter groups on the first key simulation data;

[0075] The acquired first key simulation data and first key test data are output together as first experimental data.

[0076] In this embodiment, design characteristics refer to the basic functions and performances of the soft robot, such as the flexibility, deformability, and driving mode (such as pneumatic pressure, hydraulic pressure, electroactive polymer, etc.) of the soft robot; expected application scenarios refer to the specific tasks or environments that the soft robot is designed to perform, such as minimally invasive surgery in the medical field, passing through narrow spaces in rescue operations, and performing precision operations in industrial production; expected test conditions refer to a series of test conditions set based on the design characteristics and expected application scenarios of the soft robot, including load, speed, and ambient temperature, etc.; the setting of the test equipment is predetermined, generally a mechanical testing machine; the first key test data refers to the data obtained by performing mechanical property tests on the soft robot under different test conditions, such as maximum stress, deformation, and fatigue life; the setting of the finite element analysis software is predetermined, such as ANSYS and Abaqus.

[0077] In this embodiment, the initial machine model refers to a three-dimensional model of a soft robot established using set finite element analysis software; material properties refer to the physical and chemical properties of the materials used by the soft robot, such as elastic modulus and yield strength; the geometric parameter group refers to the shape, size and structural characteristics of the soft robot, such as length, width, height, radius of curvature, wall thickness, etc.; model boundary conditions refer to a series of constraints imposed on the soft robot model in finite element analysis, which limit the displacement or deformation of the model in certain directions and simulate the constraints that the soft robot may be subject to in actual applications; loading conditions refer to external loads applied to the soft robot model in finite element simulation, such as force, pressure, temperature, etc.; the first simulation data refers to the result data obtained by calculating the soft robot model in the finite element simulation, which is used to evaluate the performance of the soft robot under different material properties and geometric parameter groups; the first key simulation data is the data obtained after importance evaluation and key data screening of the first simulation data, reflecting the key performance of the soft robot under different conditions.

[0078] The beneficial effect of the above technical solution is that by real-time acquisition and output of the first experimental data, the performance of the soft robot under different test conditions and material properties can be comprehensively evaluated, providing an important basis for the subsequent optimization design of the soft robot.

[0079] The embodiment of the present invention provides a method for predicting the performance and life of a soft robot based on artificial intelligence, which performs sensitivity assessment and key data screening on first simulation data to obtain first key simulation data, including:

[0080] Expanding the real-time parameter value of the characteristic parameter of the material corresponding to the current first simulation data to obtain a characteristic parameter threshold range of the current material characteristic parameter;

[0081] For the initial machine model, sequentially changing the characteristic parameter values ​​in the finite element simulation according to the characteristic parameter threshold range, to obtain characteristic-simulation data corresponding to the current first simulation data;

[0082] Taking each characteristic parameter value in the characteristic parameter threshold interval as an independent variable and the corresponding acquired characteristic-simulation data as a dependent variable, a characteristic-simulation data change curve is established;

[0083] Performing a curve trend analysis on the characteristic-simulation data change curve to obtain a first characteristic influence coefficient of the current characteristic parameter on the first simulation data;

[0084] After performing weighted average calculation on the first characteristic influence coefficient of all acquired characteristic parameters on the current first simulation data, a first evaluation coefficient is obtained;

[0085] The geometric parameters in the corresponding geometric parameter group of the current first simulation data are enlarged or reduced as a whole according to a number of set ratios to obtain a number of first geometric parameter groups;

[0086] For the initial machine model, according to the obtained first geometric parameter group, the geometric parameter group value in the finite element simulation is changed to obtain geometric-simulation data corresponding to the current first simulation data;

[0087] Taking the set ratio as an independent variable and the corresponding geometry-simulation data of the first geometry parameter group obtained according to the set ratio as a dependent variable, a geometry-simulation data variation curve is established;

[0088] Performing a curve trend analysis on the characteristic-simulation data variation curve to obtain a second characteristic influence coefficient of the geometric parameter group on the first simulation data, and outputting the coefficient as a second evaluation coefficient;

[0089] Calculate the average value of the first evaluation coefficient and the second evaluation coefficient to obtain a sensitive evaluation coefficient;

[0090] The corresponding first simulation data whose sensitivity evaluation coefficient is greater than the set sensitivity coefficient threshold is output as the first key simulation data.

[0091] In this embodiment, the characteristic parameter refers to the physical and chemical properties of the material, such as elastic modulus and yield strength; the real-time parameter value refers to the material characteristic parameter value set during the current finite element simulation; the characteristic parameter threshold interval is obtained by expanding the acquired real-time parameter value, and is generally expressed as , where O1 represents the real-time parameter value; the characteristic-simulation data refers to the simulation data obtained by performing finite element simulation after sequentially changing the characteristic parameter values ​​in the finite element simulation according to the characteristic parameter threshold range.

[0092] In this embodiment, the characteristic-simulation data change curve is established by taking each characteristic parameter value in the characteristic parameter threshold interval as the independent variable and the corresponding acquired characteristic-simulation data as the dependent variable; the first characteristic influence coefficient is used to characterize the influence of the material characteristic parameter on the simulation data obtained by finite element simulation, and is obtained by extracting the curve characteristics of the current characteristic-simulation data change curve and calculating the weighted average, and the curve characteristics include the slope of the curve, the sum of the deviations between the extreme points and the average value, the curvature and the acceleration; the weight assigned to the curve characteristics is obtained by solving the matrix constructed by using the hierarchical analysis method to compare the curve characteristics pairwise and score the relative importance.

[0093] In this embodiment, the first evaluation coefficient is obtained by weighted averaging the first characteristic influence coefficient of all acquired characteristic parameters on the current first simulation data, and the value range is ,The weight assigned to the first characteristic influence coefficient is obtained by solving the matrix constructed after ,pairwise comparison and relative importance scoring of the characteristic parameters using ,AHP.

[0094] In this embodiment, the set ratio is predetermined and is used to enlarge or reduce the parameters in the geometric parameter group as a whole; the first geometric parameter group refers to the geometric parameters in the acquired geometric parameter group that are enlarged or reduced as a whole according to the set ratio. For example, the geometric parameters d1, d2 and d3 in the current geometric parameter group 1 are enlarged as a whole using the set ratio r1, and the first geometric parameter group 1 is obtained as follows: , and , the geometric parameter group 1 is reduced as a whole by setting the ratio r1, and the first geometric parameter group 2 is obtained as , and .

[0095] In this embodiment, the geometry-simulation data refers to the simulation data obtained by performing finite element simulation after sequentially changing the geometry parameter values ​​in the finite element simulation according to the first geometry parameter group; the geometry-simulation data change curve is a curve established by taking the set ratio as the independent variable and the corresponding geometry-simulation data of the first geometry parameter group obtained according to the set ratio as the dependent variable; the second characteristic influence coefficient is used to characterize the influence of the geometry parameters on the simulation data obtained by finite element simulation, and is obtained by extracting the curve features of the current geometry-simulation data change curve and calculating the weighted average, and the curve features include the curve slope, the sum of the deviations of the extreme points from the average value, the curvature and the acceleration; the weight assigned to the curve features is obtained by solving the matrix constructed by using the hierarchical analysis method to compare the curve features pairwise and score the relative importance; the second evaluation coefficient refers to the second characteristic influence coefficient, and the value range is ; The sensitive evaluation coefficient refers to the average value of the first evaluation coefficient and the second evaluation coefficient; the sensitivity coefficient threshold is set in advance, generally 0.7; the first key simulation data refers to the first simulation data whose sensitive evaluation coefficient is greater than the set sensitivity coefficient threshold.

[0096] The beneficial effects of the above technical solution are: by systematically analyzing the impact of material characteristic parameters on simulation data, observing the degree of change of simulation results, characterizing the sensitivity of various types of simulation data, and then screening and outputting key simulation data, a strong data basis is provided for subsequent soft robot performance and life prediction.

[0097] The embodiment of the present invention provides a method for predicting the performance and life of a soft robot based on artificial intelligence. The calculation formula of the first characteristic influence coefficient is as follows:

[0098] ; In the formula, It is represented as the first characteristic influence coefficient of the current characteristic parameter on the first simulation data; It is expressed as the slope of the curve of the current characteristic-simulation data change curve; It is represented as the maximum value of the characteristic-simulation data in the characteristic-simulation data variation curve corresponding to the current characteristic parameter; It is represented by the minimum value of the characteristic-simulation data in the characteristic-simulation data change curve corresponding to the current characteristic parameter; It is expressed as the absolute average value of the difference between the maximum and minimum values ​​of the characteristic-simulation data change curve corresponding to all characteristic parameters; It is expressed as the total number of peaks of the characteristic-simulation data change curve corresponding to the current characteristic parameter; It is expressed as the average number of times the peak value of the characteristic-simulation data change curve corresponding to all characteristic parameters appears; e is expressed as a constant, and its value is 2.7.

[0099] The beneficial effect of the above technical solution is that by calculating the characteristic influence coefficient of the characteristic parameter on the first simulation data, a strong data basis can be provided for calculating the sensitivity evaluation coefficient of the first simulation data, thereby helping to characterize the sensitivity of various types of simulation data.

[0100] The embodiment of the present invention provides a method for predicting the performance and life of a soft robot based on artificial intelligence. The method uses the key machine data to construct a life and performance prediction model based on deep learning, predicts the performance and life of the current soft robot, and obtains key prediction results, including:

[0101] Dividing the key machine data according to a set division ratio to obtain target training data and target test data;

[0102] Using the target training data to train a neural network to obtain an initial prediction model;

[0103] Testing the initial prediction model based on the target test data to obtain a first test result;

[0104] By using the set model performance evaluation index to evaluate the model performance of the initial prediction model based on the first test result, a prediction performance coefficient is obtained;

[0105] When the predicted performance coefficient is greater than a set performance index threshold, the current initial prediction model is used as a life and performance prediction model;

[0106] When the predicted performance coefficient is not greater than the set performance index threshold, a set optimization algorithm is used to optimize the model performance parameters in the current initial prediction model to generate a life and performance prediction model;

[0107] The real-time basic design characteristic data of the current soft robot is obtained and input into the life and performance prediction model to obtain key prediction results.

[0108] In this embodiment, the performance prediction model is used to predict the mechanical properties of the soft robot, such as the maximum stress , Maximum deformation etc. The model input is the design parameters X = {h, ​​L, R, θ, E, ν}, and the output is the predicted performance index;

[0109] The commonly used deep learning model is the multi-layer perceptron (MLP). For example, a three-layer neural network is used for performance prediction. The model form is:

[0110]

[0111] Among them, W1, W2, W3 are weight matrices, b1, b2, b3 are bias terms, σ is the activation function (such as ReLU or Sigmoid), X is the input feature, and the output is the predicted maximum stress .

[0112] In this embodiment, the life prediction model is based on the fatigue behavior of the soft robot, and the fatigue data of the material (such as the SN curve) is usually used to predict the fatigue life of the soft robot;

[0113] Life prediction formulas usually use the rain flow counting method or Miner's rule to estimate fatigue life;

[0114] Fatigue life It can be estimated based on the SN curve:

[0115]

[0116] Where C is the material constant, b is the fatigue index of the material, is the maximum stress, is the predicted fatigue life.

[0117] The model optimizes the parameters C and b by learning from a large amount of design data in order to accurately predict the fatigue life of the soft robot.

[0118] In this embodiment, the setting of the division ratio is predetermined, generally target training data: target test data = 4:1; the target training data and the target test data are data obtained after dividing the set ratio of the key machine; the first test result refers to the output obtained after testing the initial prediction model using the target test data, including the robot test life and the robot test performance; the model performance evaluation index is set to quantify the standard or metric of the model performance, including accuracy, recall rate, F1 score, mean square error and root mean square error; the prediction performance coefficient is obtained by weighted averaging the index values ​​obtained by evaluating the model performance of the initial prediction model based on the set model performance evaluation index; the setting of the performance index threshold is predetermined, generally 0.8.

[0119] In this embodiment, the set optimization algorithm is a pre-set algorithm for adjusting model parameters, such as stochastic gradient descent; the model performance parameters include a weight matrix, a bias term, and a learning rate, etc.;

[0120] For example, during optimization use:

[0121]

[0122] in, is the learning rate, It is expressed as the gradient of the loss function, and θ is the parameter of the model.

[0123] Through continuous iterative optimization, the predictive performance of the model will continue to improve, ensuring that the soft robot always maintains optimal performance under changing design conditions.

[0124] In this embodiment, real-time basic design characteristic data refers to the basic design characteristic data in the current soft robot design, and the basic design characteristic data refers to the key data related to the soft robot design, including the geometric parameter group of the soft robot, the material characteristics and the working conditions, wherein the geometric parameter group includes the geometric dimensions and shape parameters of each component of the soft robot, such as wall thickness, structural length, bending radius, curvature, etc.; the material characteristics refer to the mechanical properties data of the material (such as silicone, elastomer, etc.), including elastic modulus, yield strength, stress-strain relationship, tensile strength, etc.; working conditions refer to the conditions of the soft robot in the working environment, such as working pressure, temperature, load frequency, stress application, etc.; key prediction results refer to the predicted machine life and predicted machine performance, wherein the predicted machine life refers to the service life predicted by the model according to the design characteristics of the soft robot; the predicted machine performance refers to the mechanical properties predicted by the model according to the design characteristics of the soft robot, such as maximum stress and maximum deformation.

[0125] The beneficial effect of the above technical solution is: by predicting the performance and life of the current soft robot, key prediction results are obtained, which can provide data support for subsequent soft robot design optimization, thereby reducing R&D costs and improving R&D efficiency.

[0126] The embodiment of the present invention provides a method for predicting the performance and life of a soft robot based on artificial intelligence, further comprising:

[0127] Regularly conduct a comprehensive analysis of the usage time and historical prediction performance of the life and performance prediction model to obtain a first dynamic adjustment coefficient;

[0128] The calculation formula of the first dynamic adjustment coefficient is as follows:

[0129] ; In the formula, It is represented as the first dynamic adjustment coefficient of the life and performance prediction model at the current moment; It is represented as the weight of the influence of the model usage time on the calculation of the first dynamic adjustment coefficient; It is represented by the usage time of the life and performance prediction model; It is represented by the i-th historical prediction performance coefficient of the current life and performance prediction model that is greater than the set performance index threshold, where i=1, 2, 3, , ; It is expressed as the total number of historical prediction performance coefficients greater than the set performance index threshold; It is expressed as the average value of the time interval between the acquisition time and the current time of all historical prediction performance coefficients greater than the set performance index threshold; It is represented by the jth historical prediction performance coefficient of the current life and performance prediction model that is not greater than the set performance index threshold, where j=1, 2, 3, , ; It is expressed as the total number of historical prediction performance coefficients that are not greater than the set performance index threshold; It is expressed as the average value of the time interval between the acquisition time and the current time of all historical prediction performance coefficients that are not greater than the set performance index threshold; It is represented as a set performance index threshold of the life and performance prediction model; It is expressed as the weight of the influence of the model prediction performance on the calculation of the first dynamic adjustment coefficient;

[0130] The first dynamic adjustment coefficient is used to adjust the model optimization cycle used by the current life and performance prediction model, and the model optimization data volume used at the last model optimization moment, to obtain a new model optimization cycle and a new model optimization data volume;

[0131] Determining the next model optimization time of the current life and performance prediction model according to the new model optimization cycle and the previous model optimization time;

[0132] Collecting new design data and new test data of the soft robot with a new model optimization data volume as training data, and inputting them into the life and performance prediction model at the next model optimization moment to perform model retraining and testing to obtain an optimized test result;

[0133] The optimization test results are combined with set model performance evaluation indicators for analysis, and the model performance parameters of the life and performance prediction model are iteratively optimized.

[0134] In this embodiment, the first dynamic adjustment coefficient is used to dynamically adjust the optimization period and optimization data volume of the life and performance prediction model; the historical prediction performance refers to the historical prediction effectiveness of the life and performance prediction model, which is expressed by the corresponding calculated historical prediction performance coefficient; the weights assigned to the model usage time and model prediction performance are obtained by solving the matrix constructed after pairwise comparison and relative importance scoring using the hierarchical analysis method; the model optimization period refers to the optimization time interval of the life and performance prediction model; the model optimization data volume refers to the amount of data used for model optimization; the last model optimization moment refers to the time point when the model was last optimized.

[0135] In this embodiment, the new model optimization period refers to the optimization period adjusted according to the first dynamic adjustment coefficient, such as ,in, Represented as the new optimization cycle of the model; Represented as model optimization cycle; It is expressed as the first dynamic adjustment coefficient; the new optimized data volume of the model refers to the data volume adjusted according to the first dynamic adjustment coefficient, such as ,in, Indicates the amount of new optimized data for the model; Indicates the amount of data optimized for the model; Expressed as the first dynamic adjustment coefficient; the next model optimization moment refers to the time point calculated based on the new model optimization cycle and the previous model optimization moment; the new design data refers to the new design data generated by the current soft robot during the period from the last model optimization, such as new geometric parameter groups and material properties; the new test data refers to the data obtained by testing and finite element simulation of the current soft robot under new test conditions during the period from the last model optimization; the optimization test result refers to the output result obtained after retraining and testing the model using the new design data and new test data, which is used to evaluate the performance of the model after optimization.

[0136] The beneficial effect of the above technical solution is: by dynamically adjusting the model optimization cycle and data volume, and iteratively optimizing the model, it can effectively ensure that the model always maintains a high level of prediction accuracy and stability, which in turn helps to efficiently predict the performance and life of current soft robots.

[0137] The embodiment of the present invention provides a method for predicting the performance and life of a soft robot based on artificial intelligence, and according to the key prediction results, the current soft robot design is optimized, including:

[0138] Comparing the predicted machine life and predicted machine performance in the key prediction results with the corresponding expected machine life requirements and expected machine performance requirements, respectively;

[0139] If the predicted machine life meets the expected machine life requirements, and the predicted machine performance meets the expected machine performance requirements, then there is no need to optimize the current soft robot design;

[0140] If the predicted machine life does not meet the expected machine life requirement, or the predicted machine performance does not meet the expected machine performance requirement, the design parameters in the current soft robot design are optimized using a preset optimization design scheme to obtain optimized design parameters;

[0141] The optimized design parameters are used to optimize the current soft robot design.

[0142] In this embodiment, the predicted machine life refers to the service life predicted by the model based on the design characteristics of the soft robot; the predicted machine performance refers to the mechanical properties, such as maximum stress and maximum deformation, predicted by the model based on the design characteristics of the soft robot.

[0143] In this embodiment, the expected machine life requirement refers to the predetermined expectation of the service life of the soft robot; the expected machine performance requirement refers to the predetermined expectation of the mechanical properties of the soft robot, including maximum stress, maximum deformation, stiffness, etc.; the preset optimization design scheme is used to adjust the design parameters of the soft robot to meet the expected machine life and performance requirements; the design parameters refer to the geometric parameter group and the material;

[0144] In this embodiment, the preset optimization design scheme is predetermined, specifically referring to optimizing the design parameters by minimizing or maximizing the objective function (such as performance index or life) using intelligent optimization algorithms such as genetic algorithm (GA) and particle swarm optimization (PSO);

[0145] For example, the optimization goal could be to maximize fatigue life:

[0146]

[0147] At the same time, the optimization process also needs to consider cost functions (such as material cost, manufacturing difficulty, etc.) and constraints (such as maximum load, maximum deformation, etc.).

[0148] The beneficial effect of the above technical solution is: by optimizing the current soft robot design according to the key prediction results, the design efficiency and accuracy can be effectively improved, thereby reducing R&D costs and improving R&D efficiency.

[0149] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for predicting the performance and life of a soft robot based on artificial intelligence, characterized in that: include: Step 1: collect test data and finite element simulation data of the current soft robot under different test conditions in real time, and output them as the first experimental data, wherein the first experimental data includes: first key test data and first key simulation data, establish the initial machine model of the current software robot using the set finite element analysis software, run finite element simulation on the initial machine model, obtain the first simulation data of the current software robot under different material properties and geometric parameter groups, perform sensitivity assessment and key data screening on the first simulation data, and obtain the first key simulation data, which specifically includes: Expanding the real-time parameter value of the characteristic parameter of the material corresponding to the current first simulation data to obtain a characteristic parameter threshold range of the current material characteristic parameter; For the initial machine model, sequentially changing the characteristic parameter values ​​in the finite element simulation according to the characteristic parameter threshold range, to obtain characteristic-simulation data corresponding to the current first simulation data; Taking each characteristic parameter value in the characteristic parameter threshold interval as an independent variable and the corresponding acquired characteristic-simulation data as a dependent variable, a characteristic-simulation data change curve is established; Performing a curve trend analysis on the characteristic-simulation data change curve to obtain a first characteristic influence coefficient of the current characteristic parameter on the first simulation data; After performing weighted average calculation on the first characteristic influence coefficient of all acquired characteristic parameters on the current first simulation data, a first evaluation coefficient is obtained; The geometric parameters in the corresponding geometric parameter group of the current first simulation data are enlarged or reduced as a whole according to a number of set ratios to obtain a number of first geometric parameter groups; For the initial machine model, according to the obtained first geometric parameter group, the geometric parameter group value in the finite element simulation is changed to obtain geometric-simulation data corresponding to the current first simulation data; Taking the set ratio as an independent variable and the corresponding geometry-simulation data of the first geometry parameter group obtained according to the set ratio as a dependent variable, a geometry-simulation data variation curve is established; Performing a curve trend analysis on the characteristic-simulation data variation curve to obtain a second characteristic influence coefficient of the geometric parameter group on the first simulation data, and outputting the coefficient as a second evaluation coefficient; Calculate the average value of the first evaluation coefficient and the second evaluation coefficient to obtain a sensitive evaluation coefficient; Outputting the corresponding first simulation data whose sensitivity evaluation coefficient is greater than the set sensitivity coefficient threshold as the first key simulation data; Step 2: preprocessing the first experimental data to obtain key machine data; Step 3: Using the key machine data, a life and performance prediction model is constructed based on deep learning to predict the performance and life of the current soft robot and obtain key prediction results; Step 4: Optimize the current soft robot design based on the key prediction results.

2. The method for predicting performance and life of a soft robot based on artificial intelligence according to claim 1, characterized in that: The calculation formula of the first characteristic influence coefficient is as follows: ; In the formula, It is represented as the first characteristic influence coefficient of the current characteristic parameter on the first simulation data; It is expressed as the slope of the curve of the current characteristic-simulation data change curve; It is represented as the maximum value of the characteristic-simulation data in the characteristic-simulation data variation curve corresponding to the current characteristic parameter; It is represented by the minimum value of the characteristic-simulation data in the characteristic-simulation data change curve corresponding to the current characteristic parameter; It is expressed as the absolute average value of the difference between the maximum and minimum values ​​of the characteristic-simulation data change curve corresponding to all characteristic parameters; It is expressed as the total number of peaks of the characteristic-simulation data change curve corresponding to the current characteristic parameter; It is expressed as the average number of times the peak value of the characteristic-simulation data change curve corresponding to all characteristic parameters appears; e is expressed as a constant, and its value is 2.

7.

3. The method for predicting performance and life of a soft robot based on artificial intelligence according to claim 1, characterized in that: Basic design characteristic data refers to the geometric parameter set of the soft robot, the material properties used, and the working conditions.

4. The method for predicting performance and life of a soft robot based on artificial intelligence according to claim 1, characterized in that: Based on the key prediction results, the current soft robot design is optimized, including: Comparing the predicted machine life and predicted machine performance in the key prediction results with the corresponding expected machine life requirements and expected machine performance requirements, respectively; If the predicted machine life meets the expected machine life requirements, and the predicted machine performance meets the expected machine performance requirements, then there is no need to optimize the current soft robot design; If the predicted machine life does not meet the expected machine life requirement, or the predicted machine performance does not meet the expected machine performance requirement, the design parameters in the current soft robot design are optimized using a preset optimization design scheme to obtain optimized design parameters; The optimized design parameters are used to optimize the current soft robot design.

Images