CN117428768A - Stewart parallel platform pose deviation prediction and compensation method based on LSTM-ANN network - Google Patents
Stewart parallel platform pose deviation prediction and compensation method based on LSTM-ANN network Download PDFInfo
- Publication number
- CN117428768A CN117428768A CN202311432180.6A CN202311432180A CN117428768A CN 117428768 A CN117428768 A CN 117428768A CN 202311432180 A CN202311432180 A CN 202311432180A CN 117428768 A CN117428768 A CN 117428768A
- Authority
- CN
- China
- Prior art keywords
- pose
- lstm
- deviation
- ann
- nominal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000012549 training Methods 0.000 claims abstract description 20
- 238000012360 testing method Methods 0.000 claims abstract description 18
- 210000002569 neuron Anatomy 0.000 claims description 21
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000005259 measurement Methods 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000000386 athletic effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J9/00—Programme-controlled manipulators
- B25J9/16—Programme controls
- B25J9/1602—Programme controls characterised by the control system, structure, architecture
- B25J9/161—Hardware, e.g. neural networks, fuzzy logic, interfaces, processor
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J9/00—Programme-controlled manipulators
- B25J9/16—Programme controls
- B25J9/1628—Programme controls characterised by the control loop
- B25J9/163—Programme controls characterised by the control loop learning, adaptive, model based, rule based expert control
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J9/00—Programme-controlled manipulators
- B25J9/16—Programme controls
- B25J9/1628—Programme controls characterised by the control loop
- B25J9/1653—Programme controls characterised by the control loop parameters identification, estimation, stiffness, accuracy, error analysis
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J9/00—Programme-controlled manipulators
- B25J9/16—Programme controls
- B25J9/1656—Programme controls characterised by programming, planning systems for manipulators
- B25J9/1661—Programme controls characterised by programming, planning systems for manipulators characterised by task planning, object-oriented languages
Landscapes
- Engineering & Computer Science (AREA)
- Robotics (AREA)
- Mechanical Engineering (AREA)
- Automation & Control Theory (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Numerical Control (AREA)
Abstract
The invention discloses a method for predicting and compensating pose deviation of a Stewart parallel platform based on an LSTM-ANN network, which comprises the steps of firstly, selecting a nominal pose of the Stewart parallel platform in a grid dividing mode, controlling the Stewart parallel platform to realize fixed-point movement of the corresponding pose, and obtaining actual pose data of the Stewart parallel platform by a pose measuring instrument to construct a nominal pose and actual pose deviation data set; secondly, establishing a pose deviation regression prediction model LSTM-ANN of a Stewart parallel platform combined with LSTM and ANN; furthermore, the deviation data set is divided into a training set and a testing set according to the ratio of 4:1, and the nominal pose in the training set and the corresponding deviation are respectively used as the input and the output of the LSTM-ANN model to realize the training of the model; and finally, predicting the corresponding pose deviation through the nominal pose in the test set, and pre-compensating the predicted pose deviation to the nominal pose as the input of the Stewart parallel platform controller. The method effectively solves the problems of complex traditional kinematic calibration, poor error model, poor universality and the like.
Description
Technical Field
The invention belongs to the technical field of deep learning and parallel robots, and particularly relates to a position and attitude deviation prediction and compensation method for a Stewart parallel platform based on LSTM-ANN, which is used for improving the precision of the Stewart parallel platform.
Background
Stewart parallel platform is a typical parallel robot, and can output motion with 6 degrees of freedom. In recent years, it plays an indispensable role in the fields of flight simulators, robotic processing, and underwater research. Static pose accuracy is one of the most important performance indexes of the Stewart parallel platform. The pose accuracy of a Stewart parallel platform depends on an accurate description of the inverse kinematics function in its local computer controller. Because the actual Stewart parallel platform is inevitably subject to manufacturing tolerances, link offsets, and other sources of error, the resulting pose driven by nominal joint variables differs from the desired pose, which results in the theoretical inverse kinematics of the Stewart parallel platform as represented by the design parameters not being consistent with the actual inverse kinematics of the same Stewart parallel platform. At present, the theoretical inverse kinematics function in the local computer controller of the Stewart parallel platform does not really represent the actual inverse kinematics. Thus, the actual pose of the Stewart parallel platform, obtained by controlling the motion value of the actuator, will deviate from the target pose. An effective technical approach to solving the pose deviation is to update the theoretical inverse kinematics through the kinematic calibration. The athletic school calibration includes creating a more accurate geometric model describing the relationship between the actual pose of the Stewart parallel platform and all of the drive leg lengths. And then, modifying the system control software of the Stewart parallel platform according to the established geometric model. The exercise calibration generally comprises four steps, namely, establishing a proper geometric parameter error model, measuring a set of poses of the Stewart parallel platform, identifying actual geometric parameter errors and compensating geometric errors. However, many of the system control software of the Stewart parallel platform is not open, which results in the failure of the exercise school level. Furthermore, the error sources of the Stewart platform include not only geometric errors, but also non-geometric errors such as gravity, inertia, gear backlash and the like, and it is almost impossible to build a complete error model taking all the error sources into consideration. Aiming at the problem, a model-free calibration method is gradually applied to pose calibration of the Stewart parallel platform, the Stewart parallel platform is regarded as a black box, and a mapping relation between pose errors and configuration of the Stewart parallel platform is established through an artificial neural network.
In combination, the existing Stewart parallel platform pose calibration method adopting the kinematic and dynamic models has the problems of complex kinematic calibration, unhealthy error model, weak universality and the like.
Disclosure of Invention
The invention aims to provide a Stewart parallel platform pose deviation prediction and compensation method based on an LSTM-ANN network. Aiming at the problems of complex kinematic calibration, poor error model, poor universality and the like of the prior Stewart parallel platform pose calibration method adopting a kinematic and dynamic model, the invention provides the LSTM-ANN-based Stewart parallel platform pose deviation prediction and compensation method which is flexible, efficient, accurate in prediction and high in adaptability, and has great promotion effect on improving the precision of parallel robots and establishing a digital twin model of an industrial robot.
The technical scheme of the invention is as follows: a Stewart parallel platform pose deviation prediction and compensation method based on LSTM-ANN network comprises the following steps:
s1: selecting a large number of nominal positions of the Stewart parallel platforms in a grid dividing mode, controlling the Stewart parallel platforms to realize fixed-point movement of the corresponding positions, acquiring actual positions of the Stewart parallel platforms by using a position measuring instrument, calculating position deviation between the actual positions and the nominal positions, and constructing a nominal position deviation data set and an actual position deviation data set;
s2: establishing a pose deviation regression prediction model LSTM-ANN of a Stewart parallel platform combined with LSTM and ANN, wherein the input of the model is the nominal pose of the Stewart parallel platform, and the output is pose deviation;
s3: dividing a deviation data set into a training set and a testing set according to a ratio of 4:1, selecting different numbers of the ANN neurons, training an LSTM-ANN model through training set data, and determining the most suitable number of the neurons;
s4: and taking the nominal pose of the Stewart parallel platform in the test set data as the input of the LSTM-ANN model, realizing corresponding pose deviation prediction, and pre-compensating the predicted pose deviation to the nominal pose of the Stewart parallel platform to be used as the input of the controller.
In the foregoing method for predicting and compensating pose deviation of Stewart parallel platforms based on LSTM-ANN network, in step S1, a large number of nominal poses of Stewart parallel platforms are selected by dividing grids, which can be specifically described as: dividing the position space and the gesture space of the Stewart parallel platform into 5X 5 grids, obtaining intersection points of 125 position spaces and 125 gesture spaces altogether, and taking the positions and the gestures on the 250 intersection points as nominal gestures;
the pose bias between the actual pose and the nominal pose can be described as:
ΔQ＝Q n -Q a (1)
wherein DeltaQ represents pose deviation, Q n Represents the nominal pose, Q a Representing the actual measurement pose.
In the aforementioned method for predicting and compensating pose deviation of Stewart parallel platform based on LSTM-ANN network, in step S2, the model for predicting the pose deviation by regression includes an input layer, an output layer, an LSTM layer, a full connection layer and an output layer; the input layer and the output layer are respectively composed of 6 network nodes, and respectively correspond to 6 nominal pose inputs and 6 pose deviation outputs, the LSTM layer is composed of 6 LSTM units, the output mode of the LSTM is Last, namely the output result of the Last LSTM unit is output as the LSTM layer.
In the foregoing method for predicting and compensating pose deviation of Stewart parallel platform based on LSTM-ANN network, in the step S3, the number of neurons of ANN is selected to be 10, 15, 20 and 25, the LSTM-ANN model is trained and tested by using deviation data set, and the coefficient R is determined by mean absolute error MAE 2 Determining an optimumNeuron count, MAE and R 2 The detailed calculation formula of (2) is as follows:
wherein:
wherein the predicted value is expressed asThe actual value is denoted as x j The average of the actual values is expressed as +.>n represents the number of poses used for the test.
In the foregoing method for predicting and compensating the pose deviation of the Stewart parallel platform based on the LSTM-ANN network, in step S4, the predicted pose deviation is precompensated into the nominal pose of the Stewart parallel platform, which is specifically implemented as follows: predicting pose deviation delta Q by using nominal pose of test set and LSTM-ANN model p ＝[Δx,Δy,Δz,Δα,Δβ,Δγ]Will DeltaQ p Precompensation to nominal pose Q n Then get the corrected pose Q m Using Q m Instead of Q n As input to the controller, Q m Can be expressed as:
Q m ＝Q n -ΔQ p (5)。
the invention has the beneficial effects that: compared with the prior art, the method for predicting and compensating the pose deviation of the Stewart parallel platform based on the LSTM-ANN is flexible, efficient, accurate in prediction and high in adaptability, the method predicts unknown pose deviation through the LSTM-ANN neural network, and compensates the pose deviation to the expected pose in advance, so that the static pose precision of the Stewart parallel platform is greatly improved. In addition, the traditional kinematic calibration needs to build different error models for different parallel platforms, and the process is quite complex. The method disclosed by the invention can be applied to different parallel platforms while maintaining higher precision, and has a great promotion effect on improving the precision of the parallel robot and establishing a digital twin model of the industrial robot.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a block diagram of an LSTM-ANN network of the method of the present invention;
FIG. 3 is a comparison of the result of predicting the pose deviation of the Stewart parallel platform and the true pose deviation based on the method;
fig. 4 shows the pose deviation and the true pose deviation after the compensation of the Stewart parallel platform based on the method.
Detailed Description
The invention is further illustrated by the following figures and examples, which are not intended to be limiting.
Embodiments of the invention: aiming at the problems of complex kinematic calibration, poor error model, poor universality and the like of the traditional six-degree-of-freedom Stewart parallel platform pose calibration method based on a kinematic and dynamic model, the invention provides an efficient, flexible, precise and highly-adaptive method for predicting and compensating the pose deviation of a Stewart parallel platform based on an LSTM-ANN network, which specifically comprises the following steps:
referring to fig. 1, a flow chart of the method of the present invention mainly comprises the following steps:
step S1: selecting a large number of nominal positions of the Stewart parallel platforms in a grid dividing mode, controlling the Stewart parallel platforms to realize fixed-point movement of the corresponding positions, acquiring actual positions of the Stewart parallel platforms by using a position measuring instrument, calculating position deviation between the actual positions and the nominal positions, and constructing a nominal position deviation data set and an actual position deviation data set;
step S2: establishing a pose deviation regression prediction model LSTM-ANN of a Stewart parallel platform combining LSTM and ANN, wherein the input of the model is the nominal pose of the Stewart parallel platform, the output of the LSTM layer is used as the input of the ANN, and the output of the ANN is the pose deviation;
step S3: dividing a deviation data set into a training set and a test set according to the ratio of 4:1, selecting different numbers of the ANN neurons, training LSTM-ANN models with different numbers of the neurons through training set data, and determining the most suitable number of the neurons;
step S4: and taking the nominal pose of the Stewart parallel platform in the test set data as the input of the LSTM-ANN model, realizing corresponding pose deviation prediction, and pre-compensating the predicted pose deviation to the nominal pose of the Stewart parallel platform to be used as the input of a controller, so that the pose deviation of the parallel robot can be compensated.
In the step S1, the position space and the posture space of the Stewart parallel platform are divided into a grid of 5 x 5, and since the Stewart parallel platform is a six-degree-of-freedom parallel robot, three positions and three angles of movement are possible, so that an intersection of 125 position spaces and 125 attitude spaces can be obtained. The positions and attitudes at these 250 intersections are taken as nominal poses. The fixed-point motion of the corresponding pose is realized through the controller, the actual pose of the Stewart parallel platform is measured by using the pose measuring instrument, the pose deviation between the actual pose and the nominal pose is calculated, and the nominal pose and the actual pose deviation data set of the Stewart parallel platform is constructed.
The binocular vision measuring system (movedirect XR 8) adopted in the embodiment measures the pose of the parallel platform with the measuring precision of +/-20 μm, but the method is aimed at the deviation prediction and compensation of the Stewart parallel platform, and the method is not limited to pose measuring equipment, and can also be used for measuring by other pose measuring equipment.
The pose bias between the actual pose and the nominal pose can be described as:
ΔQ＝Q n -Q a (1)
wherein DeltaQ represents pose deviation, Q n Represents the nominal pose, Q a Representing the actual measurement pose. Since the pose deviation is required to be compensated later, the actual measured pose must be subtracted from the nominal pose when the pose deviation is calculated, otherwise the pose increases after compensation.
In the step S2, the pose deviation regression prediction model LSTM-ANN comprises an input layer, an output layer, an LSTM layer, a full connection layer and an output layer. The input layer and the output layer are respectively composed of 6 network nodes, and respectively correspond to 6 nominal pose inputs and 6 pose deviation outputs, the LSTM layer is composed of 6 LSTM units, the output mode of the LSTM is Last, namely the output result of the Last LSTM unit is output as the LSTM layer.
In the step S3, different numbers of ANN neurons are selected, the LSTM-ANN model is trained through training set data, and the most suitable number of the neurons is determined. The number of neurons of ANN was selected to be 10, 15, 20, 25, and the LSTM-ANN model was trained and tested using the bias dataset, respectively, by mean absolute error (Mean Absolute Error, MAE), coefficient of determination (Determination Coefficient, R 2 ) An optimal number of neurons is determined. MAE and R 2 The detailed calculation formula of (2) is as follows:
wherein:
wherein the predicted value is expressed asThe actual value is denoted as x j The average of the actual values is expressed as +.>n represents the number of poses used for the test.
The step S4 is to pre-compensate the predicted pose deviation to the nominal pose of the Stewart parallel platform, and is specifically implemented as follows: predicting pose deviation delta Q by using nominal pose of test set and LSTM-ANN model p ＝[Δx,Δy,Δz,Δα,Δβ,Δγ]Will DeltaQ p Precompensation to nominal pose Q n Then get the corrected pose Q m Using Q m Instead of Q n As input to the controller, Q m Can be expressed as:
Q m ＝Q n -ΔQ p (5)。
the above may be summarized in the following aspects:
building a deviation data set of the nominal pose and the actual pose: dividing the position space and the gesture space of the Stewart parallel platform into a grid of 5 multiplied by 5, obtaining the intersection points of 125 position spaces and 125 gesture spaces, and taking the positions and the gestures on the 250 intersection points as the nominal gestures. The fixed-point motion of the corresponding pose is realized through the controller, the actual pose of the Stewart parallel platform is measured by using the pose measuring instrument, the pose deviation between the actual pose and the nominal pose is calculated, and a data set of the nominal pose and the actual pose deviation of the Stewart parallel platform is constructed;
building an LSTM-ANN deep learning model: building a pose deviation regression prediction model LSTM-ANN of the Stewart parallel platform by combining LSTM and ANN, wherein the input of the model is the nominal pose of the six-degree-of-freedom Stewart parallel platform, and the output is pose deviation;
determining appropriate LSTM-ANN network parameters: the bias dataset was divided into training and test sets at 4:1. Selecting the number of neurons of different ANNs, training an LSTM-ANN model through training set data, and determining the most suitable number of neurons;
predicting and compensating pose deviation of Stewart parallel platform: and taking the nominal pose of the Stewart parallel platform in the test set data as the input of the LSTM-ANN model, realizing corresponding pose deviation prediction, and pre-compensating the predicted pose deviation to the nominal pose of the Stewart parallel platform to replace the nominal pose as the input of the controller.
Referring to fig. 2, there are six network nodes in the input layer and the output layer for inputting six parameters of the nominal pose and outputting six parameters of the pose deviation, respectively, in the LSTM-ANN model structure diagram of the method of the present invention. The LSTM layer is used for learning the dependency relationship between the sequence and the related characteristics, and the LSTM layer is added with a ReLU nonlinear activation function, so that the network captures the nonlinear characteristics among the input data. The main function of the fully connected layer is to establish a connection between each neuron of the previous layer and each neuron of the present layer, which integrates the various features into a single value, and finally, the output layer outputs the predicted pose deviation. Training parameters for LSTM-ANN neural networks include: the iteration times, the learning rate, the attenuation step number and the optimizer are respectively set to 1500, 0.01, 0.1, 1200 and Adam.
In order to verify the effectiveness of the Stewart parallel platform pose deviation prediction and compensation method based on the LSTM-ANN network, 200 poses are randomly selected from 250 poses, 160 poses are used as training sets, and 40 groups of poses are used as test sets. Table 1 shows parameters of the Stewart parallel platform used in the validation test, wherein R p And R is b Representing the radius, θ, of the moving platform and the fixed platform, respectively p And theta b The angles of two adjacent joints on the movable platform and the fixed platform are respectively. Parameter l min And l min Respectively minimum and maximum lengths of telescopic legs, z p And z b Representing the distances of the joint center from the surfaces of the mobile platform and the fixed platform, respectively. Training LSTM-ANN models with different neuron numbers by training set data through MAE and R 2 The number of neurons is determined. The smaller the MAE, the better the performance of the model, R 2 Representing the fitting degree of the model to the data, R 2 The closer to 1, the better the representing effect, and the number of neurons finally determined is 15. FIG. 3 shows the result of comparing the predicted value and the actual value of the pose deviation of the LSTM-ANN model, and Table 2 shows the evaluation index of the performance of the pose deviation predicted by the model.
TABLE 1 Stewart parallel platform parameters
| R p /mm | R b /mm | θ p /° | θ b /° | l min /mm | l max /mm | z p /mm | z b /mm |
| 124 | 454 | 30 | 20 | 696 | 838 | 49 | 39 |
Referring to fig. 3, a comparison of predicted pose bias with actual pose bias using LSTM-ANN. To evaluate the predicted results, three evaluation indexes listed in table 1 were used to evaluate that the MAE for predicting pose deviation was less than 0.0498 and rmse was less than 0.0522.
TABLE 2 evaluation index of LSTM-ANN model
| Pose parameter | RMSE(mm) | MAE(mm) | R 2 |
| x | 0.0364 | 0.0395 | 0.9877 |
| y | 0.0522 | 0.0498 | 0.9609 |
| z | 0.0406 | 0.0303 | 0.8145 |
| α | 0.0128 | 0.0089 | 0.6950 |
| β | 0.0115 | 0.0092 | 0.8968 |
| γ | 0.0101 | 0.0082 | 0.9403 |
Referring to fig. 4, the pose deviation after the compensation of the Stewart platform and the pose deviation before the compensation are compared based on the method. From the figure, the method greatly improves the pose accuracy after the Stewart platform is compensated.
The above description is intended to be illustrative of the embodiments of the invention and is not to be taken in any way as limiting. One of ordinary skill in the art will be able to make a number of optimizations, improvements, modifications, etc. based on the present invention. Accordingly, the scope of the invention should be limited only by the attached claims.
Claims (5)
1. A Stewart parallel platform pose deviation prediction and compensation method based on LSTM-ANN network is characterized in that: the method comprises the following steps:
s1: selecting a large number of nominal positions of the Stewart parallel platforms in a grid dividing mode, controlling the Stewart parallel platforms to realize fixed-point movement of the corresponding positions, acquiring actual positions of the Stewart parallel platforms by using a position measuring instrument, calculating position deviation between the actual positions and the nominal positions, and constructing a nominal position deviation data set and an actual position deviation data set;
s2: establishing a pose deviation regression prediction model LSTM-ANN of a Stewart parallel platform combined with LSTM and ANN, wherein the input of the model is the nominal pose of the Stewart parallel platform, and the output is pose deviation;
s3: dividing a deviation data set into a training set and a testing set according to a ratio of 4:1, selecting different numbers of the ANN neurons, training an LSTM-ANN model through training set data, and determining the most suitable number of the neurons;
s4: and taking the nominal pose of the Stewart parallel platform in the test set data as the input of the LSTM-ANN model, realizing corresponding pose deviation prediction, and pre-compensating the predicted pose deviation to the nominal pose of the Stewart parallel platform to be used as the input of the controller.
2. The method for predicting and compensating the pose deviation of the Stewart parallel platform based on the LSTM-ANN network according to claim 1, which is characterized in that: in the step S1, a large number of nominal poses of the Stewart parallel platform are selected by a grid dividing method, which can be specifically described as: dividing the position space and the gesture space of the Stewart parallel platform into 5X 5 grids, obtaining intersection points of 125 position spaces and 125 gesture spaces altogether, and taking the positions and the gestures on the 250 intersection points as nominal gestures;
the pose bias between the actual pose and the nominal pose can be described as:
ΔQ＝Q n -Q a (1)
wherein DeltaQ represents pose deviation, Q n Represents the nominal pose, Q a Representing the actual measurement pose.
3. The method for predicting and compensating the pose deviation of the Stewart parallel platform based on the LSTM-ANN network according to claim 1, which is characterized in that: in the step S2, a pose deviation regression prediction model LSTM-ANN comprises an input layer, an output layer, an LSTM layer, a full connection layer and an output layer; the input layer and the output layer are respectively composed of 6 network nodes, and respectively correspond to 6 nominal pose inputs and 6 pose deviation outputs, the LSTM layer is composed of 6 LSTM units, the output mode of the LSTM is Last, namely the output result of the Last LSTM unit is output as the LSTM layer.
4. The method for predicting and compensating the pose deviation of the Stewart parallel platform based on the LSTM-ANN network according to claim 1, which is characterized in that: in the step S3, the number of the neurons of the ANN is selected from 10, 15, 20 and 25, the LSTM-ANN model is trained and tested by using the deviation data set, and the coefficient R is determined by mean absolute error MAE 2 Determining optimal neuron numbers, MAE and R 2 The detailed calculation formula of (2) is as follows:
wherein:
wherein the predicted value is expressed asThe actual value is denoted as x j The average of the actual values is expressed as +.>n represents the number of poses used for the test.
5. The method for predicting and compensating the pose deviation of the Stewart parallel platform based on the LSTM-ANN network according to claim 1, which is characterized in that: in the step S4, the predicted pose deviation is precompensated into the nominal pose of the Stewart parallel platform, which is specifically implemented as follows: predicting pose deviation delta Q by using nominal pose of test set and LSTM-ANN model p ＝[Δx,Δy,Δz,Δα,Δβ,Δγ]Will DeltaQ p Precompensation to nominal pose Q n Then get the corrected pose Q m Using Q m Instead of Q n As input to the controller, Q m Can be expressed as:
Q m ＝Q n -ΔQ p (5)。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311432180.6A CN117428768A (en) | 2023-10-31 | 2023-10-31 | Stewart parallel platform pose deviation prediction and compensation method based on LSTM-ANN network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311432180.6A CN117428768A (en) | 2023-10-31 | 2023-10-31 | Stewart parallel platform pose deviation prediction and compensation method based on LSTM-ANN network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117428768A true CN117428768A (en) | 2024-01-23 |
Family
ID=89554918
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311432180.6A Pending CN117428768A (en) | 2023-10-31 | 2023-10-31 | Stewart parallel platform pose deviation prediction and compensation method based on LSTM-ANN network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117428768A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117609673A (en) * | 2024-01-24 | 2024-02-27 | 中南大学 | Six-degree-of-freedom parallel mechanism forward solution method based on physical information neural network |
-
2023
- 2023-10-31 CN CN202311432180.6A patent/CN117428768A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117609673A (en) * | 2024-01-24 | 2024-02-27 | 中南大学 | Six-degree-of-freedom parallel mechanism forward solution method based on physical information neural network |
| CN117609673B (en) * | 2024-01-24 | 2024-04-09 | 中南大学 | Six-degree-of-freedom parallel mechanism forward solution method based on physical information neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110385720B (en) | Robot positioning error compensation method based on deep neural network | |
| CN110434851B (en) | 5-degree-of-freedom mechanical arm inverse kinematics solving method | |
| CN110815206B (en) | Kinematics calibration method for Stewart parallel robot | |
| CN117428768A (en) | Stewart parallel platform pose deviation prediction and compensation method based on LSTM-ANN network | |
| CN111203890B (en) | Position error compensation method of robot | |
| Ma et al. | Robust uncalibrated visual servoing control based on disturbance observer | |
| Graule et al. | Somo: Fast and accurate simulations of continuum robots in complex environments | |
| CN108638067B (en) | Strategy for preventing serious degradation of motion performance of space manipulator | |
| Varziri et al. | Kinematic calibration of a wire-actuated parallel robot | |
| CN115401699B (en) | Industrial robot precision reliability analysis method, device, equipment and storage medium | |
| CN114169230A (en) | Robot dynamics parameter identification method | |
| Yang et al. | Highly accurate manipulator calibration via extended Kalman filter-incorporated residual neural network | |
| Yang et al. | Numerical simulation of calibration algorithm based on inverse kinematics of the parallel mechanism | |
| CN114800529A (en) | Industrial robot positioning error online compensation method based on fixed-length memory window incremental learning and incremental model reconstruction | |
| Jia et al. | A novel optimal design of measurement configurations in robot calibration | |
| Zhou et al. | Adaptive hierarchical positioning error compensation for long-term service of industrial robots based on incremental learning with fixed-length memory window and incremental model reconstruction | |
| CN113799137A (en) | Mobile processing robot rapid visual positioning method based on neural network | |
| CN112720480A (en) | Robot track correction method and system based on grading errors | |
| CN115972216B (en) | Parallel robot forward motion solving method, control method, equipment and storage medium | |
| Gu et al. | Homography‐based uncalibrated visual servoing with neural‐network‐assisted robust filtering scheme and adaptive servo gain | |
| Yu | Parallel robots pose accuracy compensation using artificial neural networks | |
| CN114967472A (en) | Unmanned aerial vehicle trajectory tracking state compensation depth certainty strategy gradient control method | |
| CN113290557A (en) | Snake-shaped robot control method based on data driving | |
| CN114186477A (en) | Elman neural network-based orbit prediction algorithm | |
| CN116638524A (en) | Six-degree-of-freedom parallel robot pose calibration method based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |