Collision object identification method in vibrating process

Abstract

The invention relates to the technical field of vibration construction methods, in particular to a collision object recognition method in the vibration process. According to the method, the position and collision force data are comprehensively utilized, prediction is performed in combination with the neural network model, and high-precision recognition of the type of the collision object is achieved. According to the method, the position and collision force of the collision object are monitored in real time, the neural network model is dynamically updated through the gain matrix and the covariance matrix, and the accuracy of collision object type prediction is improved. According to the technology, the safety and efficiency in the construction process can be remarkably improved, accidental shutdown or damage caused by wrong recognition of collision objects is reduced, and powerful technical support is provided for vibration construction.

Description

Technical Field

[0001] The present application relates to the technical field of vibration construction methods, and specifically to a method for identifying collision objects during the vibration process. Background Art

[0002] The application of vibrating robots is one of the key links in the transformation and upgrading of concrete structure casting methods. Since the vibration process occurs under the cover of flowing wet concrete, the vibrating robots are very likely to collide with steel bars, formwork, etc. Different collision strategies must be implemented when colliding with different collision objects. Therefore, it is necessary to design corresponding methods to identify the stiffness of the collision object, so as to realize the identification of the collision object type.

[0003] In recent years, there are many types of intelligent robots in concrete spreading, leveling, and polishing. However, due to the narrow internal space of concrete structures, dense steel bars, and irregular binding, it is difficult for automated vibration equipment to perform well. At the same time, most of the research on vibration robots is to modify the original equipment, but the intelligence and integration levels of common vibration robots are low, and they all require manual cooperation. During the vibration process, it is inevitable that they will collide with the steel cage. Due to the difficulty in identifying the type of collision object, it is impossible to perform corresponding obstacle avoidance trajectory planning for different collision objects, which makes the vibration robots on the market generally have a low level of intelligence. Based on the above analysis, combined with the survey of intelligent vibration robots on the current market, it can be seen that the main reason for the low level of intelligence of vibration robots is that the commonly used robot collision object recognition method is to realize the perception of collision objects through visual feedback. However, the vibration construction site is complex, and the operation usually occurs inside the wet concrete, which limits the application of visual sensors. At the same time, the method of identifying collision objects often uses the least squares method, which approximates and calculates the environmental parameters of the collision object through iteration, and then determines the type of the identified object. This method has low recognition efficiency, slow response speed, and cannot directly obtain the environmental type.

[0004] In summary, there are few research results on identifying collision object types in vibration construction, and the recognition accuracy is low. It can be seen that the current method of identifying environmental objects in vibration environment needs further research. Summary of the invention

[0005] In order to avoid and overcome the technical problems existing in the prior art, the present application provides a method for identifying collision objects during vibration, which can effectively improve the accuracy of collision object type identification.

[0006] To achieve the above objectives, this application provides the following technical solutions: A method for identifying collision objects during vibration, comprising the following identification steps: S1. Obtain the position and collision force of each collision object at each moment during the vibration construction process; S2. Based on the acquired positions of the collision objects and their collision forces, the gain matrix and covariance matrix of each collision object combination at each moment are calculated; S3, in chronological order, inputting the position of each collision object and its collision force at each moment into the neural network for training, and updating the model parameters of the neural network through the gain matrix and the covariance matrix to obtain a collision force prediction model that meets the prediction accuracy requirements; S4, obtaining the real-time position of each collision object during the vibration construction process, and inputting the real-time position into the collision force prediction model to predict the corresponding real-time collision force; S5. Inputting the real-time predicted collision force into a collision object stiffness calculation model to calculate the stiffness of the collision object, and identifying the type of the collision object through the stiffness.

[0007] As a further solution of this application: the specific content of step S1 is as follows: S11, establishing a Cartesian coordinate system at the vibration construction site, scanning the vibration construction site by ultrasonic scanning technology, and inputting the scanning data into the Cartesian coordinate system to establish a three-dimensional model of the vibration construction site in the Cartesian coordinate system; S12, obtaining the position of each collision object and the position of the execution end of the robotic arm from the three-dimensional model of the vibrating construction site, and obtaining the collision force between the execution end of the robotic arm and the collision object through a force sensor installed at the execution end of the robotic arm; S13, according to the position and the corresponding collision force obtained in step S2, an input matrix and an output matrix are established, which are specifically expressed as follows: ; ; ; ; ; In the formula, express t The input matrix at time instant; Indicates that the robot arm is executing at the end t The position of the moment, subscript v Indicates the end of the robot arm execution; Indicates that the first collision object is t The location at the moment; Indicates e The collision objects t The location at the moment, , Indicates the total number of collision objects during the current vibration construction process; Indicates E The collision objectst The location at the moment; express and The distance between express and The distance between express and The distance between express t Output matrix at time instant; Indicated in t The collision force between the robot arm execution end and the first collision object at the moment; Indicated in t At the moment the robot arm executes the end e The collision force between the colliding objects; Indicated in t At the moment the robot arm executes the end E The collision force between the colliding objects; Flag for matrix transpose.

[0008] As a further solution of this application: the gain matrix is ​​expressed as follows: ; In the formula, express t The gain matrix at time; express The covariance matrix of the time instant; express t The variable forgetting factor at each moment; represents the identity matrix; at the initial moment, , represents the correction parameter of the identity matrix; The covariance matrix is ​​expressed as follows: ; In the formula, express t The covariance matrix at time instants.

[0009] As a further solution of this application: the specific content of step S3 is as follows: S31. Taking the input matrix as a feature and the output matrix as a label, the input matrix and the output matrix are combined to construct a training sample set at each moment, and each row in the training sample set represents a training sample at the current moment; the training sample set is specifically represented as follows: ; In the formula, express t The training sample set at time; S32, training sample set Input into the neural network for training and calculate the model error. If the model error is less than the set error threshold, stop training and output the current neural network, which is the collision force prediction model that meets the prediction accuracy requirements; otherwise, perform the model parameter update in step S33; S33. Based on the calculated gain matrix and covariance matrix, according to the negative feedback regulation principle, the model parameters of the neural network are updated using an update formula: the update formula is expressed as follows: ; In the formula, It means that the neural network t During the round iteration, the parameter matrix formed by the combination of the values ​​of each model parameter is iterated for one round of data at one moment, that is, t The round iteration uses t Data at the moment; It means that the neural network During the round of iterations, the parameter matrix is ​​formed by the combination of the values ​​of each model parameter; for The transposed matrix of ; at the initial iteration, , Represents the initial iteration parameters; S34. According to the operation process from step S32 to step S33, the training sample sets at each moment are sequentially input into the neural network for training according to the time sequence until a collision force prediction model is obtained.

[0010] As a further solution of this application: the specific content of step S4 is as follows: S41, according to the operation process of step S11 to step S13, an input matrix formed by combining the real-time position of each collision object and the real-time position of the execution end of the robot arm during the current vibration construction process is obtained; S42, inputting the input matrix obtained in step S41 into the collision force prediction model to predict the real-time collision force between each collision object and the execution end of the robot arm; As a further solution of this application: the specific content of step S5 is as follows: S51, input the real-time collision force obtained in step S42 into the collision object stiffness calculation model to calculate the stiffness of the collision object; the collision object stiffness calculation model is expressed as follows: ; In the formula, It represents the real-time collision force between the collision object and the end of the robot arm predicted by the collision force prediction model; Represents the estimated stiffness of the collision object; Represents the estimated damping value of the collision object; Indicates the real-time position of the end of the robot arm; Indicates the real-time position of the collision object; S52, inputting multiple real-time positions of the collision object within a set time period and multiple real-time positions corresponding to the end of the robot arm into the collision object stiffness calculation model, constructing an equation group, and solving the equation to obtain an estimated stiffness value of the collision object; S53, comparing the stiffness estimation value with the collision object stiffness table, finding the collision object type corresponding to the stiffness closest to the stiffness estimation value, and the collision object type is the type of the current collision object.

[0011] As a further solution of this application: the calculation formula of the variable forgetting factor is as follows: ; In the formula, Indicates the minimum value of the variable forgetting factor; Indicates the maximum value of the variable forgetting factor; Represents a variable function; ; ; ; ; In the formula, Represents a natural constant Power; Indicated in t At this moment, the selected time window size is The error change rate at time , the time window size is Indicates that the time window contains moment, and define the time window as the first time window; Indicates the influence weight of the error change rate; Indicated in t At this moment, the selected time window size is The data quality indicator at the time, the time window size is Indicates that the time window contains moment, and define the time window as the second time window; Indicates the impact weight of data quality indicators; In the first time window, i The error change rate weight during round iteration; In the first time window, The error during round iteration; In the first time window,i The error during round iteration; In the second time window, j The error during round iteration; In the second time window, It's your turn The average error during the iterations.

[0012] As a further solution of this application: Correction parameters The value range is .

[0013] As a further solution of this application: the initial value of the variable forgetting factor .

[0014] As a further solution of this application: initial iteration parameters The value of is 0.

[0015] Compared with the prior art, the beneficial effects of this application are: 1. This application achieves high-precision identification of the type of collision object by comprehensively using position and collision force data combined with a neural network model for prediction. This method not only monitors the position and collision force of the collision object in real time, but also dynamically updates the neural network model through the gain matrix and covariance matrix, thereby improving the accuracy of the collision object type prediction. This technology can significantly improve the safety and efficiency of the construction process, reduce unexpected downtime or damage caused by incorrect collision object identification, and provide strong technical support for vibration construction.

[0016] 2. By introducing the Cartesian coordinate system and ultrasonic scanning technology, this method can accurately construct a three-dimensional model of the vibration construction site, thereby accurately obtaining the location and collision force information of the collision object. This refined data collection method provides a high-quality data foundation for subsequent neural network training and collision object identification, further improving the accuracy and reliability of identification.

[0017] 3. The clear expressions of the gain matrix and covariance matrix provide a specific mathematical basis for updating the model parameters of the neural network. By introducing a variable forgetting factor, this method can dynamically adjust the impact of historical data on the current model parameter update, making the model more adaptable to changes in the actual construction process and improving the flexibility and accuracy of the prediction.

[0018] 4. Through the negative feedback regulation principle, this method can continuously iterate and optimize the neural network model until the predetermined prediction accuracy requirement is met. This systematic training method ensures the high performance and stability of the collision force prediction model.

[0019] 5. By obtaining the position information of the collision object in real time and inputting it into the trained collision force prediction model, the real-time collision force between each collision object and the end of the robot arm can be quickly obtained. This real-time prediction capability is of great significance for timely avoiding collision accidents during construction and ensuring construction safety.

[0020] 6. By constructing a collision object stiffness calculation model and solving it in combination with real-time collision force data, this method can accurately calculate the stiffness of the collision object and quickly identify the type of collision object by comparing the collision object stiffness table. This stiffness-based identification method has the advantages of clear physical meaning and high identification accuracy, and provides strong support for collision object management during vibration construction.

[0021] 7. The calculation formula of the variable forgetting factor comprehensively considers the error change rate and data quality indicators, and dynamically adjusts the value of the forgetting factor. This adaptive adjustment mechanism enables the model to maintain good prediction performance when facing different construction scenarios and data quality, and improves the robustness and adaptability of the method.

[0022] 8. The value range of the correction parameter is set to provide clear constraints for the update of the gain matrix. By reasonably setting the value range of the correction parameter, the model can be ensured to maintain stability during the update process and avoid the degradation of model performance due to excessive parameter adjustment.

[0023] 9. The initial value setting of the variable forgetting factor provides a reasonable starting point for model training. By setting an appropriate initial value, the model can converge to a better state quickly in the early stage of training, improving training efficiency and model performance.

[0024] 10. The value of the initial iteration parameter is 0, which simplifies the initial conditions for model training. This setting method is not only easy to implement, but also ensures that the model starts iterative optimization from the most basic state during the training process, avoiding model performance problems caused by improper initial parameter settings. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is the identification flow chart of this application.

[0026] Figure 2 Comparison chart of stiffness prediction results. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0028] See also Figure 1 In an embodiment of the present application, a method for identifying collision objects during vibration includes the following contents: 1. Get collision object data.

[0029] The collision object data includes the position and collision force of each collision object at each moment during the vibration construction process. The specific acquisition steps are as follows: S11. Establish a Cartesian coordinate system at the vibration construction site, scan the vibration construction site by ultrasonic scanning technology, and input the scanning data into the Cartesian coordinate system to establish a three-dimensional model of the vibration construction site in the Cartesian coordinate system.

[0030] S12. Obtain the position of each collision object and the position of the execution end of the robotic arm from the three-dimensional model of the vibration construction site, and simultaneously obtain the collision force between the execution end of the robotic arm and the collision object through a force sensor installed at the execution end of the robotic arm.

[0031] S13. According to the position and the corresponding collision force obtained in step S2, an input matrix and an output matrix are established, which are specifically expressed as shown in formulas (1) to (3).

[0032] (1); (2); (3); 2. Obtain the gain matrix and covariance matrix.

[0033] Based on the acquired positions of the collision objects and their collision forces, the gain matrix and covariance matrix of each collision object combination at each moment are calculated.

[0034] The gain matrix is ​​expressed as shown in formula (4), and the covariance matrix is ​​expressed as shown in formula (5).

[0035] (4); (5); The calculation formulas of the variable forgetting factor are shown in formulas (6) to (10).

[0036] (6); (7); (8); (9); (10); 3. Obtain a collision force prediction model.

[0037] In chronological order, the positions of each collision object and its collision force at each moment are input into the neural network for training, and the model parameters of the neural network are updated through the gain matrix and covariance matrix to obtain a collision force prediction model that meets the prediction accuracy requirements. The specific steps are as follows: S31. Taking the input matrix as the feature and the output matrix as the label, the input matrix and the output matrix are combined to construct the training sample set at each moment, and each row in the training sample set represents a training sample at the current moment; the specific representation of the training sample set is shown in formula (11).

[0038] (11); S32, training sample set The input is put into the neural network for training and the model error is calculated. If the model error is less than the set error threshold, the training is stopped and the current neural network is output. The neural network is the collision force prediction model that meets the prediction accuracy requirements; otherwise, the model parameter update of step S33 is performed.

[0039] S33. Based on the calculated gain matrix and covariance matrix, according to the negative feedback regulation principle, the model parameters of the neural network are updated using an update formula; the update formula is expressed as shown in formula (12).

[0040] (12); It should be noted that the data at one moment undergoes one round of iteration, i.e. t The round iteration uses t Data at the moment.

[0041] S34. According to the operation process from step S32 to step S33, the training sample sets at each moment are sequentially input into the neural network for training according to the time sequence until a collision force prediction model is obtained.

[0042] 4. Predict real-time collision forces.

[0043] The real-time position of each collision object during the vibration construction process is obtained, and the real-time position is input into the collision force prediction model to predict the corresponding real-time collision force. The specific steps are as follows: S41. According to the operation process of step S11 to step S13, an input matrix formed by combining the real-time position of each collision object and the real-time position of the execution end of the robot arm in the current vibration construction process is obtained.

[0044] S42, inputting the input matrix obtained in step S41 into the collision force prediction model to predict the real-time collision force between each collision object and the execution end of the robot arm.

[0045] 5. Determine the type of collision object.

[0046] The real-time predicted collision force is input into the collision object stiffness calculation model to calculate the stiffness of the collision object and identify the type of the collision object through the stiffness. The specific steps are as follows: S51, input the real-time collision force obtained in step S42 into the collision object stiffness calculation model to calculate the stiffness of the collision object; the collision object stiffness calculation model is expressed as shown in formula (13): (13); S52, inputting multiple real-time positions of the collision object within a set time period and multiple real-time positions corresponding to the robot arm execution end into the collision object stiffness calculation model, constructing a set of equations, and solving the stiffness estimation value of the collision object by solving the equations.

[0047] S53, comparing the stiffness estimation value with the collision object stiffness table, finding the collision object type corresponding to the stiffness closest to the stiffness estimation value, and the collision object type is the type of the current collision object.

[0048] like Figure 2 FIG. 4 is a comparison of the stiffness prediction results of the present application and other existing methods. Figure 2 In the FRLS (Forgetting Rate Least Squares) algorithm, the forgetting factor λ is a key parameter between 0 and 1, which is used to control the weight of historical data in the model update process. FRLS_λ=1 represents the collision object stiffness prediction curve when the forgetting factor is not used; FRLS_λ=0.98 means that the weight of each data point will be multiplied by 0.98 at each update, thereby gradually reducing the weight of the old data, and then forming the corresponding collision object stiffness prediction curve; FRLS_λ=0.96 means that the weight of each data point will be multiplied by 0.96 at each update, thereby gradually reducing the weight of the old data, and then forming the corresponding collision object stiffness prediction curve.

[0049] from Figure 2It can be seen that the vibration robot moves between 0-10s, collides with collision object 1 (the actual stiffness k of collision object 1 is 2000N / m) at 8s, and after the collision, the environmental stiffness is identified in real time according to the contact force at the end of the vibration robot arm and the distance to the obstacle, and a stable state is reached at about 8.3s, and the stiffness of the collision object is about 2000N / m. At 9s, it collides with collision object 2 (the actual stiffness k of collision object 2 is 4000N / m), and the stiffness of the collision object is continuously identified in real time, and a stable state is reached within 9.3s, and the stiffness of the collision object is about 4000N / m.

[0050] Compared with the traditional fixed forgetting factor algorithm, the present invention has faster recognition efficiency and also has a faster response effect when facing consecutive different collision objects. Therefore, the method of the present application can effectively improve the accuracy of stiffness prediction.

[0051] The above is only a preferred specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can make equivalent substitutions or changes within the technical scope disclosed in the present application according to the technical solution and inventive concept of the present application, which should be covered by the protection scope of the present application.

Claims

1. A method for identifying collision objects during vibration, characterized in that: The identification steps include: S1. Obtain the position and collision force of each collision object at each moment during the vibration construction process; S2. Based on the positions of the collision objects and their collision forces, the gain matrix and covariance matrix of each collision object combination at each moment are calculated; S3, in chronological order, inputting the position of each collision object and its collision force at each moment into the neural network for training, and updating the model parameters of the neural network through the gain matrix and the covariance matrix to obtain a collision force prediction model that meets the prediction accuracy requirements; S4, obtaining the real-time position of each collision object during the vibration construction process, and inputting the real-time position into the collision force prediction model to predict the corresponding real-time collision force; S5. Construct a collision object stiffness calculation model, and input the real-time collision force into the collision object stiffness calculation model to calculate the stiffness of the collision object, and identify the type of the collision object through the stiffness.

2. The method for identifying collision objects during vibration according to claim 1, characterized in that: The specific content of step S1 is as follows: S11, establishing a Cartesian coordinate system at the vibration construction site, scanning the vibration construction site by ultrasonic scanning technology, and inputting the scanning data into the Cartesian coordinate system to establish a three-dimensional model of the vibration construction site in the Cartesian coordinate system; S12, obtaining the position of each collision object and the position of the execution end of the robotic arm from the three-dimensional model of the vibrating construction site, and obtaining the collision force between the execution end of the robotic arm and the collision object through a force sensor installed at the execution end of the robotic arm; S13. According to the position of the collision object and the corresponding collision force, an input matrix and an output matrix are established, which are specifically expressed as follows: ； ； ； ； ； In the formula, express t The input matrix at time instant; Indicates that the robot arm is executing at the end t The position of the moment, subscript v Indicates the end of the robot arm execution; Indicates that the first collision object is t The location at the moment; Indicates e The collision objects t The location at the moment, , Indicates the total number of collision objects during the current vibration construction process; Indicates E The collision objects t The location at the moment; express and The distance between express and The distance between express and The distance between express t Output matrix at time instant; Indicated in t The collision force between the robot arm execution end and the first collision object at the moment; Indicated in t At the moment the robot arm executes the end e The collision force between the colliding objects; Indicated in t At the moment the robot arm executes the end E The collision force between the colliding objects; Flag for matrix transpose.

3. The method for identifying collision objects during vibration according to claim 2, characterized in that: The gain matrix is ​​expressed as follows: ； In the formula, express t The gain matrix at time; express The covariance matrix of the time instant; express t The variable forgetting factor at each moment; represents the identity matrix; at the initial moment, , represents the correction parameter of the identity matrix; The covariance matrix is ​​expressed as follows: ； In the formula, express t The covariance matrix at time instants.

4. The method for identifying collision objects during vibration according to claim 3, characterized in that: The specific content of step S3 is as follows: S31. Taking the input matrix as a feature and the output matrix as a label, the input matrix and the output matrix are combined to construct a training sample set at each moment, and each row in the training sample set represents a training sample at the current moment; the training sample set is specifically represented as follows: ； In the formula, express t The training sample set at time; S32, training sample set Input into the neural network for training, and calculate the model error of the neural network. If the model error is less than the set error threshold, stop training and output the current neural network, which is the collision force prediction model that meets the prediction accuracy requirements; otherwise, perform model parameter update in step S33; S33. Based on the calculated gain matrix and covariance matrix, according to the negative feedback regulation principle, the model parameters of the neural network are updated using an update formula; the update formula is expressed as follows: ； In the formula, It means that the neural network t During the round iteration, the parameter matrix formed by the combination of the values ​​of each model parameter is iterated for one round of data at one moment, that is, t The round iteration uses t Data at the moment; It means that the neural network During the round of iterations, the parameter matrix is ​​formed by the combination of the values ​​of each model parameter; for The transposed matrix of ; at the initial iteration, , Represents the initial iteration parameters; S34. According to the operation process from step S32 to step S33, the training sample sets at each moment are sequentially input into the neural network for training according to the time sequence until a collision force prediction model is obtained.

5. The method for identifying collision objects during vibration according to claim 4, characterized in that: The specific content of step S4 is as follows: S41, according to the operation process of step S11 to step S13, an input matrix formed by combining the real-time position of each collision object and the real-time position of the execution end of the robot arm during the current vibration construction process is obtained; S42, inputting the input matrix obtained in step S41 into the collision force prediction model to predict the real-time collision force between each collision object and the execution end of the robot arm.

6. The method for identifying collision objects during vibration according to claim 5, characterized in that: The specific content of step S5 is as follows: S51, input the real-time collision force obtained in step S42 into the collision object stiffness calculation model to calculate the stiffness of the collision object; the collision object stiffness calculation model is expressed as follows: ； In the formula, It represents the real-time collision force between the collision object and the end of the robot arm predicted by the collision force prediction model; Represents the estimated stiffness of the collision object; Represents the estimated damping value of the collision object; Indicates the real-time position of the end of the robot arm; Indicates the real-time position of the collision object; S52, inputting multiple real-time positions of the collision object within a set time period and multiple real-time positions corresponding to the end of the robot arm into the collision object stiffness calculation model, constructing an equation group, and solving the equation to obtain an estimated stiffness value of the collision object; S53, comparing the stiffness estimation value with the collision object stiffness table, finding the collision object type corresponding to the stiffness closest to the stiffness estimation value, and the collision object type is the type of the current collision object.

7. A method for identifying collision objects during vibration according to any one of claims 2 to 6, characterized in that: The calculation formula of the variable forgetting factor is as follows: ； In the formula, Indicates the minimum value of the variable forgetting factor; Indicates the maximum value of the variable forgetting factor; Represents a variable function; ； ； ； ； In the formula, Represents a natural constant Power; Indicated in t At this moment, the selected time window size is The error change rate at time , the time window size is Indicates that the time window contains moment, and define the time window as the first time window; Indicates the influence weight of the error change rate; Indicated in t At this moment, the selected time window size is The data quality indicator at the time, the time window size is Indicates that the time window contains moment, and define the time window as the second time window; Indicates the impact weight of data quality indicators; In the first time window, i The error change rate weight during round iteration; In the first time window, The error during round iteration; In the first time window, i The error during round iteration; In the second time window, j The error during round iteration; In the second time window, It's your turn The average error during the iterations.

8. The method for identifying collision objects during vibration according to claim 3, characterized in that: Correction parameters The value range is .

9. The method for identifying collision objects during vibration according to claim 7, characterized in that: Initial value of the variable forgetting factor .

10. The method for identifying collision objects during vibration according to claim 4, characterized in that: Initial Iteration Parameters The value of is 0.

Images