Nuclear decommissioning operation master-slave manipulator control system and method

Abstract

The invention provides a nuclear decommissioning operation master-slave manipulator control system and method. The system comprises an acquisition module for acquiring real-time attitude data, corresponding communication delay data and hot cell environment data of a master manipulator and a slave manipulator respectively; the data processing module preprocesses the collected data, extracts real-time posture standard feature information, communication delay standard feature information and hot cell environment feature data of the master and slave manipulators, and stores the information into a motion information database; the action prediction module constructs a slave manipulator track prediction model based on a neural network, and outputs motion action track prediction information of a slave manipulator; according to the system, in a nuclear decommissioning hot cell environment, real-time attitude data, communication delay data and hot cell environment data of a master manipulator and a slave manipulator are collected, and a high-precision slave manipulator track prediction model is constructed in combination with a neural network, so that the environmental adaptability of the system is enhanced, and the following precision and stability of the slave manipulator to the action of the master manipulator are improved; and accurate synchronous operation of the master and slave manipulators in a complex environment is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of master-slave robotic arms for nuclear decommissioning, and in particular to a control system and method for a master-slave robotic arm for nuclear decommissioning operations. Background Art

[0002] In the nuclear field, laboratories and factories engaged in nuclear scientific research and production cannot operate radioactive materials without special long-distance operation equipment of nuclear manipulators. The main difference between this type of equipment and other ordinary operation equipment is that it can separate the operator from the dangerous environment through appropriate biological shielding for operation. Its purpose is to separate the operator from the operated object and operate the radioactive material under the premise of ensuring the safety of the operator and the operating environment. Its application fields cover nuclear power, post-processing, laboratory analysis and medical fields, and it is an indispensable long-distance operation tool in the nuclear field.

[0003] The invention discloses a manipulator control system based on a master-slave teleoperated robot, which is published as CN113084775A, and comprises a potentiometer group, an AD conversion module, a master hand controller, a drive module, a communication module, a slave hand controller, a DA conversion module, a servo control module, a force sensor group and a control center; the potentiometer group, the force sensor group, the AD conversion module, the master hand controller and the drive module constitute a main control system; the manipulator control system based on the master-slave teleoperated robot collects various data on the master hand through the main control system, analyzes and processes the data through the control center, and drives the slave control system to control the slave hand to perform synchronous operation, and the slave control system collects the slave hand data in real time, and transmits it to the control center to drive the main control system to adjust the master hand, so that the master and slave manipulators are synchronized to achieve precise control.

[0004] In the existing robotic arm control system based on master-slave teleoperated robots, although the master control system collects the master hand data, the control center analyzes and processes and drives the slave control system to realize master-slave synchronous operation, and the slave control system can collect the slave hand data in real time and feed it back to the control center to adjust the master hand, but in the face of the complex and changeable hot chamber environment, the impact of environmental factors and communication time errors on the performance of the slave robot hand is not fully considered, resulting in the accuracy of the slave robot hand following the master hand's movements in the hot chamber environment, reducing the accuracy and stability of the slave robot hand following the master hand's movements. Summary of the invention

[0005] In view of this, the present invention proposes a control system and method for a master-slave robot for nuclear decommissioning operations. In a nuclear decommissioning hot chamber environment, by collecting real-time posture data, communication delay data and hot chamber environment data of the master and slave robots, a high-precision slave robot trajectory prediction model is constructed in combination with a neural network to enhance the environmental adaptability of the system, improve the following accuracy and stability of the slave robot to the master hand's movements, and ensure the precise synchronous operation of the master and slave robots in a complex environment.

[0006] The technical solution of the present invention is implemented as follows: The present invention provides a master-slave robot control system for nuclear decommissioning operations, the system comprising:

[0007] The acquisition module is used to respectively collect the real-time posture data of the master and slave manipulators and the corresponding communication delay data and the thermal chamber environment data;

[0008] The data processing module is used to pre-process the collected data, extract the standard feature information of the real-time posture of the master and slave manipulators, the standard feature information of the communication delay and the feature data of the hot chamber environment, and store them in the motion information database;

[0009] The motion prediction module is used to build a trajectory prediction model for the slave robot based on a neural network, taking the historical data and real-time posture data in the motion information database and the corresponding communication delay data and the thermal chamber environment data as input, training the trajectory prediction model for the slave robot to output the motion trajectory prediction information of the slave robot;

[0010] The execution module is used to convert the motion trajectory prediction information of the slave robot hand at the current moment into a control signal and send it to the slave hand manipulation module. The slave hand manipulation module performs corresponding actions according to the received control signal to achieve accurate following of the master hand action.

[0011] On the basis of the above technical solution, preferably, the collected data is preprocessed to obtain standard feature information of real-time postures of the master and slave manipulators, standard feature information of communication delays and feature data of the hot chamber environment, which includes:

[0012] The posture data of the master and slave manipulators include angle vectors of each joint, and the thermal chamber environment data include radiation intensity, temperature and humidity;

[0013] The collected data is denoised using low-pass filtering and normalized to obtain standard real-time posture information of the master and slave manipulators, standard communication delay information, and thermal chamber environment data;

[0014] Extracting standard feature information from the real-time posture standard information of the master and slave manipulators, the communication delay standard information and the hot room environment data, and obtaining the real-time posture standard feature information of the master and slave manipulators, the communication delay standard feature information and the hot room environment feature data;

[0015] A motion information database is constructed, and the real-time posture standard feature information of the master and slave manipulators, the communication delay standard feature information, and the thermal chamber environment feature data are associated and stored in the motion information database.

[0016] On the basis of the above technical solution, preferably, the standard feature information is extracted from the real-time posture standard information of the master and slave manipulators, the communication delay standard information and the hot room environment data to obtain the real-time posture standard feature information of the master and slave manipulators, the communication delay standard feature information and the hot room environment feature data, including:

[0017] According to the joint angle vectors of the master and slave manipulators, the position features of the master and slave manipulators are calculated, where the position feature expression is:

[0018] p m (t) = f kin (q m (t))

[0019] p s (t) = f kin (q s (t))

[0020] Where P m (t) represents the position vector of the main robot hand at time t, q m (t) represents the joint angle vector of the main robot hand at time t, P s (t) represents the position vector of the robot hand at time t, q s (t) represents the joint angle vector of the robot hand at time t, f kin is the kinematics positive solution function;

[0021] The first-order derivative of the position characteristics of the master and slave manipulators with respect to time is performed to obtain the velocity characteristics v of the master manipulator respectively. m (t) and the velocity characteristic v from the robot hand s (t); Take the second-order derivative of the position characteristics of the master and slave manipulators with respect to time, and obtain the acceleration characteristics a of the master manipulator respectively. m (t) and the acceleration characteristic a from the robot hand s (t);

[0022] Preset a fixed time step, obtain the communication delay data at each time point of the fixed time step, calculate the average delay and delay fluctuation in the fixed time step, and obtain the standard characteristic information of the communication delay. The delay fluctuation expression is:

[0023]

[0024] Among them, σΔt is the delay fluctuation, Δt i is the communication delay data at the current time point, Δt avg is the average delay, N is the number of all time points with a fixed time step;

[0025] The radiation intensity, temperature and humidity at each time point of a fixed time step are obtained, and the average radiation intensity, average temperature and average humidity in the fixed time step are calculated to obtain the environmental characteristic data of the hot room.

[0026] On the basis of the above technical scheme, preferably, the action prediction module constructs a slave robot hand trajectory prediction model based on a neural network, and the network structure of the slave robot hand trajectory prediction model includes an input layer, a multi-scale feature extraction module, a hot room environment feature enhancement module, a Transformer encoder and a multi-task output module, wherein the input layer is used to receive the real-time posture standard feature information of the master and slave robots, the communication delay standard feature information and the hot room environment feature data, the output end of the input layer is connected to the input end of the multi-scale feature extraction module, and is used to extract the spatial and temporal features of the input data, the output end of the multi-scale feature extraction module is connected to the input end of the hot room environment feature enhancement module, and is used to dynamically adjust the feature weight according to the hot room environment feature data, the output end of the hot room environment feature enhancement module is connected to the input end of the Transformer encoder, and is used to capture the global dependency through a multi-head self-attention mechanism, and the output end of the Transformer encoder is connected to the input end of the multi-task output module, and is used to respectively predict and output the position vector, velocity feature and acceleration feature of the slave operator.

[0027] On the basis of the above technical solution, preferably, the multi-scale feature extraction module includes a CNN network extraction unit, an LSTM network extraction unit and a feature fusion unit, wherein:

[0028] The CNN network extraction unit is used to extract the spatial features of the input data, and the expression is:

[0029] F cnn =CNN(X)

[0030] In the formula, F cnn is the spatial feature extracted by the CNN network extraction unit, X is the input standard feature information of the real-time posture of the master and slave manipulators, the standard feature information of the communication delay, and the feature data of the hot room environment;

[0031] The LSTM network extraction unit is used to extract the time characteristics of the input data, and the expression is:

[0032] F lstm =LSTM(X)

[0033] In the formula, F lstm Temporal features extracted for the LSTM module;

[0034] The feature fusion unit is used to fuse the features output by the CNN network extraction unit with the features output by the LSTM network extraction unit to obtain fused feature data, which is expressed as:

[0035] In the formula, F fusion It is the concatenation result of the spatial features extracted by the CNN network extraction unit and the temporal features extracted by the LSTM module.

[0036] On the basis of the above technical solution, preferably, the thermal chamber environment feature enhancement module includes a preset anti-radiation weight basic matrix and a temperature and humidity weight basic matrix, wherein:

[0037] Dynamically adjust the anti-radiation basic weight matrix according to the radiation intensity to obtain the anti-radiation weight matrix, which is expressed as:

[0038]

[0039] Where W rad (t) is the radiation resistance weight matrix, is the basic matrix of radiation resistance weight, α is the influence coefficient of radiation intensity, and γ(t) is the radiation intensity at time point t;

[0040] Dynamically adjust the temperature and humidity weight basic matrix according to temperature and humidity to obtain the temperature and humidity weight matrix, which is expressed as:

[0041]

[0042] Where W dynamic (t) is the temperature and humidity weight matrix, is the temperature and humidity weight basic matrix, β T is the temperature influence coefficient, β H is the influence coefficient of humidity, T(t) is the temperature at time point t, and H(t) is the humidity at time point t;

[0043] The fused feature data is enhanced according to the anti-radiation weight matrix and the temperature and humidity weight basic matrix to obtain enhanced feature data, which is expressed as:

[0044] F dynamic =F fusion· W rad (t) · W dynamic (t)

[0045] In the formula, F dynamic To enhance feature data.

[0046] On the basis of the above technical solution, preferably, the multi-task output module includes outputting the position vector, velocity characteristics and acceleration characteristics of the slave operator, wherein:

[0047] The expression for the predicted position vector is:

[0048] PYs (t+Δt)=W p h+b p

[0049] Where PY s (t+Δt) is the position vector predicted from the operator, W p is the weight of the predicted position vector, b p is the bias of the predicted position vector;

[0050] The expression of velocity characteristic is:

[0051] v s (t+Δt)=W v h+b v

[0052] In the formula, vY s (t+Δt) is the speed characteristic predicted from the operator, W v is the weight of the predicted speed feature, b v is the bias for predicting velocity characteristics;

[0053] The expression of acceleration characteristic is:

[0054] aY s (t+Δt)=W a h+b a

[0055] In the formula, aY s (t+Δt) is the acceleration characteristic predicted from the operator, W a is the weight of the predicted acceleration feature, b a is the bias for predicting the acceleration characteristics.

[0056] On the basis of the above technical solution, preferably, the robot hand trajectory prediction model further includes setting a loss function, and using the Adam optimization algorithm to perform model iterative training according to the loss function to adjust the parameters of the model until convergence is reached and then the iterative training is stopped, wherein the expression of the loss function is:

[0057] L=λ1||PY s (t+Δt)-P s (t)|| 2 +λ2||vY s (t+Δt)-v s (t)|| 2 +λ3||aY s (t+Δt)-a s (t)|| 2

[0058] Where L is the multi-task loss function, λ1 is the weight coefficient for position vector prediction, λ2 is the weight coefficient for velocity feature prediction, and λ3 is the weight coefficient for acceleration feature prediction.

[0059] On the basis of the above technical solution, preferably, the execution module calculates the driving torque of each joint of the slave robot hand through an inverse kinematics algorithm according to the predicted trajectory information, converts it into an analog control signal, and sends the control signal to the slave hand manipulation module. The slave hand manipulation module performs corresponding actions according to the received control signal to achieve accurate following of the master hand action.

[0060] In a second aspect, the present invention further provides a method for controlling a master-slave robot for a nuclear decommissioning operation, which is implemented by a master-slave robot control system for a nuclear decommissioning operation, and includes the following steps:

[0061] S1, collects the real-time posture data of the master and slave manipulators and the corresponding communication delay data and hot room environment data respectively;

[0062] S2, preprocessing the collected data, extracting the standard feature information of the real-time posture of the master and slave manipulators, the standard feature information of the communication delay and the feature data of the hot chamber environment, and storing them in the motion information database;

[0063] S3, constructing a master-slave robot trajectory prediction model based on a neural network, taking the historical data and real-time posture data in the motion information database and the corresponding communication delay data and the thermal chamber environment data as input, training the slave robot trajectory prediction model to output the motion action trajectory prediction information of the slave robot;

[0064] S4, converting the motion trajectory prediction information of the slave robot hand at the current moment into a control signal, and sending it to the slave hand manipulation module. The slave hand manipulation module performs corresponding actions according to the received control signal to achieve accurate following of the master hand action.

[0065] Compared with the prior art, the master-slave robot control system and method for nuclear decommissioning operation of the present invention have the following advantages:

[0066] Beneficial effects:

[0067] (1) By collecting the real-time posture data, communication delay data and thermal chamber environment data of the master and slave manipulators, the slave manipulator can maintain stable operating performance in different thermal chamber environments, enhancing the environmental adaptability of the system. Based on the neural network, a slave manipulator trajectory prediction model is constructed. The system can more accurately predict the motion trajectory of the slave manipulator, thereby improving the following accuracy and stability of the slave manipulator to the master manipulator's movements.

[0068] (2) The multi-scale feature extraction module is set up to extract spatial and temporal features through CNN and LSTM respectively, which can more comprehensively capture the information in the input data and improve the expression ability of the features. The model can more comprehensively capture the spatial and temporal features in the input data and improve the accuracy of trajectory prediction;

[0069] (3) The hot chamber environment feature enhancement module is responsible for dynamically adjusting the feature weights according to the specific characteristics of the hot chamber environment, thereby enhancing the adaptability and prediction accuracy of the model under different environmental conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0071] Figure 1 It is a structural block diagram of the master-slave robot control system for nuclear decommissioning operations of the present invention;

[0072] Figure 2 A flowchart of the master-slave robot control method for nuclear decommissioning operations of the present invention;

[0073] Figure 3 This is a network model structure diagram of the master-slave robot control system for nuclear decommissioning operations of the present invention. DETAILED DESCRIPTION

[0074] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0075] like Figure 1 As shown, a nuclear decommissioning operation master-slave robot control system of the present invention includes:

[0076] The acquisition module is used to respectively acquire the real-time posture data of the master and slave manipulators and the corresponding communication delay data and the hot room environment data; in this embodiment, the posture data of the master and slave manipulators include the angle vectors of each joint, and the hot room environment data include the radiation intensity, temperature and humidity;

[0077] It should be noted that a high-precision encoder and inertial measurement unit are installed on each joint of the master and slave manipulators to obtain the joint angle vector in real time, and the CANopen bus is used to transmit data to ensure the transmission reliability in the electromagnetic shielding environment. The round-trip time of the master-slave command is calculated by UDP timestamping, and the hardware-level time synchronization protocol is deployed to achieve microsecond accuracy. The communication delay data is the time delay for the command transmission between the master and slave manipulators. Radiation sensors and temperature and humidity sensors are installed in the hot chamber environment. The radiation sensor has a range of 0-100Sv / h and an error of ±5%. The temperature and humidity sensors are resistant to high temperature and radiation, and data is transmitted through optical fiber to avoid cable aging. Edge computing nodes are used for preliminary data packaging, which is transmitted to the main control room through 5G private network and wired optical fiber dual redundant channels.

[0078] The data processing module is used to pre-process the collected data, extract the standard feature information of the real-time posture of the master and slave manipulators, the standard feature information of the communication delay and the feature data of the hot chamber environment, and store them in the motion information database.

[0079] Among them, in this embodiment, low-pass filtering is used to denoise the collected data, and normalization is performed to obtain the real-time posture standard information of the master and slave manipulators, the communication delay standard information and the hot room environment data;

[0080] Extracting standard feature information from the real-time posture standard information of the master and slave manipulators, the communication delay standard information and the hot room environment data, and obtaining the real-time posture standard feature information of the master and slave manipulators, the communication delay standard feature information and the hot room environment feature data;

[0081] A motion information database is constructed, and the real-time posture standard feature information of the master and slave manipulators, the communication delay standard feature information, and the thermal chamber environment feature data are associated and stored in the motion information database.

[0082] It should be noted that an RC low-pass filter is integrated at the output end of each sensor to suppress high-frequency noise, and an adaptive Kalman filter is used to process the posture data. The process noise covariance is dynamically adjusted to deal with mechanical vibration interference. The communication delay data is filtered by a sliding window mean filter to eliminate the impact of sudden network jitter, where the window length is 50ms. The Z-score standardization is used to normalize the posture data to retain the anomaly detection capability. The hot room environment data is normalized to the interval [0,1] and adapted to the neural network input layer. After feature extraction, the real-time posture standard feature information of the master and slave manipulators, the communication delay standard feature information and the hot room environment feature data are obtained. The extracted real-time posture standard feature information of the master and slave manipulators, the communication delay standard feature information and the hot room environment feature data are associated to form a complete data record, and the associated data is stored in the motion information database to provide training data for the subsequent trajectory prediction model.

[0083] Among them, standard feature information is extracted from the real-time standard posture information of the master and slave manipulators, the standard communication delay information and the hot room environment data to obtain the real-time standard posture feature information of the master and slave manipulators, the standard communication delay feature information and the hot room environment feature data, including:

[0084] According to the joint angle vectors of the master and slave manipulators, the position features of the master and slave manipulators are calculated, where the position feature expression is:

[0085] p m (t) = f kin (q m (t))

[0086] p s (t) = f kin (q s (t))

[0087] Where P m (t) represents the position vector of the main robot hand at time t, q m (t) represents the joint angle vector of the main robot hand at time t, P s (t) represents the position vector of the robot hand at time t, q s (t) represents the joint angle vector of the robot hand at time t, f kin is the kinematics positive solution function;

[0088] The first-order derivative of the position characteristics of the master and slave manipulators with respect to time is performed to obtain the velocity characteristics v of the master manipulator respectively. m (t) and the velocity characteristic v from the robot hand s (t); Take the second-order derivative of the position characteristics of the master and slave manipulators with respect to time, and obtain the acceleration characteristics a of the master manipulator respectively. m (t) and the acceleration characteristic a from the robot hand s (t);

[0089] Among them, the first-order derivative of the position characteristics of the master and slave robot hands with respect to time is performed to obtain the velocity characteristics v of the master robot hand respectively. m (t) and the velocity characteristic v from the robot hand s (t), the expression is:

[0090]

[0091]

[0092] In the formula, v m (t) represents the velocity characteristics of the main robot hand at time t, v s (t) represents the velocity characteristics of the slave robot at time t;

[0093] Among them, the position characteristics of the master and slave robot hands are second-order derivatived with respect to time, and the acceleration characteristics of the master robot hand a and m (t) and the acceleration characteristic a from the robot hand s (t), the expression is:

[0094]

[0095] In the formula, a m (t) represents the acceleration characteristics of the main robot hand at time t, a s (t) represents the acceleration characteristics of the slave robot at time t;

[0096] Preset a fixed time step, obtain the communication delay data at each time point of the fixed time step, calculate the average delay and delay fluctuation in the fixed time step, and obtain the standard characteristic information of the communication delay. The delay fluctuation expression is:

[0097]

[0098] Among them, σΔt is the delay fluctuation, Δt i is the communication delay data at the current time point, Δt avg is the average delay, N is the number of all time points with a fixed time step;

[0099] The radiation intensity, temperature and humidity at each time point of a fixed time step are obtained, and the average radiation intensity, average temperature and average humidity in the fixed time step are calculated to obtain the environmental characteristic data of the hot room.

[0100] It should be noted that the position characteristics of the master and slave manipulators are calculated by the kinematic forward function, and the first-order and second-order derivatives of the position characteristics are taken to obtain the velocity and acceleration characteristics, which can more comprehensively describe the motion state of the manipulator. These characteristics provide rich information for the trajectory prediction model, which helps the model to more accurately capture the motion trend and dynamic changes of the manipulator, thereby improving the accuracy of trajectory prediction; and the extraction of the hot chamber environmental characteristic data can reflect the physical characteristics of the manipulator's operating environment. Adding these environmental characteristics to the trajectory prediction model can make the model better adapt to the manipulator's motion under different environmental conditions, further improving the prediction accuracy; at the same time, the average delay and delay fluctuation in a fixed time step are calculated to obtain the standard characteristic information of the communication delay, which helps the model understand the distribution and variation range of the communication delay, so as to consider the influence of the delay when predicting the trajectory and enhance the robustness of the system to the communication delay. By presetting a fixed time step to obtain the communication delay data and the hot chamber environment data, the periodicity and regularity of data acquisition can be ensured, which is conducive to a more stable operation of the system and reduces the delay and fluctuation caused by irregular data acquisition.

[0101] The motion prediction module is used to build a slave robot trajectory prediction model based on a neural network, taking the historical data and real-time posture data in the motion information database and the corresponding communication delay data and the thermal chamber environment data as input, training the slave robot trajectory prediction model to output the motion trajectory prediction information of the slave robot;

[0102] like Figure 3 As shown, the action prediction module in this embodiment constructs a slave robot trajectory prediction model based on a neural network. The network structure of the slave robot trajectory prediction model includes an input layer, a multi-scale feature extraction module, a hot room environment feature enhancement module, a Transformer encoder and a multi-task output module, wherein the input layer is used to receive the real-time posture standard feature information of the master and slave robots, the communication delay standard feature information and the hot room environment feature data, the output end of the input layer is connected to the input end of the multi-scale feature extraction module, and is used to extract the spatial and temporal features of the input data, the output end of the multi-scale feature extraction module is connected to the input end of the hot room environment feature enhancement module, and is used to dynamically adjust the feature weight according to the hot room environment feature data, the output end of the hot room environment feature enhancement module is connected to the input end of the Transformer encoder, and is used to capture the global dependency through a multi-head self-attention mechanism, and the output end of the Transformer encoder is connected to the input end of the multi-task output module, and is used to respectively predict and output the position vector, velocity feature and acceleration feature of the slave operator.

[0103] It should be noted that through the multi-scale feature extraction module, the model can more comprehensively capture the spatial and temporal features in the input data, thereby improving the accuracy of trajectory prediction. The hot room environment feature enhancement module can dynamically adjust the feature weights according to the environmental characteristics, so that the model can maintain a high prediction accuracy under different environmental conditions; the Transformer encoder captures global dependencies through a multi-head self-attention mechanism, making the model more robust to noise and outliers in the input sequence; the multi-task output module can provide more comprehensive and stable trajectory prediction information by predicting the position, velocity and acceleration respectively, further enhancing the robustness of the model. The model structure can process feature data from different sources and scales, has strong generalization ability, and can efficiently process input data and output prediction results.

[0104] The multi-scale feature extraction module in this embodiment includes a CNN network extraction unit, an LSTM network extraction unit and a feature fusion unit, wherein:

[0105] The CNN network extraction unit is used to extract the spatial features of the input data, and the expression is:

[0106] F cnn =CNN(X)

[0107] In the formula, F cnn is the spatial feature extracted by the CNN network extraction unit, X is the input standard feature information of the real-time posture of the master and slave manipulators, the standard feature information of the communication delay, and the feature data of the hot room environment;

[0108] The LSTM network extraction unit is used to extract the time characteristics of the input data, and the expression is:

[0109] F lstm =LSTM(X)

[0110] In the formula, F lstm Temporal features extracted for the LSTM module;

[0111] The feature fusion unit is used to fuse the features output by the CNN network extraction unit with the features output by the LSTM network extraction unit to obtain fused feature data, which is expressed as:

[0112]

[0113] In the formula, F fusion It is the concatenation result of the spatial features extracted by the CNN network extraction unit and the temporal features extracted by the LSTM module.

[0114] It can be understood that the CNN convolutional neural network can automatically learn the spatial hierarchical features in the input data through structures such as convolutional layers and pooling layers. In trajectory prediction, CNN can capture the standard feature information of the real-time posture of the master and slave manipulators and the spatial patterns in the characteristic data of the hot chamber environment; the LSTM long short-term memory network can handle the long-term dependency problems in sequence data. In trajectory prediction, LSTM can capture the temporal dynamics in the motion sequence of the master and slave manipulators, such as the changing trends of speed and acceleration, and the temporal characteristics of communication delay; the spatial features output by the CNN network extraction unit and the temporal features output by the LSTM network extraction unit are usually fused by splicing to obtain fused feature data, and then the spatial and temporal features are extracted by CNN and LSTM respectively, which can more comprehensively capture the information in the input data and improve the expression ability of the features.

[0115] The thermal chamber environment feature enhancement module in this embodiment includes a preset radiation resistance weight basic matrix and a temperature and humidity weight basic matrix, wherein:

[0116] Dynamically adjust the anti-radiation basic weight matrix according to the radiation intensity to obtain the anti-radiation weight matrix, which is expressed as:

[0117]

[0118] Where W rad (t) is the radiation resistance weight matrix, is the basic matrix of radiation resistance weight, α is the influence coefficient of radiation intensity, and γ(t) is the radiation intensity at time point t;

[0119] Dynamically adjust the temperature and humidity weight basic matrix according to temperature and humidity to obtain the temperature and humidity weight matrix, which is expressed as:

[0120]

[0121] Where W dynamic (t) is the temperature and humidity weight matrix, is the temperature and humidity weight basic matrix, β T is the temperature influence coefficient, β H is the influence coefficient of humidity, T(t) is the temperature at time point t, and H(t) is the humidity at time point t;

[0122] The fused feature data is enhanced according to the anti-radiation weight matrix and the temperature and humidity weight basic matrix to obtain enhanced feature data, which is expressed as:

[0123] F dynamic =F fusion· W rad (t) · W dynamic (t)

[0124] In the formula, F dynamic To enhance feature data.

[0125] It should be noted that the hot chamber environment feature enhancement module is responsible for dynamically adjusting the feature weights according to the specific characteristics of the hot chamber environment, thereby enhancing the adaptability and prediction accuracy of the model under different environmental conditions; among them, the anti-radiation weight basic matrix and the temperature and humidity weight basic matrix are preset. The anti-radiation weight basic matrix represents the contribution of each feature to the anti-radiation capability under basic conditions, that is, when there is no radiation or the radiation intensity is low; the temperature and humidity weight basic matrix represents the influence of temperature and humidity on each feature under basic conditions.

[0126] It is understandable that radiation intensity is an important factor affecting the safety of the hot chamber environment. As the radiation intensity increases, the contribution of the slave robot to its radiation resistance may change; therefore, it is necessary to dynamically adjust the anti-radiation weight matrix according to the radiation intensity; temperature and humidity are also important factors affecting the hot chamber environment. Temperature and humidity will affect the performance of the robot, the stability of the material, and the accuracy of the sensor. Therefore, it is necessary to dynamically adjust the temperature and humidity weight matrix according to the temperature and humidity; the fused feature data is weighted by the dynamically adjusted anti-radiation weight matrix and temperature and humidity weight matrix, so that the model can better adapt to the hot chamber environment under different radiation intensities, temperatures, and humidity conditions, which is conducive to the model to more accurately predict the motion trajectory of the slave robot, and significantly improve the prediction accuracy and stability of the model.

[0127] The multi-task output module in this embodiment includes outputting the position vector, velocity characteristics and acceleration characteristics of the slave operator, wherein:

[0128] The expression for the predicted position vector is:

[0129] PY s (t+Δt)=W p h+b p

[0130] Where PY s (t+Δt) is the position vector predicted from the operator, W p is the weight of the predicted position vector, b p is the bias of the predicted position vector;

[0131] The expression of velocity characteristic is:

[0132] v s (t+Δt)=W v h+b v

[0133] In the formula, vY s (t+Δt) is the speed characteristic predicted from the operator, W v is the weight of the predicted speed feature, b v is the bias for predicting velocity characteristics;

[0134] The expression of acceleration characteristic is:

[0135] aY s (t+Δt)=W a h+b a

[0136] In the formula, aY s (t+Δt) is the acceleration characteristic predicted from the operator, W a is the weight of the predicted acceleration feature, b a is the bias for predicting the acceleration characteristics.

[0137] It should be noted that the position vector is the key information describing the position of the slave operator in space. The enhanced feature data is linearly transformed through the fully connected layer to obtain the predicted position vector; the speed feature is the key information describing the speed and direction of the slave operator's movement. The enhanced feature data is linearly transformed through the fully connected layer to obtain the predicted speed feature; the acceleration feature is the key information describing the speed of change of the slave operator's movement speed. The enhanced feature data is linearly transformed through the fully connected layer to obtain the predicted acceleration feature; this module can simultaneously predict the position vector, speed feature and acceleration feature of the slave operator, providing comprehensive information for the control of the slave operator.

[0138] Among them, the robot hand trajectory prediction model in this embodiment also includes setting a loss function, and using the Adam optimization algorithm to perform model iterative training according to the loss function to adjust the parameters of the model until convergence is reached and then stop the iterative training, wherein the expression of the loss function is:

[0139] L=λ1||PY s (t+Δt)-P s (t)|| 2 +λ2||vY s (t+Δt)-v s (t)|| 2 +λ3||aY s (t+Δt)-a s (t)|| 2

[0140] Where L is the multi-task loss function, λ1 is the weight coefficient for position vector prediction, λ2 is the weight coefficient for velocity feature prediction, and λ3 is the weight coefficient for acceleration feature prediction.

[0141] It should be noted that λ1||PY s (t+Δt)-P s (t)|| 2 is the position vector prediction loss, which measures the predicted position vector PY s (t+Δt) and the true position vector P s (t), λ1 is the weight coefficient of position vector prediction, which is used to adjust the proportion of this item in the total loss; λ2||vY s (t+Δt)-v s (t)|| 2 is the speed feature prediction loss, which measures the predicted speed feature vY s (t+Δt) and true velocity characteristic v s (t) between the mean square error; λ3||aY s (t+Δt)-a s (t)|| 2 is the acceleration feature prediction loss, which measures the predicted acceleration feature aY s (t+Δt) and the true acceleration characteristic a s The multi-task loss function takes into account the prediction errors of position, velocity and acceleration at the same time, so that the model can learn the motion law of the robot arm more comprehensively, and by adjusting the weight coefficients λ1, λ2 and λ3, it can flexibly control the proportion of different prediction tasks in the total loss, so as to adapt to different application requirements.

[0142] The execution module is used to convert the motion trajectory prediction information of the slave robot hand at the current moment into a control signal and send it to the slave hand manipulation module. The slave hand manipulation module performs corresponding actions according to the received control signal to achieve accurate following of the master hand action.

[0143] Among them, the execution module in this embodiment calculates the driving torque of each joint of the slave robot hand through the inverse kinematics algorithm according to the predicted trajectory information, converts it into an analog control signal, and sends the control signal to the slave hand manipulation module. The slave hand manipulation module performs corresponding actions according to the received control signal to achieve accurate following of the master hand action.

[0144] When it is necessary to explain, the execution module first receives the motion trajectory prediction information of the current moment output by the trajectory prediction model of the slave robot, including the position vector, velocity characteristics and acceleration characteristics. The execution module calculates the driving torque of each joint of the slave robot through the inverse kinematics algorithm according to the predicted trajectory information. The execution module converts the calculated joint driving torque into an analog control signal for driving each joint of the slave robot; the execution module sends the converted analog control signal to the slave manipulation module so that the slave manipulation module performs corresponding actions according to these signals; thus, the system has a high degree of flexibility. By adjusting the trajectory prediction model and the inverse kinematics algorithm, it can adapt to the configuration and application scenarios of different master and slave robots, and realize the flexible following of the slave robot to the master hand action.

[0145] like Figure 2 As shown, in a second aspect, the present invention also provides a method for controlling a master-slave robot for nuclear decommissioning operation, which is implemented by a master-slave robot control system for nuclear decommissioning operation, and includes the following steps:

[0146] S1, collects the real-time posture data of the master and slave manipulators and the corresponding communication delay data and hot room environment data respectively;

[0147] S2, preprocessing the collected data, extracting the standard feature information of the real-time posture of the master and slave manipulators, the standard feature information of the communication delay and the feature data of the hot chamber environment, and storing them in the motion information database;

[0148] S3, constructing a master-slave robot trajectory prediction model based on a neural network, taking the historical data and real-time posture data in the motion information database and the corresponding communication delay data and the thermal chamber environment data as input, training the slave robot trajectory prediction model to output the motion action trajectory prediction information of the slave robot;

[0149] S4, converting the motion trajectory prediction information of the slave robot hand at the current moment into a control signal, and sending it to the slave hand manipulation module. The slave hand manipulation module performs corresponding actions according to the received control signal to achieve accurate following of the master hand action.

[0150] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0151] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system and module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0152] In the embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0153] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0154] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0155] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, ROM, RAM, disk or optical disk, etc., which can store program codes.

[0156] In addition, it should be noted that in the system and method of the present invention, it is obvious that each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent schemes of the present invention. Moreover, the steps of performing the above-mentioned series of processing can naturally be performed in chronological order according to the order of description, but it is not necessary to perform them in chronological order, and some steps can be performed in parallel or independently of each other. For those of ordinary skill in the art, it is understood that all or any steps or components of the method and apparatus of the present invention can be implemented in any computing device (including processors, storage media, etc.) or a network of computing devices in hardware, firmware, software or a combination thereof, which can be achieved by those of ordinary skill in the art using their basic programming skills after reading the description of the present invention.

[0157] Therefore, the purpose of the present invention can also be achieved by running a program or a group of programs on any computing system. The computing system can be a well-known general-purpose system. Therefore, the purpose of the present invention can also be achieved by simply providing a program product containing a program code for implementing a method or device. That is to say, such a program product also constitutes the present invention, and a storage medium storing such a program product also constitutes the present invention. Obviously, the storage medium can be any well-known storage medium or any storage medium developed in the future. It should also be pointed out that in the device and method of the present invention, it is obvious that each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent schemes of the present invention. In addition, the steps of performing the above-mentioned series of processing can naturally be performed in chronological order according to the order of description, but it is not necessary to perform them in chronological order. Some steps can be performed in parallel or independently of each other.

[0158] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A master-slave robot control system for nuclear decommissioning operations, characterized in that: The system comprises: The acquisition module is used to respectively collect the real-time posture data of the master and slave manipulators and the corresponding communication delay data and the thermal chamber environment data; The data processing module is used to pre-process the collected data, extract the standard feature information of the real-time posture of the master and slave manipulators, the standard feature information of the communication delay and the feature data of the hot chamber environment, and store them in the motion information database; The motion prediction module is used to build a trajectory prediction model for the slave robot based on a neural network, taking the historical data and real-time posture data in the motion information database and the corresponding communication delay data and the thermal chamber environment data as input, training the trajectory prediction model for the slave robot to output the motion trajectory prediction information of the slave robot; The execution module is used to convert the motion trajectory prediction information of the slave robot hand at the current moment into a control signal and send it to the slave hand manipulation module. The slave hand manipulation module performs corresponding actions according to the received control signal to achieve accurate following of the master hand action.

2. The nuclear decommissioning operation master-slave robot control system according to claim 1, characterized in that: The collected data is preprocessed to obtain standard feature information of real-time postures of the master and slave manipulators, standard feature information of communication delays, and feature data of the hot chamber environment, including: The posture data of the master and slave manipulators include the angle vectors of each joint, and the thermal chamber environment data include radiation intensity, temperature and humidity; The collected data is denoised using low-pass filtering and normalized to obtain standard real-time posture information of the master and slave manipulators, standard communication delay information, and thermal chamber environment data; Extracting standard feature information from the real-time posture standard information of the master and slave manipulators, the communication delay standard information and the hot room environment data, and obtaining the real-time posture standard feature information of the master and slave manipulators, the communication delay standard feature information and the hot room environment feature data; A motion information database is constructed, and the real-time posture standard feature information of the master and slave manipulators, the communication delay standard feature information, and the thermal chamber environment feature data are associated and stored in the motion information database.

3. The nuclear decommissioning operation master-slave robot control system according to claim 2, characterized in that: The method extracts standard feature information from the real-time posture standard information of the master and slave manipulators, the communication delay standard information and the hot room environment data to obtain the real-time posture standard feature information of the master and slave manipulators, the communication delay standard feature information and the hot room environment feature data, including: According to the joint angle vectors of the master and slave manipulators, the position features of the master and slave manipulators are calculated, where the position feature expression is: p m (t)=f kin (q m (t)) p s (t)=f kin (q s (t)) Where P m (t) represents the position vector of the main robot hand at time t, q m (t) represents the joint angle vector of the main robot hand at time t, P s (t) represents the position vector of the robot hand at time t, q s (t) represents the joint angle vector of the robot hand at time t, f kin is the kinematics positive solution function; The first-order derivative of the position characteristics of the master and slave manipulators with respect to time is performed to obtain the velocity characteristics v of the master manipulator respectively. m (t) and the velocity characteristic v from the robot hand s (t); Take the second-order derivative of the position characteristics of the master and slave manipulators with respect to time, and obtain the acceleration characteristics a of the master manipulator respectively. m (t) and the acceleration characteristic a from the robot hand s (t); Preset a fixed time step, obtain the communication delay data at each time point of the fixed time step, calculate the average delay and delay fluctuation in the fixed time step, and obtain the standard characteristic information of the communication delay. The delay fluctuation expression is: Among them, σΔt is the delay fluctuation, Δt i is the communication delay data at the current time point, Δt avg is the average delay, N is the number of all time points with a fixed time step; The radiation intensity, temperature and humidity at each time point of a fixed time step are obtained, and the average radiation intensity, average temperature and average humidity in the fixed time step are calculated to obtain the environmental characteristic data of the hot room.

4. The nuclear decommissioning operation master-slave robot control system according to claim 3, characterized in that: The action prediction module constructs a slave robot trajectory prediction model based on a neural network. The network structure of the slave robot trajectory prediction model includes an input layer, a multi-scale feature extraction module, a hot room environment feature enhancement module, a Transformer encoder and a multi-task output module, wherein the input layer is used to receive real-time posture standard feature information of the master and slave robots, communication delay standard feature information and hot room environment feature data, the output end of the input layer is connected to the input end of the multi-scale feature extraction module, and is used to extract the spatial and temporal features of the input data, the output end of the multi-scale feature extraction module is connected to the input end of the hot room environment feature enhancement module, and is used to dynamically adjust the feature weight according to the hot room environment feature data, the output end of the hot room environment feature enhancement module is connected to the input end of the Transformer encoder, and is used to capture global dependencies through a multi-head self-attention mechanism, and the output end of the Transformer encoder is connected to the input end of the multi-task output module, and is used to respectively predict and output the position vector, velocity feature and acceleration feature of the slave operator.

5. The nuclear decommissioning operation master-slave robot control system according to claim 4, characterized in that: The multi-scale feature extraction module includes a CNN network extraction unit, an LSTM network extraction unit and a feature fusion unit, wherein: The CNN network extraction unit is used to extract the spatial features of the input data, and the expression is: F cnn =CNN(X) In the formula, F cnn is the spatial feature extracted by the CNN network extraction unit, X is the input standard feature information of the real-time posture of the master and slave manipulators, the standard feature information of the communication delay, and the feature data of the hot room environment; The LSTM network extraction unit is used to extract the time characteristics of the input data, and the expression is: F lstm =LSTM(X) In the formula, F lstm Temporal features extracted for the LSTM module; The feature fusion unit is used to fuse the features output by the CNN network extraction unit with the features output by the LSTM network extraction unit to obtain fused feature data, which is expressed as: In the formula, F fusion It is the concatenation result of the spatial features extracted by the CNN network extraction unit and the temporal features extracted by the LSTM module.

6. The nuclear decommissioning operation master-slave robot control system according to claim 5, characterized in that: The thermal chamber environment feature enhancement module includes a preset anti-radiation weight basic matrix and a temperature and humidity weight basic matrix, wherein: Dynamically adjust the anti-radiation basic weight matrix according to the radiation intensity to obtain the anti-radiation weight matrix, which is expressed as: Where W rad (t) is the radiation resistance weight matrix, is the basic matrix of radiation resistance weight, α is the influence coefficient of radiation intensity, and γ(t) is the radiation intensity at time point t; Dynamically adjust the temperature and humidity weight basic matrix according to temperature and humidity to obtain the temperature and humidity weight matrix, which is expressed as: Where W dynamic (t) is the temperature and humidity weight matrix, is the temperature and humidity weight basic matrix, β T is the temperature influence coefficient, β H is the influence coefficient of humidity, T(t) is the temperature at time point t, and H(t) is the humidity at time point t; The fused feature data is enhanced according to the anti-radiation weight matrix and the temperature and humidity weight basic matrix to obtain enhanced feature data, which is expressed as: F dynamic =F fusion· W rad (t) · W dynamic (t) In the formula, F dynamic To enhance feature data.

7. The nuclear decommissioning operation master-slave robot control system according to claim 6, characterized in that: The multi-task output module includes outputting the position vector, velocity feature and acceleration feature of the slave operator, wherein: The expression for the predicted position vector is: PY s (t+Δt)=W p h+b p Where PY s (t+Δt) is the position vector predicted from the operator, W p is the weight of the predicted position vector, b p is the bias of the predicted position vector; The expression of velocity characteristic is: vY s (t+Δt)=W v h+b v In the formula, vY s (t+Δt) is the speed characteristic predicted from the operator, W v is the weight of the predicted speed feature, b v is the bias for predicting velocity characteristics; The expression of acceleration characteristic is: aY s (t+Δt)=W a h+b a In the formula, aY s (t+Δt) is the acceleration characteristic predicted from the operator, W a is the weight of the predicted acceleration feature, b a is the bias for predicting the acceleration characteristics.

8. The nuclear decommissioning operation master-slave robot control system according to claim 7, characterized in that: The robot hand trajectory prediction model also includes setting a loss function, and using the Adam optimization algorithm to perform model iterative training according to the loss function to adjust the parameters of the model until convergence is reached and then the iterative training is stopped, wherein the expression of the loss function is: L=λ1||PY s (t+Δt)-P s (t)|| 2 +λ2||vY s (t+Δt)-v s (t)|| 2 +λ3||aY s (t+Δt)-a s (t)|| 2 Where L is the multi-task loss function, λ1 is the weight coefficient for position vector prediction, λ2 is the weight coefficient for velocity feature prediction, and λ3 is the weight coefficient for acceleration feature prediction.

9. The nuclear decommissioning operation master-slave robot control system according to claim 1, characterized in that: The execution module calculates the driving torque of each joint of the slave robot hand through the inverse kinematics algorithm based on the predicted trajectory information, converts it into an analog control signal, and sends the control signal to the slave hand manipulation module. The slave hand manipulation module performs corresponding actions according to the received control signal to achieve accurate following of the master hand action.

10. A method for controlling a master-slave robot for nuclear decommissioning operation, implemented by the master-slave robot control system for nuclear decommissioning operation according to any one of claims 1 to 9, characterized in that: The following steps are involved: S1, collects the real-time posture data of the master and slave manipulators and the corresponding communication delay data and hot room environment data respectively; S2, preprocessing the collected data, extracting the standard feature information of the real-time posture of the master and slave manipulators, the standard feature information of the communication delay and the feature data of the hot chamber environment, and storing them in the motion information database; S3, constructing a master-slave robot trajectory prediction model based on a neural network, taking the historical data and real-time posture data in the motion information database and the corresponding communication delay data and the thermal chamber environment data as input, training the slave robot trajectory prediction model to output the motion action trajectory prediction information of the slave robot; S4, converting the motion trajectory prediction information of the slave robot hand at the current moment into a control signal, and sending it to the slave hand manipulation module. The slave hand manipulation module performs corresponding actions according to the received control signal to achieve accurate following of the master hand action.

Images