Industrial robot intelligent control system

Abstract

The invention discloses an industrial robot intelligent control system which comprises a master control terminal, the master control terminal is provided with an industrial-grade multi-core processor and is integrated with a motion control subsystem and an environment sensing subsystem, the motion control subsystem is connected with the master control terminal through a first communication interface, and the environment sensing subsystem is connected with the master control terminal through a second communication interface. According to the motion control subsystem, a distributed axis controller array and reinforcement learning strategy switching algorithm is adopted, so that the trajectory tracking precision and the energy consumption efficiency are improved; and the environment sensing subsystem realizes real-time anomaly detection through multi-mode sensor fusion and an LSTM-GAN fault prediction model. The system supports containerized deployment and edge computing collaboration and integrates AR human-computer interaction, and the adaptive capacity and safety of the industrial robot are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent control of industrial robots, and particularly to an intelligent control system for industrial robots. Background Art

[0002] An intelligent control system for industrial robots refers to a system used to achieve precise control and motion coordination of the mechanical structure of industrial robots. It controls each part of the mechanical structure by receiving instructions from operators or computers to achieve predetermined work tasks. The intelligent control system plays a crucial role in industrial robots, ensuring the efficient and accurate execution of various operations by the robots.

[0003] The currently used intelligent control systems still have some deficiencies in certain aspects of intelligence. For example, traditional PID control is difficult to cope with vibrations and thermal deformations during high-speed movement, resulting in a decrease in the repeat positioning accuracy of the end effector. Moreover, modules such as vision and force control operate independently, lacking a multi-sensor spatio-temporal alignment and fusion mechanism. This makes the fixed control mode unable to adapt to diverse process requirements, and the switching depends on manual experience. Summary of the Invention

[0004] (I) Technical Problems to be Solved

[0005] Aiming at the deficiencies of the prior art, the present invention integrates a motion control subsystem and an environment perception subsystem through a main control terminal to achieve high-precision dynamic compensation and intelligent decision-making.

[0006] (II) Technical Solutions

[0007] To achieve the above object, the present invention provides the following technical solution: An intelligent control system for industrial robots, including a main control terminal, which is configured with an industrial-grade multi-core processor and integrated with a motion control subsystem and an environment perception subsystem. The motion control subsystem is connected to the main control terminal through a first communication interface and includes:

[0008] An axis controller array, where six independent drive modules are respectively installed adjacent to the drive units of each joint of the robotic arm. Each group includes a joint motor, a harmonic reducer, an absolute encoder, and a current closed-loop feedback unit;

[0009] A dynamic compensation module, which is distributed inside each joint drive unit and includes a vibration accelerometer, a temperature sensor, and a torque sensor;

[0010] A hybrid control module, which integrates a trajectory control unit and a point-to-point control unit, and is built-in with a control strategy switching algorithm based on reinforcement learning;

[0011] The environment perception subsystem is connected to the main control terminal through a second communication interface and includes:

[0012] Multimodal perception network, integrating a depth vision sensor, lidar, a force sensing array, and an inertial measurement unit;

[0013] and

[0014] an inertial measurement unit;

[0015] Anomaly detection module, including a time series data analysis unit and a fault prediction model, and the fault prediction model adopts an LSTM-GAN hybrid neural network architecture.

[0016] As a preferred solution, the master control terminal is equipped with a real-time operating system and supports a containerized deployment module, and the containerized deployment module is used to dynamically load the algorithm containers of the motion control subsystem and the environment perception subsystem, and each algorithm container realizes resource isolation and fast hot switching through lightweight virtualization technology.

[0017] As a preferred solution, the dynamic compensation module fuses vibration, temperature, and torque data through an extended Kalman filter algorithm to generate a six-degree-of-freedom joint compensation matrix, and the compensation matrix is injected into the current closed-loop of the axis controller array through feedforward control, and the compensation error range is less than ±0.05°.

[0018] As a preferred solution, after the dynamic compensation module generates a compensation matrix through a Kalman filter, it is injected into the current loop through a feedforward channel, and its compensation formula is: T comp = K v .a + K t .ΔT + K f .K ext , where K v , K t , K f are adaptive gain coefficients.

[0019] As a preferred solution, the specific process of the reinforcement learning control strategy switching algorithm is as follows:

[0020] A1. The input state space includes the joint angular velocity deviation, the end effector pose error, and the collision risk score of the environment perception subsystem;

[0021] A2. The output action is the hybrid weight coefficient of trajectory control and point-to-point control;

[0022] A3. The reward function is designed as a weighted combination of energy consumption efficiency, trajectory tracking accuracy, and safety threshold;

[0023] A4. The priority experience replay mechanism is adopted during the training process to accelerate convergence.

[0024] As a preferred solution, the training method of the LSTM-GAN fault prediction model is:

[0025] a) The input of the generator is the time series data of historical joint temperature, vibration spectrum, and motor current, and the output is the predicted fault probability;

[0026] b) The input of the discriminator is the joint feature vector of real fault data and generated data, and the output is the data authenticity score;

[0027] c) The Wasserstein distance is used to optimize the adversarial training process to avoid mode collapse.

[0028] As a preferred solution, in the multi-modal perception network, the data of each sensor is fused by a spatio-temporal alignment module. The spatio-temporal alignment module uses the timestamp synchronization and spatial coordinate transformation algorithm, and it adopts the multi-sensor calibration technology, including: the joint calibration of the depth vision sensor and the lidar, generating the external parameter matrix through the checkerboard target;

[0029] The spatial mapping of the force sensor array and the manipulator kinematic model, establishing the conversion model from six-dimensional force / torque to joint torque.

[0030] As a preferred solution, the control system integrates a human-computer interaction module, which includes:

[0031] An augmented reality (AR) projection unit, through the holographic diffraction diaphragm installed at the end of the manipulator, projects the planned trajectory and safety restricted area in 3D form into the working space;

[0032] A voice command parsing unit, which supports the process parameter modification command of natural language processing (NLP), and matches the action command library through the bidirectional attention mechanism.

[0033] As a preferred solution, the control system further includes an edge computing node, which is deployed inside the manipulator base and integrates an FPGA acceleration module for real-time processing of the point cloud data of the environmental perception subsystem and the visual SLAM mapping task, and dynamically allocates the computing load of the master control terminal and the edge node through the adaptive task dispatching algorithm.

[0034] (III) Beneficial effects

[0035] Compared with the prior art, the present invention provides an intelligent control system for an industrial robot, which has the following beneficial effects:

[0036] The system of the present invention integrates a motion control subsystem and an environment perception subsystem through a master control terminal to achieve high-precision dynamic compensation and intelligent decision-making. The motion control subsystem adopts a distributed axis controller array and a reinforcement learning strategy switching algorithm to improve the trajectory tracking accuracy and energy consumption efficiency; the environment perception subsystem realizes real-time anomaly detection through multi-modal sensor fusion and an LSTM-GAN fault prediction model. The system supports containerized deployment and edge computing collaboration, and integrates AR human-computer interaction, significantly improving the adaptive ability and safety of industrial robots. Description of the Drawings

[0037] Figure 1 It is a block diagram of the control system of the present invention;

[0038] Figure 2 It is a flowchart of the control strategy switching algorithm of the control system of the present invention;

[0039] Figure 3 It is a flowchart of the training method of the fault prediction model of the control system of the present invention. Detailed Embodiments

[0040] In order to better understand the purpose, structure and function of the present invention, the intelligent control system of an industrial robot of the present invention will be further described below in conjunction with the drawings and specific embodiments.

[0041] Embodiment 1

[0042] Refer to Figures 1-3 , the intelligent control system of an industrial robot of the present invention includes:

[0043] A master control terminal, which is configured with an industrial multi-core processor and integrated with a motion control subsystem and an environment perception subsystem. The motion control subsystem is connected to the master control terminal through a first communication interface, and it includes:

[0044] An axis controller array, six groups of independent drive modules are respectively installed adjacent to the joint drive units of the robotic arm, and each group includes a joint motor, a harmonic reducer, an absolute encoder and a current closed-loop feedback unit;

[0045] A dynamic compensation module, which is distributedly installed inside each joint drive unit, including a vibration accelerometer, a temperature sensor and a torque sensor;

[0046] A hybrid control module, which integrates a trajectory control unit and a point-to-point control unit, and is built-in with a control strategy switching algorithm based on reinforcement learning;

[0047] Specifically, the master control terminal is equipped with a real-time operating system and supports a containerized deployment module. The containerized deployment module is used to dynamically load the algorithm containers of the motion control subsystem and the environment perception subsystem. Each algorithm container realizes resource isolation and fast hot swapping through lightweight virtualization technology. As the core of the present invention, the master control terminal is responsible for central decision-making and resource scheduling. It uses a multi-core processor (such as NVIDIA Jetson AGX Orin) to run a real-time operating system (RTOS) to ensure low-latency response. It is internally equipped with a containerized deployment module that dynamically loads algorithm containers through lightweight virtualization technology to achieve hot swapping and resource isolation of motion control and perception algorithms. The motion control subsystem connected to it by signal is provided with an axis controller array to drive the movement of the robotic arm joints. Each group of drive modules independently controls one joint, including a joint motor that provides power output, a harmonic reducer that eliminates transmission backlash through a flexible gear phase adjustment technology, and an absolute encoder that real-time feedbacks the joint angle to form a current closed-loop control.

[0048] Specifically, the dynamic compensation module of the present invention compensates for vibration, temperature drift, and load disturbance in real time, and the sensor cluster (vibration accelerometer, temperature sensor, torque sensor) collects data. The dynamic compensation module fuses vibration, temperature, and torque data through an extended Kalman filter algorithm to generate a six-degree-of-freedom joint compensation matrix. The compensation matrix is injected into the current closed-loop of the axis controller array through feedforward control, and the compensation error range is less than ±0.05°. After the dynamic compensation module generates the compensation matrix through the Kalman filter, it is injected into the current loop through the feedforward channel, and its compensation formula is: T comp =K v .a + K t .ΔT + K f .K ext , where K v , K t , K f are adaptive gain coefficients.

[0049] Specifically, the hybrid control module it sets is used to dynamically switch between trajectory control and point-to-point control strategies, and uses the DDPG reinforcement learning algorithm to output the hybrid weight coefficient for the input state space (angular velocity deviation, pose error, collision risk score).

[0050] Furthermore, its hybrid control module is built-in with a control strategy switching algorithm based on reinforcement learning. The specific process of its reinforcement learning control strategy switching algorithm is as follows:

[0051] A1. The input state space includes the joint angular velocity deviation, the pose error of the end effector, and the collision risk score of the environment perception subsystem;

[0052] A2. The output action is the hybrid weight coefficient of trajectory control and point-to-point control;

[0053] A3. The reward function is designed as a weighted combination of energy consumption efficiency, trajectory tracking accuracy, and safety threshold;

[0054] A4. During the training process, a prioritized experience replay mechanism is adopted to accelerate convergence.

[0055] Embodiment 2

[0056] The environmental perception subsystem of the present invention is connected to the main control terminal through the second communication interface, and it includes: a multi-modal perception network, integrating a depth vision sensor, a lidar, a force sensor array, and

[0057] an inertial measurement unit;

[0058] An anomaly detection module, including a time series data analysis unit and a fault prediction model, and the fault prediction model adopts an LSTM-GAN hybrid neural network architecture.

[0059] Furthermore, in the multi-modal perception network of the present invention, the data of each sensor is fused with heterogeneous data through a spatio-temporal alignment module. The spatio-temporal alignment module adopts a timestamp synchronization and space coordinate transformation algorithm, and it adopts a multi-sensor calibration technology, including: joint calibration of the depth vision sensor and the lidar, generating an external parameter matrix through a checkerboard target;

[0060] Spatial mapping of the force sensor array and the kinematic model of the robotic arm, establishing a conversion model from six-dimensional force / torque to joint torque

[0061] Even further, in the present invention, the training method of the LSTM-GAN fault prediction model is as follows:

[0062] a) The input of the generator is the time series data of historical joint temperature, vibration spectrum, and motor current, and the output is the predicted fault probability;

[0063] b) The input of the discriminator is the joint feature vector of real fault data and generated data, and the output is the data authenticity score;

[0064] c) The Wasserstein distance is used to optimize the adversarial training process to avoid mode collapse.

[0065] Embodiment 3

[0066] The control system of the present invention further includes an edge computing node, which is deployed inside the robotic arm base and integrated with an FPGA acceleration module for real-time processing of the point cloud data and visual SLAM mapping tasks of the environmental perception subsystem, and dynamically allocating the computing loads of the master control terminal and the edge node through an adaptive task dispatching algorithm. Among them, the FPGA acceleration module is used for real-time processing of point cloud downsampling (such as VoxelGrid filtering) and visual SLAM mapping, and its adaptive task dispatching algorithm is based on the Q-learning framework to dynamically allocate the computing tasks of the master control terminal and the edge node and optimize the system response delay.

[0067] The system of the present invention is also provided with a human-computer interaction module, which includes:

[0068] An augmented reality (AR) projection unit, which projects the planned trajectory and safety restricted area in 3D form onto the working space through a holographic diffraction diaphragm installed at the end of the robotic arm;

[0069] A voice command parsing unit, which supports process parameter modification commands for natural language processing (NLP) and matches the action command library through a bidirectional attention mechanism.

[0070] Among them, the augmented reality (AR) projection unit projects the planned trajectory onto the working space through DLP projection technology, enabling the visualization of operation instructions and safety warnings. Its voice command parsing unit maps the voice commands to the parameters of the predefined action library based on the NLP model with a bidirectional attention mechanism, enabling natural language interaction.

[0071] In the system of the present invention, the system communication interface realizes data interaction, specifically as follows: The master control terminal is connected to the motion control subsystem, then issues commands to the axis controller to drive the joints, then realizes dynamic compensation and real-time correction, and the hybrid control module performs strategy switching;

[0072] Among them, the environmental perception subsystem is responsible for collecting data, performing spatio-temporal alignment and fusion, the anomaly detection module is responsible for predicting faults and feeding back the fault signals to the master control terminal to adjust the control strategy, and high-load tasks (such as SLAM) are processed by the edge computing node, and the processing results are also returned to the master control terminal through the communication interface. During the processing, the system can also be provided with a self-check unit to perform multi-level diagnosis at startup. For example: First-level diagnosis: Detect the wear state through the acoustic fingerprint characteristics of the harmonic reducer; Second-level diagnosis: Judge the encoder offset based on the harmonic analysis of the motor current; Third-level diagnosis: Test the robustness of the perception subsystem by generating adversarial samples with GAN.

[0073] Furthermore, the master control terminal of the present invention uses NVIDIA Jetson AGX Orin as a multi-core processor, runs the Ubuntu real-time kernel, and deploys algorithm modules such as motion planning and perception fusion through Docker containers. In the axis controller array of its motion control subsystem, the harmonic reducer of each group of drive modules adopts the flexible gear phase adjustment technology to suppress the transmission clearance, while the space-time alignment module of the environmental perception subsystem realizes the space-time synchronization of sensor data through the TF2 library under the ROS framework.

[0074] It can be understood that the present invention is described through some embodiments. Those skilled in the art know that without departing from the spirit and scope of the present invention, various changes or equivalent replacements can be made to these features and embodiments. Additionally, under the teaching of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application belong to the scope protected by the present invention.

Claims

1. An intelligent control system for an industrial robot, characterized in that, Including a master control terminal, which is configured with an industrial multi-core processor and integrated with a motion control subsystem and an environment perception subsystem. The motion control subsystem is connected to the master control terminal through a first communication interface and includes: An axis controller array, where six groups of independent drive modules are respectively installed adjacent to the drive units of each joint of the robotic arm. Each group includes a joint motor, a harmonic reducer, an absolute encoder, and a current closed-loop feedback unit; A dynamic compensation module, which is distributed inside each joint drive unit and includes a vibration accelerometer, a temperature sensor, and a torque sensor; A hybrid control module, which integrates a trajectory control unit and a point-to-point control unit, and has a control strategy switching algorithm based on reinforcement learning built-in; The environment perception subsystem is connected to the master control terminal through a second communication interface and includes: A multi-modal perception network, which integrates a depth vision sensor, a lidar, a force sensor array, and An inertial measurement unit; An anomaly detection module, which includes a time series data analysis unit and a fault prediction model. The fault prediction model adopts an LSTM-GAN hybrid neural network architecture.

2. An intelligent control system for an industrial robot according to claim 1, wherein The master control terminal is equipped with a real-time operating system and supports a containerized deployment module. The containerized deployment module is used to dynamically load the algorithm containers of the motion control subsystem and the environment perception subsystem. Each algorithm container realizes resource isolation and fast hot swapping through lightweight virtualization technology.

3. An intelligent control system for an industrial robot according to claim 1, characterized in that, The dynamic compensation module fuses vibration, temperature, and torque data through an extended Kalman filter algorithm to generate a six-degree-of-freedom joint compensation matrix. The compensation matrix is injected into the current closed-loop of the axis controller array through feedforward control, and the compensation error range is less than ±0.05°.

4. An intelligent control system for an industrial robot according to claim 3, wherein , After the dynamic compensation module generates a compensation matrix through Kalman filtering, it is injected into the current loop through the feedforward channel, and its compensation formula is: T comp = K v .a + K t .ΔT + K f .K ext , where K v , K t , K f are adaptive gain coefficients.

5. An intelligent control system for an industrial robot according to claim 1, characterized in that ,The specific process of the reinforcement learning control strategy switching algorithm is as follows: A1. The input state space includes the joint angular velocity deviation, the end effector pose error, and the collision risk score of the environment perception subsystem; A2. The output action is the hybrid weight coefficient of trajectory control and point-to-point control; A3. The reward function is designed as a weighted combination of energy consumption efficiency, trajectory tracking accuracy, and safety threshold; A4. During the training process, a priority experience replay mechanism is adopted to accelerate convergence.

6. The intelligent control system of an industrial robot according to claim 1, characterized in that ,The training method of the LSTM-GAN fault prediction model is: a) The input of the generator is the time series data of historical joint temperature, vibration spectrum, and motor current, and the output is the predicted fault probability; b) The input of the discriminator is the joint feature vector of real fault data and generated data, and the output is the data authenticity score; c) The Wasserstein distance is used to optimize the adversarial training process to avoid mode collapse.

7. An intelligent control system for an industrial robot according to claim 1, characterized in that ,In the multi-modal perception network, the data of each sensor is fused through a spatio-temporal alignment module. The spatio-temporal alignment module adopts a timestamp synchronization and space coordinate transformation algorithm, and it adopts a multi-sensor calibration technology, including: the joint calibration of the depth vision sensor and the lidar, and the external parameter matrix is generated through a checkerboard target; The spatial mapping of the force sensor array and the kinematic model of the robotic arm to establish a conversion model from six-dimensional force / torque to joint torque.

8. An intelligent control system for an industrial robot according to claim 1, characterized in that ,The control system integrates a human-computer interaction module, which includes: An augmented reality (AR) projection unit projects the planned trajectory and safety restricted area in 3D form into the working space through a holographic diffraction diaphragm installed at the end of the robotic arm. A voice command parsing unit supports process parameter modification commands for natural language processing (NLP) and matches the action instruction library through a bidirectional attention mechanism.

9. An intelligent control system for an industrial robot according to claim 8, characterized in that , The control system further includes an edge computing node deployed inside the robotic arm base, integrated with an FPGA acceleration module, for real-time processing of the point cloud data and visual SLAM mapping tasks of the environmental perception subsystem, and dynamically allocating the computing loads of the master control terminal and the edge node through an adaptive task dispatching algorithm.

Images