Control methods for industrial robot trajectory optimization based on intelligent algorithms

Abstract

This invention relates to the field of industrial robot control technology and discloses a control method for industrial robot trajectory optimization based on intelligent algorithms. The method includes collecting joint position information, tool center point coordinates, and motion time sequences of the robot when performing various tasks, and storing them in a trajectory database; after cleaning and filtering the data, extracting a feature set containing path point sequences, velocity distributions, and acceleration profiles; constructing an intelligent optimization algorithm model with a neural network structure, and training it with the feature set to learn trajectory optimization strategies; parsing the target position coordinates and motion constraints of the current task to obtain initial trajectory parameters; inputting the initial parameters into the trained model, outputting an optimized trajectory sequence containing a list of path points and velocity curves, and then generating control commands to drive the robot's motion. This method can be adapted to different tasks, improve trajectory rationality and motion stability, and is well-suited to actual industrial production.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial robot control, in particular to a control method for trajectory optimization of an industrial robot based on an intelligent algorithm. BACKGROUND

[0002] In the intelligent manufacturing system, industrial robots have been deeply integrated into core production links such as automobile welding, 3C electronic assembly, mechanical part transportation, and precision instrument processing, due to their multi-degree-of-freedom motion capability and continuous operation stability. In these application scenarios, the motion trajectory of the industrial robot is directly related to the smoothness of the work process and the quality of the final product. For example, in automobile body welding, trajectory deviation may cause misplacement of welding points, affecting the structural strength of the body; in electronic component assembly, insufficient trajectory precision may cause component damage or improper installation, increasing production loss.

[0003] Currently, industrial robot trajectory planning relies mainly on traditional algorithms. These algorithms usually construct trajectories based on pre-set mathematical models, requiring technicians to input a large number of fixed parameters in advance. However, in actual industrial production, the target positions, load weights, and motion speed limits of different tasks vary significantly, making it difficult for traditional algorithms to adapt to diversified working conditions. For example, when a robot switches from "heavy part transportation" to "light component assembly" tasks on the same production line, the traditional algorithm needs to adjust the entire set of parameters, which not only consumes a lot of time but also may cause trajectory connection to be sluggish due to insufficient parameter matching.

[0004] The mechanical structure characteristics of industrial robots also pose challenges to trajectory precision. The connecting rod components and drive motors of the robot may experience slight flexible deformation during long-term operation, especially when the robot is moving at high speed or the load changes. This deformation may cause uncontrolled deviation of the end effector. Traditional algorithms cannot sense and compensate for such deviations in real time, resulting in deviations between the trajectory and the pre-set path. In addition, traditional trajectory planning often ignores the correlation between motion time sequences and speed and acceleration, focusing only on the positional accuracy of the path points. This may result in sudden changes in speed during the motion process, exacerbating equipment wear and tear and possibly causing mechanical vibrations, affecting the service life and operation stability of the robot.

[0005] Existing trajectory planning schemes lack effective utilization of historical motion data. Industrial robots generate a large amount of trajectory data during long-term task execution, but most of this data is only stored as records and is not used to optimize subsequent trajectory planning. This results in "starting from scratch" for each new task, which not only reduces trajectory planning efficiency but also makes it difficult to continuously improve trajectory precision, failing to meet the increasing demands of modern industry for production efficiency and product quality. SUMMARY

[0006] The present application aims to provide a control method for trajectory optimization of an industrial robot based on an intelligent algorithm to solve the problems presented in the background art.

[0007] To achieve the above-mentioned purpose, the present application provides a control method for trajectory optimization of an industrial robot based on an intelligent algorithm, which comprises:

[0008] Collecting motion trajectory data generated by the industrial robot when performing various work tasks, the motion trajectory data including joint position information, tool center point coordinates and motion time series, and storing the motion trajectory data in a trajectory database;

[0009] Cleaning and screening the motion trajectory data in the trajectory database, extracting a feature set of the motion trajectory data, the feature set including path point sequences, speed distribution and acceleration profile;

[0010] Building an intelligent optimization algorithm model, the intelligent optimization algorithm model adopting a neural network structure, using the feature set to train the intelligent optimization algorithm model, and making the intelligent optimization algorithm model learn trajectory optimization strategies;

[0011] Obtaining task description information of the current industrial robot, the task description information including target position coordinates and motion constraint conditions, and analyzing the task description information to obtain initial trajectory parameters;

[0012] Inputting the initial trajectory parameters into the trained intelligent optimization algorithm model, the intelligent optimization algorithm model outputting an optimized motion trajectory sequence, the optimized motion trajectory sequence including path point lists and speed curves;

[0013] Generating control instructions according to the optimized motion trajectory sequence, and driving joint motors of the industrial robot through a controller of the industrial robot to make the industrial robot move according to the optimized motion trajectory sequence.

[0014] Preferably, the step of collecting motion trajectory data generated by the industrial robot when performing various work tasks comprises:

[0015] Collecting joint angle data through an encoder sensor installed at a joint of the industrial robot;

[0016] Capturing spatial coordinate data of a tool center point through a vision system;

[0017] Recording joint angle data and tool center point coordinate data corresponding to each motion timestamp;

[0018] Associating the joint angle data, the tool center point coordinate data and the motion timestamp data with a task identifier, the task identifier including a task type and a workpiece number;

[0019] Storing the associated motion trajectory data into a data table of the trajectory database.

[0020] Preferably, the step of cleaning and screening the motion trajectory data in the trajectory database comprises:

[0021] reading the motion trajectory data from the data table of the trajectory database;

[0022] checking the integrity and consistency of the motion trajectory data, and removing records containing missing values or outliers;

[0023] screening the motion trajectory data according to a preset precision threshold, and retaining records with path errors less than the threshold;

[0024] extracting path point sequences from the screened records, the path point sequences being composed of tool center point coordinates in time sequence;

[0025] calculating the speed values and acceleration values between the path point sequences to form speed distribution and acceleration profile.

[0026] Preferably, the step of constructing the intelligent optimization algorithm model comprises:

[0027] initializing the layer structure and parameters of the neural network model, the neural network model including an input layer, a hidden layer and an output layer;

[0028] taking the path point sequences, speed distribution and acceleration profile in the feature set as training input data;

[0029] setting an optimization objective function, the optimization objective function including shortest path and highest smoothness;

[0030] adjusting the parameters of the neural network model using a gradient descent algorithm to minimize the optimization objective function;

[0031] verifying the performance of the trained intelligent optimization algorithm model on the test data set, and saving the model parameters.

[0032] Preferably, the step of obtaining the task description information of the current industrial robot comprises:

[0033] receiving task instructions from user input or upper system;

[0034] parsing target position coordinates in the task instructions, the target position coordinates including three-dimensional space coordinates;

[0035] extracting motion constraint conditions in the task instructions, the motion constraint conditions including maximum speed limit, acceleration limit and obstacle avoidance area;

[0036] combining the target position coordinates and the motion constraint conditions into initial trajectory parameters.

[0037] Preferably, the step of generating the optimized motion trajectory sequence comprises:

[0038] input the initial trajectory parameters into the trained intelligent optimization algorithm model;

[0039] the intelligent optimization algorithm model calculates the path point list through forward propagation;

[0040] interpolate the path point list to generate a continuous velocity curve;

[0041] adjust the velocity curve to meet the motion constraints;

[0042] output the optimized motion trajectory sequence, including the path point list and the velocity curve.

[0043] Preferably, the step of generating control instructions comprises:

[0044] convert the optimized motion trajectory sequence into joint space trajectory, calculate the target angle of each joint through inverse kinematics;

[0045] generate the velocity control signal of each joint according to the velocity curve;

[0046] send the target angle and the velocity control signal to the joint drive of the industrial robot;

[0047] the joint drive controls the motor to rotate, realizing the motion of the industrial robot.

[0048] Preferably, the method further comprises the following steps:

[0049] monitor the actual motion trajectory of the industrial robot in real time, and obtain the actual joint angle and tool center point coordinates through sensors;

[0050] compare the actual motion trajectory with the optimized motion trajectory sequence, and calculate the trajectory tracking error;

[0051] if the trajectory tracking error exceeds the allowed range, adjust the velocity control signal in the control instruction;

[0052] repeat the monitoring and adjusting process until the task is completed.

[0053] Preferably, the step of adjusting the velocity control signal in the control instruction comprises:

[0054] calculate the velocity correction according to the size and direction of the trajectory tracking error;

[0055] superimpose the velocity correction on the original velocity control signal;

[0056] limit the adjusted velocity control signal not to exceed the maximum speed limit in the motion constraints;

[0057] send the adjusted velocity control signal to the joint drive.

[0058] Preferably, the method further comprises the following steps:

[0059] After the task is completed, record the actual motion trajectory data and its performance indicators;

[0060] Add the actual motion trajectory data to the trajectory database;

[0061] Update the training set of the intelligent optimization algorithm model with the new data, retrain the model;

[0062] Save the updated model parameters for subsequent tasks.

[0063] Compared with the prior art, the beneficial effects of the present application are:

[0064] By collecting the motion trajectory data of the industrial robot performing various work tasks, covering joint position information, tool center point coordinates and motion time series, and storing these data in the trajectory database, the trajectory information under different working conditions can be accumulated in full, providing rich actual data support for subsequent trajectory optimization. These data are not isolated pieces of information, but cover the real motion state of the robot under different task types, different load conditions and different motion speeds, which can make the subsequent optimization process more in line with the actual production scene and avoid theoretical planning that is divorced from reality.

[0065] After cleaning and screening the data in the trajectory database, a feature set is extracted, including path point sequence, speed distribution and acceleration profile, which can remove outliers and redundant information in the data, so that the extracted features more accurately reflect the core characteristics of the trajectory. For example, by screening out erroneous joint position data caused by temporary sensor failure, and by cleaning up chaotic motion time series, the feature data used for subsequent model training is ensured to be consistent and reliable, so that the trajectory rules learned by the model are more in line with the real motion logic of the robot, and the interference of invalid data on the optimization result is reduced.

[0066] An intelligent optimization algorithm model using a neural network structure is constructed, and the feature set is used for training, so that the model learns the trajectory optimization strategy, and fully utilizes the advantages of neural networks in handling nonlinear problems. Compared with the limitations of traditional algorithms relying on fixed mathematical models, neural networks can autonomously mine the potential correlations between trajectory parameters from a large amount of feature data, such as the matching relationship between path point sequence and speed distribution, and the influence law of acceleration profile on motion stability, and thus form a more adaptive optimization logic. This data learning-based optimization method does not require manual adjustment of complex parameters by technical personnel, and can autonomously adapt to the needs of different tasks, reducing the dependence on human experience.

[0067] The description information of the current task is acquired, including target position coordinates and motion constraint conditions, and initial trajectory parameters are parsed to ensure that the initial trajectory planning is closely combined with the actual requirements of the current task. In actual production, the target positions of different tasks differ greatly, and the motion constraint conditions are also different. For example, some tasks require rapid motion to improve efficiency, and some tasks require low-speed and smooth motion to ensure accuracy. The initial trajectory parameters obtained by parsing the task description information can accurately meet the specific requirements of the current task, lay a foundation for subsequent optimization that meets the actual situation, and avoid deviation of the optimization direction caused by disconnection between the initial parameters and the task requirements.

[0068] The initial trajectory parameters are input into the trained model, and the optimized motion trajectory sequence containing the path point list and the speed curve is output, which can make the trajectory meet the task requirements while optimizing the motion state. The optimized path point list can ensure that the joints move coordinately during the motion of the robot, reducing unnecessary path detours; the optimized speed curve can avoid sudden changes in speed, making the robot motion more stable, reducing mechanical vibration, and reducing equipment wear and tear. This optimization is not a single-dimensional adjustment, but a comprehensive optimization that takes into account the rationality of the path and the stability of the motion, making the robot efficient and stable when performing tasks.

[0069] According to the optimized motion trajectory sequence, control instructions are generated, and the joints motors are driven by the controller to make the robot move according to the optimized trajectory, which can effectively convert the optimization results into actual work behavior. The generated control instructions are accurately matched with the hardware structure of the robot, which can ensure that the joint motors work cooperatively according to the optimized speed and position requirements, avoiding motion deviation caused by incompatible instructions and hardware. In actual work, this process can make the motion of the robot more accurately meet the task target, reduce work errors, improve the smoothness of the production process, reduce the probability of equipment failure caused by unreasonable trajectory, reduce maintenance downtime, and ensure the continuity of production. BRIEF DESCRIPTION OF DRAWINGS

[0070] Figure 1 The working principle diagram of the control method of the industrial robot trajectory optimization based on intelligent algorithm is described in the present application;

[0071] Figure 2 The flowchart of motion trajectory data acquisition is described;

[0072] Figure 3 The flowchart of the optimized motion trajectory sequence generation is described. DETAILED DESCRIPTION

[0073] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.

[0074] Please refer to Figure 1 The present application provides a control method for trajectory optimization of an industrial robot based on an intelligent algorithm. The method comprises: motion trajectory data generated by an industrial robot when performing various work tasks is collected by a system, the motion trajectory data including joint position information, tool center point coordinates, and motion time sequences, and the data is stored in a trajectory database. The motion trajectory data in the trajectory database is processed by cleaning and screening to extract a feature set, the feature set including path point sequences, speed distribution, and acceleration profiles. An intelligent optimization algorithm model is constructed using a neural network structure, the feature set is used to train the intelligent optimization algorithm model, and the intelligent optimization algorithm model learns trajectory optimization strategies. Task description information of a current industrial robot is obtained, the task description information including target position coordinates and motion constraint conditions, and the task description information is analyzed to obtain initial trajectory parameters. The initial trajectory parameters are input into the trained intelligent optimization algorithm model, the intelligent optimization algorithm model outputs optimized motion trajectory sequences, and the optimized motion trajectory sequences include path point lists and speed curves. Control instructions are generated according to the optimized motion trajectory sequences, each joint motor of the industrial robot is driven through a controller of the industrial robot, and the industrial robot moves according to the optimized motion trajectory sequences.

[0075] Embodiment 1: Please refer to Figure 2The collection of industrial robot motion trajectory data is realized by integrating encoder sensors at the joint parts of the robot. The encoder sensors use absolute optical encoders with a resolution of millions of pulses per revolution, which can capture small changes in joint angles in real time. The joint angle data is output in digital signal form and transmitted to the data acquisition card through the industrial field bus. The data acquisition card filters and digitizes the signal to eliminate electrical noise interference. The vision system is deployed above the robot work unit. The vision system consists of a high-frame-rate industrial camera and machine vision software. The industrial camera is connected to the processing unit through a gigabit Ethernet network. The machine vision software calibrates and coordinates the captured images and calculates the three-dimensional coordinates of the tool center point in the base coordinate system. Each motion timestamp is generated by a high-precision clock chip. The timestamp data, joint angle data, and tool center point coordinate data are strictly time-aligned through a hardware synchronization signal, which is distributed uniformly by the robot controller. The process of associating motion trajectory data with task identifiers is completed in the data preprocessing module. The task identifier includes task type classification code and workpiece number serial number. The task type classification code is defined according to the international standard process code, and the workpiece number serial number corresponds to the material tracking number in the production management system. The association operation uses a database transaction processing mechanism to ensure the atomicity and consistency of data records. The trajectory database uses a time series database architecture. The data table design includes time index fields, joint angle array fields, tool center point coordinate fields, and task identifier fields. The database engine optimizes the storage efficiency of large-scale time series data through partition table technology. The data acquisition frequency is dynamically adjusted according to the robot motion speed. At high speed, the sampling rate is in the kilohertz level, and at low speed, it is reduced to the hundred hertz level, balancing data accuracy and storage overhead.

[0076] The motion trajectory data cleaning process in the trajectory database starts from the data quality inspection link. The data quality inspection module scans each record in the data table to detect whether the joint angle data exceeds the mechanical limit value and whether the tool center point coordinates are within the robot workspace envelope. Records containing missing values are identified by database query statements. The statistical process control method is used for anomaly value detection. The standard deviation of the data in the moving window is calculated, and data points deviating from the mean value by more than three standard deviations are removed. The preset precision threshold is derived from the technical specifications provided by the robot manufacturer. Path error calculation is completed by comparing the deviation of the actual tool center point trajectory from the theoretical trajectory. The theoretical trajectory is generated by the computer-aided manufacturing system. The screened records are transferred to the feature extraction stage. The path point sequence reconstruction uses a time alignment algorithm to resample the tool center point coordinates at millisecond-level time intervals to generate a uniform time-step spatial path point array. The speed distribution calculation is based on the time differentiation of the path point sequence. The displacement vector between adjacent path points is divided by the time interval to obtain the instantaneous velocity vector. The length of the velocity vector in the three-dimensional space constitutes the velocity distribution curve. The acceleration profile is derived by the time differentiation of the velocity distribution curve. The central difference method is used to calculate the acceleration value. The smoothing process uses a moving average filter to reduce numerical noise. The feature extraction module outputs a standardized feature set. The path point sequence is converted into a homogeneous transformation matrix representation in the relative base coordinate system. The velocity distribution and acceleration profile are normalized to eliminate the influence of dimensional differences on model training. The entire data cleaning and screening process is embedded in the real-time operating system of an industrial computer. The multi-thread parallel processing improves the computing efficiency. The processing log is recorded to the system audit database for quality traceability.

[0077] The hardware configuration of the data acquisition system includes a signal conditioning circuit that amplifies and isolates the original encoder signal to prevent ground loop interference. The camera intrinsic matrix and extrinsic matrix are updated regularly during the calibration process of the vision system using Zhang Zhengyou's calibration method to compensate for optical system drift caused by mechanical vibration. The motion trajectory data storage uses a redundant disk array architecture. The data backup strategy follows the industrial data management specification. Hot backup data is used for real-time queries, and cold backup data is used for long-term archiving. The trajectory database access interface provides both structured query language and application programming interface, supporting third-party analysis tool integration. Machine learning algorithms are introduced for anomaly value processing in trajectory data cleaning. The Isolation Forest model identifies abnormal patterns in high-dimensional data, assisting traditional statistical methods to improve detection sensitivity. Path error calculation integrates multi-sensor data. The spatial position measured by the laser tracker serves as a reference value to verify the accuracy of the vision system. Feature set construction considers kinematic constraints. The curvature continuity of the path point sequence is ensured through spline interpolation. The derivative continuity of the velocity distribution and acceleration profile meets the robot motion smoothness requirements. The data preprocessing pipeline realizes automated scheduling. The monitoring interface displays data quality indicators and processing progress. The operator can manually intervene in the screening threshold parameters.

[0078] The collection accuracy of industrial robot motion trajectory data is affected by mechanical transmission error. The transmission error compensation model is embedded in the data collection link. The compensation model is established based on robot calibration data, and the joint angle reading is corrected in real time. The tool center point coordinate calculation involves a complex coordinate system transformation chain, which includes the robot base coordinate system, the flange coordinate system, the tool coordinate system, and the vision sensor coordinate system. Each coordinate system transformation is realized through a homogeneous transformation matrix. The time synchronization system uses the IEEE 1588 precision clock protocol, and the clock synchronization error of each node in the network is controlled within microseconds. The management of task identifiers is integrated with the production order system. The task type classification code is mapped to the process route library of the manufacturing execution system, and the workpiece number serial number is associated with the bill of materials of the product life cycle management system. The motion trajectory data storage scheme design supports a distributed architecture, with edge computing nodes responsible for local data caching and cloud platforms responsible for historical data archiving. The data cleaning rule engine supports visual configuration, and users can customize integrity check rules and anomaly detection algorithms. The precision threshold parameters are dynamically adjusted according to the processing technology requirements. Strict threshold values are used for precision assembly tasks, and loose threshold values are used for rough machining tasks. The feature extraction of the path point sequence includes geometric property analysis, calculation of the curvature tensor and torsion distribution of the path point, and description of the spatial twist degree of the trajectory. The speed distribution analysis considers kinematic constraints, compares the speed extreme points with the robot dynamics limit value, and the acceleration profile analysis identifies the jerk mutation points.

[0079] Embodiment 2: The construction of the intelligent optimization algorithm model is based on a deep neural network architecture, which contains a full connection structure of an input layer, three hidden layers, and an output layer. The number of input layer neurons matches the dimension of the feature set, which contains the coordinate data of the path point sequence, the scalar value of the speed distribution, and the vector components of the acceleration profile. The hidden layers use rectified linear unit activation functions, which introduce non-linear transformation capabilities, and the number of neurons in the hidden layers follows the geometric pyramid decreasing rule. The output layer is designed as a multi-dimensional regression output, and the number of output layer neurons corresponds to the dimension of the optimized trajectory parameters. The weight initialization of the neural network model uses the Xavier initialization method, and the bias term is initialized to zero. The optimization objective function is defined as a multi-objective weighted sum, which includes a path length term and a smoothness term. The path length term is calculated by the cumulative sum of the Euclidean distance between trajectory points, and the smoothness term is measured by the integral of acceleration variation. The adaptive moment estimation algorithm is used in the training process, which dynamically adjusts the learning rate. The training data is divided into batch input, and the batch size is set to 256 samples. The loss function calculation uses the mean square error form, which calculates the deviation between the predicted trajectory and the ideal trajectory. The backpropagation algorithm calculates the gradient of the loss function with respect to the weights, and the gradient clipping technique prevents gradient explosion. The number of training iterations is set to 10000 cycles, and the early stopping method monitors the change of the validation set loss value. The validation set is randomly extracted from 20% of the feature set, and the test set uses independently collected trajectory data. The training platform is equipped with a graphics processing unit accelerator, which performs parallel matrix operations through the CUDA architecture. The model parameters are saved in binary file format, including weight matrices and bias vectors.

[0080] The task description information acquisition module integrates multiple input interfaces, including user interface interface and manufacturing execution system interface. The user interface interface provides a graphical parameter input form, and the user inputs the X, Y and Z component values of the target position coordinates. The manufacturing execution system interface adopts OPCUA communication protocol, and the manufacturing execution system interface receives production order data in real time. The task instruction parser analyzes the syntax structure of the input data, which conforms to the predefined job description language specification. The target position coordinates are extracted through regular expression matching, and the coordinate values are converted into floating point number format. The motion constraint condition analysis identifies keywords, and the maximum speed limit parameter is mapped from the robot technical specification table. The acceleration limit value is determined according to the mechanical structure strength. The obstacle avoidance area definition adopts the boundary box representation method, which is composed of the minimum coordinate point and the maximum coordinate point. The initial trajectory parameter combination process performs data integrity check, which checks the existence of necessary parameters. The target position coordinates and motion constraint conditions are packaged into structured data objects, and the structured data objects adopt JSON lightweight data exchange format. The coordinate system conversion module unifies the target position coordinates to the robot base coordinate system, and the coordinate system conversion involves homogeneous transformation matrix multiplication operation. The motion constraint condition is coded into a mathematical constraint expression, which includes inequality constraints and equality constraints. The initial trajectory parameter generation module outputs a standardized data stream, which is transmitted to the intelligent optimization algorithm model through network socket. The parameter verification link checks whether the coordinate value is within the robot workspace, and the workspace check is realized through inverse kinematics solution feasibility. The reasonableness of the motion constraint condition is verified by comparing the user input value with the physical limit value of the robot, and the value exceeding the physical limit is automatically adjusted to the allowed range.

[0081] The hidden layer design of the neural network model considers the gradient flow characteristics, and a residual connection structure is added between the hidden layers to alleviate the gradient vanishing problem of the deep network. A batch normalization layer is inserted before each hidden layer to stabilize the distribution changes of the training process. A learning rate scheduler adopts a cosine annealing strategy, which periodically adjusts the learning rate during the training process. A weight decay term is added to the loss function to control the model complexity and prevent overfitting. Training data enhancement techniques apply random noise to improve the robustness of the model. The model evaluation indicators include the average trajectory error and the maximum deviation value, and the average trajectory error calculates the average distance between the predicted trajectory and the true trajectory. The task description information acquisition system implements an identity authentication mechanism to ensure the security of operation permissions. The task instruction version management records the modification history and supports the operation rollback function. The input of the target position coordinates supports multiple format conversions, including Cartesian coordinates and joint angle coordinates. The graphical editing tool of the motion constraint condition provides a visual interface, and the graphical editing tool adjusts the parameter values by dragging the slider. The data structure of the initial trajectory parameters includes a metadata field that records the creation timestamp and operator number. The parameter parsing error handling mechanism records exception logs and provides an error recovery process. The network communication layer implements data encryption transmission using the TLS security protocol.

[0082] The deployment environment of the intelligent optimization algorithm model configures a model server, which provides a REST application programming interface. The inference request includes a session identifier, which associates the context of multiple calls. The input data preprocessing pipeline performs feature scaling, which normalizes the input values to zero mean and unit variance. The model output post-processing program applies a constraint satisfaction algorithm to correct trajectory points that violate motion constraints. The inference result caching mechanism stores frequently used trajectory patterns to reduce the model computation load. The model performance monitoring dashboard displays response time indicators and alerts abnormal states. The task description information acquisition process includes a manual review step to verify the reasonableness of parameters for high-risk tasks. The task instruction syntax definition uses the Backus-Naur form, which describes the formal syntax rules of the instructions. The coordinate transformation chain includes the conversion from the tool coordinate system to the world coordinate system, and the coordinate transformation chain accumulates the product of each transformation matrix. The motion constraint condition compiler generates intermediate representation code to optimize the constraint evaluation order. The initial trajectory parameter verification suite runs unit tests to ensure code quality. The data transmission protocol defines a binary message format to improve transmission efficiency. The error code system classifies various exception types to guide fault diagnosis.

[0083] The hyperparameter tuning of the neural network model uses a Bayesian optimization method that builds a Gaussian process surrogate model. The activation function selection considers the phenomenon of dead neurons, and the activation function selection compares the rectified linear unit and the exponential linear unit. The loss function weight coefficients are determined by Pareto optimization, and the loss function weight coefficients balance multiple optimization objectives. The training data sampling strategy handles class imbalance, and the training data sampling strategy resamples minority class samples. Model ensemble techniques combine multiple neural network models, and model ensemble techniques vote to determine the final output. Feature importance analysis identifies key input variables, and feature importance analysis guides feature engineering improvements. The task description information acquisition system implements a load balancing mechanism, and the load balancing mechanism distributes requests to multiple processing nodes. The instruction parser supports a plug-in architecture extension, and the plug-in architecture extension supports the addition of new instruction syntax. The motion constraint condition database stores historical constraint templates, and the motion constraint condition database supports fast retrieval. The initial trajectory parameter generation algorithm includes fault tolerance processing, and the fault tolerance processing skips damaged input data records. System configuration management maintains parameter configuration files, and system configuration management supports environment-specific settings. Audit tracking records each parameter modification operation, and audit tracking meets the requirements of the quality management system. The performance analysis tool profiles algorithm time consumption, and the performance analysis tool identifies computational bottlenecks.

[0084] In embodiment 3, refer to Figure 3 , the generation process of the optimized motion trajectory sequence starts from the input stage of the initial trajectory parameters, which are transmitted to the intelligent optimization algorithm model through a data interface. The data interface uses a high-speed serial communication protocol to ensure the real-time nature of data transmission. The intelligent optimization algorithm model loads pre-trained neural network parameters, which are read from non-volatile memory into the memory of the computing unit. Forward propagation calculations are performed layer by layer according to the network hierarchy. The initial trajectory parameters received by the input layer are processed through linear transformation and activation functions. Linear transformation uses weight matrix multiplication to achieve linear transformation, and the activation function selects the S-shaped function to introduce nonlinearity. The calculation of the hidden layer involves large-scale matrix operations, which are accelerated through the parallel computing cores of the graphics processing unit. The output layer generates a path point list, which contains a sequence of three-dimensional coordinate values represented in floating-point format for precision.

[0085] The interpolation of the path point list adopts a cubic spline interpolation algorithm, which ensures the second-order derivative continuity of the curve between path points. The interpolation node spacing is adaptively determined according to the motion accuracy requirement, and the node density is increased in the high-curvature area and the node number is reduced in the flat area. The velocity curve generation is based on the time parameterization of the path point list, which is initialized by the uniform motion assumption and then adjusted according to the motion constraint conditions. The process of satisfying the kinematic constraint of the velocity curve uses an optimization algorithm to iteratively adjust the time allocation so that the acceleration profile does not exceed the mechanical system limit. The output format of the motion trajectory sequence includes the correspondence between the time stamp and the spatial coordinates, and the time stamp marks the arrival time of each path point with millisecond precision.

[0086] The path smoothness optimization is realized by the following formula:

[0087]

[0088] Wherein: represents the curvature value at the path point, represents the velocity vector, represents the acceleration vector. The curvature value is used to evaluate the smoothness of the path, and a smaller curvature value corresponds to a smoother motion trajectory. The velocity vector is calculated by the difference between path points, and the acceleration vector is derived by the second difference of the velocity vector.

[0089] The generation stage of the control command starts from the conversion of the motion trajectory sequence to the joint space trajectory, which is obtained by inverse kinematics calculation. The inverse kinematics calculation adopts a combination of analytical solution and numerical solution, the analytical solution handles the special nature of the geometric configuration of the robot, and the numerical solution handles the calculation difficulty near the singular position. The target angle calculation of each joint considers the physical limitations of the mechanical structure, and the joint angle limit check prevents overtravel damage. The generation of the speed control signal is based on the pseudo-inverse calculation of the Jacobian matrix, which maps the operation space speed to the joint space speed. The sending period of the control signal is synchronized with the sampling period of the robot controller, which is usually set to the order of milliseconds. The control algorithm of the joint driver adopts a double-closed-loop structure of position loop and speed loop, the position loop adjusts the joint angle to the target value, and the speed loop stabilizes the joint motion process. The motor drive signal is amplified through a power amplifier, which uses pulse width modulation technology to control the motor current. The feedback data in the actual motion process is collected in real time through the encoder, and the encoder data is compared with the target value to generate an error signal. The guarantee of trajectory tracking accuracy is realized through a feedforward compensation algorithm, which predicts the system dynamic characteristics to correct the control quantity in advance. The vibration suppression in the motion process adopts input shaping technology, which filters out the frequency components that excite mechanical resonance.

[0090] The forward propagation computation of the intelligent optimization algorithm model includes a normalization step that scales the input parameters to the numerical range used during model training. The dropout regularization technique for hidden layers is turned off during the inference phase, which prevents the model from overfitting. The post-processing of the path point list applies curvature continuity constraints that ensure smooth motion of the robot without abrupt changes. The optimization of the velocity profile takes into account energy consumption metrics that integrate the square of the acceleration term to evaluate motion efficiency. The time reparameterization of the motion trajectory sequence uses a monotonic function that preserves the order of the path points while optimizing the time allocation. The singular region handling of the inverse kinematics computation employs a damped least squares method that adds a damping factor to ensure numerical stability when the Jacobian matrix is singular. The generation of joint angle trajectories takes into account the motion coordination of each axis, which is achieved through a time synchronization mechanism. The amplitude limiting of the velocity control signals prevents exceeding the driver's capabilities, with the limiting value being dynamically adjusted based on the motor torque characteristics. The transmission of control commands uses a deterministic Ethernet protocol that guarantees the determinism of communication delays. The self-tuning function of the driver parameters adapts to changes in load, with an online algorithm that identifies the mechanical system inertia.

[0091] The optimization process of the path point list includes obstacle avoidance constraints that are implemented using the artificial potential field method. The generation of the velocity profile takes into account jerk limits that reduce mechanical system impact. The verification of the motion trajectory sequence is performed in a virtual controller environment that simulates the actual robot motion process. The verification of control commands includes a logical check step that detects the continuity of joint angle changes. The data recording function during actual motion saves running parameters, which are used for subsequent analysis and optimization. The output post-processing of the intelligent optimization algorithm model includes a trajectory smoothing algorithm that applies convolution filtering to remove high-frequency noise. The time optimization of the path point list uses a dynamic programming method that finds the optimal solution for time allocation. The satisfaction of constraints for the velocity profile is solved using quadratic programming, which converts the constraint conditions into optimization objectives. The compressed storage of the motion trajectory sequence uses differential encoding, which reduces storage space requirements. The generation efficiency of control commands is improved through code optimization, which uses compilation techniques to increase execution speed.

[0092] The real-time performance of inverse kinematics calculation is guaranteed by using a lookup table method, which pre-calculates joint angles for common positions. The interpolation of joint space trajectories uses a quintic polynomial, which guarantees the continuity of acceleration. The generation of velocity control signals takes into account torque limits, which are determined based on a motor heating model. The reliability of driver communication is guaranteed by a redundancy check, which detects transmission errors. The system safety monitoring module detects abnormal motion states and triggers an emergency stop signal. The optimization process of the motion trajectory sequence involves a multi-objective trade-off that considers both the shortest time and the minimum energy. The quality assessment of the path point list uses multiple indicators, including path length and curvature extrema. The optimization algorithm for the velocity curve uses a gradient descent method, which iteratively improves the velocity distribution. The verification of control instructions is performed on a hardware-in-the-loop simulation platform, which connects to the actual controller. The analysis of motion performance includes vibration spectrum analysis, which identifies resonance frequency points.

[0093] The inference process of the intelligent optimization algorithm model uses batch processing technology, which optimizes multiple trajectory segments simultaneously. The generation of the path point list takes into account the workspace boundary, which is constrained by a penalty function method. The smoothing of the velocity curve uses a phase plane method, which optimizes the relationship between velocity and acceleration. The visualization of the motion trajectory sequence helps with debugging and uses a three-dimensional graphical interface. The issuance of control instructions uses a double buffering mechanism, which avoids instruction update conflicts. The real-time performance of the system is guaranteed by an interrupt service program, which responds to timer interrupts. The accuracy of inverse kinematics calculation is improved by an iterative method, which gradually approaches the exact solution. The generation of joint trajectories includes transition segment design, which smoothly connects different trajectory segments. The filtering of velocity control signals eliminates high-frequency noise, which is achieved using a digital filter. The online adjustment of driver parameters adapts to mechanical wear and is based on model reference adaptive control.

[0094] Example 4: The actual trajectory of the industrial robot is monitored by a multi-sensor data fusion system, which integrates joint encoders, laser trackers, and vision measurement units. Joint encoders are installed at the back end of the servo motors of the robot's six joints. The joint encoders use multi-turn absolute value encoders with a resolution of 24-bit accuracy. The laser tracker is placed in the corner of the robot's work cell. The laser tracker measures the spatial position of the tool center point through the interferometer principle. The vision measurement unit consists of a high-speed camera and marker points, which are attached to the surface of the robot's end effector. The sensor data acquisition frequency is set to 1000 Hz, and the acquisition card uses a synchronization signal to ensure time alignment. Actual joint angle data is read from the encoder digital interface, and tool center point coordinates are calculated through a coordinate transformation chain. The comparison of the actual trajectory and the optimized trajectory sequence is performed in the trajectory tracking error calculation module, which calculates the position error and the direction error in real time. The position error is defined as the Euclidean distance between the actual tool center point coordinates and the target coordinates, and the direction error is calculated through the angle of the quaternion. The allowed range of trajectory tracking error is set according to the process requirements, with a precision assembly task allowing an error range of 0.1 mm and a welding task allowing an error range of 1 mm. The error comparison algorithm runs on a real-time operating system, with a calculation period synchronized with the control period. Error data is stored in a ring buffer, which can accommodate up to 10 seconds of motion data. The error overrun detection uses a sliding window statistics with a window size of 100 sampling points.

[0095] The adjustment of the velocity control signal is based on a proportional-integral-derivative control algorithm, and the parameters of the proportional-integral-derivative control algorithm are tuned according to the robot dynamics model. The proportional coefficient handles the current error, the integral coefficient accumulates the historical error, and the derivative coefficient predicts the error trend. The velocity correction amount calculation uses an incremental formula to avoid integral saturation phenomenon. The correction amount limiter prevents excessive adjustment, and the limit threshold is set according to the motor torque capacity. The adjusted velocity control signal is sent through the field bus, which uses the EtherCAT protocol to ensure real-time performance. The signal update period is consistent with the controller period, with a typical value of 1 ms. The monitoring data recording system saves the complete motion process, and the data recording system includes timestamps, actual coordinates, target coordinates, and error values. The recording file is stored in binary format to reduce storage space occupation. The data playback function supports motion process reproduction, and the playback function is used for offline analysis and debugging. The system status monitoring interface displays the real-time error curve, and the monitoring interface color coding distinguishes between normal and overrun states. The alarm function is triggered when the overrun is continuous, and the alarm function prompts the operator through audio and visual signals.

[0096] The trajectory tracking error calculation considers coordinate system conversion error, which is compensated by hand-eye calibration. The timestamp alignment of sensor data uses a hardware trigger signal, which is distributed by the main controller. The error statistics report generates periodic reports, including maximum error, average error, and standard deviation. The error compensation algorithm adapts to different motion stages, increasing the differential coefficient in high-speed stages and enhancing the integral effect in low-speed stages, as shown in Table 1.

[0097] Table 1: Trajectory tracking error classification and processing strategy

[0098]

[0099] The velocity control signal adjustment process includes a feedforward compensation link that calculates the ideal velocity based on the kinematic model. The friction compensation module offsets Coulomb friction and viscous friction, and the friction parameters are obtained through experimental identification. The inertia compensation considers load changes, and the inertia compensation parameters are adjusted according to the mass properties. The resonance suppression filter eliminates mechanical vibrations, and the resonance suppression filter is designed using a notch filter. The temperature compensation corrects thermal deformation errors, and the temperature sensor monitors the motor temperature rise. The backlash compensation eliminates transmission backlash, and the backlash compensation value is calibrated by laser measurement. The actual motion trajectory monitoring system includes data validity check, which identifies sensor failures. The abnormal data filtering uses a median filter algorithm, and the median filter window size is configurable. The data fusion algorithm uses Kalman filtering to estimate the optimal trajectory state. The coordinate system is unified to the robot base coordinate system, and the coordinate system transformation is realized through a homogeneous matrix. The time synchronization accuracy reaches the microsecond level, and the PTP precise time protocol is used for time synchronization.

[0100] The trajectory tracking error analysis distinguishes between systematic errors and random errors, and the systematic errors are eliminated through parameter calibration, while the random errors are suppressed through filtering algorithms. The error trend prediction uses time series analysis to identify periodic errors. The error compensation amount calculation considers the robot configuration, and the error characteristics differ in different poses. The error mapping model establishes the relationship between error and joint angle, and the error mapping model is trained through a neural network. The error compensation table stores compensation values for common positions, and the error compensation table is established through laser tracker measurement. The velocity control signal adjustment algorithm includes an adaptive mechanism that adjusts parameters based on error changes. The fuzzy logic controller handles nonlinear characteristics, and the rule base of the fuzzy logic controller contains expert experience. The neural network controller learns the optimal adjustment strategy, and the neural network controller adjusts the control parameters online. The genetic algorithm optimizes the controller parameters, and the genetic algorithm searches for the optimal combination of parameters. The reinforcement learning algorithm improves the control strategy autonomously, and the reinforcement learning algorithm is guided by a reward function.

[0101] The real-time performance of the monitoring system is verified by benchmark tests that measure the delay from data acquisition to control output. System resource monitoring tracks CPU usage, which is an indicator of real-time performance. Data integrity checks verify the transmission process using cyclic redundancy checks. System logs record the operating status, which are used for fault diagnosis. Performance counters count the error handling time, which evaluates the efficiency of the algorithm. Actual motion trajectory data is used for online calibration, which updates the robot kinematic parameters. The tool center point calibration program is executed periodically using the three-point method. Workpiece coordinate system calibration compensates for positioning errors, which is achieved by feature matching. Kinematic parameter identification uses the least squares method to improve model accuracy. Dynamic parameter identification is performed using excitation trajectories to improve control performance. The trajectory tracking error handling process contains multiple safety levels, the first level performs error detection, the second level implements adjustment strategies, and the third level triggers safety protection. Error thresholds are set hierarchically to avoid false alarms. The adjustment strategy is gradually strengthened to ensure stability. The safety protection mechanism takes effect immediately to prevent equipment damage. The system state machine manages the operating modes, which include normal, adjustment, and protection modes.

[0102] In example 5, after the industrial robot completes the automobile door welding task, the actual motion trajectory data recording system starts running, and the actual motion trajectory data recording system reads the joint angle sampling values from the controller memory buffer. The encoder data of each joint is recorded at a frequency of 1000 Hz, and the encoder data includes motor rotation angle and torque current values. The tool center point coordinates are calculated by the hand-eye vision system, which captures the reflective marker points on the welding torch tip. The performance index calculation module runs synchronously, which calculates the total time, average speed, and root mean square value of acceleration of the welding path. The data timestamp is synchronized with the factory-level network time server, which uses GPS signals to ensure global clock consistency. The recording file uses a segmented storage format, which is naturally segmented according to the welding seam. The actual motion trajectory data is added to the trajectory database, which includes a data cleaning step to remove abnormal points caused by sensor jitter. The table structure of the trajectory database includes a task number field, a timestamp field, a joint angle array field, and a tool center point coordinate field. New data records are inserted through database transaction operations, which ensure data integrity constraints. Index update operations optimize query performance, which are targeted at the task type and time range fields. The data archiving strategy keeps recent data in high-speed storage and migrates historical data to nearline storage devices. Database consistency checks verify foreign key relationships, which ensure that the task number matches the main table.

[0103] The training set update of the intelligent optimization algorithm model adopts an incremental learning mode, which retains the original model parameters and fuses new data. The training set is constructed by extracting the last 100 task data from the trajectory database, which includes successful welding cases and manually corrected trajectories. The feature extraction process recalculates the curvature features of the path point sequence, which is calculated by the three-point method. The speed distribution feature is standardized, and the speed distribution feature is normalized to the maximum rated speed of the robot. The training sample label is generated by the expert system score, which is based on the welding quality detection result. The data balancing algorithm handles class imbalance, and the data balancing algorithm resamples the minority class samples. The learning rate of the retrained intelligent optimization algorithm model adopts an annealing strategy, with an initial value of 0.001 and a 10% decay every 1000 iterations. The training loss function combines trajectory smoothness and time optimality, with the trajectory smoothness term calculating the integral of acceleration change and the time optimality term evaluating the total movement time. The gradient clipping threshold is dynamically adjusted, and the gradient clipping threshold is adaptively changed according to the loss function value. The validation set is divided using time sequence segmentation, which ensures that the model verifies future data. The early stopping method monitors the validation set loss curve, and the early stopping method terminates the training when there is no improvement for 200 consecutive iterations. Model checkpoints are saved regularly, and model checkpoints include weight fast and optimizer state. The updated model parameters are deployed using a blue-green deployment strategy, which keeps the old version of the model as a fallback solution. Parameter encryption transmission ensures security, and parameter encryption uses the AES-256 algorithm to encrypt the model file. The version control system records the parameter change history, and the version control system marks the data source and performance indicators of each model version. The model inference service hot reloads new parameters, and the hot reload process does not interrupt the execution of the welding task. The health check interface verifies the availability of the model, and the health check interface sends test data to verify the input and output formats.

[0104] The actual motion trajectory data record includes environmental parameters, including workshop temperature and humidity and workpiece positioning error. The welding quality detection result is associated with the trajectory data, and the welding quality detection evaluates the weld quality grade through a visual detection system. A data time alignment algorithm compensates for transmission delay, and the time alignment algorithm uses an interpolation method to synchronize data sources with different sampling rates. An abnormal situation marker records welding spatter events, and the abnormal situation marker helps to filter high-quality training data. A data compression algorithm reduces storage space, and the data compression algorithm uses lossy compression to retain key features. A query interface of the trajectory database supports complex condition filtering, and the query interface allows querying by workpiece model and welding parameter combination. A data backup strategy implements the 3-2-1 principle, and the data backup strategy includes local backup, off-site backup, and cloud backup. Data desensitization processing protects production process secrets, and the data desensitization processing hides key dimension parameters. Access permission control limits the scope of data use, and the access permission control assigns query permissions based on roles. Data lifecycle management automatically cleans up expired data, and the data lifecycle management deletes historical data according to the preservation strategy.

[0105] The training process of the intelligent optimization algorithm model uses a distributed computing framework, which trains multiple model variants in parallel on a GPU cluster. The hyperparameter optimization adopts a Bayesian optimization method, which explores the optimal combination of learning rate and batch size. The training data enhancement technique applies random perturbations, and the training data enhancement expands sample diversity by adding Gaussian noise. Model structure optimization tries different network depths, and model structure optimization compares the effects of three-layer and five-layer neural networks. Regularization techniques prevent overfitting, and regularization techniques include L2 weight decay and dropout. Training curve visualization monitors the convergence process, and training curve visualization displays changes in loss function and evaluation metrics. Verification testing after model parameter deployment includes regression testing, which uses historical data to verify model output consistency. Stress testing evaluates model inference speed, and stress testing simulates high-concurrency request scenarios. Boundary testing checks the processing of extreme inputs, and boundary testing inputs abnormal trajectory parameters to verify robustness. Integration testing verifies cooperation with the controller, and integration testing runs a complete welding process in a simulation environment. Performance benchmarking compares version differences, and performance benchmarking records inference delay and memory usage indicators. Data mining of actual motion trajectory data discovers potential optimization patterns, and data mining algorithms cluster similar motion trajectories. Association rule analysis finds parameter combination rules, and association rule analysis discovers the correlation between speed parameters and weld quality. Trend analysis predicts equipment performance degradation, and trend analysis establishes a model of precision change over time. Anomaly detection identifies abnormal motion patterns, and anomaly detection marks trajectory segments that deviate from the norm. Statistical analysis generates quality reports, and statistical analysis calculates statistical quantities of each dimension indicator.

[0106] The update cycle of the intelligent optimization algorithm model is determined according to performance monitoring, which tracks the performance of the model on the production line. Model degradation detection monitors prediction bias, and model degradation detection statistics the difference between the actual trajectory and the predicted trajectory. The trigger condition sets the performance threshold, and the trigger condition starts the update when the evaluation index decreases by a certain percentage. The rollback mechanism ensures system reliability, and the rollback mechanism restores the old version when the model update fails. Version compatibility handles data format changes, and version compatibility keeps the interface consistent between new and old models. Update notification informs related systems, and update notification broadcasts version changes through a message queue. Long-term accumulation of actual motion trajectory data forms a knowledge base, which contains optimization trajectory patterns for different workpiece models. Transfer learning technology applies cross-task optimization, and transfer learning technology migrates welding knowledge to cutting tasks. Meta-learning algorithm extracts the commonality of optimization strategies, and meta-learning algorithm learns trajectory optimization methods from multiple tasks. Knowledge graph constructs process parameter relationships, and knowledge graph represents the constraint relationship between parameters. Intelligent retrieval supports similar case queries, and intelligent retrieval recommends historical trajectories according to new task characteristics. Continuous learning system continuously improves performance, and continuous learning system forms a closed loop from data to optimization.

[0107] It should be noted that the relational terms, such as first and second, and the like, are used solely to distinguish one from another entity or action, without necessarily requiring or implying any actual relationship or order between or among such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0108] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A control method for trajectory optimization of an industrial robot based on intelligent algorithms, characterized in that, The method comprises the following steps: Collecting motion trajectory data generated by an industrial robot during execution of multiple work tasks, the motion trajectory data comprising joint position information, tool center point coordinates and motion time series, and storing the motion trajectory data in a trajectory database; Cleaning and screening the motion trajectory data in the trajectory database, extracting a feature set of the motion trajectory data, the feature set comprising a path point sequence, a speed distribution and an acceleration profile; Building an intelligent optimization algorithm model, the intelligent optimization algorithm model adopting a neural network structure, training the intelligent optimization algorithm model using the feature set, and enabling the intelligent optimization algorithm model to learn a trajectory optimization strategy; Obtaining task description information of a current industrial robot, the task description information comprising target position coordinates and motion constraint conditions, and analyzing the task description information to obtain initial trajectory parameters; Inputting the initial trajectory parameters into the trained intelligent optimization algorithm model, the intelligent optimization algorithm model outputting an optimized motion trajectory sequence, the optimized motion trajectory sequence comprising a path point list and a speed curve; Generating control instructions according to the optimized motion trajectory sequence, and driving joint motors of the industrial robot through a controller of the industrial robot to enable the industrial robot to move according to the optimized motion trajectory sequence; The step of collecting motion trajectory data generated by an industrial robot during execution of multiple work tasks comprises: Collecting joint angle data through an encoder sensor installed at a joint of the industrial robot; Capturing spatial coordinate data of a tool center point through a vision system; Recording joint angle data and tool center point coordinate data corresponding to each motion timestamp; Associating the joint angle data, the tool center point coordinate data and the motion timestamp data with a task identifier, the task identifier comprising a task type and a workpiece number; Storing the associated motion trajectory data in a data table of the trajectory database; The step of building an intelligent optimization algorithm model comprises: Initializing a layer structure and parameters of a neural network model, the neural network model comprising an input layer, a hidden layer and an output layer; Taking the path point sequence, the speed distribution and the acceleration profile in the feature set as training input data; Setting an optimization objective function, the optimization objective function comprising a shortest path and a highest smoothness; Adjusting parameters of the neural network model using a gradient descent algorithm to minimize the optimization objective function; Verifying performance of the trained intelligent optimization algorithm model on a test data set, and saving model parameters; The step of obtaining task description information of a current industrial robot comprises: Receiving task instructions from user input or an upper system; Analyzing target position coordinates in the task instructions, the target position coordinates comprising three-dimensional spatial coordinates; Extracting motion constraint conditions in the task instructions, the motion constraint conditions comprising maximum speed limits, acceleration limits and obstacle avoidance regions; Combining the target position coordinates and the motion constraint conditions into initial trajectory parameters.

2. The control method of trajectory optimization of an industrial robot based on intelligent algorithm as claimed in claim 1, wherein, The step of cleaning and screening motion trajectory data in a trajectory database comprises: Reading motion trajectory data from a data table of the trajectory database; Checking completeness and consistency of the motion trajectory data, and removing records containing missing values or abnormal values; Screening the motion trajectory data according to a preset precision threshold, and retaining records with path errors less than the threshold. Extracting path point sequences from the filtered records, which are composed of tool center point coordinates in chronological order; Calculating velocity and acceleration values between path point sequences to form velocity distribution and acceleration profile.

3. The control method of trajectory optimization of an industrial robot based on intelligent algorithm as claimed in claim 1, wherein, The generation step of the optimized motion trajectory sequence includes: Input initial trajectory parameters into the trained intelligent optimization algorithm model; The intelligent optimization algorithm model calculates the path point list through forward propagation; Interpolate the path point list to generate a continuous velocity curve; Adjust the velocity curve to meet the motion constraints; Output the optimized motion trajectory sequence, including the path point list and the velocity curve.

4. The control method of trajectory optimization of an industrial robot based on intelligent algorithm as claimed in claim 3, wherein, The step of generating control instructions includes: Convert the optimized motion trajectory sequence to joint space trajectory, calculate the target angle of each joint through inverse kinematics; Generate velocity control signals for each joint according to the velocity curve; Send the target angle and velocity control signals to the joint drive of the industrial robot; The joint drive controls the motor to rotate, realizing the motion of the industrial robot.

5. The control method of trajectory optimization of an industrial robot based on intelligent algorithm as claimed in claim 4, wherein, Further comprising the following steps: Real-time monitoring of the actual motion trajectory of the industrial robot, obtaining the actual joint angle and tool center point coordinates through sensors; Compare the actual motion trajectory with the optimized motion trajectory sequence, calculate the trajectory tracking error; If the trajectory tracking error exceeds the allowed range, adjust the velocity control signal in the control instruction; Repeat the monitoring and adjustment process until the task is completed.

6. The control method of trajectory optimization of an industrial robot based on intelligent algorithm as claimed in claim 5, wherein, The step of adjusting the velocity control signal in the control instruction includes: According to the size and direction of the trajectory tracking error, calculate the velocity correction amount; Add the velocity correction amount to the original velocity control signal; Limit the adjusted velocity control signal not to exceed the maximum speed limit in the motion constraints; Send the adjusted velocity control signal to the joint drive.

7. The control method of trajectory optimization of an industrial robot based on intelligent algorithm as claimed in claim 6, wherein, Further comprising the following steps: After the task is completed, record the actual motion trajectory data and its performance indicators; Add the actual motion trajectory data to the trajectory database; Use the new data to update the training set of the intelligent optimization algorithm model, retrain the model; Save the updated model parameters for subsequent tasks.

Images