Early safety warning method for a robot arm

Abstract

The application discloses an early safety warning method of a mechanical arm and belongs to the technical field of robot motion monitoring. The application learns and analyzes motion characteristics of the mechanical arm in a long-term operation process as main observation parameters through an AI module, compares operation results obtained through analysis with subsequent operation results of the mechanical arm, judges whether the motion characteristics of the mechanical arm in the subsequent operation process deviate from the normal operation results recorded in the AI module, judges whether the mechanical arm itself or operation conditions have subtle changes that are difficult to detect, such as wear of some components, performance decline and the like, makes a warning for these changes, so as to realize predictive maintenance, early identification of changes in the motion characteristics of the mechanical arm, and the technical effect of early warning is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of robot motion monitoring, and particularly to an early safety warning method for a robotic arm. Background Art

[0002] A multi-degree-of-freedom robotic arm is an electromechanical (control) integrated device, and its motion performance is affected by many factors. There are factors from its own body, such as the servo joint part, which integrates components such as a reducer, a brake, a motor, an encoder, and a servo driver (some are not integrated). The quality, wear over time, aging, lifespan, and even damage of these components will affect the deterioration of the robotic arm's motion performance; there are also influences from the load side during the operation of the robotic arm.

[0003] Therefore, when the robotic arm operates in the industrial field for a long time, its motion characteristics will inevitably change over time. Sometimes this change is caused by the aging of the robotic arm components, changes in external operating objects and conditions, and sometimes it is the early symptom of component damage. To enable the robotic arm to continuously work normally and with high performance, it is necessary to identify and warn of the change phenomenon of the robotic arm's motion characteristics as early as possible, so as to carry out maintenance in advance or troubleshoot faults, thereby improving the working reliability and performance sustainability of the robotic arm. Especially for robotic arms with fixed bases in the industrial field, their working types and load conditions are generally fixed, regular or periodically repeated. Once the motion characteristics of the robotic arm change, it must indicate that there is a problem in a certain link or the working condition has changed. Although it is a small problem, if it can be discovered early, it may avoid the occurrence of greater problems.

[0004] In the prior art, most robots have a fault protection function for the servo joint drive status, including that the robot controller also monitors the robot body during operation through relevant sensors or other data and gives early warnings in a timely manner.

[0005] The patent with the application number 202210803690.9 discloses a warning method, which generates a real-time operation dynamic factor of the robotic arm and a periodic accuracy inspection factor of the robotic arm through the processing of its information. By comparing and analyzing the real-time operation dynamic signal and the periodic accuracy inspection factor of the robotic arm, the functions of real-time maintenance reminder and periodic maintenance reminder of the robotic arm are realized as a whole, thereby assisting the staff to maintain the robotic arm.

[0006] The patent with the application number 202010186762.0 discloses a warning method, which uses the preprocessed robotic arm operation data and the constructed mechanism characteristics to monitor the operation state of the robotic arm in the closed-loop operation process of the robotic arm digital twin integrated simulation model in real time.

[0007] However, both of these technologies can only identify problems or malfunctions after they have become quite obvious, which is a drawback as they cannot achieve a real early warning effect.

[0008] Therefore, there is an urgent need in the market for an early warning system that can provide warnings before a malfunction or problem becomes obvious, so as to enable early maintenance or upkeep of the robotic arm. Summary of the Invention

[0009] This invention discloses an early safety warning method for a robotic arm, comprising using an AI module, wherein the AI ​​module is embedded within the robot controller and controlled by the robot controller, and the method includes the following steps:

[0010] S1, the AI ​​module learns the motion characteristics of the robotic arm during its operation;

[0011] S2, the AI ​​module defines the learning results as the motion characteristic results of the robotic arm's standard operation.

[0012] S3, the AI ​​module performs secondary learning on the motion characteristics of the robotic arm during subsequent operation;

[0013] S4, the AI ​​module defines the secondary learning result as the motion characteristic result of the robotic arm to be warned during operation;

[0014] S5 compares the motion characteristic results of the robotic arm during standard operation with the motion characteristic results of the robotic arm during operation requiring warning.

[0015] If the deviation of the comparison result exceeds the set value, the AI ​​module will issue a warning signal;

[0016] If the deviation of the comparison result does not exceed the set value, the AI ​​module determines that the robotic arm is in a safe operating state.

[0017] Preferably, the motion characteristic is the jerk of the joint servo drive system of the robotic arm.

[0018] Preferably, step S1 includes:

[0019] S1A1, the AI ​​module receives speed commands from the robot controller for the joint servo motors of the robotic arm and feedback on the actual operating speed values ​​of the joint servo motors;

[0020] S1A2, calculates the jerk of the speed command and the jerk of the speed value;

[0021] S1A3, calculates the absolute value of the error between the jerk of the speed command and the jerk of the speed value;

[0022] S1A4, standardizes the data in steps S1A2 and S1A3;

[0023] S1A5 imports the data processed in step S1A4 into the AI ​​module for learning and training.

[0024] Preferably, step S2 further includes refining the learning results by using the least squares method to reduce the data layer by layer in stages.

[0025] Preferably, step S1 includes:

[0026] S1B1, the AI ​​module receives speed commands from the robot controller for the joint servo motors of the robotic arm and feedback on the actual operating speed values ​​of the joint servo motors.

[0027] S1B2, calculates the jerk of the speed command and the jerk of the speed value;

[0028] S1B3 performs frequency domain conversion on the jerk of the speed command and the jerk of the speed value in the frequency domain;

[0029] S1B4 performs data normalization on the converted data;

[0030] S1B5 imports the data processed in step S1A4 into the AI ​​module for learning and training.

[0031] Preferably, step S4 further includes fitting the motion characteristic curve to the secondary learning result using the least squares method.

[0032] Preferably, step S1B3 first calculates the difference between the jerk of the speed command and the jerk of the speed value, and then performs frequency domain conversion on the difference.

[0033] Preferably, the method is applied to safety warning of the robotic arm of a single-arm or multi-arm robot.

[0034] This invention solves the technical problem that in the prior art, early warning for robotic arms only exists when a fault or problem occurs, and cannot actually achieve the early warning effect. The technical solution of this invention uses an AI module to learn and train the motion characteristics of the robotic arm to form a learning result, and then learns and trains the subsequent motion characteristics of the robotic arm to form a secondary learning result. By comparing the two learning results, it is possible to identify changes in the motion characteristics of the robotic arm at an early stage, thereby achieving the technical effect of early warning. Attached Figure Description

[0035] The accompanying drawings used in the following description of the embodiments or prior art will be briefly introduced. Obviously, the drawings described below are only one embodiment of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 This diagram illustrates the data interaction between the AI ​​module and the robot controller.

[0037] Figure 2 This is a diagram of the internal structure of the AI ​​module;

[0038] Figure 3 This is a block diagram illustrating the principle structure of Example 1;

[0039] Figure 4 This is the control flowchart for Example 1;

[0040] Figure 5 This is a comparison chart of the two learning results in Example 1;

[0041] Figure 6 This is a block diagram illustrating the principle structure of Example 2;

[0042] Figure 7 This is a comparison chart of the two learning results in Example 2;

[0043] Figure 8 This is a schematic diagram of the principle structure of Example 3. Detailed Implementation

[0044] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments and accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] Example 1

[0046] In a specific embodiment 1, such as Figure 1 , Figure 2 , Figure 3 , Figure 4 and Figure 5 As shown, an early safety warning method for a robotic arm includes an AI module, which is embedded in the robot controller and controlled by the robot controller.

[0047] In this embodiment 1, the AI ​​module receives servo joint motion commands and joint servo speed data from the robot controller. The AI ​​module outputs the prediction results to an external HMI device and then to the robot controller. The AI ​​module uses an embedded processor with an embedded artificial neural network core. This embedded processor can be an ARM processor, an ARM-based processor, or an FPGA chip, etc. It runs an artificial neural network algorithm internally to learn and train the motion characteristic data of the robotic arm.

[0048] The subject of this study is a robot with a 6-DOF robotic arm.

[0049] The most significant and likely factor influencing the motion characteristics of a robotic arm comes from the joint servo drive system. Therefore, this paper analyzes the motion characteristics of a specific joint servo drive system by starting with the motion equations of the servo drive system.

[0050]

[0051] In the formula,

[0052] Te is the electromagnetic torque. TL is the load torque, J is the system inertia, B is the motor viscous friction coefficient, and ω is the motor speed.

[0053] Taking the differential of both sides of the above equation,

[0054]

[0055] further,

[0056]

[0057] Let the right-hand side of the above equation be a function f(J, T). e T L ,ω,B),

[0058] The above formula can then be written as:

[0059]

[0060] In the above formula, the left side is the second derivative with respect to the velocity, which is called the jerk ε. It can be considered that the jerk ε is a function of five parameters: system inertia J, electromagnetic torque Te, load torque TL, motor speed ω, and motor viscous friction coefficient B.

[0061] Further calculations,

[0062]

[0063] Then there is

[0064] ε=f(J,T e T L , ω, B)

[0065] From a physical perspective, the jerk ε of a servo drive system characterizes the flexible impact performance of the actuator and reflects the flexibility of the motion system's speed. As can be seen from the above formula, any change in any parameter of the system, such as the system inertia J, electromagnetic torque Te, load torque TL, motor speed ω, or motor viscous friction coefficient B, will inevitably cause a change in the system's jerk. Therefore, the jerk parameter is chosen to characterize the motion characteristics of the robotic arm.

[0066] This embodiment does not consider the short-term dynamic characteristics of the servo drive system. During the movement of the robot arm, parameters such as load size, electromagnetic torque, and load inertia change dynamically. The changes in jerk caused by these instantaneous dynamic changes are all considered short-term dynamic characteristics. This embodiment studies the changes in parameters represented by jerk under long-term operation of the robot arm, with the jerk parameter of the servo drive system as the main parameter. Because the robot performs periodic repetitive actions, the long-term study collects and analyzes data within the repetition period of these actions.

[0067] This embodiment designs an embedded AI module to learn the above data over a long period of time, obtaining the learning results under normal operation. Then, in subsequent long-term operation, the obtained results are compared with the learning results under normal operation. If there is a certain deviation, it indicates that the robotic arm itself or the operating conditions have undergone a qualitative change, thereby issuing an early warning for predictive maintenance.

[0068] This embodiment includes the following steps:

[0069] After receiving the speed commands and actual speed values ​​from the servo drive systems of each joint from the robot controller, the embedded AI module calculates the jerk of the speed command and the jerk of the feedback speed value. Then, it calculates the absolute value of the error between the jerk of the speed command and the jerk of the feedback speed value for each joint. This absolute value is then combined with 18 parameter values ​​(18 input nodes in total), including the jerk of the speed command and the jerk of the speed value. After data normalization, the AI ​​module learns and trains using its artificial neural network algorithm.

[0070] This embodiment uses the jerk and its error of the servo drive system as the main input data for long-term learning and training. During formal production operations on the robot production line, the jerk parameters during normal operation are modeled and learned continuously for a certain period of time to obtain the standard operating motion characteristics of the robotic arm under normal robot operation. Then, under the control of the robot controller, the training results are saved. The AI ​​learning module then continues learning under the control of the robot controller. After a long period of learning, the results are compared with the previous standard operating motion characteristics. If they are basically consistent, it means that the robotic arm system's working condition remains unchanged and requires no maintenance. If there is a significant deviation, a warning message is issued, requiring maintenance or investigation of the cause.

[0071] The AI ​​learning time and the duration of each learning phase can be set, but it is generally at least in days. For example, if the data obtained from 100 days of normal learning is considered the normal result, then the learning results obtained each day thereafter can be compared and analyzed with this normal result; however, to increase the reliability of the results, in Figure 4 Based on the control process shown, continuous comparisons can be made over several consecutive days. If similar situations with similar deviations persist, it indicates that there must be an anomaly lurking in the system, requiring an early warning.

[0072] To reduce the large amount of storage space occupied by AI learning results data, it is necessary to process the learning results data in stages to reduce the data volume. In this technical solution, the least squares method is used to refine the data layer by layer in stages.

[0073] Figure 5 This is the data result of AI learning on the motion characteristics of a robotic arm in normal operation. The initial data was in hours (0 position on the Z-axis in the figure). To refine the data, the least squares method was used to fit a trend curve of data change. Data on the curve and within a certain range of error were retained, while other data were discarded, resulting in data in days. Figure 5 The data point at position "1" on the Z-axis is shown in the figure. Further, as the collection time increases, many more data points are added to the refined data set, as shown by the data point at position "2" on the Z-axis. Based on this data set, the least squares method is used to fit the data curve. This data curve should basically coincide with the previous data curve. Similarly, data outside the error range of this data curve are removed to form the refined data, ultimately forming a data set in 10-day units.

[0074] By adopting the above data reduction measures, or similar layer-by-layer reduction measures, the data can be kept within a reasonable range of system computing and storage resources over time, without causing the final data to become distorted.

[0075] Example 2

[0076] In a specific embodiment 2, such as Figure 1 , Figure 2 , Figure 6 and Figure 7 As shown, an early safety warning method for a robotic arm includes using an AI module, which is embedded inside the robot controller and controlled by the robot controller.

[0077] The subject of this study is a robot with a 6-DOF robotic arm.

[0078] In this embodiment, the motion characteristic learned by the AI ​​module is the accelerometer of the servo drive system.

[0079] The principle of this embodiment is the same as that of Embodiment 1. Both involve training the AI ​​module to learn the motion characteristics of the robotic arm to form a learning result, and then training the subsequent motion characteristics of the robotic arm to form a secondary learning result. By comparing the two learning results, it is possible to identify changes in the motion characteristics of the robotic arm in the early stage, thereby achieving the technical effect of early warning.

[0080] In this embodiment, the jerk values ​​are calculated based on the speed values ​​of the servo drive systems of each joint. The jerk values ​​are then converted in the frequency domain and normalized before being input into the neural network model of the AI ​​module for learning and training. As the running time increases, the motion characteristic data of the servo drive systems of each joint are gradually obtained, and the motion characteristic results of standard operation are based on this. The AI ​​module continues to learn and train the motion characteristics of the robotic arm and obtains secondary learning results. The two are compared to determine whether the system motion characteristics have changed.

[0081] Figure 7 The results show the comparison of motion characteristics in the frequency domain.

[0082] Under normal operating conditions, through continuous AI learning and training, a set of data points in the normal state is obtained. Based on this, mathematical methods can be used to fit a trend curve of the data changes, which is then stored in the system memory as the result of the normal state. If, over time, the motion state of the servo system changes significantly, possibly due to qualitative physical changes in the robotic arm joints or changes in robot operating conditions, resulting in a set of data points deviating from the normal data range, and this deviation from the original normal trend increases, mathematical methods, such as the least squares method, can be used to fit a trend curve for this abnormal state. When the deviation between this trend curve and the trend curve in the normal state exceeds a certain set value, the system is considered to have a problem, requiring an early warning for preventative maintenance or cause investigation.

[0083] Example 3

[0084] In a preferred embodiment 3, such as Figure 1 , Figure 2 , Figure 6 , Figure 7 and Figure 8 As shown, an early safety warning method for a robotic arm includes using an AI module, which is embedded inside the robot controller and controlled by the robot controller.

[0085] The subject of this study is a robot with a 6-DOF robotic arm.

[0086] In this embodiment, the motion characteristic learned by the AI ​​module is the accelerometer of the servo drive system.

[0087] The principle of this embodiment is the same as that of Embodiments 1 and 2. In both embodiments, the AI ​​module learns and trains the motion characteristics of the robotic arm to form a learning result, and then learns and trains the subsequent motion characteristics of the robotic arm to form a secondary learning result. By comparing the two learning results, the changes in the motion characteristics of the robotic arm can be identified early, thereby achieving the technical effect of early warning.

[0088] This embodiment is based on embodiment 2. In order to reduce the data processing volume of the AI ​​module in embodiment 2, reduce the number of neuron input units of the AI ​​module, and improve the learning efficiency and speed of the AI ​​module's neurons, it is an optimization of embodiment 2.

[0089] In this embodiment, before training, the difference between the jerk of the speed command and the jerk of the speed value of each joint servo drive system is calculated. After frequency domain conversion and data normalization, six data input neurons are generated for training. The training results are then obtained. During normal robot operation, the learning results from several days or even months are extracted and stored as normal operation data. Subsequently, as time and operation continue, the learned data is compared with the normal operation data to determine whether the robot has undergone a qualitative change or its operating state has changed, enabling early warning and maintenance.

[0090] It should be noted that the technical solution of this invention is aimed at robot operation scenarios involving highly repetitive and periodic operations. In such consistent operational situations, it provides early warnings by comparing early and late AI learning data. If the robot's operation content, operating mode, or operating conditions change—for example, from picking and stacking workpieces to removing them from a stack—then the robot's normal operation learning data needs to be relearned as comparison data, and the previous comparison data becomes meaningless.

Claims

1. An early safety warning method for a robotic arm, comprising an AI module, wherein the AI ​​module is embedded within a robot controller and controlled by the robot controller; characterized in that, The method includes the following steps: S1, the AI ​​module learns the motion characteristics of the robotic arm during its operation; S2, the AI ​​module defines the learning results as the motion characteristic results of the robotic arm's standard operation. S3, the AI ​​module performs secondary learning on the motion characteristics of the robotic arm during subsequent operation; S4, the AI ​​module defines the secondary learning result as the motion characteristic result of the robotic arm to be warned during operation; S5 compares the motion characteristic results of the robotic arm during standard operation with the motion characteristic results of the robotic arm during operation requiring warning. If the deviation of the comparison result exceeds the set value, the AI ​​module will issue a warning signal; If the deviation of the comparison result does not exceed the set value, the AI ​​module determines that the robotic arm is in a safe operating state; The motion characteristic refers to the jerk of the joint servo drive system of the robotic arm.

2. The early safety warning method for a robotic arm according to claim 1, characterized in that, Step S1 includes: S1A.1, the AI ​​module receives speed commands and feedback values ​​of the actual running speed of the joint servo motors from the robot controller. S1A.2, calculates the jerk of the speed command and the jerk of the speed value; S1A.3, calculate the absolute value of the error between the jerk of the speed command and the jerk of the speed value; S1A.4, standardize the data from steps S1A.2 and S1A.3; S1A.5: Import the data processed in step S1A.4 into the AI ​​module for learning and training.

3. The early safety warning method for a robotic arm according to claim 2, characterized in that, The S2 step also includes refining the learning results by using the least squares method to reduce the data layer by layer in stages.

4. The early safety warning method for a robotic arm according to claim 1, characterized in that, Step S1 includes: S1B.1, the AI ​​module receives speed commands and feedback values ​​of the actual running speed of the joint servo motors from the robot controller; S1B.2, calculates the jerk of the speed command and the jerk of the speed value; S1B.3 performs frequency domain conversion on the jerk of the speed command and the jerk of the speed value in the frequency domain; S1B.4, Perform data normalization on the converted data; S1B.5: Import the data processed in step S1B.4 into the AI ​​module for learning and training.

5. The early safety warning method for a robotic arm according to claim 4, characterized in that, The S4 step also includes fitting the motion characteristic curve to the second learning result using the least squares method.

6. The early safety warning method for a robotic arm according to claim 5, characterized in that, Step S1B.3 first calculates the difference between the jerk of the speed command and the jerk of the speed value, and then performs frequency domain conversion on the obtained difference.

7. The early safety warning method for a robotic arm according to any one of claims 1-6, characterized in that, The method is applied to safety early warning of robotic arms in single-arm or multi-arm robots.

Images