Collision detection method for high-reduction-ratio heavy-load mechanical arm

Abstract

According to the collision detection method for the high-reduction-ratio and large-load mechanical arm, the problems of sensitivity and reliability of collision detection of the high-reduction-ratio mechanical arm are solved by combining a two-stage detection mechanism with tail end vibration characteristics and joint torque errors; according to the first-stage detection, tail end vibration signals are captured in real time through a three-axis acceleration sensor, and rapid recognition of slight collision is achieved; in the second-stage detection, joint torque errors are calculated based on current feedback, the nonlinear effect of the speed reducer is compensated in combination with a kinetic model, and the accuracy of violent collision detection is improved. High-precision collision identification and multi-level safety control of the mechanical arm and an external object under the heavy load working condition are achieved, and meanwhile the anti-interference capacity and reliability of a detection system under the complex working condition are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of robot control, and particularly to a collision detection method for a high reduction ratio and large load manipulator, which is used to realize the collision recognition and safety control of the manipulator and external objects under heavy load conditions. Background Art

[0002] At present, the manipulator collision detection technology is mainly realized based on motor current feedback or joint torque sensors combined with dynamic models. The traditional method converts the motor current signal into joint torque through the inverse dynamics model and compares it with the theoretical value to judge the collision event. For example, the method based on the external torque observer estimates the joint external torque through momentum deviation and compensates for friction to improve the detection accuracy.

[0003] However, such methods have significant defects in the scenario of high reduction ratio (such as harmonic reducer with large reduction ratio): the non-linear characteristics of the reducer (such as friction, elastic deformation) lead to a serious misalignment of the mapping relationship between the motor terminal current and the actual joint torque; at the same time, mechanical inertia and signal processing delay cause obvious hysteresis of the current feedback after the collision occurs and cannot respond in time.

[0004] To solve the above problems, some studies have tried to introduce acceleration sensors or vision systems. For example, the method based on vibration characteristics extracts the modal frequency of the manipulator and uses neural networks to identify collision events; while the vision scheme based on multiple binocular depth cameras realizes the accurate positioning of the collision position through three-dimensional point cloud analysis. However, such methods still have limitations: single sensors are vulnerable to environmental noise interference and cannot meet the detection requirements for both minor collisions (such as assembly contacts) and severe collisions. For example, acceleration sensors are prone to misjudgment in complex vibration environments, and the accuracy of vision systems decreases under dynamic occlusion or light changes.

[0005] There is also a part of research on torque detection methods based on dynamic models, and its typical technical solutions are as follows: install torque sensors at the joints of the manipulator or calculate joint torque through motor current, and compare the actual torque with the theoretical value in real time. When the error exceeds the threshold, a collision is determined. However, this method has the following problems in the scenario of high reduction ratio: (1) Signal distortion: The non-linear characteristics of the reducer lead to an unstable current-torque mapping relationship, especially with significant errors under heavy load conditions; (2) Insufficient sensitivity: Only relying on the torque threshold, it is not sensitive enough to minor collisions such as end contacts and is prone to missed detections; (3) Weak anti-interference ability: Environmental vibration or motion shock is prone to cause misjudgment, especially with poor stability in the complex electromagnetic environment of the industrial site.

[0006] Therefore, there is an urgent need for a collision detection method for a high reduction ratio and large load manipulator. Summary of the Invention

[0007] The object of the present invention is to overcome the above deficiencies and provide a collision detection method for a high reduction ratio and large load robotic arm, which overcomes the problems of inaccurate motor current feedback, response lag in collision detection of a high reduction ratio and large load robotic arm, and the inability of a single sensor detection mechanism to balance the sensitivity requirements for mild and severe collisions, realizes high-precision collision recognition and multi-level safety control of the robotic arm and external objects under heavy load conditions, and at the same time improves the anti-interference ability and reliability of the detection system under complex working conditions.

[0008] The object of the present invention is achieved as follows: A collision detection method for a high reduction ratio and large load robotic arm, including the following: S1. First-level detection: End acceleration vibration detection; S1.1. Sensor configuration: Integrate a 3-axis acceleration sensor inside the end arm body of the robotic arm, install it close to the end effector, and directly sense the vibration acceleration signal generated by the end collision; S1.2. Detection logic: S1.1.1. Collect the three-axis vibration signals of the acceleration sensor in real time, filter out the low-frequency noise of the normal movement of the robotic arm through a band-pass filter, and retain the collision characteristic frequency band; S1.1.2. Calculate the root mean square (RMS) value or kurtosis value of the vibration signal. When it exceeds the preset mild collision threshold, trigger the first-level detection response and generate a mild collision warning signal; S2. Second-level detection: Joint torque error compensation detection; S2.1. Signal processing: S2.1.1. Calculate the expected torque: Based on the kinematic and dynamic models of the robotic arm, input the joint positions, velocities, and end load parameters, and calculate the theoretical expected torque T of each joint under collision-free conditions des ; S2.1.2. Obtain the actual torque: Collect the motor current signal I through a motor current sensor, and use a nonlinear compensation model considering the friction, backlash, and elastic deformation of the reducer T act = f ( I , θ , θ ^ ) to calculate the actual joint torque, where θ is the joint angle and θ^ is the joint angular velocity; S2.2. Detection logic: S2.2.1. Calculate the joint torque error. When the error exceeds the severe collision threshold dynamically adjusted according to the load condition, it is determined as a severe collision event; S2.2.2. Introduce the sliding window filtering algorithm to perform time series analysis on the torque error signal, eliminate short-term interference, and avoid misjudgment; S3. Two-stage detection fusion and safety control strategy; S3.1. Hierarchical response mechanism: S3.1.1. When the first-stage detection is triggered, the controller immediately enters the contact protection mode, reduces the end movement speed, and monitors the change of the acceleration signal in real time to prevent a minor contact from evolving into a severe collision; S3.1.2. When the second-stage detection is triggered, the controller executes an emergency braking strategy, cuts off the power output, records the collision position, and simultaneously generates an obstacle avoidance path through the teaching system; S3.2. Multi-source information fusion: Input the vibration signal of the acceleration sensor and the joint torque error signal into the fuzzy logic controller, and improve the detection robustness under complex working conditions through the preset fusion rules; S4. System architecture and data flow; S4.1. Hardware layer: It includes a 3-axis acceleration sensor at the end, a motor current sensor, a joint position encoder, and a drive controller; S4.2. Algorithm layer: Integrate a vibration signal preprocessing module, a dynamic model calculation module, a two-stage threshold comparator, and a fusion decision-making unit; S4.3. Control layer: Output speed control instructions or braking signals according to the detection results to achieve a closed-loop control from collision detection to safety response. Furthermore, in step S1.1, the vibration acceleration signal includes linear acceleration and angular acceleration, and the sensor coordinate system is aligned with the end coordinate system for collecting the end linear acceleration and angular acceleration signals.

[0009] Furthermore, in step S1.1.2, calculate the root mean square (RMS) value of the vibration signal, and the formula is: ; where a x,i , a y,i , a z,i are the three-axis acceleration values at the i-th sampling point, and N is the number of sampling points in the sliding window.

[0010] Furthermore, in step S1.1.2, when the calculated RMS value exceeds the preset threshold, send a minor collision warning signal to the controller, and the controller reduces the end speed of the robotic arm to 20% of the original speed.

[0011] Furthermore, in step S2.2.1, calculate the joint torque error: .

[0012] Further, the preset fusion rule in step S3.2: "High vibration signal + high torque error" is determined as a severe collision, and "Low vibration signal + low torque error" is determined as normal contact.

[0013] Further, the fuzzy logic controller in step S3.2 defines three fuzzy sets: Low L, Medium M, and High H, and makes decisions according to the following rules: 3.21. If the RMS is H and the ΔT is H, it is determined as a severe collision; 3.22. If the RMS is M and the ΔT is L, it is determined as normal contact; 3.23. In other cases, maintain the current state.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention provides a collision detection method for a high reduction ratio and large load robotic arm, which brings significant beneficial effects to the collision detection of the high reduction ratio and large load robotic arm through a two-stage composite detection mechanism and a multi-source information fusion strategy.

[0015] In terms of detection accuracy, the first-stage end acceleration vibration detection can capture the high-frequency vibration generated by a slight collision, solving the problem of missed detection in traditional methods; the second stage corrects the distorted mapping of current and torque under a high reduction ratio through dynamic model compensation, improving the accuracy of severe collision detection.

[0016] In terms of response speed, the end acceleration sensor avoids the signal transmission delay caused by a high reduction ratio and can achieve a millisecond-level response.

[0017] In terms of safety and reliability, the hierarchical response mechanism can give an early warning and reduce the speed for protection in case of a slight collision, and perform an emergency brake in case of a severe collision. At the same time, the fuzzy logic controller fuses multi-source signals, effectively reducing misjudgment under complex working conditions.

[0018] In addition, the present invention is compatible with existing sensor hardware, does not require additional complex equipment, reduces the system transformation cost, and can be widely applied to scenarios such as heavy-load industrial automation and logistics handling, improving the operation safety and working efficiency of the robotic arm. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a schematic diagram of the sensor installation of the present invention.

[0020] Figure 2 It is a logical schematic diagram of the robotic arm dynamics calculation controller of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] To better understand the technical solution of the present invention, the following will be described in detail in conjunction with relevant diagrams. It should be understood that the following specific embodiments are not intended to limit the specific implementation forms of the technical solution of the present invention, but only the implementation forms that the technical solution of the present invention can adopt. It should be noted first that the expressions regarding the positional relationships of various components herein, such as component A is located above component B, are based on the relative positions of the components in the diagrams and are not intended to limit the actual positional relationships of the components.

[0022] See Figure 1-2 , Figure 1 The sensor installation schematic diagram of the collision detection method for the high reduction ratio and large load robotic arm in Embodiment 1 is drawn. As shown in the figure, a collision detection method for a high reduction ratio and large load robotic arm involved in Embodiment 1 solves the problems of accuracy and reliability in collision detection of a high reduction ratio and large load robotic arm through a two-stage composite detection mechanism and a multi-source information fusion strategy, including the following content: S1. First-stage detection: End acceleration vibration detection (for minor collisions); Utilize the characteristics of high-frequency vibration signals generated during the collision of the robotic arm end, capture the dynamic acceleration changes of the end arm body through an acceleration sensor, and achieve rapid identification of contact-type minor collisions; S1.1. Sensor configuration: Integrate a 3-axis acceleration sensor inside the end arm body of the robotic arm, install it close to the end effector, and directly sense the vibration acceleration signals (including linear acceleration and angular acceleration) generated by the end collision; S1.2. Detection logic: S1.1.1. Real-time collect the three-axis vibration signals of the acceleration sensor, filter out the low-frequency noise of the normal movement of the robotic arm (such as the inertial force caused by joint movement) through a band-pass filter, and retain the collision characteristic frequency band (such as high-frequency vibration of 50 - 500 Hz); S1.1.2. Calculate the root mean square (RMS) or kurtosis value of the vibration signal. When it exceeds the preset minor collision threshold, trigger the first-stage detection response and generate a minor collision warning signal; The acceleration sensor is directly installed in the collision-sensitive area (end), avoiding the transmission delay of the signal to the motor end caused by the high reduction ratio, and can detect contact within sub-millisecond time after the collision, solving the problem of missed detection of minor collisions (such as workpiece assembly contact, low-speed touch) by traditional methods. S2. Second-stage detection: Joint torque error compensation detection (for severe collisions); Aiming at the non-linear characteristics of the high reduction ratio reducer, compensate for the mapping deviation between the motor current and the joint torque through a dynamic model, and combine the error analysis of the actual torque and the expected torque to achieve accurate judgment of severe collisions; S2.1. Signal processing: S2.1.1. Expected torque calculation: Based on the manipulator kinematics and dynamics models, input the parameters of the positions, velocities of each joint and the end load, and calculate the theoretical expected torque T of each joint under the collision-free condition. des ; S2.1.2. Actual torque acquisition: Collect the motor current signal I through the motor current sensor, and use the non-linear compensation model T act = f ( I , θ , θ ^ ) to calculate the actual torque of the joint, where θ is the joint angle and θ^ is the joint angular velocity; S2.2. Signal detection logic: S2.2.1. Calculate the joint torque error . When the error exceeds the severe collision threshold dynamically adjusted according to the load condition, it is determined as a severe collision event; S2.2.2. Introduce the sliding window filtering algorithm to perform time series analysis on the torque error signal, exclude short-term interferences (such as start-up shocks), and avoid misjudgment; Compensate the non-linear effect of the reducer through the dynamics model, correct the distorted mapping relationship between the motor current and the joint torque under high reduction ratios, improve the torque detection accuracy to a level adaptable to heavy load conditions, and solve the problems of lag and inaccuracy of traditional methods during severe collisions. S3. Two-stage detection fusion and safety control strategy S3.1. Hierarchical response mechanism: S3.1.1. When the first-stage detection is triggered (minor collision warning), the controller immediately enters the contact protection mode, reduces the end movement speed and monitors the change of the acceleration signal in real time to avoid the minor contact evolving into a severe collision; S3.1.2. When the second-stage detection is triggered (severe collision determination), the controller executes the emergency braking strategy, cuts off the power output and records the collision position. At the same time, generates an obstacle avoidance path through the teaching system; S3.2. Multi-source information fusion: Input the vibration signal of the acceleration sensor and the joint torque error signal into the fuzzy logic controller. Through the preset fusion rules (such as "high vibration signal + high torque error" is determined as a severe collision, "low vibration signal + low torque error" is determined as normal contact), improve the detection robustness under complex conditions (such as vibration environment, load mutation). S4. System architecture and data flow; S4.1. Hardware layer: Includes the end 3-axis acceleration sensor, motor current sensor, joint position encoder and drive controller; S4.2, Algorithm layer: Integrate the vibration signal preprocessing module, the dynamic model calculation module, the two-stage threshold comparator, and the fusion decision-making unit; S4.3, Control layer: Output speed control instructions or braking signals according to the detection results to achieve closed-loop control from collision detection to safety response. Through the above solution, the present invention uses the direct perception of the end vibration to solve the sensitivity problem of slight collision detection, and solves the accuracy problem of severe collision detection under high reduction ratios through the torque error analysis of dynamic compensation, and finally realizes the hierarchical detection and reliable control of different collision levels.

[0023] Specific implementation cases: 1. Sensor installation: Integrate a 3-axis MEMS acceleration sensor (model ADXL345) inside the flange at the end of the robotic arm. The sensor coordinate system is aligned with the end coordinate system to collect the linear acceleration and angular acceleration signals at the end.

[0024] 2. Implementation steps for the first-stage end acceleration vibration detection: 2.1. Signal acquisition and preprocessing: 2.11. The acceleration sensor collects three-axis acceleration data at a sampling frequency of 10 kHz, and filters out low-frequency noise through a band-pass filter (passband frequency 50 - 500 Hz) inside the FPGA; 2.12. Calculate the root mean square (RMS) value of the filtered signal. The formula is: ; where a x,i , a y,i , a z,i are the three-axis acceleration values at the i-th sampling point, and N is the number of sampling points in the sliding window (take 200); 2.2. Collision determination: When the calculated RMS value exceeds the preset threshold (such as 0.5g, which can be adjusted according to the working conditions), send a slight collision warning signal to the controller, and the controller reduces the speed of the end of the robotic arm to 20% of the original speed.

[0025] 3. Implementation steps for the second-stage joint torque error compensation detection: 3.1. Desired torque calculation: The dynamic equation obtained by the Newton-Euler recursive method can be used to calculate the driving torque of each joint for the single robotic arm dynamics: ; In the formula is the inertial force proportional to the generalized acceleration , and each component in is a quadratic form of the generalized velocity. is the gravity term, and are the driving torques of each joint; 3.2. Actual torque calculation: Collect the motor current signal I, and use the pre-calibrated non-linear compensation model T act = f ( I , θ , θ ^ ) to calculate the actual joint torque; this model is established by fitting experimental data, considering the friction characteristics, backlash of the harmonic reducer, and the influence of temperature on the transmission efficiency; 3.3. Collision determination: Calculate the torque error , when ΔT exceeds the dynamic threshold (the threshold is adaptively adjusted according to the load weight and movement speed), trigger the determination of a severe collision; at the same time, use the sliding window algorithm to perform mean filtering on ΔT for 10 consecutive sampling periods to avoid misjudgment caused by instantaneous interference.

[0026] 4. Two-stage detection fusion and control process: 4.1. Information fusion: The RMS value of the acceleration sensor and the joint torque error ΔT are input to the fuzzy logic controller. The fuzzy controller defines three fuzzy sets: low (L), medium (M), and high (H), and makes decisions according to the following rules: 4.11. If RMS is H and ΔT is H, then it is determined as a severe collision; 4.12. If RMS is M and ΔT is L, then it is determined as normal contact; 4.13. In other cases, maintain the current state; 4.2. Response control: 4.21. When a minor collision warning occurs, the controller switches the robotic arm to the low-speed compliant mode, allowing the end effector to continue moving with a lower stiffness; 4.22. After a severe collision is determined, the controller immediately cuts off the motor drive current, triggers the brake device to brake, and displays the collision position and type through the teach pendant to prompt the operator to handle it.

[0027] Working principle: The present invention provides a collision detection method for a high reduction ratio and large load robotic arm. Through a two-stage detection mechanism (acceleration sensor + current feedback), and by combining the end vibration characteristics and joint torque error, the present invention solves the problems of sensitivity and reliability in collision detection of high reduction ratio robotic arms. Among them, the first stage uses a 3-axis acceleration sensor to capture the end vibration signal in real time to achieve rapid identification of minor collisions; the second stage calculates the joint torque error based on current feedback, and combines the dynamic model to compensate for the non-linear effect of the reducer to improve the accuracy of severe collision detection. This composite detection strategy not only overcomes the limitations of a single sensor, but also enhances the system robustness through multi-source information fusion.

[0028] The above are only specific application examples of the present invention, and do not constitute any limitation to the protection scope of the present invention. Any technical solutions formed by equivalent transformation or equivalent substitution fall within the scope of the present invention's rights protection.

Claims

1. A collision detection method for a high reduction ratio and large load mechanical arm, characterized in that: Includes the following: S1, first level detection: terminal acceleration vibration detection; S1.1, Sensor configuration: A 3-axis acceleration sensor is integrated in the end arm of the robot arm and installed close to the end effector to directly sense the vibration acceleration signal generated by the end collision. S1.2, detection logic: S1.1.

1. Collect the three-axis vibration signal of the acceleration sensor in real time, filter out the low-frequency noise of the normal movement of the robot arm through a bandpass filter, and retain the collision characteristic frequency band; S1.1.

2. Calculate the effective value RMS or kurtosis value of the vibration signal. When it exceeds the preset slight collision threshold, trigger the first level detection response and generate a slight collision warning signal. S2, second level detection: joint torque error compensation detection; S2.1, Signal processing: S2.1.

1. Calculation of expected torque: Based on the kinematics and dynamics model of the robot arm, input the position, speed and end load parameters of each joint to calculate the theoretical expected torque T of each joint under non-collision conditions. des ; S2.1.

2. Actual torque acquisition: The motor current signal I is collected through the motor current sensor, and a nonlinear compensation model that considers the friction, backlash and elastic deformation of the reducer is used. T act = f ( I , θ , θ ^ ), calculate the actual torque of the joint, where θ is the joint angle and θ^ is the joint angular velocity; S2.2, signal detection logic: S2.2.

1. Calculate the joint torque error. When the error exceeds the severe collision threshold value dynamically adjusted according to the load condition, it is determined as a severe collision event. S2.2.2, introduce sliding window filtering algorithm to perform time series analysis on torque error signal, eliminate short-term interference and avoid misjudgment; S3, two-level detection fusion and security control strategy; S3.

1. Hierarchical response mechanism: S3.1.

1. When the first level detection is triggered, the controller immediately enters the contact protection mode, reduces the end motion speed and monitors the acceleration signal changes in real time to prevent a slight contact from evolving into a violent collision; S3.1.2, when the second level detection is triggered, the controller executes the emergency braking strategy, cuts off the power output and records the collision position, and generates an obstacle avoidance path through the teaching system; S3.2, Multi-source information fusion: The vibration signal of the acceleration sensor and the joint torque error signal are input into the fuzzy logic controller, and the detection robustness under complex working conditions is improved through the preset fusion rules; S4, system architecture and data flow; S4.1, hardware layer: including the terminal 3-axis acceleration sensor, motor current sensor, joint position encoder and drive controller; S4.2, algorithm layer: integrated vibration signal preprocessing module, dynamic model calculation module, two-level threshold comparator and fusion decision unit; S4.3, control layer: output speed control instructions or braking signals according to the detection results to achieve closed-loop control from collision detection to safety response.

2. The collision detection method of a high reduction ratio and large load mechanical arm according to claim 1, characterized in that: In step S1.1, the vibration acceleration signal includes linear acceleration and angular acceleration, and the sensor coordinate system is aligned with the terminal coordinate system to collect the terminal linear acceleration and angular acceleration signals.

3. The collision detection method for a high reduction ratio and large load mechanical arm according to claim 1, characterized in that: In step S1.1.2, the effective value RMS of the vibration signal is calculated using the formula: ； Among them, a x,i 、a y,i 、a z,i is the three-axis acceleration value of the i-th sampling point, and N is the number of sliding window sampling points.

4. The collision detection method for a high reduction ratio and large load mechanical arm according to claim 1, characterized in that: In step S1.1.2, when the calculated RMS value exceeds the preset threshold, a slight collision warning signal is sent to the controller, and the controller reduces the speed of the robot end to 20% of the original speed.

5. The collision detection method for a high reduction ratio and large load mechanical arm according to claim 1, characterized in that: The joint torque error is calculated in step S2.2.1: .

6. The collision detection method for a high reduction ratio and large load mechanical arm according to claim 1, characterized in that: The fusion rule preset in step S3.2: "high vibration signal + high torque error" is judged as a severe collision, and "low vibration signal + low torque error" is judged as normal contact.

7. The collision detection method of a high reduction ratio and large load mechanical arm according to claim 1, characterized in that: In step S3.2, the fuzzy logic controller defines three fuzzy sets: low L, medium M, and high H, and makes decisions according to the following rules: 3.

21. If RMS is H and ΔT is H, it is judged as a violent collision; 3.

22. If RMS is M and ΔT is L, it is judged as normal contact; 3.

23. In other cases, maintain the current status.

Images