A method, device, and medium for joint diagnosis in a state-aware robotic dog.

Abstract

This invention discloses a method, device, and medium for joint diagnosis in robotic dogs based on state perception, relating to the field of joint state monitoring technology for robotic dogs. The method includes: synchronously collecting joint angles, angular velocities, driving currents, driving voltages, joint temperatures, body posture, and foot contact states during the robotic dog's operation to construct a state perception sequence; identifying support, swing, steering, and acceleration / deceleration motion states based on the state perception sequence and calculating corresponding health reference torques; generating state-related residuals based on the difference between the actual output torque and the health reference torque; fusing residual energy, temperature rise deviation, and symmetrical joint coordination deviation to obtain a health index, thereby determining the location and severity of abnormal joints. This invention can adapt to the diagnostic needs of dynamic changing operating conditions, improving the accuracy of early fault identification, coordination degradation judgment, and abnormal classification and location.

Description

Technical Field

[0001] This invention relates to the field of robot dog joint status monitoring technology, and in particular to a robot dog joint diagnosis method, device and medium based on status perception. Background Technology

[0002] In the field of joint status monitoring of legged robots such as robot dogs, conventional diagnostic methods mostly rely on single-parameter threshold judgment or offline detection methods. Specifically, such methods usually collect a certain type of physical quantity among drive current, joint temperature or vibration signal and compare it with a pre-set fixed threshold, or use periodic shutdown to manually inspect the joints. It has been widely used in the field of robot technology and can preliminarily determine whether there are overloads, overheating or jamming phenomena in the joints, providing a certain basis for the basic operation and maintenance of robot dogs.

[0003] However, conventional methods have two limitations in dynamic and varying operating conditions. First, they fail to consider the significant differences in joint load characteristics under different motion states (such as support, swing, turning, and acceleration / deceleration), resulting in a single diagnostic benchmark and insufficient adaptability. Second, conventional methods often rely on the instantaneous values ​​of single physical quantities for judgment, lacking the fusion processing of multi-source heterogeneous sensory information (such as angle, angular velocity, temperature, body posture, and foot contact state), making it difficult to effectively distinguish between transient disturbances and persistent abnormalities, especially in scenarios with symmetrical joint coordination degradation or early temperature rise accumulation, where recognition capabilities are limited. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a state-aware robot dog joint diagnosis method to solve the problems of insufficient adaptability to complex dynamic motion states and limited ability to fuse multi-source perception information and identify early progressive anomalies in existing technologies.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a state-aware joint diagnosis method for a robot dog, comprising: synchronously collecting the angle, angular velocity, driving current, driving voltage, joint temperature, body posture, and foot contact state of each joint during the operation of the robot dog, and constructing a state-aware sequence for each joint; identifying the current support, swing, turning, and acceleration / deceleration motion states of the robot dog based on the state-aware sequence, and calculating the health reference torque of each joint in combination with the corresponding motion states; obtaining the state-related residual based on the difference between the actual output torque of each joint and the health reference torque, and calculating the health index of each joint in combination with the residual energy, temperature rise deviation, and symmetrical joint coordination deviation to determine the abnormal joint location and abnormality level; and outputting the abnormal joint location and abnormality level of the robot dog.

[0007] As a preferred embodiment of the state-aware robot dog joint diagnosis method of the present invention, the step of collecting the angle, angular velocity, driving current, driving voltage, joint temperature, body posture, and foot contact state of each joint includes: establishing a unified sampling time base and generating time markers corresponding to each sampling time based on the unified sampling time base; collecting the joint angle, driving current, driving voltage, and joint temperature of each joint under each time marker; calculating the joint angular velocity of each joint based on the change in joint angle between adjacent sampling times and the sampling period; collecting the pitch angle, roll angle, and yaw angle of the robot dog body as the body posture; collecting the contact signal of the foot corresponding to each mechanical leg, and determining the foot contact state corresponding to each mechanical leg based on the comparison result of the foot contact force and the contact threshold.

[0008] As a preferred embodiment of the state-aware robot dog joint diagnosis method of the present invention, the construction of the state-aware sequence of each joint includes: performing time alignment, outlier correction, and missing value compensation on the collected joint angles, joint angular velocities, driving currents, driving voltages, joint temperatures, body postures, and foot contact states of each joint to obtain the effective state data corresponding to each joint at each sampling time; combining the joint angles, joint angular velocities, driving currents, driving voltages, joint temperatures, body postures, and foot contact states corresponding to each joint at the same sampling time to construct the state-aware vector corresponding to the joint at the sampling time; and arranging the state-aware vectors corresponding to each joint at multiple consecutive sampling times in chronological order to construct the state-aware sequence of the joint.

[0009] As a preferred embodiment of the state-aware robot dog joint diagnosis method of the present invention, the step of identifying the current support, swing, turning, and acceleration / deceleration motion states of the robot dog based on the state-aware sequence includes: determining whether each mechanical leg is in a support candidate state or a swing candidate state based on the foot contact state in the state-aware sequence; performing consistency statistics on the foot contact states at consecutive sampling times to determine whether the corresponding mechanical leg is in a support state or a swing state; calculating the yaw angle change rate based on the body yaw angle at adjacent sampling times, and determining whether the robot dog is in a turning state based on the comparison result of the yaw angle change rate and a turning threshold; calculating the body linear acceleration based on the body linear velocity at adjacent sampling times, and determining whether the robot dog is in an acceleration / deceleration state based on the comparison result of the body linear acceleration and an acceleration / deceleration threshold; and combining the determination of the support state, swing state, turning state, and acceleration / deceleration state to generate the target motion state of each joint corresponding to the current sampling time.

[0010] As a preferred embodiment of the state-aware robot dog joint diagnosis method of the present invention, the step of calculating the health reference torque of each joint in combination with the corresponding motion state includes: obtaining the target motion state corresponding to each joint at the current sampling time; calling the position load coefficient, velocity damping coefficient, attitude coupling coefficient and contact compensation coefficient corresponding to the target motion state according to the target motion state; calculating the health reference torque of each joint at the current sampling time by combining the joint angle, joint angular velocity, body attitude state quantity and foot contact state of each joint; smoothly updating the health reference torque of consecutive adjacent sampling times to obtain the smoothed health reference torque corresponding to each joint; and using the smoothed health reference torque as the normal load torque reference value of the corresponding joint in the current target motion state.

[0011] As a preferred embodiment of the state-aware robot dog joint diagnosis method of the present invention, the state association residual includes: reading the drive current of the target joint at the current sampling time; calculating the actual output torque of the target joint at the current sampling time based on the drive current and the torque constant of the drive motor corresponding to the target joint; calling the healthy reference torque corresponding to the target joint in the current motion state; calculating the difference between the actual output torque and the healthy reference torque to obtain the state association residual of the target joint at the current sampling time.

[0012] As a preferred embodiment of the state-aware robot dog joint diagnosis method of the present invention, the health index of each joint includes: accumulating the state correlation residuals of the target joint at continuous sampling times to obtain the residual energy of the target joint; comparing the joint temperature of the target joint at the current sampling time with the corresponding health reference temperature under the current motion state to obtain the temperature rise deviation of the target joint; performing difference analysis on the state correlation residuals of the target joint and the state correlation residuals of the corresponding joints in symmetrical positions to obtain the symmetrical joint coordination deviation of the target joint; and fusing the state correlation residuals, residual energy, temperature rise deviation, and symmetrical joint coordination deviation to obtain the health index of the target joint.

[0013] As a preferred embodiment of the state-aware robot dog joint diagnosis method of the present invention, the step of determining the abnormal joint location and abnormality level includes: comparing the health index of each joint at the current sampling time, and determining the joints whose health index exceeds the abnormality judgment threshold as candidate abnormal joint locations; performing continuous time-series limit judgment on the health index corresponding to the candidate abnormal joint locations; determining that the target joint is in a warning abnormal state when the health index of the target joint exceeds the warning threshold for consecutive sampling times; determining that the target joint is an abnormal joint location when the health index of the target joint exceeds the fault threshold for consecutive sampling times; and classifying the abnormality level of the abnormal joint according to the health index value range corresponding to the abnormal joint; the abnormality level includes mild abnormality, moderate abnormality, and severe abnormality.

[0014] In a second aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the state-aware robot dog joint diagnosis method as described in the first aspect of the present invention.

[0015] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any step of the state-aware robot dog joint diagnosis method as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: by synchronously collecting multi-dimensional state data and constructing a state perception sequence, a comprehensive perception of the robot dog's joint operating state is achieved; by identifying motion states such as support, swing, steering, acceleration, and deceleration and calculating the health reference torque, the accuracy of joint anomaly judgment under dynamic changing working conditions is improved; by fusing state correlation residuals, residual energy, temperature rise deviation, and symmetrical joint coordination deviation to calculate the health index, transient disturbances and persistent anomalies are effectively distinguished, enhancing the ability to identify early faults and coordination degradation; by continuously judging the health index over-limit and classifying the anomaly level, the precise location and severity grading of abnormal joints are achieved. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a state-aware robot dog joint diagnosis method.

[0019] Figure 2 Build a flowchart for state-aware sequences.

[0020] Figure 3 Flowchart for calculating healthy reference torque.

[0021] Figure 4 Flowchart for determining the location and severity of abnormal joints. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides a state-aware robot dog joint diagnosis method, including the following steps: S1. During the operation of the robot dog, the angle, angular velocity, driving current, driving voltage, joint temperature, body posture and foot contact status of each joint are collected synchronously to construct the state perception sequence of each joint.

[0026] Furthermore, in this embodiment, the robot dog includes a main body, multiple joint drive parts disposed on each mechanical leg, and detection devices disposed corresponding to each joint drive part.

[0027] Each joint drive part corresponds to at least one of the hip joint, the intermediate joint of the leg, and the distal joint.

[0028] Furthermore, each joint drive part is equipped with an angle acquisition device, a current acquisition device, a voltage acquisition device, and a temperature acquisition device, the main body is equipped with an attitude acquisition device, and each foot position is equipped with a contact status acquisition device.

[0029] Among them, the angle acquisition device is used to acquire joint angles, the current acquisition device is used to acquire drive current, the voltage acquisition device is used to acquire drive voltage, the temperature acquisition device is used to acquire joint temperature, the posture acquisition device is used to acquire body posture, and the contact state acquisition device is used to acquire foot contact state.

[0030] Furthermore, the angle acquisition device is preferably an absolute encoder or a magnetic encoder; the current acquisition device is preferably a bus current sampling circuit or a phase current sampling circuit located inside the servo driver; the voltage acquisition device is preferably a DC bus voltage sampling circuit inside the driver; the temperature acquisition device is preferably a thermistor or a temperature probe attached to the joint drive housing; the attitude acquisition device is preferably an inertial measurement unit; and the contact state acquisition device is preferably a foot force-sensitive device, a grounding switch, or a contact determination structure based on foot reaction force.

[0031] Furthermore, after the robot dog enters the operating state of walking, turning, traversing undulating terrain, or standing still, the control processor establishes a unified sampling time base and synchronously collects the operating status of each joint and the overall status of the robot body at the sampling period.

[0032] Specifically, the control processor generates a current sampling time marker at the beginning of each sampling period, and simultaneously reads the sampling time marker. Joint angles of each joint Drive current Drive voltage and joint temperature Simultaneously read the aircraft's pitch angle Roll angle of the machine body Yaw angle and the The contact state of the foot end corresponding to each mechanical leg .

[0033] It should be noted that, in order to ensure the consistency of various types of collected data in the time dimension, encoder sampling, driver electrical parameter sampling, temperature sampling, inertial measurement component sampling, and foot contact signal sampling are all triggered by the same clock source; for sampling channels with fixed communication delays, delay compensation is performed first before participating in the subsequent state perception sequence construction.

[0034] Furthermore, regarding the first Joint angles of each joint After continuous sampling, the joint angular velocity is calculated based on the joint angle changes at adjacent sampling times. .

[0035] Specifically, the initial estimate of the joint angular velocity is obtained by subtracting the joint angle at the current sampling moment from the joint angle at the previous sampling moment and dividing by the sampling period; then, the initial estimate of the joint angular velocity is smoothed to obtain the... The joint in the first The joint angular velocity at each sampling time.

[0036] The formula for calculating the joint angular velocity is as follows: ; in, Indicates the first The joint in the first Joint angular velocity at each sampling time, Indicates the first The joint angular velocity of each joint at the previous sampling time. Indicates the first The joint in the first Joint angle at each sampling time, Indicates the first The joint angle of each joint at the previous sampling time. Indicates the sampling period. This represents the smoothing coefficient for angular velocity.

[0037] It should be noted that the angular velocity smoothing coefficient is obtained by differentially calculating the joint angle sequence collected by the robot dog under no-load and standard gait conditions, and calibrating the coefficient value corresponding to the minimum fluctuation variance of the filtered angular velocity relative to the original differential angular velocity of the encoder and the dynamic response lag meeting the preset control requirements.

[0038] Furthermore, when acquiring the aircraft's attitude, the attitude acquisition device outputs the aircraft's pitch angle, roll angle, and yaw angle, and uses these parameters as attitude quantities reflecting the aircraft's motion state.

[0039] Specifically, the attitude acquisition device is preferably an inertial measurement unit, which includes a gyroscope and an accelerometer; the control processor fuses the gyroscope output and the accelerometer output to obtain the body pitch angle. Roll angle of the machine body and the yaw angle of the aircraft .

[0040] Furthermore, when collecting the foot contact status, the vertical contact force of the foot is detected by a foot force-sensitive device, or the foot touch signal is output by a ground switch to determine whether each foot is in a grounding state at the current sampling time.

[0041] Specifically, when the Vertical contact force at the foot of a mechanical leg Greater than or equal to the contact threshold At that time, the judgment of the first The mechanical leg foot end is at the first The sampling time is in contact state; when the first sampling time is in contact state; Vertical contact force at the foot of a mechanical leg Less than the contact threshold At that time, the judgment of the first The mechanical leg foot end is at the first Each sampling moment is in a non-contact state.

[0042] The formula for determining the contact state of the foot is expressed as follows: ; in, Indicates the first One mechanical leg in the first Foot contact state at each sampling time This indicates that they are in contact. This indicates that the device is in a non-contact state. Indicates the first One mechanical leg in the first The vertical contact force at the foot at each sampling moment. Indicates the first The contact threshold of a mechanical leg.

[0043] It should be noted that the contact threshold is obtained by collecting vertical contact force data of the foot in two working conditions: the robot dog is standing on a flat ground and swinging off the ground. The value is then calibrated by taking the boundary value between the contact force distribution intervals of the two working conditions or the value corresponding to the minimum classification error. The value range is usually 5% to 15% of the rated support load of the foot.

[0044] It should be noted that, even without a separate foot force-sensitive device, the foot contact state can be determined by combining changes in joint current, leg kinematic constraints, and changes in body vertical acceleration, as long as a contact state marker corresponding to the foot landing condition can be output.

[0045] Furthermore, after obtaining the joint angles, joint angular velocities, joint drive currents, joint drive voltages, joint temperatures, body posture, and foot contact status, the collected raw state data is preprocessed to obtain valid state data under a unified time scale.

[0046] Specifically, the preprocessing includes outlier removal, amplitude limiting, time alignment, and missing value compensation. For any state variable, when the change between the current sample value and the previous sample value exceeds the upper limit of the allowable change, the current sample value is marked as an outlier sample value and corrected using the previous valid sample value.

[0047] The state quantity correction formula is expressed as follows: ; in, Indicates the corrected number Joint state quantities Indicates the first The joint in the first The original state quantity at each sampling time. Indicates the first The original state of each joint at the previous sampling time. Indicates the first The upper limit of permissible changes in the state variables corresponding to each joint.

[0048] It should be noted that the upper limit of allowable variation is obtained by continuously collecting the time-series data of the corresponding state quantities under the normal operating conditions of the robot dog, and statistically analyzing the normal fluctuation upper limit of the state difference between adjacent sampling times, combined with the sensor measurement error and the maximum physical response speed of the mechanism for calibration. The value range is usually set according to the state quantity category as follows: joint angle 0.5°~3° / sampling cycle, joint angular velocity 5° / s~30° / s / s per sampling cycle, drive current 0.1A~0.8A / sampling cycle, drive voltage 0.2V~1V / sampling cycle, joint temperature 0.1℃~0.5℃ / sampling cycle, and body attitude angle 0.2°~2° / sampling cycle.

[0049] Furthermore, after completing the preprocessing, the first... The joint in the first The joint angle, joint angular velocity, driving current, driving voltage, joint temperature, body posture, and foot contact state at each sampling moment are combined to generate the first... The joint in the first The state-aware vector at each sampling time.

[0050] Among them, the The joint in the first The state-aware vector at each sampling time point is represented as: ; in, Indicates the first The joint in the first The state-aware vector at each sampling time.

[0051] Furthermore, the first The state-aware vectors of each joint at multiple consecutive sampling times are arranged in the order of sampling time to construct the state-aware vector of the first joint. A sequence of state perceptions for each joint.

[0052] S2. Identify the current support, swing, steering, and acceleration / deceleration motion states of the robot dog based on the state perception sequence, and calculate the health reference torque of each joint in combination with the corresponding motion states.

[0053] Furthermore, in this embodiment, after acquiring the state perception sequence of each joint, the control processor performs joint analysis on the joint angle, joint angular velocity, driving current, driving voltage, joint temperature, body posture, and foot contact state at consecutive sampling times, such as two sampling times, in order to identify the motion state of the robot dog at the current sampling time.

[0054] Furthermore, the motion states include support state, swing state, turning state, and acceleration / deceleration state; among them, the support state is used to characterize the corresponding mechanical leg foot being in a ground-supporting state, the swing state is used to characterize the corresponding mechanical leg foot being in a swinging-back state, the turning state is used to characterize the robot dog's overall directional adjustment motion process, and the acceleration / deceleration state is used to characterize the robot dog's overall speed change motion process.

[0055] Furthermore, the control processor is in the first The state-sensing sequence of the joint corresponding to each sampling time. As the basis for state recognition, the system comprehensively judges the foot contact state, body posture changes, joint angular velocity changes, and body linear velocity changes in the local state window to obtain the robot dog's state in the [missing information]. The target motion state at each sampling time.

[0056] Specifically, based on the first One mechanical leg in the first Foot contact state at each sampling time For continuous Consistency statistics were performed on the foot contact state within each sampling time; when continuous When the foot contact state is in contact state for all sampling moments, the corresponding mechanical leg is determined to enter the support state; when the foot contact state is in contact state for all sampling moments, the corresponding mechanical leg is determined to enter the support state. If the foot is in a non-contact state for all sampling times, the corresponding mechanical leg is determined to enter the swinging state.

[0057] Among them, continuous At each sampling time By collecting the foot contact state sequence of the robot dog during at least one complete gait cycle under normal operating conditions, and determining the minimum number of sampling points at which the foot contact state of the same mechanical leg remains continuously unchanged within a complete gait cycle, the following method was used: .

[0058] The support state determination condition is expressed as follows: ; in, Indicates the first One mechanical leg in the first Foot contact state at each sampling time This indicates the length of the continuous determination window for contact status.

[0059] It should be noted that, Correspondingly, the oscillation state determination condition is expressed as: ; It should be noted that by applying consistency constraints to the foot contact state at continuous sampling times, misjudgments of support and swing states caused by uneven ground, contact bounce, or sensor jitter can be suppressed.

[0060] Furthermore, based on the identification of the support state and the swing state, the control processor further identifies whether the robot dog is in a turning state according to the rate of change of the body's yaw angle.

[0061] Specifically, through the first Yaw angle of the aircraft at each sampling time With the Yaw angle of the aircraft at each sampling time Perform differential processing and combine it with the sampling period. Computer body yaw angle change rate The calculation formula is expressed as: ; in, Indicates the robot dog in the... The rate of change of the aircraft's yaw angle at each sampling time.

[0062] Furthermore, when the absolute value of the rate of change of the aircraft's yaw angle is greater than or equal to the turning threshold... At that time, it was determined that the robot dog was in the first... At each sampling moment, the aircraft is in a turning state; when the absolute value of the rate of change of the aircraft's yaw angle is less than the turning threshold... At that time, it was determined that the robot dog was in the first... The sampling time is not in a turning state.

[0063] It should be noted that the steering threshold is determined by collecting samples of the yaw angle change rate of the robot dog under normal steering and normal straight-line conditions, and statistically analyzing the distribution boundaries of the two types of samples. It is a critical value that can distinguish between steering and non-steering states. The value range is usually 0.08 rad / s to 0.35 rad / s, preferably 0.12 rad / s to 0.25 rad / s.

[0064] Furthermore, while identifying the turning state, the control processor identifies whether the robot dog is in an acceleration or deceleration state based on the speed change of the robot dog in the forward direction.

[0065] Specifically, the control processor obtains the robot dog's position in the [missing information] based on body displacement information, leg kinematics calculation results, or inertial measurement unit output results. linear velocity of the organism at each sampling time And calculate the body linear acceleration based on the body linear velocity at adjacent sampling times. , is represented as: ; in, Indicates the robot dog in the... The linear acceleration of the organism at each sampling time, Indicates the robot dog in the... The linear velocity of the organism at each sampling time. Indicates the robot dog in the... The linear velocity of the organism at each sampling time.

[0066] Furthermore, when the absolute value of the linear acceleration of the machine body is greater than or equal to the acceleration / deceleration threshold... At that time, it was determined that the robot dog was in the first... At each sampling moment, the body is in an acceleration / deceleration state; when the absolute value of the linear acceleration is less than the acceleration / deceleration threshold... At that time, it was determined that the robot dog was in the first... At any given sampling time, the device is not in a state of acceleration or deceleration.

[0067] It should be noted that the acceleration / deceleration threshold is determined by collecting samples of the robot dog's linear acceleration under normal constant speed, normal acceleration, and normal deceleration conditions, and statistically analyzing the distribution range of the linear acceleration under each condition. This threshold is the critical value that can distinguish between acceleration / deceleration and non-acceleration / deceleration states. The value range is usually 0.15 m / s² to 0.80 m / s², preferably 0.20 m / s² to 0.50 m / s².

[0068] Furthermore, after determining the support state, swing state, steering state, and acceleration / deceleration state, the control processor generates the next step according to a preset priority. The target motion state label at each sampling time.

[0069] Specifically, when the When one mechanical leg meets the support state determination condition and the robot dog simultaneously meets the turning state determination condition, the first mechanical leg will... The mechanical leg corresponds to the joint in the first The motion state at the sampling time is marked as the steering support state; when the sampling time is 1... When the first mechanical leg satisfies the swing state determination condition and the robot dog simultaneously satisfies the turning state determination condition, the first... The mechanical leg corresponds to the joint in the first The motion state at the sampling time is marked as the turning and swinging state; when the sampling time is 1... When the first mechanical leg meets the support state determination condition and the robot dog simultaneously meets the acceleration / deceleration state determination condition, the first... The mechanical leg corresponds to the joint in the first The motion state at the sampling time is marked as the acceleration / deceleration support state; when the... When the first mechanical leg satisfies the swing state determination condition and the robot dog simultaneously satisfies the acceleration / deceleration state determination condition, the first... The mechanical leg corresponds to the joint in the first The motion state at each sampling moment is marked as the acceleration / deceleration swing state; if the conditions for determining the turning state or acceleration / deceleration state are not met, only the support state or swing state is retained as the target motion state.

[0070] It should be noted that when both the steering state and the acceleration / deceleration state are in effect, the steering state takes precedence.

[0071] Furthermore, after identifying the target motion state, the control processor calculates the health reference torque of each joint based on the target motion state; whereby the health reference torque is used to characterize the normal load torque that each joint should theoretically output under the current target motion state.

[0072] Specifically, regarding the first The joint, in the... At each sampling time, based on the joint angle of the joint. Joint angular velocity A healthy reference torque calculation model is constructed by considering the body posture state and foot contact state.

[0073] Furthermore, the first The joint in the first Healthy reference torque at each sampling time It consists of joint position, joint velocity, body posture coupling, and foot contact compensation.

[0074] The formula for calculating the healthy reference torque is as follows: ; in, Indicates the first The joint in the first The health reference torque at each sampling time Indicates the first Each joint in the target motion state The load factor at the location below, Indicates the first The joint in the first Joint angle at each sampling time, Indicates the first Each joint in the target motion state The velocity damping coefficient at the bottom, Indicates the first The joint in the first Joint angular velocity at each sampling time, Indicates the first Each joint in the target motion state The attitude coupling coefficient under the following conditions Indicates the first The body attitude state at each sampling time. Indicates the first Each joint in the target motion state The contact compensation coefficient below, Indicates the first One mechanical leg in the first The foot contact state at each sampling time.

[0075] It should be noted that the position load coefficient, velocity damping coefficient, attitude coupling coefficient, and contact compensation coefficient are obtained by collecting samples of joint angles, joint angular velocities, body posture, foot contact state, and actual output torque of each joint under different motion states, and by using least squares fitting to identify the parameters of the health reference torque calculation model.

[0076] Furthermore, the body attitude state quantity It is composed of the aircraft pitch angle, aircraft roll angle, and aircraft yaw angle, and is represented as follows: ; in, Indicates the first The body attitude state at each sampling time. Indicates the first The pitch angle of the aircraft at each sampling time. Indicates the first The roll angle of the organism at each sampling time. Indicates the first The aircraft yaw angle at each sampling time. , and These represent the pitch angle weighting coefficient, roll angle weighting coefficient, and yaw angle weighting coefficient, respectively.

[0077] It should be noted that the pitch angle weighting coefficient, roll angle weighting coefficient, and yaw angle weighting coefficient are determined by collecting pitch angle, roll angle, and yaw angle samples and corresponding joint output torque samples of the robot dog under different attitude disturbance conditions, and then normalizing them after multivariate regression fitting analysis to determine the contribution of each attitude angle to the joint load change. The values ​​of the pitch angle weighting coefficient, roll angle weighting coefficient, and yaw angle weighting coefficient are usually taken in the range of 0.10 to 0.60, and the normalized values ​​satisfy the following conditions: Preferably, the pitch angle weighting coefficient is 0.35 to 0.50, the roll angle weighting coefficient is 0.30 to 0.45, and the yaw angle weighting coefficient is 0.10 to 0.25.

[0078] Furthermore, the control processor can establish a health reference parameter table for each joint under different target motion states based on the normal operation samples of the robot dog; during actual operation, it can directly call the position load coefficient, velocity damping coefficient, attitude coupling coefficient and contact compensation coefficient in the corresponding parameter table according to the identified target motion state to complete the calculation of health reference torque.

[0079] Specifically, when the The joint in the first When a sampling moment is identified as a support state, the corresponding support state is invoked. , , and Calculate the healthy reference torque; when the first The joint in the first When a sampling moment is identified as a swing state, the corresponding swing state is called. , , and Calculate the healthy reference torque; when the first The joint in the first When a sampling moment is identified as a steering state or an acceleration / deceleration state, the steering state parameter group or the acceleration / deceleration state parameter group is called to calculate the healthy reference torque.

[0080] Furthermore, to improve the smoothness of the healthy reference torque over continuous time, in the first... After the healthy reference torque is calculated at each sampling time, the healthy reference torque at consecutive adjacent sampling times is updated smoothly.

[0081] The smooth update formula is expressed as follows: ; in, Indicates the first The joint in the first The healthy reference torque after smoothing at each sampling time point Indicates the first The healthy reference torque of each joint after smoothing at the previous sampling time. Indicates the first The joint in the first The health reference torque is calculated at each sampling time. This represents the torque smoothing coefficient.

[0082] It should be noted that the torque smoothing coefficient is determined by jointly calibrating the fluctuation amplitude of the healthy reference torque at adjacent sampling times under the normal operation sample of the robot dog and the tracking error of the actual output torque, and selecting the coefficient value that suppresses the fluctuation of the healthy reference torque sequence and minimizes the tracking error.

[0083] It should be noted that by smoothly updating the healthy reference torque, the sudden change in reference torque caused by the instant of state switching can be avoided, thereby improving the continuity and stability of subsequent state-related residual calculations.

[0084] S3. Based on the difference between the actual output torque of each joint and the healthy reference torque, the state-related residual is obtained. Combined with the residual energy, temperature rise deviation and symmetrical joint coordination deviation, the health index of each joint is calculated to determine the location and level of abnormal joints.

[0085] Furthermore, in this embodiment, after obtaining the healthy reference torque corresponding to each joint, the control processor continues to perform torque mapping calculation on the drive current of each joint at the current sampling time to obtain the actual output torque of each joint, and constructs the state association residual based on the difference between the actual output torque and the healthy reference torque.

[0086] Furthermore, regarding the first The joint, in the... At each sampling moment, the control processor reads the drive current corresponding to the joint. And combined with the torque constant of the drive motor corresponding to the joint Calculate the first The joint in the first The actual output torque at each sampling time.

[0087] The formula for calculating the actual output torque is as follows: ; in, Indicates the first The joint in the first The actual output torque at each sampling time Indicates the first Each joint corresponds to the torque constant of the drive motor. Indicates the first The joint in the first The driving current at each sampling moment.

[0088] It should be noted that the torque constant is obtained by reading the factory calibration parameters of the drive motor, or by measuring the ratio of the output torque to the drive current of the drive motor under constant current drive conditions.

[0089] Furthermore, in obtaining the actual output torque and health reference torque Then, the control processor calculates the difference between the two to obtain the first value. The joint in the first The state-related residuals at each sampling time.

[0090] The formula for calculating the state-related residual is as follows: ; in, Indicates the first The joint in the first State-related residuals at each sampling time Indicates the first The joint in the first The actual output torque at each sampling time Indicates the first The joint in the first The health reference torque at each sampling time.

[0091] Furthermore, when the state-related residuals When the absolute value of is small, it indicates that the th The actual output torque of each joint under the current target motion state is close to the theoretical healthy output level; when the state-related residual... When the absolute value of continues to increase, it indicates that the th Some joints exhibit a tendency for actual output to deviate from the healthy operating benchmark.

[0092] It should be noted that it is difficult to accurately distinguish between transient fluctuations such as ground impacts and attitude disturbances and persistent joint anomalies based solely on the state correlation residual at a single sampling moment. Therefore, in this embodiment, the state correlation residual is further accumulated over time to obtain the residual energy.

[0093] Furthermore, the control processor controls the first Energy accumulation calculation is performed on the state correlation residuals of each joint at consecutive sampling times to obtain the energy of the first joint. The joint in the first The residual energy at each sampling time.

[0094] Specifically, the residual energy at the current sampling time is obtained by weighted accumulation of the squared residual value associated with the state at the current sampling time and the residual energy at the previous sampling time, and is expressed as: ; in, Indicates the first The joint in the first The residual energy at each sampling time. Indicates the first The residual energy of each joint at the previous sampling time. Indicates the first The joint in the first State-related residuals at each sampling time This represents the residual energy forgetting factor.

[0095] It should be noted that the residual energy forgetting factor is obtained by recursively fitting the state-related residual sequence on the normal operation samples of the robot dog, and taking the highest accuracy of residual energy in distinguishing between continuous anomalies and transient disturbances as the optimization objective. The value range is usually 0.80 to 0.98.

[0096] It should be noted that residual energy is used to characterize the degree of continuous accumulation of state-related residuals over time. When a joint has only a single short-term perturbation, the growth of residual energy is limited. When a joint has a continuous deviation, the residual energy will gradually increase, thereby enhancing the ability to identify chronic anomalies and progressive failures.

[0097] Furthermore, while calculating the residual energy, the control processor, according to the first... The difference between the joint temperature of each joint under the current target motion state and the healthy reference temperature is calculated. Temperature rise deviation of each joint.

[0098] Specifically, the control processor, based on the state recognition result, calls the first... Each joint in the target motion state The corresponding health reference temperature and will the The joint in the first Joint temperature at each sampling time The temperature rise deviation is obtained by calculating the difference between the temperature and the health reference temperature.

[0099] The formula for calculating the temperature rise deviation is as follows: ; in, Indicates the first The joint in the first Temperature rise deviation at each sampling time, Indicates the first The joint in the first Joint temperature at each sampling time, Indicates the first Each joint in the target motion state The following is a healthy reference temperature.

[0100] Furthermore, the health reference temperature is obtained by collecting historical temperature data of the corresponding joint under normal working conditions and corresponding exercise conditions, and by calculating the historical temperature data through statistical averaging or recursive averaging.

[0101] Furthermore, while calculating the temperature rise deviation, the control processor calculates the first... Symmetrical joint coordination deviation of each joint.

[0102] Specifically, in the symmetrical structure of the robot dog, for the first... Each joint determines its corresponding joint on the symmetrical robotic leg. ; will the first The joint in the first State-related residuals at each sampling time With corresponding joints In the State-related residuals at each sampling time Compare and obtain the first Symmetrical joint coordination deviation of each joint.

[0103] The formula for calculating the symmetrical joint coordination deviation is as follows: ; in, Indicates the first The joint in the first Symmetrical joint coordination deviation at each sampling time Indicates the first The joint in the first State-related residuals at each sampling time Indicates the relationship with the first The corresponding joints in symmetrical positions are in the 1st... The state-related residuals at each sampling time.

[0104] Furthermore, after obtaining the state-related residual, residual energy, temperature rise deviation, and symmetric joint coordination deviation respectively, the control processor performs fusion calculation on the above multiple abnormal characterization quantities to obtain the first... The joint in the first The health index at each sampling time.

[0105] Specifically, the control processor first determines the... Each joint in the target motion state Residual baseline quantity Residual energy reference quantity and temperature difference reference quantity The state-related residuals, residual energy, and temperature rise deviations are normalized; then, combined with the symmetric joint cooperative deviation, the various abnormal characterization quantities are weighted and fused to generate the first... The joint in the first The health index at each sampling time.

[0106] The formula for calculating the health index is as follows: ; in, Indicates the first The joint in the first Health index at each sampling time , , and These represent the weights of the state-related residual term, the residual energy term, the temperature rise deviation term, and the symmetric joint coordination deviation term, respectively. Indicates the first Each joint in the target motion state The residual baseline quantity, Indicates the first Each joint in the target motion state The residual energy benchmark quantity below, Indicates the first Each joint in the target motion state The temperature difference reference quantity.

[0107] It should be noted that the weights of the state-related residual term, residual energy term, temperature rise deviation term, and symmetry joint coordination deviation term are obtained by normalizing each abnormal characterization quantity on normal and abnormal samples of the robot dog, and then using a multi-parameter joint calibration method aimed at maximizing the accuracy of abnormal joint recognition and the consistency of abnormal level classification. The weights of the state-related residual term... Weights of residual energy terms The value range is typically 0.20 to 0.35, and the weight of the temperature rise deviation term is... The value range is typically 0.10 to 0.25, representing the weight of the symmetrical joint coordination deviation term. The value range is usually 0.15 to 0.30.

[0108] It should be noted that the benchmark quantities in the health index calculation formula are obtained by collecting the state correlation residuals, residual energy and temperature rise deviation data of the corresponding joints under normal working conditions and under corresponding motion conditions, and then statistically averaging the state correlation residuals, residual energy and temperature rise deviation data respectively.

[0109] Furthermore, after calculating and obtaining the health index of each joint, the control processor determines the location of abnormal joints based on the comparison of the magnitude of the health index of each joint.

[0110] Specifically, the control processor in the At each sampling time, the health index of all joints is sorted, and the joint with the largest health index that exceeds the anomaly judgment threshold is selected as the candidate position of the abnormal joint; when the health index of multiple joints exceeds the anomaly judgment threshold, the corresponding multiple joints are marked as candidate positions of the abnormal joint at the same time.

[0111] It should be noted that the anomaly detection threshold is determined by statistically analyzing the health index distribution of each joint of the robot dog under normal operating conditions, and using the lowest misjudgment rate of the health index in distinguishing between normal and abnormal joints as the criterion. The value range is usually 1.0 to 2.5.

[0112] Furthermore, to avoid misjudgment of abnormal joints caused by short-term fluctuations at a single sampling time, the first... The health index of each joint is continuously exceeded at each sampling time to determine if it exceeds the limit.

[0113] Specifically, when the Health index of each joint continuous The number of sampling times is greater than or equal to the warning threshold. At that time, the judgment of the first The first joint is in an abnormal warning state; when the first joint is in an abnormal warning state; Health index of each joint continuous The sampling time is greater than or equal to the fault threshold. At that time, the judgment of the first The joint is in a faulty or abnormal state.

[0114] It should be noted that the warning threshold is determined by statistically analyzing the health index distribution of each joint of the robot dog under normal operating conditions and early abnormality samples, using a high detection rate of early abnormalities and a controllable false alarm rate as criteria. The value range is typically 1.2–2.0. The fault threshold is determined by statistically analyzing the health index distribution of each joint of the robot dog under normal operating conditions and fault / abnormality samples, using the highest accuracy in identifying established faults and a clear distinction between the fault and the warning threshold as criteria. The value range is typically 2.0–3.5, and also meets the following requirements: .

[0115] It should be noted that the number of sampling times and By collecting the health index sequence of the corresponding joints of the robot dog under normal operating conditions, and determining the minimum number of consecutive sampling points required to prevent false alarms when the health index continuously exceeds the warning threshold and fault threshold respectively, the following method was used: and .

[0116] Furthermore, after determining the location of the abnormal joint, the control processor determines the abnormality level based on the range of health index values ​​and their main contributing factors.

[0117] Specifically, when the Health index of each joint When it is in the mildly abnormal range, the first... The abnormality level of the first joint is mild; when the first joint... Health index of each joint When the condition is in the moderately abnormal range, the first... The abnormality level of the first joint is moderate; when the first joint... Health index of each joint When the condition is in the severely abnormal range, the first... The abnormality level of each joint is severe.

[0118] Furthermore, the rules for classifying anomaly levels, specifically, when... When, it is judged as a mild abnormality; when When, it is judged as moderately abnormal; when At that time, it was determined to be a severe abnormality; among them, This represents the threshold for severe anomalies, and .

[0119] It should be noted that the severe abnormality threshold is determined by statistically analyzing the health index distribution of each joint of the robot dog under severe abnormality samples, and using the highest accuracy in identifying severe abnormal states and the clear level boundary between the threshold and the warning threshold and the fault threshold as the criteria. The value range is usually 3.5~5.0.

[0120] Furthermore, while determining the anomaly level, the control processor can also assist in judging the anomaly type based on the anomaly characteristics with the largest proportion in the health index; when the state-related residual term and residual energy term have the largest proportion, it is preferentially judged as a load deviation anomaly; when the temperature rise deviation term has the largest proportion, it is preferentially judged as a thermal anomaly; when the symmetrical joint coordination deviation term has the largest proportion, it is preferentially judged as a left-right coordination imbalance anomaly.

[0121] Finally, the abnormal joint locations and abnormality levels of the robot dog are output.

[0122] This embodiment also provides a computer device applicable to the state-aware robot dog joint diagnosis method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the state-aware robot dog joint diagnosis method proposed in the above embodiment.

[0123] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0124] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the state-aware joint diagnosis method for a robot dog as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0125] In summary, this invention achieves comprehensive perception of the robot dog's joint operating status by synchronously collecting multi-dimensional state data and constructing a state perception sequence; it improves the accuracy of joint anomaly judgment under dynamic changing conditions by identifying motion states such as support, swing, steering, acceleration and deceleration and calculating health reference torque; it effectively distinguishes between transient disturbances and persistent anomalies by fusing state correlation residuals, residual energy, temperature rise deviations and symmetrical joint coordination deviations to calculate a health index, enhancing the ability to identify early faults and coordination degradation; and it achieves precise location and severity quantification of abnormal joints by continuously time-series exceeding the limits of the health index and classifying abnormal levels.

[0126] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for diagnosing joints in a robot dog based on state awareness, characterized in that, include: During the operation of the robot dog, the angle, angular velocity, driving current, driving voltage, joint temperature, body posture and foot contact status of each joint are collected synchronously to construct the state perception sequence of each joint. The robot dog is identified in its current support, swing, turning and acceleration / deceleration motion states based on the state perception sequence, and the health reference torque of each joint is calculated in combination with the corresponding motion states. Based on the difference between the actual output torque of each joint and the healthy reference torque, the state-related residual is obtained. Combined with the residual energy, temperature rise deviation and symmetrical joint coordination deviation, the health index of each joint is calculated to determine the location and level of abnormal joints. Output the location and severity of any abnormal joints in the robot dog.

2. The state-aware robot dog joint diagnosis method as described in claim 1, characterized in that, The acquisition of angles, angular velocities, driving currents, driving voltages, joint temperatures, body postures, and foot contact states for each joint includes: Establish a unified sampling time base, and generate time stamps corresponding to each sampling time based on the unified sampling time base; At each time marker, the joint angle, driving current, driving voltage, and joint temperature of each joint are collected. The joint angular velocity of each joint is calculated based on the change in joint angle at adjacent sampling times and the sampling period. The pitch, roll, and yaw angles of the robot dog's body are collected as its attitude. The contact signals of the corresponding foot end of each robotic leg are collected, and the contact state of the corresponding foot end of each robotic leg is determined based on the comparison results of the foot end contact force and the contact threshold.

3. The state-aware robot dog joint diagnosis method as described in claim 2, characterized in that, The construction of the state-aware sequence for each joint includes: The collected joint angles, joint angular velocities, driving currents, driving voltages, joint temperatures, body postures, and foot contact states of each joint are time-aligned, outlier corrections are made, and missing value compensations are performed to obtain the effective state data of each joint at each sampling time. The joint angle, joint angular velocity, driving current, driving voltage, joint temperature, body posture and foot contact state corresponding to each joint at the same sampling time are combined to construct the state perception vector of the joint at the sampling time. The state-aware vectors corresponding to each joint at multiple consecutive sampling times are arranged in chronological order to construct the state-aware sequence of the joint.

4. The state-aware robot dog joint diagnosis method as described in claim 1, characterized in that, The process of identifying the robot dog's current support, swing, turning, and acceleration / deceleration motion states based on the state-aware sequence includes: Based on the foot contact state in the state-aware sequence, determine whether each mechanical leg is in a support candidate state or a swing candidate state. Consistency statistics are performed on the foot contact state at continuous sampling times to determine whether the corresponding mechanical leg is in a supporting or swinging state. The rate of change of yaw angle is calculated based on the yaw angle of the aircraft at adjacent sampling times, and the robot dog is determined to be in a turning state based on the comparison result of the rate of change of yaw angle and the turning threshold. The machine dog is calculated based on the linear velocity of the machine body at adjacent sampling times, and the machine dog is determined to be in an acceleration or deceleration state based on the comparison result of the linear acceleration of the machine body with the acceleration and deceleration threshold. By combining the determination of support state, swing state, turning state and acceleration / deceleration state, the target motion state of each joint corresponding to the current sampling time is generated.

5. The state-aware robot dog joint diagnosis method as described in claim 1 or 4, characterized in that, The calculation of the health reference torque for each joint based on the corresponding motion state includes: Obtain the target motion state of each joint at the current sampling time; The position load coefficient, velocity damping coefficient, attitude coupling coefficient, and contact compensation coefficient corresponding to the target motion state are invoked according to the target motion state. By combining the joint angles, joint angular velocities, body posture states, and foot contact states of each joint, the health reference torque of each joint at the current sampling moment is calculated. The healthy reference torque at consecutive adjacent sampling times is smoothly updated to obtain the smoothed healthy reference torque for each joint. The smooth and healthy reference torque is used as the reference value of the normal load torque of the corresponding joint under the current target motion state.

6. The state-aware robot dog joint diagnosis method as described in claim 5, characterized in that, The state-related residuals include: Read the drive current of the target joint at the current sampling moment; Based on the driving current and the torque constant of the driving motor corresponding to the target joint, calculate the actual output torque of the target joint at the current sampling moment; Call the healthy reference torque corresponding to the target joint in the current motion state; The difference between the actual output torque and the healthy reference torque is calculated to obtain the state-related residual of the target joint at the current sampling time.

7. The state-aware robot dog joint diagnosis method as described in claim 1, characterized in that, The health index of each joint includes: The residual energy of the target joint is obtained by accumulating the state correlation residuals of the target joint at continuous sampling times. The joint temperature of the target joint at the current sampling time is compared with the corresponding healthy reference temperature under the current motion state to obtain the temperature rise deviation of the target joint. The state association residual of the target joint is compared with the state association residual of the corresponding joint in a symmetrical position to obtain the symmetrical joint coordination deviation of the target joint. The health index of the target joint is obtained by fusing state-related residuals, residual energy, temperature rise deviation, and symmetric joint coordination deviation.

8. The state-aware robot dog joint diagnosis method as described in claim 7, characterized in that, Determining the location and severity of abnormal joints includes: The health index of each joint at the current sampling time is compared, and joints whose health index exceeds the abnormality judgment threshold are identified as candidate abnormal joints. Continuous temporal limit determination is performed on the health index corresponding to the candidate locations of abnormal joints. When the health index of the target joint exceeds the warning threshold for consecutive sampling times, the target joint is determined to be in an abnormal warning state. When the health index of the target joint exceeds the fault threshold for consecutive sampling times, the target joint is determined to be an abnormal joint location. The abnormality level of the abnormal joint is classified according to the range of health index values ​​corresponding to the abnormal joint. The abnormality levels include mild abnormality, moderate abnormality, and severe abnormality.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the state-aware robot dog joint diagnosis method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the state-aware robot dog joint diagnosis method according to any one of claims 1 to 8.

Images