A Drift Detection and Update Method Based on Spatial Intelligent Pose Residual Trend Decomposition
By employing a spatial intelligent pose residual trend decomposition method and utilizing Student's t-distribution and mathematical decomposition techniques, the problems of false detection and source tracing in robotic arm pose drift detection were solved. This enabled quantitative source tracing of dynamic drift and local parameter updates, improving detection adaptability and computational efficiency.
Patent Information
- Application Number
- CN202611619029.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-20
- Publication Date
- 2026-11-17
AI Technical Summary
Existing methods for detecting pose drift in robotic arms cannot effectively distinguish between random noise and dynamic drift, cannot trace the cause of drift, and cannot differentiate between normal wear and sudden faults, leading to false detections and failure to detect mechanical faults in a timely manner.
By using a spatial intelligent pose residual trend decomposition method, and employing Student's t-distribution and mathematical decomposition techniques, we can distinguish between drift trends and noise, quantify the influence of various physical parameters, and achieve directional tracing and local updating of drift parameters.
It effectively distinguishes between dynamic drift and noise, enables quantitative traceability of drift parameters, reduces false detection rate, and performs local parameter updates without stopping the system, thereby improving detection adaptability and computational efficiency.
Smart Images

Figure CN123103708A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a drift detection and update method based on spatial intelligent pose residual trend decomposition, belonging to the field of spatial intelligence and robot dynamics modeling technology. Background Technology
[0002] Existing methods for detecting drift in robotic arm pose have the following shortcomings: Traditional residual thresholding cannot distinguish between random noise and dynamic drift. Existing residual analysis and drift monitoring of robotic arm dynamics generally adopt a fixed residual threshold comparison mechanism to determine whether the actual pose residual of the robotic arm exceeds the predetermined judgment threshold. This ignores the temporal variation law and statistical distribution characteristics of the residual, and cannot distinguish between random noise and continuous parameter drift caused by component aging. As a result, some non-drift noise interference is misjudged as model parameter drift, leading to false alarm and false detection problems.
[0003] It can only determine the overall abnormality of the damage, but cannot accurately identify the specific issue. Existing robotic arm dynamic drift monitoring methods can only determine whether there is residual drift in the overall robotic arm, but do not have the ability to trace the source of drift residuals. They cannot identify the specific physical parameters that cause drift. Furthermore, traditional drift monitoring methods lack a quantitative mechanism for physical parameters, and cannot quantify the degree of differential contribution of different dynamic parameters to the end-effector pose residuals. They also cannot distinguish between highly sensitive drift parameters and low-sensitivity invalid parameters, resulting in a lack of quantitative basis for residual source tracing. It can only rely on manual inspection or identification, which has a long inspection cycle and high labor costs.
[0004] Unable to distinguish between normal wear drift and fault-induced drift: In actual operation, deviations in the physical parameters of a robotic arm can be caused by normal aging drift due to long-term wear of components such as bearings and reducers, or by abnormal drift caused by sudden failures such as mechanical impact, component jamming, lubrication failure, part damage, or loose assembly. Traditional drift monitoring methods cannot identify the essential differences between these two types of drift. Regardless of whether it is normal aging or abnormal failure, they uniformly perform model adaptive correction, forcibly fitting the residual deviation caused by the failure by modifying the model parameters. This approach transfers the mechanical faults of the robotic arm design to the dynamic model, masking the real mechanical faults. As a result, the faults cannot be detected in time, posing safety hazards and failing to guarantee the normal operation of the robotic arm. Summary of the Invention
[0005] To address the shortcomings of existing technologies, the present invention aims to provide a drift detection and update method based on spatial intelligent pose residual trend decomposition, which is intended to solve the problems of drift false detection, drift source tracing, and directional update of model parameters in robotic arm dynamics.
[0006] To achieve the above objectives, the present invention is implemented through the following technical solution: a drift detection and update method based on spatial intelligent pose residual trend decomposition includes: Acquire transmission data of the robotic arm joints, calculate the actual three-dimensional position and actual posture of the robotic arm end effector; use the initial dynamic model to predict the end effector pose of the robotic arm, and calculate the comprehensive spatial residual; Mathematically decompose the residual of the integrated space, screen out the robotic arms with drift trend as the robotic arms to be inspected, analyze the residual of the physical parameters of the robotic arms to be inspected with respect to the integrated space, and screen out the first abnormal parameter. Compare the parameter changes of the part corresponding to the first abnormal parameter with the theoretical wear changes of the part to determine whether the parameter drift of the part belongs to normal wear; if not, an alarm is triggered. If so, check whether the linkage between parts is compliant; If compliant, no action is taken; if non-compliant, a comprehensive drift score for the first anomaly parameter is calculated. Set different drift tolerance thresholds; compare the comprehensive drift trend score of the first abnormal parameter with the drift tolerance threshold to filter the second abnormal parameter; Based on the theoretical wear changes of each part, determine whether the wear of the part corresponding to the second abnormal parameter is normal wear; if it is normal wear, update the initial dynamic model. If the wear and tear is not normal, an alarm will be triggered; Test the unupdated initial dynamic model and the updated initial dynamic model to check whether the drift trend has been eliminated; if it has been eliminated, archive the entire process of drift judgment and physical parameter modification. If the problem persists, continue adjusting.
[0007] Furthermore, the calculation steps for the actual 3D position and actual attitude are as follows: Obtain transmission data of the robotic arm joints; Arrange the robot arm joint sequence, and extract the link length, link torsion angle, joint offset and joint rotation angle of all joints from the transmission data of the robot arm joints according to the joint order in the robot arm joint sequence. Construct joints from the first joint to the second joint, from the second joint to the third joint, and so on, until the first joint... The joint to the first The homogeneous transformation matrix of each joint is obtained to obtain the matrix. To matrix ; Obtain the link length and torsion angle from the robot arm base to the first joint, construct the homogeneous transformation matrix of the first joint, and obtain the matrix. ; Multiply matrices in sequence To matrix The total homogeneous transformation matrix is obtained. ; In the matrix Extracting the elements from the first, second, and third rows of the fourth row yields the actual three-dimensional position of the robotic arm's end effector. ; In the matrix In the middle, extract the top left (3×3) submatrix to obtain the actual pose of the robotic arm's end effector. ; Add timestamps to the actual 3D position and actual orientation of each robotic arm end effector; Simulate the motion changes of each joint of the robotic arm and construct an initial dynamic model; Based on the initial dynamic model, the integrated spatial residual is calculated.
[0008] Furthermore, the calculation steps for the integrated spatial residuals are as follows: Calculate the output torque of the robotic arm using the initial dynamics model; Set hysteresis parameters According to the lag class parameters Given the output torque of the robotic arm, calculate the effective output torque of the robotic arm; based on the effective output torque of the robotic arm, calculate the angles of each joint in the robotic arm; based on the angles of each joint of the robotic arm, calculate the total homogeneous transformation matrix from the base to the end effector, and extract the predicted position and predicted posture of the end effector corresponding to the initial dynamic model; Add timestamps to the predicted position and orientation of the robotic arm's end effector to record the robotic arm's real-time physical parameters; Calculate the residuals between the position and attitude of the robotic arm's end effector to obtain the comprehensive spatial residuals; Based on the timestamp, summarize all the actual 3D positions, actual postures, predicted positions, and predicted postures of the robotic arm end effector. Calculate the Euclidean distance between all the actual 3D positions and all the predicted positions of the robotic arm end effector to obtain the position residual of the robotic arm end effector. Calculate the attitude residual at the end of the robotic arm; Based on the 3σ principle, outliers in the position and attitude residuals of the robotic arm end are screened and removed to obtain the comprehensive spatial residual. Mathematical decomposition of the integrated spatial residuals is used to screen out robotic arms to be inspected.
[0009] Furthermore, the steps for calculating the attitude residuals are as follows: The actual posture corresponding to the end effector of the robotic arm is denoted as... Predicted posture is recorded as ; Calculate the attitude residual rotation matrix Calculate the matrix traces ;according to Calculate matrix The rotation angle is obtained by taking the cosine of the corresponding rotation angle. ; According to the rotation angle Calculate the antisymmetric matrix ; Constructing a matrix rotation angle Rodriguez rotation formula, , as well as Substituting into the Rodriguez rotation formula, calculate the matrix. unit axis of rotation ; For matrix Perform a logarithmic mapping and calculate the matrix. rotation vector ; calculate The vector 2 norm is used to obtain the attitude residual of the robotic arm end effector.
[0010] Furthermore, the screening steps for the robotic arms to be inspected are as follows: Arrange all position residuals in chronological order of timestamps to obtain the position residual sequence; Set the sliding window size for the location residual sequence. The first position residual sample is obtained by extracting a sample from the position residual sequence. A linear regression model for the residual samples at the first position is established, and the residuals are decomposed into three parts: the baseline mean, the linear trend, and the random noise. Without including random noise, the least squares method is used to fit the regression intercept. With regression slope Minimize the sum of squared residuals of the samples to obtain the standard regression intercept. With standard regression slope ; use and Fitting from the 1st to the 2nd Random noise of timestamps 1 to 2, making the 1st to 3rd timestamps 1st to Random noise with timestamps follows a normal distribution with zero mean, thus obtaining standard random noise; by For degrees of freedom, calculate the standard deviation of the standard random noise, and obtain... ; The average time to calculate the residual sample at the first location Calculate the standard regression slope Sampling standard error .
[0011] Furthermore, the screening process for the robotic arms to be inspected also includes: Calculate the test statistic for the residual sample at the first position. ; Based on test statistic Constructing degrees of freedom Using the student t-distribution, a target confidence level is set. The two-sided quantiles corresponding to the target confidence level are extracted from the student t-distribution to obtain the drift threshold for the residual sample at the first position. ; Comparing absolute values With threshold Size: like This indicates that the residual sample at the first position has a drifting trend; like If the result is negative, it indicates that the residual sample at the first position does not have a drift trend and belongs to temporary jitter noise, so it is not processed. Distinguish between other position residuals and attitude residuals in the integrated spatial residuals, and filter out temporary jitter noise and drift trends; The robotic arm exhibiting a drift trend is selected as the robotic arm to be inspected. The physical parameters in the initial dynamic model corresponding to the robotic arm to be inspected are extracted, the influence of each physical parameter on the comprehensive spatial residual is quantified, and the first abnormal parameter is screened.
[0012] Furthermore, the screening steps for the first abnormal parameter are as follows: Obtain all physical parameters from the initial dynamic model corresponding to the robotic arm under inspection. ~ ;in, Indicates the number of physical parameters; Calculate the influence coefficient of the first physical parameter on the overall spatial residual. ; Backup the initial dynamic model corresponding to the robotic arm to be inspected to obtain model a and model b; set up positive disturbance With reverse disturbance Backup model b to obtain model b (1) With model b (2) The positive disturbances will be respectively With reverse disturbance Add model b (1) With model b (2) middle; Get current time According to the end of the robotic arm under inspection Time and The position and attitude changes at each moment are calculated using model a and model b. (1) With model b (2)The corresponding position residuals and attitude residuals; model a and model b respectively. (1) With model b (2) The corresponding position residuals and attitude residuals are added together to obtain model a and model b. (1) With model b (2) The total spatial residual, denoted as , and .
[0013] Furthermore, the screening steps for the first abnormal parameter also include: calculate Relative influence coefficient ; according to ,calculate Influence coefficient ; Calculate the influence coefficients of the remaining physical parameters; Arrange all influence coefficients corresponding to the robotic arm under inspection in ascending order to obtain the sequence. ~ ; Calculate sequence ~ The upper quartile of the parameter is used to identify the physical parameter whose influence coefficient is greater than the upper quartile. The part corresponding to the first abnormal parameter is taken as the target part; the normal parameter change curve of the target part corresponding to the first abnormal parameter is obtained and denoted as curve c; Based on the real-time parameter changes of the first abnormal parameter, construct the curve of the first abnormal parameter changing over time, denoted as curve d; By comparing the time spans of curves c and d, we can determine whether the cosine similarity between curves c and d meets the standard. If it does not meet the standard, it indicates that the target part is experiencing abnormal wear. If the cosine similarity meets the standard, it indicates that the target part is under normal wear. Check whether the mutual coupling and linkage between the parts conforms to the motion law set by the robotic arm. If it matches, then no action is taken; If it does not meet the requirements, the comprehensive drift score of physical parameters will be calculated from three dimensions: data statistics, wear matching degree, and coupling verification.
[0014] Furthermore, the calculation steps for the overall drift score are as follows: Obtain the standard regression slope of the univariate linear regression model corresponding to the first outlier parameter, and use it as the score of the first outlier parameter in the data statistics dimension, denoted as score a1; Obtain the influence coefficient corresponding to the first abnormal parameter, and use it as the score of the first abnormal parameter in the wear matching degree dimension, denoted as score a2; Calculate the score for the coupling verification dimension: Obtain curves c and d corresponding to the first anomaly parameter, unify the time span of curves c and d, and divide curves c and d into several equally spaced time intervals. Calculate the derivatives of curves c and d in each time interval, extract the sign of the highest-order term of curves c and d, and increment the coupling by 1 if the signs of the highest-order terms of the two curves are the same, otherwise increment it by zero. Calculate the coupling verification dimension score for the first anomaly parameter. The scores a1 to a3 corresponding to the first abnormal parameter are added together to obtain the comprehensive drift score of the first abnormal parameter.
[0015] Furthermore, the update steps for the initial dynamic model are as follows: Different drift tolerance thresholds are set, the comprehensive drift trend score of the first abnormal parameter is compared with the corresponding drift tolerance threshold, the first abnormal parameter whose comprehensive drift trend score exceeds the drift tolerance threshold is selected, and the second abnormal parameter is obtained. Determine whether the wear of the part corresponding to the second abnormal parameter is normal wear; If the wear is abnormal, record it; If it is normal wear, then update the initial dynamic model: Obtain the physical parameters corresponding to the second abnormal parameter as the physical parameters to be updated; Alternatively, take the standard regression slope corresponding to the second abnormal parameter. ; Set sampling time The robotic arm corresponding to the second abnormal parameter is taken as the quasi-robotic arm. During the sampling time, the actual position and actual posture of the quasi-robotic arm end are recorded. The reference position and reference posture of the quasi-robotic arm end are predicted using the initial dynamic model corresponding to the second abnormal parameter. Based on the actual position, actual posture, reference position, and reference posture of the quasi-robotic arm end effector within the sampling time, the correction offset of the second abnormal parameter is calculated. ; Add a correction offset to the second abnormal parameter corresponding to the quasi-robotic arm. Update the initial dynamic model corresponding to the second abnormal parameter.
[0016] Compared with the prior art, the beneficial effects of the present invention are: This invention distinguishes between dynamic drift and random noise disturbance: It divides the integrated spatial residual by scanning windows and then introduces the significance test of the Student's t-distribution to determine whether there is a continuously changing trend component within each scanning window. In the actual determination of the trend component, only those that meet the significance condition of the Student's t-distribution (i.e., the two-sided quantile corresponding to a 95% confidence level) are judged as drift trends. Even if there are some peak data within the scanning window, they will not be judged as trend components in a "one-size-fits-all" manner, reducing the probability of false detection and improving the adaptability of the detection method.
[0017] Achieving directional source tracing of drift parameters: This invention calculates the influence coefficient of each physical parameter on the spatial residual through central difference calculation, analyzes the strength of the influence of each physical parameter on the comprehensive spatial residual, and quantifies the possibility of drift for each parameter. At the same time, this invention also integrates data statistics, wear matching degree, and coupling verification to calculate the comprehensive drift trend score of each dynamic parameter, realizing quantitative source tracing of drift, with good physical interpretability, and providing a reference for the directional update of the subsequent initial dynamic model.
[0018] No full model update is required; only local numerical updates are performed on the physical parameters of abnormal wear. This invention does not perform full parameter retraining of the initial dynamic model. After confirming the second abnormal parameter in step S3, only the dynamic parameters corresponding to the second abnormal parameter are modified. The other parameters related to the robot arm's inertia, friction, transmission, and hysteresis remain unchanged, preserving the overall framework and calculation process of the model. Only local replacements are made at the parameter numerical level, resulting in a small amount of computation. The update can be completed online during the normal operation of the robot arm without stopping for calibration, balancing the correction effect and computational efficiency. Attached Figure Description
[0019] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a schematic diagram of the drift detection and directional update method of the present invention; Figure 2 This is a functional flowchart of step S2 of the present invention; Figure 3 This is a functional flowchart of steps S3 and S4 of the present invention. Detailed Implementation
[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0021] Please see Figure 1 Drift detection and update methods based on spatial intelligent pose residual trend decomposition include: Step S1: Obtain the transmission data of the robotic arm joints, and calculate the actual three-dimensional position and actual posture of the robotic arm end effector using "Forward Kinematics"; The "initial dynamics model" is used to predict the end-effector pose of the robotic arm; the actual value and the predicted value of the end-effector pose are compared, and the position residual and attitude residual are calculated respectively to obtain the comprehensive spatial residual; It should be noted that forward kinematics means: calculating the position and orientation of the end effector (such as a gripper or welding torch) in space based on the joint parameters (angle or displacement) of the robotic arm; The initial dynamic model represents the dynamic model obtained by the robotic arm during the factory debugging phase through calibration experiments and parameter identification training. This model stores the initial dynamic parameters of the robotic arm (e.g., joint friction coefficient, transmission ratio, response delay constant, etc.). The model is input with the current joint state and control commands of the robotic arm to predict the position and attitude (pose) of the end effector of the robotic arm in three-dimensional space at the next moment.
[0022] The specific steps of step S1 are as follows: The transmission data of the robotic arm joints is obtained in the form of DH parameters; It should be noted that DH parameters are data used to describe the spatial geometric relationship between two adjacent links in a robotic arm. Each pair of adjacent links corresponds to four parameters: link length, link torsion angle, joint offset, and joint rotation angle. The link length refers to the length of the common perpendicular of two adjacent joint axes. Linkage torsion angle represents the angle between two adjacent joint axes of rotation; Joint offset represents the offset between two adjacent links along the vertical axis (i.e., the Z-axis direction). Joint rotation angle refers to the rotation angle of the joint axis. Arrange all the joints of the robotic arm in order from the base of the robotic arm to the end of the robotic arm to obtain the joint sequence of the robotic arm; Calculate the actual three-dimensional position and actual orientation of the robotic arm's end effector: Based on the joint sequence in the robotic arm joint sequence, extract the 1st, 2nd, and so on, from the transmission data of the robotic arm joints. Link length of each joint , ~ ; Linkage torsion angle , ~ ; Joint offset , ~ ; Joint angle , ~ ;in, Indicates the number of joints in the robotic arm; Construct the homogeneous transformation matrix from the first joint to the second joint to obtain the matrix. : ; Construct the homogeneous transformation matrix from the second joint to the third joint to obtain the matrix. : ; in, and These represent the joint rotation angle and joint offset of the third joint, respectively. And so on, constructing the first The joint to the first The homogeneous transformation matrix of each joint is obtained to obtain the matrix. ; Obtain the link length from the robot arm base to the first joint. Torque of connecting rod Construct the homogeneous transformation matrix of the first joint to obtain the matrix. : ; matrix ,matrix Until the matrix Multiplying them sequentially yields the total homogeneous transformation matrix, denoted as matrix . ; In the matrix In the above, extract the elements from the first, second, and third rows of the fourth row to obtain the three-dimensional position vector of the robotic arm's end effector, denoted as . ; The shape is (1×3), namely the horizontal axis (X-axis), the vertical axis (Y-axis), and the vertical axis (Z-axis); This indicates the actual three-dimensional position of the robotic arm's end effector; In the matrix In the middle, extract the top left (3×3) submatrix to obtain the attitude rotation matrix of the robotic arm end effector, denoted as matrix. ; This indicates the actual posture of the robotic arm's end effector; Add timestamps to the actual 3D position and actual orientation of each robotic arm end effector; Based on the RNE model (i.e., Recursive Newton-Euler Algorithm), the motion changes of each joint of the robotic arm are simulated to obtain the initial dynamic model; The working process of the initial dynamic model is as follows: Obtain the motion trajectory programmed by the robotic arm and extract the range of motion of each joint in the robotic arm; During the actual movement of the robotic arm, the physical parameters of each joint are recorded. These physical parameters include: inertial parameters, friction parameters, and transmission parameters. Inertial parameters: link weight, link center of mass position, link product of inertia, and link moment of inertia relative to the center of mass (i.e., the principal moments of inertia of the link relative to the center of mass on the horizontal axis (X-axis), vertical axis (Y-axis), and vertical axis (Z-axis); Friction parameters: Coulomb coefficient of friction of the joint, viscous friction coefficient of the joint, and static friction coefficient of the joint; Transmission parameters: actual output torque and actual transmission ratio of the motor, moment of inertia of the motor, viscous damping of the motor rotor, and flexible equivalent stiffness of the joints. It should be noted that the physical parameters listed above are divided into directly measurable and non-directly measurable parts. Directly measurable parts include: connecting rod weight, moment of inertia of motor rotor, viscous damping of motor rotor, etc. Non-directly measurable parts include: connecting rod inertia product, moment of inertia, friction parameters, etc. The angles and angular accelerations of each joint of the robotic arm are acquired in real time during actual activities, and the parameters of the physical parameters that cannot be directly measured are fitted by the dynamic equations. The dynamic equations are as follows: ; in, This represents the total torque of the joint. The inertial torque of the joint is calculated using forward kinematics based on the weight of the link, the position of the link's center of mass, the product of inertia of the link, the moment of inertia, and the angle of the joint. This represents the Coriolis eccentric torque of the joint, which is generated by doing... The partial derivatives of each element with respect to the angle of each joint are obtained by combining the angular velocity of each joint with the Christofer symbolic formula. This represents the gravitational torque of the joint, which is calculated from the weight of the link, the position of the link's center of mass, the gravitational acceleration, and the angle of the joint. The gravitational potential energy corresponding to each joint is calculated, and the negative gradient of the gravitational potential energy of each joint with respect to the joint angle is calculated. This represents the frictional torque of the joint, which is calculated using the Stribeck friction model based on the joint's Coulomb friction coefficient, viscous friction coefficient, and static friction coefficient. This represents the total torque of the joint. This torque is calculated by converting the actual output torque of the motor and the moment of inertia of the motor according to the actual transmission ratio of the motor, and then superimposing the elastic torque calculated by the viscous damping of the motor rotor and the flexible equivalent stiffness of the joint. The elastic torque is calculated as follows: based on the actual rotation angle of the motor and the actual rotation angle of the joint per unit time, calculate the angle difference between the motor and the joint, and multiply the angle difference between the motor and the joint by the flexible equivalent stiffness of the joint to obtain the stiffness term torque. Based on the actual angular velocity of the motor and the actual angular velocity of the joint per unit time, calculate the angular velocity difference between the motor and the joint, and multiply the angular velocity difference between the motor and the joint by the viscous damping of the motor rotor to obtain the damping torque. Adding the stiffness term torque to the damping term torque yields the elastic torque; Substitute the physical parameters calculated from the dynamic equations into the recursive Newton-Euler dynamic model to output the output torque of the robotic arm; Set hysteresis parameters Lag parameters Represents: the response delay constant of the joint, which represents the lag time from the issuance of the control command to the actual generation of torque by the joint, i.e., the detection object for drift testing in this invention; Based on lag class parameters Given the output torque of the robotic arm, calculate the effective output torque of the robotic arm; based on the effective output torque of the robotic arm, establish a system of differential equations including the differential equations of motor rotor motion (representing the relationship between the motor rotor angle and time) and the differential equations of link rigid body motion (representing the relationship between the link angle of the robotic arm joint and time), and calculate the angles of each joint in the robotic arm; input the angles of each joint of the robotic arm into the DH forward kinematics model, calculate the total homogeneous transformation matrix from the base to the end effector, and extract the predicted position and predicted attitude of the end effector corresponding to the initial dynamic model; Add timestamps to the predicted position and orientation of the robotic arm end effector. Each time the initial dynamic model is used to predict the position and orientation of the robotic arm end effector, the real-time physical parameters of the robotic arm are recorded. These parameters are used in the subsequent step S2 to determine whether the parameter drift of the part belongs to normal wear. Calculate the residuals between the position and attitude of the robotic arm's end effector to obtain the comprehensive spatial residuals; Based on the timestamp, summarize all the actual 3D positions, actual postures, predicted positions, and predicted postures of the robotic arm end effector. Calculate the Euclidean distance between all the actual 3D positions and all the predicted positions of the robotic arm end effector to obtain the position residual of the robotic arm end effector. The attitude residuals at the end of the robotic arm are calculated using the geodesic distance of the SO(3) manifold. The actual posture corresponding to the end effector of the robotic arm is denoted as... Predicted posture is recorded as ; Calculate the attitude residual rotation matrix : Where T represents transpose; Extracting the matrix The main diagonal elements are used to calculate the matrix. traces ;Will Substitute into the trace identity and calculate the matrix. The matrix is obtained by taking the cosine of the rotation angle. rotation angle ; Constructing a matrix rotation angle The Rodriguez rotation formula is as follows: Where T denotes the transpose, and represents the (3×3) identity matrix. express The antisymmetric matrix; Indicates the unit axis of rotation; According to the rotation angle ,calculate : ; Will , as well as Substituting into the Rodriguez rotation formula, calculate the matrix. unit axis of rotation ; according to and For the matrix Perform a logarithmic mapping and calculate the matrix. rotation vector ; calculate The vector 2 norm is used to obtain the attitude residual of the robotic arm's end effector; Based on the 3σ principle, outliers in the position and attitude residuals of the robotic arm end are screened and removed to obtain the comprehensive spatial residual.
[0023] Step S2: Perform mathematical decomposition on the integrated spatial residual to distinguish between temporary jitter noise and drift trend; select the robotic arm with drift trend as the robotic arm to be inspected, quantify the influence of each physical parameter in the "initial dynamic model" of the robotic arm to be inspected on the integrated spatial residual, screen the physical parameters that affect the integrated spatial residual, and obtain the first abnormal parameter; By comparing the parameter changes of the part corresponding to the first abnormal parameter with the theoretical wear changes of the part, it can be determined whether the parameter drift of the part belongs to normal wear. If the wear and tear is not normal, record it and trigger an alarm; If it is normal wear and tear, check whether the mutual coupling and linkage between the parts conforms to the motion law set by the robotic arm. If it meets the criteria, no action is taken; if it does not meet the criteria, the comprehensive drift score of the first abnormal parameter is calculated from three dimensions: data statistics, wear matching degree, and coupling verification. Please see Figure 2 The specific steps of step S2 are as follows: In the integrated spatial residuals, distinguish between temporary jitter noise and drift trend in the location residuals: Arrange all position residuals in chronological order of timestamps to obtain the position residual sequence; Set the sliding window size for the location residual sequence. ,by To extract samples of a fixed length from the positional residual sequence, we obtain the first to the second... The position residual of each timestamp ~ ;in, Indicates the start timestamp of the location residual sequence; Position residual ~ As the first position residual sample; Establish ~ The univariate linear regression model decomposes the residuals into three parts: the baseline mean term, the linear trend term, and the random noise term. ;in, This represents the position residual corresponding to the i-th timestamp, where i represents the time sequence number of the timestamp, and the value of i ranges from 1 to... ; This represents the regression intercept, corresponding to the baseline mean term; This represents the regression slope, corresponding to the linear trend term; This represents the random noise corresponding to the i-th timestamp, and the corresponding random noise term. Without including random noise, the least squares method is used to fit the regression intercept. With regression slope Minimize samples ~ The sum of squared residuals is used to obtain the standard regression intercept. With standard regression slope ; use and Fitting from the 1st to the 2nd Random noise of timestamps 1 to 2, making the 1st to 3rd timestamps 1st to The random noise at each timestamp follows a normal distribution with zero mean, yielding the first to the second timestamps. Standard random noise of timestamps ~ ; by For degrees of freedom, calculate standard random noise. ~ The standard deviation is obtained. ; calculate arrive average time Calculate the standard regression slope Sampling standard error : ; Calculate the test statistic for the residual sample at the first position. : ; Based on test statistic Constructing degrees of freedom The t-distribution of students; Set a target confidence level, and extract the two-sided quantile corresponding to the target confidence level (usually set to 95%) from the Student t-distribution. Use this quantile as the drift threshold for the first position residual sample, denoted as the threshold value. ; Comparing absolute values With threshold Size: like This indicates that the residual sample at the first position has a drifting trend; like If the result is negative, it indicates that the residual sample at the first position does not have a drift trend and belongs to temporary jitter noise, so it is not processed. Repeat the steps to distinguish temporary jitter noise or drift trend of the first position residual sample, distinguish other position residuals and attitude residuals in the comprehensive spatial residual, and filter out temporary jitter noise and drift trend; Robotic arms exhibiting a drift trend are selected as the robotic arms to be inspected (i.e., the position residual or attitude residual corresponding to the robotic arm to be inspected exhibits a drift trend, or both the position residual and attitude residual corresponding to the robotic arm to be inspected exhibit a drift trend). Extract the physical parameters from the initial dynamic model corresponding to the robot arm under test, quantify the influence of each physical parameter in the initial dynamic model on the comprehensive spatial residual, and screen the first abnormal parameter; Obtain all physical parameters from the initial dynamic model corresponding to the robotic arm under test, denoted as... ~ (i.e., the 1st to the 2nd) (one physical parameter); among which, This indicates the number of physical parameters in the initial dynamic model corresponding to the robotic arm under inspection; Calculate the influence coefficient of the first physical parameter on the overall spatial residual. ; Backup the initial dynamic model corresponding to the robotic arm to be inspected to obtain model a and model b; set up positive disturbance With reverse disturbance ; and The absolute values are the same, but the signs are opposite; Backup model b to obtain model b (1) With model b (2) The positive disturbances will be respectively With reverse disturbance Add model b (1) With model b (2) In model a, no positive perturbation is added. With reverse disturbance ; Get current time To obtain the end effector of the robotic arm under inspection The actual position and orientation at time t, using model a and model b. (1) With model b (2) Predict the end effector of the robotic arm under inspection respectively. Position and posture at any moment; Obtain the end effector of the robotic arm under inspection The actual position and attitude at time t are used to calculate the values for model a and model b, respectively. (1) With model b (2) The corresponding position residuals and attitude residuals; model a and model b respectively. (1) With model b (2) The corresponding position residuals and attitude residuals are added together to obtain model a and model b. (1) With model b (2) The total spatial residual, denoted as , and ; Calculate physical parameters Local influence coefficient : ; Calculate physical parameters Relative influence coefficient : ; Based on the chain rule, according to Calculate physical parameters The contribution percentage of the regression slope to the robotic arm under inspection is used to obtain physical parameters. Influence coefficient ; repeat The calculation process involves calculating the influence coefficients of the remaining physical parameters in the initial dynamic model corresponding to the robotic arm under inspection. Summarize all influence coefficients corresponding to the robotic arm under inspection ~ ,Will ~ Sort in ascending order to obtain the sequence. ~ ; Using the IQR quartile method, calculate the sequence ~ The upper quartile of the parameter is used to identify the physical parameter whose influence coefficient is greater than the upper quartile. The part corresponding to the first abnormal parameter is taken as the target part; the actual working temperature of the robotic arm is obtained as the standard temperature; the physical parameter type corresponding to the first abnormal parameter is taken as the target type. Obtain the target part and the corresponding parameter change curve of the target type at standard temperature, denoted as curve c. The horizontal axis of the curve represents the usage time of the target part, and the vertical axis represents the corresponding parameter change of the target type. Based on the real-time parameter changes of the first abnormal parameter of the robotic arm under inspection during operation, a curve of the first abnormal parameter changing over time is constructed and denoted as curve d; the horizontal and vertical axes of curve d and curve c are the same. The time span of curve c and curve d is unified, and the cosine similarity between curve c and curve d is calculated. If the cosine similarity is less than 0.5 (cosine similarity less than 0.5 indicates that the cosine similarity between curve c and curve d does not meet the standard), it means that the part corresponding to the first abnormal parameter is abnormally worn. The position of the target part is recorded, and the relevant technicians of the robotic arm are notified to carry out maintenance and an alarm is triggered. If the cosine similarity is greater than or equal to 0.5 (cosine similarity greater than or equal to 0.5 indicates that the cosine similarity between curve c and curve d meets the standard), it means that the part corresponding to the first abnormal parameter is under normal wear. Check whether the mutual coupling and linkage between the parts conforms to the motion law set by the robotic arm. If it matches, then no action is taken; If it does not meet the requirements, the comprehensive drift score of physical parameters will be calculated from three dimensions: data statistics, wear matching degree, and coupling verification. The calculation process for the overall drift score is as follows: Obtain the standard regression slope of the univariate linear regression model corresponding to the first outlier parameter, and use it as the score of the first outlier parameter in the data statistics dimension, denoted as score a1; Obtain the influence coefficient corresponding to the first abnormal parameter, and use it as the score of the first abnormal parameter in the wear matching degree dimension, denoted as score a2; Calculate the score for the coupling verification dimension: Obtain curves c and d corresponding to the first anomaly parameter, unify the time span of curves c and d, and divide curves c and d into several equally spaced time intervals. Calculate the derivatives of curves c and d in each time interval, extract the sign of the highest-order term of curves c and d, and increment the coupling by 1 if the signs of the highest-order terms of the two curves are the same, otherwise increment it by zero. Calculate the coupling verification dimension score for the first anomaly parameter. Divide the curve c Equal time intervals (that is, dividing the horizontal axis of curve c into 12 equal time intervals) (parts), divided in curve d There are equidistant time intervals, and the number of time intervals and the span of the time intervals are the same for curves c and d. Calculate the temporary score for the first time interval. : Calculate the derivative of curve c in the first time interval, denoted as derivative c. (1) Calculate the derivative of curve d in the first time interval, denoted as derivative d. (1) ; Extracting the derivative function c (1) With derivative d (1) The sign of the highest-order term (i.e., the positive or negative sign of the highest-order term) is used to determine whether the two signs are the same. If they are the same, then... The value is 1; if they are different, then The value is zero; Repeated determination The steps for corresponding values are to determine the 2nd to the 3rd. Temporary scores corresponding to each time interval ~ ; Calculated average ~ The score of the first anomaly parameter in the coupling verification dimension is denoted as score a3. The scores a1 to a3 corresponding to the first abnormal parameter are added together to obtain the comprehensive drift score of the first abnormal parameter.
[0024] Step S3: Set different drift tolerance thresholds according to the importance of the robotic arm task (precision assembly / normal assembly); The comprehensive drift trend score of the first abnormal parameter is compared with the corresponding drift tolerance threshold. The first abnormal parameter with a comprehensive drift trend score exceeding the drift tolerance threshold is selected to obtain the second abnormal parameter. Based on the theoretical wear changes of each part, determine whether the wear corresponding to the second abnormal parameter is normal wear; If it is normal wear, then in the "Initial Dynamics Model", modify the corresponding parameters of the part that is drifting and update the initial dynamics model; If the wear is not normal, record it and trigger an alarm; Please see Figure 3 The specific steps of step S3 are as follows: Different drift tolerance thresholds are set according to the importance of the robotic arm task (precision assembly / normal assembly); The comprehensive drift trend score of the first abnormal parameter is compared with the corresponding drift tolerance threshold. The first abnormal parameter with a comprehensive drift trend score exceeding the drift tolerance threshold is selected to obtain the second abnormal parameter. Repeat step S2 to determine whether the part corresponding to the first abnormal parameter is normal wear, and determine whether the wear corresponding to the second abnormal parameter is normal wear; If it is normal wear, then in the "Initial Dynamics Model", modify the corresponding parameters of the part that is drifting and update the initial dynamics model; If abnormal wear occurs, record it and trigger an alarm; The steps to update the initial dynamic model are as follows: Obtain the physical parameters corresponding to the second abnormal parameter as the physical parameters to be updated; Obtain the univariate linear regression model corresponding to the second outlier parameter, and extract the standard regression slope corresponding to the second outlier parameter. ; Set sampling time The robotic arm corresponding to the second abnormal parameter is taken as the quasi-robotic arm, and the actual position and actual posture of the end of the quasi-robotic arm are recorded during the sampling time. During the sampling time, the initial dynamic model corresponding to the second abnormal parameter is used synchronously to record the real-time changes of the second abnormal parameter and predict the position and attitude of the quasi-robotic arm end effector, so as to obtain the reference position and reference attitude of the quasi-robotic arm end effector during the sampling time. Based on the actual position, actual posture, reference position and reference posture of the quasi-robotic arm end effector during the sampling time, calculate the total spatial residual corresponding to the second abnormal parameter (i.e. (actual position - reference position) + (actual posture - reference posture)). The sampling time is divided into several sub-time intervals. Within each sub-time interval, the change in the total spatial residual and the change in the second anomaly parameter are calculated. Within each sub-time interval, the local partial derivative of the change in the total spatial residual with respect to the change in the second anomaly parameter (i.e., change in total spatial residual / change in the second anomaly parameter) is calculated. The average of the local partial derivatives across all sub-time intervals is calculated to obtain the total partial derivative of the total spatial residual with respect to the second anomaly parameter, denoted as […]. ; Obtain the second outlier parameter with respect to the standard regression slope percentage Calculate the correction offset of the second anomaly parameter. : ; Add a correction offset to the second abnormal parameter corresponding to the quasi-robotic arm. Update the initial dynamic model corresponding to the second abnormal parameter.
[0025] Step S4: Using a brand new dataset that has not been involved in parameter tuning (i.e., a dataset that has not undergone the processing of steps S1 to S3 above), the drift judgment rules of steps S1 to S3 are followed to test the unupdated initial dynamic model and the updated initial dynamic model to check whether the drift trend has been eliminated (i.e., the total spatial residual of the "updated initial dynamic model" is reduced relative to the total spatial residual of the "updated initial dynamic model", and the comprehensive drift trend score of all physical parameters of the "updated initial dynamic model" does not exceed the drift tolerance threshold corresponding to each physical parameter). If eliminated, the entire process of drift judgment and physical parameter modification in steps S1 to S3 is archived as a reference for the next adjustment of the initial dynamic model or robotic arm. If not eliminated, reset the sampling time. Continue adjusting the physical parameters in the initial dynamic model until the drift trend is eliminated.
[0026] Compared to the problems of noise false detection, lack of source tracing, residual coupling, full retraining, fault masking and lack of closed-loop tracing in traditional robotic arm dynamic residual monitoring, this invention adopts a spatial intelligent pose residual trend decomposition and statistical significance verification mechanism to replace the single threshold residual judgment method. By using least squares trend decomposition and T test, it distinguishes between random noise and dynamic drift, effectively solving the problems of traditional solutions being susceptible to working condition disturbances and serious false alarms and false detections, and improving the statistical reliability and industrial scene adaptability of drift recognition. Furthermore, this invention relies on finite difference quantification to assess the impact of various physical parameters on the end-effector position and attitude residuals of the robotic arm. It integrates multi-dimensional evaluation indicators such as statistical significance, physical wear matching degree, and mechanical coupling consistency to construct a comprehensive parameter drift score, enabling directional tracing of drift parameters and overcoming the shortcomings of traditional techniques that cannot locate the source of residuals and have low investigation efficiency. At the same time, this invention establishes a decoupled identification logic for geometric kinematic residuals and dynamic drift residuals, performing local directional numerical updates only on drift parameters that exceed the standard due to normal wear. This achieves online lightweight model updates while retaining the initial high-precision calibration parameters of the equipment, avoiding model contamination and accuracy degradation.
[0027] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to specific implementation methods. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The present invention is limited only to the technical solutions and their full scope and equivalents.
Claims
1. A drift detection and update method based on spatially intelligent pose residual trend decomposition, characterized in that, The method includes: Acquire transmission data of the robotic arm joints, calculate the actual three-dimensional position and actual posture of the robotic arm end effector; use the initial dynamic model to predict the end effector pose of the robotic arm, and calculate the comprehensive spatial residual; Mathematically decompose the residual of the integrated space, screen out the robotic arms with drift trend as the robotic arms to be inspected, analyze the residual of the physical parameters of the robotic arms to be inspected with respect to the integrated space, and screen out the first abnormal parameter. Compare the parameter changes of the part corresponding to the first abnormal parameter with the theoretical wear changes of the part to determine whether the parameter drift of the part belongs to normal wear; if not, an alarm is triggered. If so, check whether the linkage between parts is compliant; If compliant, no action is taken; if non-compliant, a comprehensive drift score for the first anomaly parameter is calculated. Set different drift tolerance thresholds; compare the comprehensive drift trend score of the first abnormal parameter with the drift tolerance threshold to filter the second abnormal parameter; Based on the theoretical wear changes of each part, determine whether the wear of the part corresponding to the second abnormal parameter is normal wear; if it is normal wear, update the initial dynamic model. If the wear and tear is not normal, an alarm will be triggered; Test the unupdated initial dynamic model and the updated initial dynamic model to check whether the drift trend has been eliminated; if it has been eliminated, archive the entire process of drift judgment and physical parameter modification. If the problem persists, continue adjusting.
2. The drift detection and update method based on spatially intelligent pose residual trend decomposition according to claim 1, characterized in that, The calculation steps for the actual 3D position and actual attitude are as follows: Obtain transmission data of the robotic arm joints; Arrange the robot arm joint sequence, and extract the link length, link torsion angle, joint offset and joint rotation angle of all joints from the transmission data of the robot arm joints according to the joint order in the robot arm joint sequence. Construct homogeneous transformation matrices of the first joint to the second joint, the second joint to the third joint, and so on until the nth joint to the (n+1)th joint, to obtain matrices to matrices ; Obtain the link length and torsion angle from the robot arm base to the first joint, construct the homogeneous transformation matrix of the first joint, and obtain the matrix. ; Multiply matrices in sequence To matrix The total homogeneous transformation matrix is obtained. ; In the matrix Extracting the elements from the first, second, and third rows of the fourth row yields the actual three-dimensional position of the robotic arm's end effector. ; In the matrix In the middle, extract the top left (3×3) submatrix to obtain the actual pose of the robotic arm's end effector. ; Add timestamps to the actual 3D position and actual orientation of each robotic arm end effector; Simulate the motion changes of each joint of the robotic arm and construct an initial dynamic model; Based on the initial dynamic model, the integrated spatial residual is calculated.
3. The drift detection and update method based on spatial intelligent pose residual trend decomposition according to claim 2, characterized in that, The calculation steps for the integrated spatial residual are as follows: Calculate the output torque of the robotic arm using the initial dynamics model; Set hysteresis parameters According to the lag class parameters Given the output torque of the robotic arm, calculate the effective output torque of the robotic arm; Calculate the angles of each joint in the robotic arm based on the effective output torque of the robotic arm; Based on the angles of each joint of the robotic arm, the total homogeneous transformation matrix from the base to the end effector is calculated, and the predicted position and predicted posture of the end effector corresponding to the initial dynamic model are extracted. Add timestamps to the predicted position and orientation of the robotic arm's end effector to record the robotic arm's real-time physical parameters; Calculate the residuals between the position and attitude of the robotic arm's end effector to obtain the comprehensive spatial residuals; Based on the timestamp, summarize all the actual 3D positions, actual postures, predicted positions, and predicted postures of the robotic arm end effector. Calculate the Euclidean distance between all the actual 3D positions and all the predicted positions of the robotic arm end effector to obtain the position residual of the robotic arm end effector. Calculate the attitude residual at the end of the robotic arm; Based on the 3σ principle, outliers in the position and attitude residuals of the robotic arm end are screened and removed to obtain the comprehensive spatial residual. Mathematical decomposition of the integrated spatial residuals is used to screen out robotic arms to be inspected.
4. The drift detection and update method based on spatial intelligent pose residual trend decomposition according to claim 3, characterized in that, The steps for calculating the attitude residuals are as follows: The actual posture corresponding to the end effector of the robotic arm is denoted as... Predicted posture is recorded as ; Calculate the attitude residual rotation matrix Calculate the matrix traces ;according to Calculate matrix The rotation angle is obtained by taking the cosine of the corresponding rotation angle. ; According to the rotation angle Calculate the antisymmetric matrix ; Constructing a matrix rotation angle Rodriguez rotation formula, , as well as Substituting into the Rodriguez rotation formula, calculate the matrix. unit axis of rotation ; For matrix Perform logarithmic mapping and calculate the matrix. rotation vector ; calculate The vector 2 norm is used to obtain the attitude residual of the robotic arm end effector.
5. The drift detection and update method based on spatial intelligent pose residual trend decomposition according to claim 3, characterized in that, The screening steps for the robotic arms to be inspected are as follows: Arrange all position residuals in chronological order of timestamps to obtain the position residual sequence; Set the sliding window size for the location residual sequence. The first position residual sample is obtained by extracting a sample from the position residual sequence. A linear regression model for the residual samples at the first position is established, and the residuals are decomposed into three parts: the baseline mean, the linear trend, and the random noise. Without including random noise, the least squares method is used to fit the regression intercept. With regression slope The standard regression intercept was obtained. With standard regression slope ; use and Fitting from the 1st to the 2nd Random noise of timestamps 1 to 2, making the 1st to 3rd timestamps 1st to Random noise with timestamps follows a normal distribution with zero mean, thus obtaining standard random noise; by For degrees of freedom, calculate the standard deviation of the standard random noise, and obtain... ; Calculating the average time of the first position residual sample Calculate the standard regression slope Sampling standard error .
6. The drift detection and update method based on spatial intelligent pose residual trend decomposition according to claim 5, characterized in that, The screening process for the robotic arms to be inspected also includes: Calculate the test statistic for the residual sample at the first position. ; Based on test statistic Constructing degrees of freedom Using the student t-distribution, a target confidence level is set. The two-sided quantiles corresponding to the target confidence level are extracted from the student t-distribution to obtain the drift threshold for the residual sample at the first position. ; Comparing absolute values With threshold Size: like This indicates that the residual sample at the first position has a drifting trend; like If the result is negative, it indicates that the residual sample at the first position does not have a drift trend and belongs to temporary jitter noise, so it is not processed. Distinguish between other position residuals and attitude residuals in the integrated spatial residuals, and filter out temporary jitter noise and drift trends; The robotic arm exhibiting a drift trend is selected as the robotic arm to be inspected. The physical parameters in the initial dynamic model corresponding to the robotic arm to be inspected are extracted, the influence of each physical parameter on the comprehensive spatial residual is quantified, and the first abnormal parameter is screened.
7. The drift detection and update method based on spatial intelligent pose residual trend decomposition according to claim 6, characterized in that, The steps for filtering the first abnormal parameter are as follows: Obtain all physical parameters from the initial dynamic model corresponding to the robotic arm under inspection. ~ ;in, Indicates the number of physical parameters; Calculate the influence coefficient of the first physical parameter on the overall spatial residual. ; Backup the initial dynamic model corresponding to the robotic arm to be inspected to obtain model a and model b; set up positive disturbance With reverse disturbance Backup model b to obtain model b (1) With model b (2) The positive disturbances will be respectively With reverse disturbance Add model b (1) With model b (2) middle; Get current time According to the end of the robotic arm under inspection Time and The position and attitude changes at each moment are calculated using model a and model b. (1) With model b (2) The corresponding position residuals and attitude residuals; model a and model b respectively. (1) With model b (2) The corresponding position residuals and attitude residuals are added together to obtain model a and model b. (1) With model b (2) The total spatial residual, denoted as , and .
8. The drift detection and update method based on spatial intelligent pose residual trend decomposition according to claim 7, characterized in that, The first step in filtering abnormal parameters also includes: calculate Relative influence coefficient ; according to ,calculate Influence coefficient ; Calculate the influence coefficients of the remaining physical parameters; Arrange all influence coefficients corresponding to the robotic arm under inspection in ascending order to obtain the sequence. ~ ; Calculate sequence ~ The upper quartile of the parameter is used to identify the physical parameter whose influence coefficient is greater than the upper quartile. The part corresponding to the first abnormal parameter is taken as the target part; the normal parameter change curve of the target part corresponding to the first abnormal parameter is obtained and denoted as curve c; Based on the real-time parameter changes of the first abnormal parameter, construct the curve of the first abnormal parameter changing over time, denoted as curve d; By comparing the time spans of curves c and d, we can determine whether the cosine similarity between curves c and d meets the standard. If it does not meet the standard, it indicates that the target part is experiencing abnormal wear. If the cosine similarity meets the standard, it indicates that the target part is under normal wear. Check whether the mutual coupling and linkage between the parts conforms to the motion law set by the robotic arm. If it matches, then no action is taken; If it does not meet the requirements, the comprehensive drift score of physical parameters will be calculated from three dimensions: data statistics, wear matching degree, and coupling verification.
9. The drift detection and update method based on spatial intelligent pose residual trend decomposition according to claim 8, characterized in that, The calculation steps for the overall drift score are as follows: Obtain the standard regression slope of the univariate linear regression model corresponding to the first outlier parameter, and use it as the score of the first outlier parameter in the data statistics dimension, denoted as score a1; Obtain the influence coefficient corresponding to the first abnormal parameter, and use it as the score of the first abnormal parameter in the wear matching degree dimension, denoted as score a2; Calculate the score for the coupling verification dimension: Obtain curves c and d corresponding to the first anomaly parameter, unify the time span of curves c and d, and divide curves c and d into several equally spaced time intervals. Calculate the derivatives of curves c and d in each time interval, extract the sign of the highest-order term of curves c and d, and increment the coupling by 1 if the signs of the highest-order terms of the two curves are the same, otherwise increment it by zero. Calculate the coupling verification dimension score for the first anomaly parameter. The scores a1 to a3 corresponding to the first abnormal parameter are added together to obtain the comprehensive drift score of the first abnormal parameter.
10. The drift detection and update method based on spatial intelligent pose residual trend decomposition according to claim 1, characterized in that, The steps for updating the initial dynamic model are as follows: Different drift tolerance thresholds are set, the comprehensive drift trend score of the first abnormal parameter is compared with the corresponding drift tolerance threshold, the first abnormal parameter whose comprehensive drift trend score exceeds the drift tolerance threshold is selected, and the second abnormal parameter is obtained. Determine whether the wear of the part corresponding to the second abnormal parameter is normal wear; If the wear is abnormal, record it; If it is normal wear, then update the initial dynamic model: Obtain the physical parameters corresponding to the second abnormal parameter as the physical parameters to be updated; Alternatively, take the standard regression slope corresponding to the second abnormal parameter. ; Set sampling time The robotic arm corresponding to the second abnormal parameter is taken as the quasi-robotic arm. During the sampling time, the actual position and actual posture of the quasi-robotic arm end are recorded. The reference position and reference posture of the quasi-robotic arm end are predicted using the initial dynamic model corresponding to the second abnormal parameter. Based on the actual position, actual posture, reference position, and reference posture of the quasi-robotic arm end effector within the sampling time, the correction offset of the second abnormal parameter is calculated. ; The correction offset is superimposed on the second abnormal parameter corresponding to the quasi-robotic arm. Update the initial dynamic model corresponding to the second abnormal parameter.
