Multi-working-condition robot joint anomaly detection method and system

Abstract

The present application relates to the technical field of robot anomaly detection, and relates to a multi-working-condition robot joint anomaly detection method and system, comprising a working condition data acquisition module, a coupling matrix generation module, a feature data generation module and a joint anomaly discrimination module; from the perspective of multi-physical signal fusion and dynamics modeling, the limitations of traditional single signal detection being susceptible to coupling interference are overcome, the accuracy of abnormal joint identification and classification is significantly improved, potential faults are warned in advance, and the risk of unplanned downtime is reduced; on the other hand, through decoupling analysis and a working condition adaptive threshold strategy, the adaptability of the detection system to robot variable working conditions and variable loads is enhanced, the reliability and scientificity of the detection process are ensured, so that the robot operation safety, maintenance efficiency and service life are improved, and reliable data support and decision basis for realizing state-based intelligent maintenance and health management are provided.

Description

Technical Field

[0001] This invention belongs to the field of robot anomaly detection technology, and relates to a method and system for detecting joint anomalies in robots under multiple working conditions. Background Technology

[0002] Existing joint anomaly detection methods typically analyze each joint as an independent unit, employing a "single-joint, single-model" detection architecture. However, in real-world "multi-condition" operating environments, robots are constantly in motion, with their spatial posture continuously changing. For serial industrial robots (such as six-axis robotic arms), the movement of one joint is transmitted to other joints through mechanical linkages, forming cross-joint dynamic coupling. When the detection system indicates an anomaly in joint J3, this abnormal signal may actually originate from vibration of joint J2 transmitted through the mechanical structure, or from system-wide resonance induced under specific postures, rather than a fault in joint J3 itself.

[0003] There are still some areas for optimization in the current robot anomaly detection, specifically in the following aspects:

[0004] As the robot's posture changes from a contracted state to an extended state, the rotational inertia matrix and structural stiffness matrix of the entire robot change significantly. This means that the system's natural frequency dynamically drifts with the pose, making it difficult for a fixed-parameter detection model to adapt to the dynamic characteristics across the entire operating range.

[0005] The transmission efficiency of abnormal vibration signals in mechanical structures changes under different operating conditions. For example, in a posture close to a kinematic singularity, a weak fault vibration generated by a joint may be amplified by the robotic arm structure and transmitted to the end joint; while in a folded posture, the same vibration signal may be significantly attenuated, resulting in inconsistent signal representation. Summary of the Invention

[0006] In view of the problems existing in the prior art, the present invention provides a method and system for detecting joint anomalies in robots under multiple working conditions, in order to solve the above-mentioned technical problems.

[0007] To achieve the above and other objectives, the technical solution adopted by the present invention is as follows:

[0008] This invention provides a method for detecting joint anomalies in a robot under multiple working conditions, the method comprising the following steps:

[0009] Step S1: Acquire real-time multi-dimensional sensing data of each joint of the robot. The multi-dimensional sensing data includes joint motor current, joint vibration signal and joint encoder position data; perform forward kinematics calculation of the robot based on the joint encoder position data, construct the robot spatial link posture model in real time, and generate real-time working condition data of the robot containing the current spatial pose features.

[0010] Step S2: Perform dynamic analysis of the robot's variable inertia and stiffness based on the robot's real-time working condition data, and construct a spatial pose vibration transmission model that dynamically drifts with the pose; use the spatial pose vibration transmission model to quantify the vibration wave transmission efficiency between adjacent joints of the robot, and generate the joint vibration mode coupling coefficient matrix corresponding to the current working condition.

[0011] Step S3: Use the joint vibration mode coupling coefficient matrix to decouple the joint vibration signal from structural transmission noise, remove the coupled vibration components transmitted from other joints to the target joint through mechanical links, extract the intrinsic vibration feature vector characterizing the health status of the target joint, and combine it with the joint motor current to generate multidimensional decoupled joint feature data.

[0012] Step S4: Based on the robot's real-time operating status data, call the preset dynamic operating condition benchmark threshold library, use the dynamic operating condition benchmark threshold library to perform adaptive anomaly discrimination on the multi-dimensional decoupled joint feature data, locate the specific joint where the anomaly occurs and identify the anomaly type, thereby generating a robot joint anomaly detection result report.

[0013] Another aspect of the present invention provides a multi-condition robot joint anomaly detection system, the system comprising:

[0014] Working condition data acquisition module: acquires real-time multi-dimensional sensing data of each joint of the robot; performs forward kinematics calculation of the robot based on the joint encoder position data, constructs the robot spatial link posture model in real time, and generates real-time working condition status data of the robot;

[0015] Coupling matrix generation module: Based on the robot's real-time working condition data, perform dynamic analysis of the whole machine's variable inertia and variable stiffness, and construct a spatial pose vibration transmission model that dynamically drifts with the pose; use the spatial pose vibration transmission model to quantify the vibration wave transmission efficiency between adjacent joints of the robot, and generate the joint vibration mode coupling coefficient matrix corresponding to the current working condition.

[0016] Feature data generation module: The joint vibration signal is decoupled from the structural transmission noise by using the inter-joint vibration mode coupling coefficient matrix, the intrinsic vibration feature vector characterizing the health status of the target joint body is extracted, and multi-dimensional decoupled joint feature data is generated by combining the joint motor current.

[0017] Joint anomaly detection module: Based on the robot's real-time operating status data, it calls the preset dynamic operating condition benchmark threshold library to perform adaptive anomaly detection on multi-dimensional decoupled joint feature data, locate the specific joint where the anomaly occurs and identify the anomaly type, thereby generating a robot joint anomaly detection result report.

[0018] As described above, the multi-condition robot joint anomaly detection method and system provided by the present invention have at least the following beneficial effects:

[0019] (1) The multi-condition robot joint anomaly detection method and system provided by the present invention collects real-time multi-dimensional sensing information of each joint during robot operation, integrates joint motor current, joint vibration signal and joint encoder position data, and calculates the forward kinematics of the robot in real time based on the encoder position data to construct its spatial link posture model, thereby generating real-time working condition state data of the robot that reflects the current posture characteristics, effectively realizing multi-dimensional synchronous perception and digital modeling of the robot's operating state; on this basis, dynamic analysis of the whole machine's variable inertia and variable stiffness is carried out based on the real-time working condition state data, and a spatial posture vibration transmission model that changes dynamically with posture is established, thereby quantitatively evaluating the vibration wave transmission efficiency between adjacent joints and generating a joint vibration model that fits the current working condition. The system employs a mode coupling coefficient matrix to reveal the transmission law and coupling mechanism of vibration energy during the dynamic operation of the robot. It then uses this matrix to decouple the original joint vibration signal from structural transmission noise, filtering out cross-joint coupled vibration components transmitted to the target joint via mechanical links. This extracts intrinsic vibration feature vectors that directly characterize the joint's health state, and combines this with joint motor current data to generate a multi-dimensional decoupled joint feature dataset, significantly improving the independence and interpretability of the state features. Furthermore, by combining real-time robot operating condition data and dynamically calling a preset dynamic operating condition benchmark threshold library, the system adaptively identifies anomalies in the multi-dimensional decoupled joint features, accurately locates abnormal joints, identifies their anomaly types, and finally outputs a robot joint anomaly detection result report.

[0020] (2) On the one hand, this invention overcomes the limitations of traditional single signal detection which is susceptible to coupling interference from the perspective of multi-physical signal fusion and dynamic modeling, significantly improving the accuracy of abnormal joint identification and classification, which helps to provide early warning of potential faults and reduce the risk of unplanned downtime; on the other hand, through decoupling analysis and working condition adaptive threshold strategy, it enhances the adaptability of the detection system to the robot's changing working conditions and loads, ensuring the reliability and scientific nature of the detection process, thereby improving the robot's operational safety, maintenance efficiency and service life, while also providing reliable data support and decision-making basis for realizing state-based intelligent maintenance and health management. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0022] Figure 1 This is a schematic diagram showing the connections between the steps of the method of the present invention.

[0023] Figure 2This is a schematic diagram showing the connections of the various modules in the system of the present invention. Detailed Implementation

[0024] Example 1

[0025] Please see Figure 1 As shown, a method for detecting joint anomalies in a multi-condition robot includes the following steps:

[0026] Step S1: Acquire real-time multi-dimensional sensing data of each joint of the robot. The multi-dimensional sensing data includes joint motor current, joint vibration signal and joint encoder position data; perform forward kinematics calculation of the robot based on the joint encoder position data, construct the robot spatial link posture model in real time, and generate real-time working condition status data of the robot containing the current spatial pose features.

[0027] In one possible design, step S1 includes the following steps:

[0028] Step S11: Synchronously collect the robot's raw operating data using accelerometers, current transformers, and absolute encoders installed at each joint of the robot;

[0029] Step S12: Perform timestamp alignment and preprocessing on the raw running data to obtain multidimensional sensing data;

[0030] Step S13: Extract joint angle information from multidimensional sensing data, establish a robot link coordinate system using the DH parameter method, and calculate the spatial coordinates of the robot end effector and the center of gravity of each link.

[0031] Step S14: Calculate the robot's instantaneous velocity and acceleration based on the spatial coordinates and the rate of change of joint angles. Combine the spatial coordinates, instantaneous velocity, and acceleration to generate real-time working status data of the robot.

[0032] In the specific implementation process, firstly, multi-source signals are synchronously acquired by deploying Hall current sensors, MEMS triaxial accelerometers, and absolute multi-turn encoders at the drive ends of each joint of the robot, generating time-synchronized multi-dimensional sensing data. Then, based on the joint angle information in this multi-dimensional sensing data, a step-by-step recursive analysis of the robot's link spatial pose is performed. This analysis uses an improved DH parameter method, first constructing the homogeneous transformation matrix of adjacent joint coordinate systems, calculated using the following formula: ,in Let be the transformation matrix of the (i-1)th joint coordinate system relative to the ith joint coordinate system. Let be the rotation angle of the i-th joint. The length of the link. For the link torsion angle, This is the link offset distance. These are all structural parameters calibrated at the robot's factory, and this matrix describes the relative geometric relationships between adjacent links. Subsequently, the chain rule of homogeneous transformation matrices is used to calculate the absolute spatial position of the center of gravity of each link in the base coordinate system in real time, generating spatial link attitude model data. The formula for calculating the center of gravity coordinates is... in Let i be the position vector of the centroid of the i-th link in the base coordinate system. , Let be the cumulative transformation matrix from the base to the i-th joint. This is the position vector of the center of gravity of the i-th link in its own local coordinate system, in meters, and is retrieved from the robot dynamics parameter library.

[0033] Finally, based on the temporal variation characteristics of joint angles, the motion state differential is calculated, and the spatial pose features and motion state features are assembled into a high-dimensional tensor to generate real-time robot operating state data. The assembly formula for this data is as follows: ,in This represents the robot's real-time operating status data, in multi-dimensional matrix form, where N is the number of robot degrees of freedom. The sampling period, the second term The first term represents the calculated angular velocity of the i-th joint, in radians per second; the second term represents the angular acceleration of the i-th joint, in radians per second squared. This is the spatial coordinate transpose of the centroid.

[0034] Step S2: Perform dynamic analysis of the robot's variable inertia and stiffness based on the robot's real-time working condition data, and construct a spatial pose vibration transmission model that dynamically drifts with the pose; use the spatial pose vibration transmission model to quantify the vibration wave transmission efficiency between adjacent joints of the robot, and generate the joint vibration mode coupling coefficient matrix corresponding to the current working condition.

[0035] In one possible design, step S2 includes the following steps:

[0036] Step S21: Based on the link spatial position in the robot's real-time working condition data, call the robot's rigid body dynamics parameters to calculate the robot's overall mass distribution matrix and rotational inertia matrix in real time;

[0037] In the specific implementation process, based on the joint angles and link spatial position information in the robot's real-time working condition data, a pre-set robot rigid body dynamics parameter database is invoked. This database contains the mass, center of mass position vector, and local inertia tensor of each link obtained through offline CAD model import or parameter identification experiments. These parameters are used to construct the link velocity Jacobian matrix, and the construction formula is as follows: ,in Let be the Jacobian matrix of the k-th link. Let be the unit vector of the rotation axis of the (k-1)th joint, and be a dimensionless parameter. Let be the spatial position vector of the centroid of the k-th link, in meters. is the spatial position vector of the origin of the (k-1)th joint coordinate system, in meters;

[0038] Next, based on the constructed Jacobian matrix and rigid body dynamic parameters, a spatial rotational transformation of the inertia tensor and a calculation of the overall system kinetic energy are performed to generate the rotational inertia matrix. The calculation formula is as follows: ,in Let N be the generalized inertia matrix of the robot in joint space, expressed in kilogram-meter squared, where N is the total number of robot joints. Let K be the mass of the k-th link, in kilograms. and These represent the linear velocity component (in meters per radian) and the angular velocity component of the Jacobian matrix, respectively. The unit is meters per radian; The units are dimensionless; Let be the inertia tensor of the k-th link in its local coordinate system, expressed in kilograms per square meter. Let be the rotation matrix from the local coordinate system of the k-th link to the base coordinate system, and let be a dimensionless matrix. The term maps local inertia to the global coordinate system through similarity transformation.

[0039] Step S22: Based on the rotational inertia matrix and combined with the robot joint flexibility coefficient, perform system natural frequency drift analysis to obtain real-time system modal frequency data that varies with pose.

[0040] In practical implementation, based on the rotational inertia matrix that dynamically changes with pose, and combined with the torsional stiffness characteristics of the robot joint deceleration mechanism, the elastic potential energy matrix of the system is constructed (i.e., the stiffness matrix). The formula for constructing the stiffness matrix is ​​as follows: ,in Let be the stiffness matrix of the robot system, expressed in Newton-meters per radian, and be an N×N diagonal matrix. The equivalent torsional stiffness coefficient of the i-th joint reducer is expressed in Newton-meters per radian. This parameter is obtained by consulting the technical specifications of the robot reducer.

[0041] Next, based on the generalized inertia matrix and the system stiffness matrix, the characteristic equation of the undamped free vibration differential equation of the robot's multi-degree-of-freedom system is established. Then, the modal frequencies are calculated in real time using the generalized eigenvalue solving algorithm in numerical linear algebra. The formula for calculating the system modal frequencies is as follows: ,in Let be the r-th natural frequency at the current spatial pose q, in radians per second. This is the inverse of the generalized inertia matrix, with units of 1 / kg·m². , where · represents the operation function for extracting the r-th eigenvalue of the matrix. This formula reflects the rotational inertia caused by the robot's posture extension. When the stiffness increases, Natural frequency A nonlinear drift towards lower frequencies will occur. Finally, the calculated natural frequencies are arranged in ascending order to generate real-time system modal frequency data containing the distribution from the fundamental frequency to higher-order modal frequencies.

[0042] Step S23: Based on the real-time system modal frequency data and the geometric length of the link, analyze the attenuation and amplification characteristics of vibration energy in the link structure using the transfer function method, and construct a spatial orientation vibration transmission model;

[0043] During implementation, the vibration frequency response ratio is dynamically calculated based on real-time system modal frequency data and link geometric length data in the robot body parameter library. The calculation formula is as follows: ,in Here, is the frequency tuning ratio, and is a dimensionless parameter. The currently monitored external excitation frequency, in radians per second, is extracted from the joint vibration signal by FFT transformation. The r-th natural frequency of the system, expressed in radians per second, is derived from real-time system modal frequency data. This ratio reflects the proximity of the current excitation frequency to the resonance point after the system's "configuration change." Next, using the frequency tuning ratio combined with a single-degree-of-freedom forced vibration model from mechanical vibration theory, a spatial pose vibration transfer function incorporating dynamic amplification and geometric attenuation characteristics is constructed. The formula for calculating this transfer function is: ),in Let be the vibration transmissibility, and be a dimensionless parameter. The first part of the formula is the dynamic amplification factor, which describes the resonant amplification effect of the structure on vibration energy when the excitation frequency is close to the natural frequency. The material damping ratio is a dimensionless parameter, typically ranging from 0.01 to 0.1. The latter part of the formula is the geometric attenuation factor, which describes the energy dissipation of the vibration wave propagating in a solid link. Here, represents the material loss factor, which is a dimensionless parameter. Let be the geometric length of the k-th link, in meters. The value represents the sound of vibration waves propagating in the metal material of the link, expressed in meters per second. Finally, the calculated vibration transmissibility is combined with the link topology of the robot to construct a spatial pose vibration transmission model. This model is essentially a frequency domain filter bank that changes in real time with the pose q, describing how vibration energy at a specific frequency is "amplified" or "absorbed" in the physical structure of the link under a specific extended or folded posture.

[0044] Step S24: Calculate the vibration response transfer ratio between any two joints using the spatial pose vibration transfer model, and combine the transfer ratios of all joint pairs to generate the joint vibration mode coupling coefficient matrix.

[0045] In one possible design, step S24 includes the following steps:

[0046] Step S241: Select the i-th joint as the vibration source joint and the j-th joint as the tested joint, and extract the link transmission path parameters between the two joints under the current working condition.

[0047] Step S242: Input the unit pulse excitation to the joint position of the vibration source in the spatial pose vibration transmission model, and simulate and calculate the amplitude of the response signal at the tested joint;

[0048] Step S243: Calculate the ratio of the response signal amplitude to the unit pulse excitation to obtain the single-path vibration transfer factor;

[0049] Step S244: Traverse all joint combinations of the robot, calculate the corresponding single-path vibration transfer factor, and arrange them according to row and column rules to generate the joint vibration mode coupling coefficient matrix.

[0050] In the specific implementation process, the transmission path is analyzed by traversing all joints based on the robot's real-time working condition data. The analysis logic is set with the i-th joint as the vibration source and the j-th joint as the test object. The link transmission path parameters under the current spatial pose are extracted. First, the spatial Euclidean distance between the two joints and the angle between the link axis are calculated to generate path geometric feature data. Then, the dynamic impedance of the path is evaluated using system modal frequency data to generate path dynamic impedance data. Subsequently, response simulation and transfer factor calculation under unit impulse excitation are performed based on the path geometric feature data and path dynamic impedance data. The formula for calculating the single-path vibration transfer factor is as follows: ,in For single-path vibration transfer factor, The angle between the axis of the i-th joint and the axis of the j-th joint in the current spatial pose is derived from the forward kinematics data of the robot and is used to characterize the directional efficiency of the vibration wave propagation along the axis. The equivalent modal mass at the current pose, in kilograms, is extracted from the moment of inertia matrix. The first-order natural frequency of the system under the current operating conditions, expressed in radians per second; Let be the structural damping ratio, and be a dimensionless parameter. The linear transmission distance between the two joints. The characteristic attenuation length is calculated based on the sound velocity and frequency of the connecting rod material. This is a normalization constant, expressed in N·s·kg / m, used to eliminate dimensions and ensure the transfer factor is between 0 and 1. Finally, based on the calculated single-path vibration transfer factor, the inter-joint vibration mode coupling coefficient matrix is ​​assembled using a row-column mapping method. Fill the element in the i-th row and j-th column of the matrix, for the main diagonal elements Set it to 1, and finally generate the inter-joint vibration mode coupling coefficient matrix.

[0051] Step S3: Use the joint vibration mode coupling coefficient matrix to decouple the joint vibration signal from structural transmission noise, remove the coupled vibration components transmitted from other joints to the target joint through mechanical links, extract the intrinsic vibration feature vector characterizing the health status of the target joint, and combine it with the joint motor current to generate multidimensional decoupled joint feature data.

[0052] In one possible design, step S3 includes the following steps:

[0053] Step S31: Perform frequency domain transformation on the joint vibration signal to obtain the original joint vibration spectrum data;

[0054] Step S32: Based on the inter-joint vibration mode coupling coefficient matrix, identify and calculate the crosstalk vibration components received by the current joint from adjacent joints and distal joints;

[0055] Step S33: Subtract the crosstalk vibration component from the original joint vibration spectrum data, perform spectrum purification and reconstruction, and obtain the decoupled intrinsic vibration spectrum of the target joint;

[0056] Step S34: Perform inverse time-domain transformation and feature extraction on the intrinsic vibration spectrum of the target joint, calculate the effective value, kurtosis index and energy centroid, and generate the intrinsic vibration feature vector;

[0057] Step S35: The intrinsic vibration feature vector and the frequency domain feature of the joint motor current are fused to generate multidimensional decoupled joint feature data.

[0058] In one possible design, step S32 includes the following steps:

[0059] Step S321: Obtain the raw vibration intensity data of all joints except the target joint;

[0060] Step S322: Index the row vector corresponding to the target joint in the inter-joint vibration mode coupling coefficient matrix to obtain the input coupling coefficient set of the target joint;

[0061] Step S323: Use the input coupling coefficient set to perform weighted convolution operation on the original vibration intensity data of all other joints to simulate the interference waveform caused by structural transmission;

[0062] Step S324: Linearly superimpose the simulated interference waveforms to calculate the crosstalk vibration component at the current joint.

[0063] In the specific implementation process, the joint vibration signal is processed by short-time Fourier transform based on the Hanning window, transforming the discrete acceleration sequence in the time domain into a complex spectrum in the frequency domain, thereby obtaining the original joint vibration spectrum data. The purpose of this transform is to deconstruct the aliased time-domain waveform into frequency components. Next, inverse analysis of multi-source interference is performed based on the inter-joint vibration mode coupling coefficient matrix. This process first locks the target joint i to be analyzed, extracts the i-th row vector from the coupling matrix as the input coupling coefficient set, and the elements in this vector... The efficiency of vibration transmission from other joints j to joint i was characterized. Then, using the principle of equivalent time-domain convolution through frequency-domain multiplication, the crosstalk vibration component at the current joint was calculated. The calculation formula is as follows: ,in To calculate the frequency The crosstalk vibration component spectrum is given below, in meters per second squared, where N is the total number of robot joints. Let be the vibration transfer factor that varies with pose q, and be a dimensionless parameter. The measured vibration amplitude spectrum of joint j, the source of the interference, is expressed in meters per second squared.

[0064] Subsequently, the calculated crosstalk vibration components were removed from the original joint vibration spectrum data using the principle of spectral subtraction, and the spectrum was purified and reconstructed. The formula for calculating the intrinsic vibration spectrum is as follows: ,in The intrinsic vibration amplitude of the target joint after decoupling is expressed in meters per second squared. The total vibration amplitude measured for the target joint. This is the over-subtraction factor, a dimensionless parameter, typically taken as 1.0 to 1.2, used to compensate for model errors; is the spectral limit parameter, which is a dimensionless parameter used to prevent negative values ​​or non-physical zero spectra after subtraction.

[0065] Then, the intrinsic vibration spectrum after decoupling is subjected to feature reduction and dimensionality extraction. The kurtosis index characterizing the impact characteristics and the energy centroid characterizing the frequency band variation are calculated respectively. The formula for calculating the intrinsic kurtosis index is as follows: In the formula The time-domain intrinsic signal is obtained by inverse Fourier transform, with units of meters per second squared, and M being the number of sampling points; the formula for calculating the energy centroid is... In the formula The frequency point is measured in Hertz. This indicator is used to capture high-frequency resonance band shifts caused by bearing wear or tooth surface pitting.

[0066] Finally, the above vibration characteristics and the frequency domain characteristics of the joint motor current are heterogeneously fused to construct multidimensional decoupled joint feature data. The formula for constructing the fusion vector is as follows: ,in This is the effective value of the motor current, in amperes. Let be the total harmonic distortion of the current, and be a dimensionless parameter. , The amplitude of the h-th harmonic current. This represents the amplitude of the fundamental current.

[0067] Step S4: Based on the robot's real-time operating status data, call the preset dynamic operating condition benchmark threshold library, use the dynamic operating condition benchmark threshold library to perform adaptive anomaly discrimination on the multi-dimensional decoupled joint feature data, locate the specific joint where the anomaly occurs and identify the anomaly type, thereby generating a robot joint anomaly detection result report.

[0068] In one possible design, step S4 includes the following steps:

[0069] Step S41: Based on the speed and load information in the robot's real-time operating status data, divide the robot's operating conditions into intervals and determine the current operating condition level.

[0070] Step S42: Index the corresponding dynamic working condition benchmark threshold library from the preset database according to the working condition level. The benchmark threshold library contains the upper and lower limits of health characteristics under the working condition.

[0071] Step S43: Input the multidimensional decoupled joint feature data into the dynamic working condition benchmark threshold library for comparison and analysis, and calculate the feature deviation.

[0072] Step S44: When the feature deviation exceeds the preset safety margin, it is determined to be an abnormal state, and the abnormal joint number is output;

[0073] Step S45: Based on the spectral distribution characteristics of the feature deviation, match the fault mode library to generate a robot joint anomaly detection result report.

[0074] In one possible design, step S43 includes the following steps:

[0075] Step S431: Extract the intrinsic vibration kurtosis value and current harmonic distortion rate from the multidimensional decoupled joint feature data;

[0076] Step S432: Obtain the mean and standard deviation of the health benchmark from the dynamic operating condition benchmark threshold library;

[0077] Step S433: Calculate the distance between the intrinsic vibration kurtosis value and the current harmonic distortion rate relative to the healthy baseline mean using the Mahalanobis distance algorithm;

[0078] Step S434: Normalize the calculated distance to obtain the feature deviation.

[0079] In the specific implementation process, a multi-dimensional gridded working space is mapped based on the robot's real-time working condition data. A weighted normalization algorithm is used to calculate the current working condition pressure index. The calculation formula is as follows: ,in The working pressure index is a dimensionless integer used as a database index key. This refers to the working condition resolution coefficient. and These are the current joint angular velocity and joint torque, respectively, in radians per second and Newton-meters. and These are the rated parameters. This refers to the current arm span of the robot, in meters. For maximum wingspan, , and These are the weighting coefficients, and the sum of the three is 1.

[0080] Next, the calculated working pressure index is used. A rapid index is performed on the multivariate statistical distribution parameters from a pre-set dynamic operating condition baseline threshold library to extract the health baseline mean vector for that specific operating condition. and health baseline covariance matrix The dimension is N×N, where The natural correlation between vibration characteristics and current characteristics under healthy conditions was recorded.

[0081] Then, the generated multidimensional decoupled joint feature data is input into a statistical discriminant model based on Mahalanobis distance to calculate the feature deviation, in order to overcome the deficiency of traditional Euclidean distance in ignoring the correlation between variables. The Mahalanobis distance calculation formula is as follows: ,in The Mahalanobis distance value is a dimensionless parameter. This data represents the currently measured multidimensional decoupled joint feature data, including intrinsic kurtosis, energy centroid, and current distortion rate. The inverse of the covariance matrix is ​​then calculated. The feature deviation is obtained by normalization, and the formula is: ,in The characteristic deviation is a dimensionless parameter. For features with p degrees of freedom and a confidence level of p, the feature dimension is p. The chi-square distribution critical value is used to precisely quantify the degree to which the current state deviates from the "normal state under this working condition" using statistical methods.

[0082] Finally, when the feature deviation When the score is greater than 1, a joint abnormality is determined. Fault mode matching is then performed based on the spectral distribution characteristics of the feature vectors to establish a fault similarity scoring model. The calculation formula is as follows: ,in Let be the cosine similarity between the current signal and the k-th fault mode (such as bearing outer ring damage, gear tooth breakage), and be a dimensionless parameter. This is the decoupled intrinsic vibration spectrum vector, in meters per second squared. The feature frequency template vector of the k-th fault in the fault mode library is selected by... The pattern corresponding to the maximum value is used as the recognition result, and finally a robot joint anomaly detection result report containing the abnormal joint number, the degree of deviation, and the specific fault type is generated.

[0083] It should be added that the following steps are included after step S4:

[0084] The robot analyzes the joint anomaly detection report and extracts the severity level and fault type. If the severity level is minor, a deceleration command is generated to adjust the robot's motion planning trajectory to avoid the resonant frequency range. If the severity level is severe, an emergency stop command is generated and the faulty axis is locked with a brake.

[0085] Example 2

[0086] like Figure 2 As shown, a multi-condition robot joint anomaly detection system includes a condition data acquisition module, a coupling matrix generation module, a feature data generation module, and a joint anomaly discrimination module.

[0087] The above modules are connected via wired connections to enable data transmission between them;

[0088] Working condition data acquisition module: acquires real-time multi-dimensional sensing data of each joint of the robot; performs forward kinematics calculation of the robot based on the joint encoder position data, constructs the robot spatial link posture model in real time, and generates real-time working condition status data of the robot;

[0089] Coupling matrix generation module: Based on the robot's real-time working condition data, perform dynamic analysis of the whole machine's variable inertia and variable stiffness, and construct a spatial pose vibration transmission model that dynamically drifts with the pose; use the spatial pose vibration transmission model to quantify the vibration wave transmission efficiency between adjacent joints of the robot, and generate the joint vibration mode coupling coefficient matrix corresponding to the current working condition.

[0090] Feature data generation module: The joint vibration signal is decoupled from the structural transmission noise by using the inter-joint vibration mode coupling coefficient matrix, the intrinsic vibration feature vector characterizing the health status of the target joint body is extracted, and multi-dimensional decoupled joint feature data is generated by combining the joint motor current.

[0091] Joint anomaly detection module: Based on the robot's real-time operating status data, it calls the preset dynamic operating condition benchmark threshold library to perform adaptive anomaly detection on multi-dimensional decoupled joint feature data, locate the specific joint where the anomaly occurs and identify the anomaly type, thereby generating a robot joint anomaly detection result report.

[0092] It should be noted that the interval and threshold sizes are set for ease of comparison. The size of the threshold depends on the amount of sample data and the base number set by those skilled in the art for each set of sample data, as long as it does not affect the proportional relationship between the parameter and the quantized value. Furthermore, the above formulas are all dimensionless calculations, and the formulas are derived from software simulations using a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

Claims

1. A method for detecting joint anomalies in a multi-condition robot, characterized in that, Includes the following steps: Step S1: Acquire real-time multi-dimensional sensing data of each joint of the robot. The multi-dimensional sensing data includes joint motor current, joint vibration signal and joint encoder position data; perform forward kinematics calculation of the robot based on the joint encoder position data, construct the robot spatial link posture model in real time, and generate real-time working condition data of the robot containing the current spatial pose features. Step S1 includes the following steps: Step S11: Synchronously collect the robot's raw operating data using accelerometers, current transformers, and absolute encoders installed at each joint of the robot; Step S12: Perform timestamp alignment and preprocessing on the raw running data to obtain multidimensional sensing data; Step S13: Extract joint angle information from multidimensional sensing data, establish a robot link coordinate system using the DH parameter method, and calculate the spatial coordinates of the robot end effector and the center of gravity of each link. Step S14: Calculate the robot's instantaneous velocity and acceleration based on the spatial coordinates and the rate of change of joint angles. Combine the spatial coordinates, instantaneous velocity, and acceleration to generate real-time working status data of the robot. Step S2: Perform dynamic analysis of the robot's variable inertia and stiffness based on the robot's real-time working condition data, and construct a spatial pose vibration transmission model that dynamically drifts with the pose; use the spatial pose vibration transmission model to quantify the vibration wave transmission efficiency between adjacent joints of the robot, and generate the joint vibration mode coupling coefficient matrix corresponding to the current working condition. Step S2 includes the following steps: Step S21: Based on the link spatial position in the robot's real-time working condition data, call the robot's rigid body dynamics parameters to calculate the robot's overall mass distribution matrix and rotational inertia matrix in real time; Step S22: Based on the rotational inertia matrix and combined with the robot joint flexibility coefficient, perform system natural frequency drift analysis to obtain real-time system modal frequency data that varies with pose. Step S23: Based on the real-time system modal frequency data and the geometric length of the link, analyze the attenuation and amplification characteristics of vibration energy in the link structure using the transfer function method, and construct a spatial orientation vibration transmission model; Step S24: Calculate the vibration response transfer ratio between any two joints using the spatial pose vibration transfer model, and combine the transfer ratios of all joint pairs to generate the vibration mode coupling coefficient matrix between joints. Step S3: Use the joint vibration mode coupling coefficient matrix to decouple the joint vibration signal from structural transmission noise, remove the coupled vibration components transmitted from other joints to the target joint through mechanical links, extract the intrinsic vibration feature vector characterizing the health status of the target joint, and combine it with the joint motor current to generate multidimensional decoupled joint feature data. Step S3 includes the following steps: Step S31: Perform frequency domain transformation on the joint vibration signal to obtain the original joint vibration spectrum data; Step S32: Based on the inter-joint vibration mode coupling coefficient matrix, identify and calculate the crosstalk vibration components received by the current joint from adjacent joints and distal joints; Step S33: Subtract the crosstalk vibration component from the original joint vibration spectrum data, perform spectrum purification and reconstruction, and obtain the decoupled intrinsic vibration spectrum of the target joint; Step S34: Perform inverse time-domain transformation and feature extraction on the intrinsic vibration spectrum of the target joint, calculate the effective value, kurtosis index and energy centroid, and generate the intrinsic vibration feature vector; Step S35: Perform feature fusion between the intrinsic vibration feature vector and the frequency domain feature of the joint motor current to generate multidimensional decoupled joint feature data; Step S4: Based on the robot's real-time working condition data, call the preset dynamic working condition benchmark threshold library, use the dynamic working condition benchmark threshold library to perform adaptive anomaly discrimination on the multi-dimensional decoupled joint feature data, locate the specific joint where the anomaly occurs and identify the anomaly type, thereby generating a robot joint anomaly detection result report. Step S4 includes the following steps: Step S41: Based on the speed and load information in the robot's real-time operating status data, divide the robot's operating conditions into intervals and determine the current operating condition level. Step S42: Index the corresponding dynamic operating condition benchmark threshold library from the preset database according to the operating condition level. The benchmark threshold library contains the upper and lower limits of health characteristics under the operating condition level. Step S43: Input the multidimensional decoupled joint feature data into the dynamic working condition benchmark threshold library for comparison and analysis, and calculate the feature deviation. Step S44: When the feature deviation exceeds the preset safety margin, it is determined to be an abnormal state, and the abnormal joint number is output; Step S45: Based on the spectral distribution characteristics of the feature deviation, match the fault mode library to generate a robot joint anomaly detection result report.

2. The multi-condition robot joint anomaly detection method according to claim 1, characterized in that, Step S24 includes the following steps: Step S241: Select the i-th joint as the vibration source joint and the j-th joint as the tested joint, and extract the link transmission path parameters between the two joints under the current working condition. Step S242: Input the unit pulse excitation to the joint position of the vibration source in the spatial pose vibration transmission model, and simulate and calculate the amplitude of the response signal at the tested joint; Step S243: Calculate the ratio of the response signal amplitude to the unit pulse excitation to obtain the single-path vibration transfer factor; Step S244: Traverse all joint combinations of the robot, calculate the corresponding single-path vibration transfer factor, and arrange them according to row and column rules to generate the joint vibration mode coupling coefficient matrix.

3. The multi-condition robot joint anomaly detection method according to claim 1, characterized in that, Step S32 includes the following steps: Step S321: Obtain the raw vibration intensity data of all joints except the target joint; Step S322: Index the row vector corresponding to the target joint in the inter-joint vibration mode coupling coefficient matrix to obtain the input coupling coefficient set of the target joint; Step S323: Use the input coupling coefficient set to perform weighted convolution operation on the original vibration intensity data of all other joints to simulate the interference waveform caused by structural transmission; Step S324: Linearly superimpose the simulated interference waveforms to calculate the crosstalk vibration component at the current joint.

4. The multi-condition robot joint anomaly detection method according to claim 1, characterized in that, Step S43 includes the following steps: Step S431: Extract the intrinsic vibration kurtosis value and current harmonic distortion rate from the multidimensional decoupled joint feature data; Step S432: Obtain the mean and standard deviation of the health benchmark from the dynamic operating condition benchmark threshold library; Step S433: Calculate the distance between the intrinsic vibration kurtosis value and the current harmonic distortion rate relative to the healthy baseline mean using the Mahalanobis distance algorithm; Step S434: Normalize the calculated distance to obtain the feature deviation.

5. A multi-condition robot joint anomaly detection system, characterized in that, It is implemented based on the multi-condition robot joint anomaly detection method according to any one of claims 1-4, and the joint anomaly detection system includes: Working condition data acquisition module: acquires real-time multi-dimensional sensing data of each joint of the robot; performs forward kinematics calculation of the robot based on the joint encoder position data, constructs the robot spatial link posture model in real time, and generates real-time working condition status data of the robot; Coupling matrix generation module: Based on the robot's real-time working condition data, perform dynamic analysis of the whole machine's variable inertia and variable stiffness, and construct a spatial pose vibration transmission model that dynamically drifts with the pose; use the spatial pose vibration transmission model to quantify the vibration wave transmission efficiency between adjacent joints of the robot, and generate the joint vibration mode coupling coefficient matrix corresponding to the current working condition. Feature data generation module: The joint vibration signal is decoupled from the structural transmission noise by using the inter-joint vibration mode coupling coefficient matrix, the intrinsic vibration feature vector characterizing the health status of the target joint body is extracted, and multi-dimensional decoupled joint feature data is generated by combining the joint motor current. Joint anomaly detection module: Based on the robot's real-time operating status data, it calls the preset dynamic operating condition benchmark threshold library to perform adaptive anomaly detection on multi-dimensional decoupled joint feature data, locate the specific joint where the anomaly occurs and identify the anomaly type, thereby generating a robot joint anomaly detection result report.

Images