A robot fault diagnosis method and system based on evolution of kinetic parameters

Abstract

This invention belongs to the field of robot fault diagnosis, and specifically discloses a robot fault diagnosis method and system based on the evolution of dynamic parameters, which includes: based on the robot dynamics model, obtaining: τ=YP=Y b P b τ is the joint driving torque vector, Y is the regression matrix describing the robot's motion process, and P is the robot's parameter set. b Let P be the largest linearly independent set of column vectors in matrix Y. b The minimum identifiable parameter set P is defined; the parameters in parameter set P are classified and their correlation with various faults is marked; the excitation trajectory is run under fault-free conditions of the robot, and τ and Y are recorded. b This yields a fault-free parameter set; during robot inspection, the excitation trajectory is run, and τ and Y are recorded. b The current parameter set is obtained; the current parameter set is compared with the corresponding parameters in the fault-free parameter set, and the robot fault category is determined based on the deviation of each parameter. This invention can realize robot fault location and accurate analysis.

Description

Technical Field

[0001] This invention belongs to the field of robot fault diagnosis, and more specifically, relates to a robot fault diagnosis method and system based on the evolution of dynamic parameters. Background Technology

[0002] Robot fault diagnosis plays a crucial role in maintaining the effective operation of robotic equipment in engineering practice, encompassing both fault detection and fault identification. Currently, industrial robots are widely used in numerous manufacturing enterprises. However, due to the complex nature of robots as nonlinear, strongly coupled systems, traditional fault diagnosis techniques used in ordinary rotating machinery face many challenges in their application to industrial robots.

[0003] Existing robot fault diagnosis technologies can be categorized into three types: model-based, data-driven, and knowledge-based. Model-based methods essentially utilize the concept of analytical redundancy, using an explicit mathematical model of the system as a reference and then constructing a residual observer for fault diagnosis. In practice, model-based methods offer advantages in real-time performance and are suitable for diagnosing linear systems, but the challenge lies in generating torque residuals insensitive to noise, external disturbances, and model uncertainties, requiring sophisticated model building. When an accurate explicit model cannot be established, another approach, data-driven, can be used for robot fault diagnosis. Data-driven methods can be model-independent, primarily predicting fault occurrence through multivariate statistical processes, such as Principal Component Analysis (PCA) and Partial Least Squares (PLS). They transform raw data of variables into features characterizing specific faults through time-domain or frequency-domain analysis or prior knowledge of signal distribution. Knowledge-based methods require knowing and establishing the correlation between identified behaviors and fault characteristics. A drawback of these methods is their reliance on prior fault isolation knowledge, as it is difficult to establish a complete prior knowledge base. Hybrid fault diagnosis methods are any combination of two or three of the above methods. Most existing fault diagnosis technologies employ model-observer combined with statistical methods or data-driven machine learning methods. These methods are mostly black-box or gray-box predictions, meaning that the mechanism of fault diagnosis cannot be clearly described. Furthermore, they require a large amount of prior fault data to construct a mapping relationship between faults and data, resulting in relatively weak generalization ability and reliability. Summary of the Invention

[0004] In view of the above-mentioned defects or improvement needs of the existing technology, the present invention provides a robot fault diagnosis method and system based on dynamic parameter evolution. Its purpose is to monitor the changes in the internal physical parameters of the robot and realize the fault location and accurate and reliable analysis of the robot.

[0005] To achieve the above objectives, according to a first aspect of the present invention, a robot fault diagnosis method based on dynamic parameter evolution is proposed, comprising the following steps:

[0006] Based on the robot dynamics model, we obtain: τ=YP=Y b P b Where τ is the joint driving torque vector, Y is the regression matrix describing the robot's motion process, and P is the robot's parameter set. b Let P be the largest linearly independent set of column vectors in matrix Y. b The minimum identifiable parameter set;

[0007] The parameters in parameter set P are located and classified, and the correlation between each parameter and various types of faults is marked.

[0008] With matrix Y b The objective function is to minimize the condition number, and the excitation trajectory is determined accordingly.

[0009] Run the excitation trajectory under fault-free conditions and record τ and Y. b Then, the corresponding P is calculated, and the fault-free parameter set is obtained;

[0010] Detect the robot: run the excitation trajectory and record τ and Y. b The calculated P is the current parameter set; the current parameter set is compared with the corresponding parameters in the fault-free parameter set, and the robot fault category is determined based on the deviation of each parameter and the correlation between each parameter and various types of faults.

[0011] As a further preferred option, the parameters in parameter set P are classified, including:

[0012] For any parameter in parameter set P, first determine the robot joint corresponding to the parameter; for each robot joint, the parameter categories include the moment of inertia parameter of the link, the first-order mass moment and mass parameter of the link, the equivalent moment of inertia parameter of the joint transmission chain, and the friction parameter of the joint.

[0013] As a further preferred option, the single-joint friction model used in the robot dynamics model is:

[0014]

[0015] Among them, F cj Let F be the Coulomb friction parameter. vj Let α be the viscous friction parameter. j For the empirical parameters of the robot, B j f is the bias parameter for the asymmetry of forward and reverse friction forces; j For the frictional force of the joint, The velocity of the joint;

[0016] Then for robot joint j, the friction parameter P of the joint... jf =[F cj F vj α j B j ] T .

[0017] As a further preferred method, the excitation trajectory is run multiple times under fault-free conditions of the robot, and τ and Y are recorded. b This allows us to calculate multiple sets of P, which are then used as the fault-free parameter set.

[0018] As a further optimization, the current parameter set is compared with the corresponding parameters in the fault-free parameter set to determine the deviation of each parameter; for any parameter i, its deviation d i The calculation formula is as follows:

[0019]

[0020] Where, μ0(P) i ), σ0(P i ) represent the mean and standard deviation of parameter i in the fault-free parameter set, respectively; μ1(P i ) is the current mean value obtained after processing parameter i in the current parameter set through a moving average filter.

[0021] As a further preferred embodiment, for any parameter i, the deviation d is obtained. i Then, the parameter changes are quantified using the following confidence function to determine whether the parameters are abnormal; the confidence function is:

[0022]

[0023] Wherein, c(d) i ) represents the quantification result of parameter changes, L u L d These are the upper and lower bounds, which are set according to different parameter characteristics.

[0024] As a further preferred option, when determining the excitation trajectory, matrix Y is used. b The objective function is to minimize the condition number. By changing the constraints, multiple sets of excitation trajectories are obtained. After running the excitation trajectories, multiple sets of τ and Y are obtained accordingly. b Thus, P can be obtained using the least squares method. b And then according to P b Find P.

[0025] As a further preferred embodiment, the change of constraint conditions is: changing the constraint range of robot joint rotation angle, speed, acceleration, and torque limits.

[0026] According to a second aspect of the present invention, a robot fault diagnosis system based on dynamic parameter evolution is provided, comprising a processor for executing the above-described robot fault diagnosis method based on dynamic parameter evolution.

[0027] According to a third aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the above-described robot fault diagnosis method based on the evolution of dynamic parameters.

[0028] In summary, compared with the prior art, the above-described technical solutions conceived by this invention mainly possess the following technical advantages:

[0029] 1. This invention diagnoses faults by identifying parameter changes during each robot's specific trajectory movement. By monitoring changes in the robot's internal physical parameters, it achieves early-stage fault location and accurate, reliable analysis. This method boasts strong analytical capabilities and portability, making it easily applicable to robots with different kinematic chains. Furthermore, it requires only measurements from sensors used by the robot itself, eliminating the need for additional sensors.

[0030] 2. The method of the present invention provides a full range of dynamic parameters for the performance indicators of faults, which can more comprehensively and accurately diagnose mechanical defects in the early stages. It can separate various faults affecting the drive chain, such as wear, motor overload and overheating, or load changes, and can be used for state-based maintenance strategies to prevent robot failures.

[0031] 3. This invention uses matrix Y b The minimum condition number determines the excitation trajectory for the objective function. A small condition number means that the matrix calculation is less sensitive to error and has good numerical stability, which can reduce parameter identification error and enable the excitation trajectory to identify more accurate parameters. Furthermore, by changing the constraints of robot joint angle, velocity, acceleration, and torque limit, multiple sets of local optimal trajectories can be obtained, which can cover a wider range of motion spaces and prevent overfitting.

[0032] 4. When assessing parameter deviation, a moving average filter is applied to provide the current mean value of each parameter. The use of a moving average filter can reduce the influence of noise or outliers, thereby improving the accuracy of diagnosis. Furthermore, by constructing a confidence function, the parameter changes can be quantified, thereby determining the fault condition. Attached Figure Description

[0033] Figure 1 This is a flowchart of a robot fault diagnosis method based on dynamic parameter evolution according to an embodiment of the present invention;

[0034] Figure 2 This is a schematic diagram illustrating the parameter classification of a six-axis industrial robot model according to an embodiment of the present invention;

[0035] Figure 3 This is a schematic diagram illustrating the quantitative analysis of robot parameters in an embodiment of the present invention. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0037] Robot dynamics models can describe the relationship between robot motion and force. By measuring motion and torque data, the process parameters of the robot under a given motion can be identified based on the robot dynamics model. When the robot is identified under a specific motion trajectory, the robot's process parameters are consistent with the robot's physical parameters. Therefore, fault diagnosis can be performed based on the parameter changes identified each time the robot runs a specific trajectory.

[0038] This invention provides a robot fault diagnosis method based on the evolution of dynamic parameters, such as... Figure 1 As shown, it includes the following steps:

[0039] S1, based on the robot dynamics model, constructs the relationship between the robot's physical parameters and classifies the parameters.

[0040] The robot dynamics model expression is:

[0041]

[0042] Where q, and The vectors representing the angular position, velocity, and acceleration of the robot joints are given in radians. M(q) is the inertia matrix. It is the Coriolis and centripetal effect matrix, and G(q) is the gravity vector. τ represents the friction torque vector, and τ is the driving torque vector of the joint.

[0043] Let formula (1) be:

[0044]

[0045]

[0046] Equation (1) to Equation (3) is called the linearization of the dynamic equations. The linearization can be achieved through the Newton-Euler equations. The Y matrix is ​​the regression matrix describing the robot's motion process, and P is the process parameter matrix related to the robot's physical parameters.

[0047] In this embodiment, the robot's physical parameter set is set as follows:

[0048] P = [P] I P g P r P f ] T (4)

[0049] Among them, P I P represents the moment of inertia parameter for all links. g Let P be the first-order mass moment and the mass parameter of all links. r For the equivalent rotational inertia parameter of all joint transmission chains, P f These are the friction parameters for all joints.

[0050] Furthermore, for a certain joint j, the above parameters can be expanded to specifically represent: P jI =[Ij xx I jxy I jxz I jyy I jyz I jzz ] T P jg =[m j r jx m j r jy m j r jz m j ] T P jr =I aj Among them, I jxx I jyy I jzz I represents the moment of inertia about the x-axis, y-axis, and z-axis, respectively. jxy I jxz I jyz The product of inertia; r is the distance between the centers of mass, [r jx r jy r jz ] T Let m be the vector of the center of mass in the link coordinate system at joint j. j For quality.

[0051] To make the robot dynamics model more accurate, the single-joint friction model used in this invention is as follows:

[0052]

[0053] In the formula, f j F is the frictional force at joint j. cLet F be the Coulomb friction parameter. v Here, α is the viscous friction parameter, and B is an empirical parameter for different robots. j This is the bias parameter for the asymmetry of the forward and reverse friction forces. Therefore, for a certain joint j, the friction parameter of the joint is expressed as P. jf =[F cj F vj α j B j ] T .

[0054] As shown above, in this embodiment, for each robot joint, there are 6 link rotational inertia parameters, 4 link first-order mass moments and link mass parameters, 1 transmission chain equivalent rotational inertia parameter, and 4 friction parameters. Therefore, for a serial robot with n rotary joints, the number of parameters in the parameter set P is n×(6+4+1+4). These parameters have physical meaning and are physically interpretable, meaning they reflect the robot's physical state.

[0055] The parameters in parameter set P can then be classified. In this embodiment, parameters 1-15 are classified as 1-axis sub-dynamic systems, parameters 16-30 as 2-axis sub-dynamic systems, and so on. Each sub-dynamic system corresponds to each joint of the robot and is categorized into inertia-related, mass-related, transmission chain-related, and friction-related parameters. For a six-axis industrial robot, its model parameters are defined as follows: Figure 2 As shown. The above definition allows for rapid fault location. For example, if the parameter value of number 57 is abnormal, it indicates a friction-related fault in one of the robot's four axes.

[0056] After establishing and linearizing the robot's dynamics model, it needs to be used with the known regression matrix. To solve for the robot parameters P using τ, we need to find... The largest linearly independent set of column vectors Then, the corresponding parameters in P are eliminated, thus forming the minimum identifiable parameter set P of the recombined system. b :

[0057]

[0058] The minimum identifiable parameter set can be determined using numerical or analytical methods.

[0059] S2, determine the excitation trajectory.

[0060] In the identification of robot dynamics model parameters, it is necessary to design specific excitation trajectories to stimulate the robot's dynamic characteristics, and the regression matrix... The local optimal excitation trajectory under the current constraints is found when the condition number is minimized. The condition number is defined as the matrix Y.b The norm multiplied by matrix Y b The norm of the inverse matrix: cond(Y) b )=||Y b ||·||Y b -1 ||.

[0061] In this embodiment, the excitation trajectory is described in the form of a Fourier series, the motion period is set to T = 6s, the sampling time is ts = 0.004s, and the expression for each set of excitation trajectories for the j-th joint is as follows:

[0062]

[0063] in, It is the offset term, ω f The fundamental frequency of the Fourier series is 2π / T, where t is the motion time. Let $\mathbf{ ... Therefore, each trajectory has 6 × 9 = 54 parameters to be optimized.

[0064] By establishing the objective function Min(cond(Y) b By solving the nonlinear multivariate function under constraints of robot joint angle, velocity, acceleration, and torque limits, trajectory parameters can be obtained. By changing the constraints, three sets of locally optimal trajectories r (r = 1, 2, 3, r ∈ R) are obtained as excitation trajectories. Selecting three sets can cover a wider range of motion spaces and prevent overfitting.

[0065] S3, establish a fault-free parameter set.

[0066] Run the above three sets of excitation trajectories under fault-free conditions, and record the three sets of τ and Y values. b Then, the corresponding P is calculated; the excitation trajectory is run multiple times to obtain multiple P to form a fault-free parameter set.

[0067] Specifically, when the robot runs along the excitation trajectory, the feedback position q and feedback torque τ of each axis of the robot can be obtained. The velocity can be obtained by performing differential and filtering on the feedback position data. Acceleration data, Then, by inputting the position, velocity, and acceleration data into the dynamic model, the regression matrix Y of each trajectory can be obtained. b .

[0068] Furthermore, based on the three sets of τ and Y b Identify P b Parameter P bIt can be identified using linear regression methods such as least squares or weighted least squares.

[0069] P b =(Y b T ·Y b ) -1 ·Y b T ·τ (8)

[0070] For a certain link j, its moment of inertia and mass should both be non-negative. In this embodiment, the constraint conditions for each parameter are set as follows: I j >0, m j >0, F cj >0, F vj >0.

[0071] Then, considering the above physical feasibility constraints, the minimum parameter P b This can be obtained by solving the following constrained optimization problem:

[0072] Arg(P b min(||Y) b ·P b -τ||2)

[0073] Subject to: I j >0, m j >0, F cj >0, F vj >0,k=1,2,3,...,n (9)

[0074] Because the minimum parameter P b It can be formed by a linear combination of robot parameters P, and there exists a constant matrix C such that:

[0075] CP = P b (10)

[0076] Therefore, the minimum parameter P b After solving, the robot's full parameter set P is obtained using formula (10).

[0077] S4, the robot periodically runs the excitation trajectory to obtain the current parameter set in order to monitor the robot's status.

[0078] Specifically, the robot periodically runs three sets of excitation trajectories and records three sets of τ and Y values. b The calculated P is the current parameter set; the specific calculation method is the same as in S3, and will not be repeated here.

[0079] S5. Compare the current parameter set with the corresponding parameters in the fault-free parameter set, and determine the robot fault category based on the deviation of each parameter and the correlation between each type of parameter and each type of fault.

[0080] After identifying the value of each parameter, a comparative analysis is performed to determine whether the values ​​are within the normal range. This process is called quantitative fault analysis.

[0081] Specifically, the mean μ0(P) of each parameter in the fault-free parameter set i ) and standard deviation σ0(P i ), i = 1, 2, 3...n×15. During the diagnostic process, based on the current parameter set, a moving average filter is applied to give the current mean μ1(P) of each parameter. i The use of a moving average filter can reduce the influence of noise or outliers, thereby improving the accuracy of diagnosis. Subscript 1 indicates the diagnostic status.

[0082] Then, the deviation coefficient d is defined. i Represents the physical parameter P i Compared to its fault-free nominal value P i0 Regarding the deviation of its standard deviation:

[0083]

[0084] Then construct the confidence function c(d) i )∈[0,1]:

[0085]

[0086] Among them, L u L d These are the upper and lower bounds, respectively. The settings for these bounds differ depending on the parameters, and are chosen based on experience. The uncertainty will vary depending on the parameters. The obtained c(d) i This can be considered a quantitative result representing parameter changes, such as Figure 3 As shown, when c(d) i If c(d) approaches 1, the parameter is considered abnormal and malfunctioning. i If the value approaches 0, the parameter is considered normal and the robot is not malfunctioning.

[0087] The fault analysis proceeds in reverse order, starting from the 6-axis dynamics system and working backwards to the 1-axis dynamics system. Expanding the theoretical formula (6) (at this point, friction parameters are not included), its structure is shown in formula (13), indicating that the inertial parameters of robot joint j only affect the joint torques τ1…τ before joint j. j And without affecting the joint torque τ after joint j. j+1 …τ nIn other words, the inertial parameter data of sub-dynamic system 6 will not be affected by inertial-related faults in sub-dynamic systems 1, 2, 3, 4, and 5. This analysis sequence utilizes the characteristics of theoretical formulas, which also helps to quickly and accurately locate faults.

[0088]

[0089] Each sub-dynamic system categorizes faults and establishes mapping relationships based on the physical meaning of parameters. Fault categories include joint wear, joint backlash, motor overload and overheating, base loosening, link flexibility changes, static load overload, and dynamic load overload. Table 1 shows the correlation between different types of faults and parameters in industrial robot dynamics; of course, fault classifications include, but are not limited to, those shown in Table 1. Based on the correlations in Table 1, quantitative analysis can be combined to perform accurate fault diagnosis of the robot.

[0090] Table 1. Fault classification and fault-parameter correlation of sub-dynamic systems.

[0091]

[0092] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A robot fault diagnosis method based on dynamic parameter evolution, characterized in that, Includes the following steps: Based on the robot dynamics model, we obtain: ;in, It is the driving torque vector of the joint. It is a regression matrix describing the robot's motion process. For the robot's parameter set, For matrix The largest linearly independent set of column vectors in the set. The minimum identifiable parameter set; For parameter set The parameters in the dataset are located and categorized, and the correlation between each parameter and various types of faults is marked; specifically, this includes: [the following is a list of parameters, likely related to fault location and classification]. For any parameter in the parameters, first determine the corresponding robot joint; for each robot joint, the parameter categories include the moment of inertia parameter of the link, the first-order mass moment and mass parameter of the link, the equivalent moment of inertia parameter of the joint transmission chain, and the friction parameter of the joint; based on the physical meaning of the parameters, classify the faults and establish the mapping relationship. The fault categories include joint wear, joint backlash, motor overload and overheating, base loosening, link flexibility change, static load overload, and dynamic load overload; With matrix The objective function is to minimize the condition number, and the excitation trajectory is determined accordingly. Run the excitation trajectory under fault-free conditions and record it. and Then calculate the corresponding This yields a set of fault-free parameters; Inspect the robot: run the excitation trajectory and record it. and The calculation yields the result at this time. This is the current parameter set; by comparing the current parameter set with the corresponding parameters in the fault-free parameter set, the robot fault category is determined based on the deviation of each parameter and the correlation between each parameter and various types of faults.

2. The robot fault diagnosis method based on dynamic parameter evolution as described in claim 1, characterized in that, In the robot dynamics model, the single-joint friction model used is as follows: in, Here are the Coulomb friction parameters. For viscous friction parameters, For the empirical parameters of the robot, This is the bias parameter for the asymmetry of the forward and reverse friction forces; For the frictional force of the joint, The velocity of the joint; Then for robot joints j Friction parameters of the joint .

3. The robot fault diagnosis method based on dynamic parameter evolution as described in claim 1, characterized in that, The robot was run multiple times on the excitation trajectory under fault-free conditions, and the results were recorded. and Thus, the corresponding multiple sets are calculated. This is used as the fault-free parameter set.

4. The robot fault diagnosis method based on dynamic parameter evolution as described in claim 3, characterized in that, By comparing the current parameter set with the corresponding parameters in the fault-free parameter set, the deviation of each parameter is determined; for any parameter i Its deviation The calculation formula is as follows: in, , These are the parameters in the fault-free parameter set. i The mean and standard deviation; To apply a moving average filter to the parameters in the current parameter set i The current mean obtained after processing.

5. The robot fault diagnosis method based on dynamic parameter evolution as described in claim 4, characterized in that, For any parameter i The deviation was obtained. Then, the parameter changes are quantified using the following confidence function to determine whether the parameters are abnormal; the confidence function is: in, The quantification results of parameter changes, These are the upper and lower bounds, which are set according to different parameter characteristics.

6. The robot fault diagnosis method based on dynamic parameter evolution as described in any one of claims 1-5, characterized in that, When determining the excitation trajectory, a matrix is ​​used. The objective function is to minimize the condition number. By changing the constraints, multiple sets of excitation trajectories are obtained. After running the excitation trajectories, multiple sets of corresponding excitation trajectories are obtained. and Thus, the least squares method is used to obtain And then according to Seeking .

7. The robot fault diagnosis method based on dynamic parameter evolution as described in claim 6, characterized in that, The changed constraint conditions are: changing the constraint range of robot joint rotation angle, speed, acceleration, and torque limits.

8. A robot fault diagnosis system based on dynamic parameter evolution, characterized in that, Includes a processor for executing the robot fault diagnosis method based on the evolution of dynamic parameters as described in any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the robot fault diagnosis method based on the evolution of dynamic parameters as described in any one of claims 1-7.

Images