Method for removing and implementing vibration strong interference in force signal of polishing robot

Abstract

The application discloses a method for removing and realizing strong vibration interference in a force signal of a polishing robot, two filter algorithms are connected in series to form a double Kalman filter, a first filter realizes filtering of Gaussian white noise and high-frequency vibration spectrum peak group noise; a single parameter is introduced, a gradually time-varying noise variance is designed in an exponential weighting mode to describe the colored noise characteristics, and an improved Kalman filter is designed to form a second filter to realize filtering and elimination of the colored noise. The application realizes effective filtering and elimination of complex noise composed of Gaussian white noise, vibration noise and colored noise superimposed in force measurement information in robot force control polishing.

Description

Technical Field

[0001] This invention relates to the field of robotics, specifically to a method for removing and implementing strong vibration interference in force signals of a grinding robot. Background Technology

[0002] Robots achieve their actions (such as grinding, polishing, and deburring) by controlling the contact force of their robotic arms. Controlling this contact force first requires force or torque sensors to acquire force or torque signals. However, these signals are often affected by environmental factors (natural and working environments), resulting in complex noise. For example, during grinding operations, the force sensor at the end of the robotic arm is affected by the vibration of the grinding wheel (e.g., the material of the flexible workpiece, the contact environment, and the contact depth). The force signal it acquires is submerged in the vibration interference environment containing white and colored noise, directly impacting the force control effect based on this signal.

[0003] Because of its low computational cost, the classic Kalman filter algorithm is widely used for filtering various real-time signals. However, its filtering effect deteriorates when faced with environmental noise such as process noise, observation noise, and vibration. Therefore, many studies have improved the Kalman filter algorithm. In the research direction of adaptive Kalman filter algorithms, the main methods include correlation, covariance matching, maximum likelihood, and Bayesian methods. In the research direction of intelligent Kalman filter algorithms, radial basis function neural networks and reinforcement learning methods are used to compensate for Kalman filter algorithm errors. Although these improved Kalman filter algorithms can improve filtering accuracy, they inevitably increase the computational load. Considering the complex noise in the force signal of robots during grinding operations, finding an improved Kalman filter algorithm that can extract real and effective force signals in real time from environmental noise containing white and colored noise such as vibration, with low computational complexity, is an urgent problem to be solved in robot grinding operations. Summary of the Invention

[0004] The present invention aims to address the problem of complex noise in the force signals collected by robots during grinding operations, and provides a method for removing and implementing strong vibration interference in the force signals of grinding robots.

[0005] To solve the above problems, the present invention is achieved through the following technical solution:

[0006] A method for removing and implementing strong vibration interference in the force signal of a grinding robot includes the following steps:

[0007] Step 1: The force sensor at the end of the robot arm collects force signals in real time, obtaining the force signal observation value z at the current time k. k ;

[0008] Step 2: Use the classic Kalman filter algorithm to remove Gaussian white noise and high-frequency vibration spectrum peak group noise from the force signal observation value at time k obtained in Step 1, to obtain the intermediate state estimate of the force signal at time k. The state update formula for the classic Kalman filter algorithm is as follows:

[0009]

[0010] Step 3: Use the improved Kalman filter algorithm to remove the intermediate state estimate of the force signal at time k obtained in Step 2. The colored noise in the signal is used to obtain the final state estimate of the force signal at time k. This is to remove strong vibration interference from the force signal; the state update formula of the improved Kalman filter algorithm is as follows:

[0011]

[0012] Among them, z k The force signal observation value at time k is the current time. This is the estimated intermediate state value of the force signal at time k. The intermediate state estimate of the force signal at the previous time k-1 is given at the current time. This is the estimated final state value of the force signal at time k. The final state estimate of the force signal at the previous time k-1 is given by the current time. Let Variance be the intermediate estimate error of the time step k-1 before the current time step. is the final estimated error variance of the previous time k-1; Q is the system process noise variance, R is the observation noise variance, and s is a manually set parameter, where 1 < s < 1.1.

[0013] In the above scheme, the estimated intermediate state of the force signal at the previous time k-1 is... initial value The final state estimate of the force signal at the previous time k-1. initial value The variance of the intermediate estimation error at the previous time k-1. initial value And the final estimate error variance of the previous time k-1 before the current time. initial value All are given in advance.

[0014] In step 2, the update formula for the intermediate estimation error variance of the classic Kalman filter algorithm is:

[0015]

[0016] in, Let Variance be the intermediate estimate error at time k. Let Q be the variance of the intermediate estimation error at the previous time k-1, Q be the variance of the system process noise, and R be the variance of the observation noise.

[0017] In step 3, the update formula for the final estimation error variance of the improved Kalman filter algorithm is:

[0018]

[0019] in, Let be the variance of the final estimate error at time k. Let Q be the variance of the final estimated error from the previous time k-1, let R be the variance of the system process noise, let s be the variance of the observation noise, and let s be a parameter set manually.

[0020] Compared with existing technologies, this invention addresses the problem of complex noise in force signals, which consists of the superposition of Gaussian white noise, vibration noise, and colored noise. It proposes a dual Kalman filter (DKF) based on a series structure. The first filter employs the classic Kalman filtering algorithm, primarily filtering out Gaussian white noise and high-frequency vibration spectrum peak noise. To address the colored noise formed by the superposition of Gaussian white noise and low-frequency vibration noise, which the first filter fails to filter, a single parameter 's' is introduced, and a gradually time-varying noise variance is designed using an exponential weighting method to describe the characteristics of the colored noise. This improved Kalman filtering algorithm serves as the second filter. The two filters are connected in series to form a dual Kalman filter, effectively filtering and eliminating the complex noise composed of the superposition of Gaussian white noise, vibration noise, and colored noise in the force measurement information during robot force-controlled grinding. This effectively eliminates strong interference signals in the force sensor output during the robot grinding process. Attached Figure Description

[0021] Figure 1 This is a flowchart of the Dual Kalman Filter (DKF).

[0022] Figure 2 This is a schematic diagram of colored noise cascade filtering.

[0023] Figure 3 The time-domain effect of filtering out strong vibration interference in a force signal using the dual Kalman filter (DKF) algorithm is shown.

[0024] Figure 4 The time-domain effect of filtering out strong vibration interference in a force signal using the dual Kalman filter (DKF) algorithm is shown. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific examples.

[0026] A method for removing and implementing strong vibration interference in the force signal of a grinding robot, such as... Figure 1 As shown, the specific process is as follows:

[0027] Step 1: The force sensor at the end of the robot arm collects force signals in real time to obtain the force signal observation value z at the current time k. k .

[0028] Step 2: Use the classic Kalman filter algorithm to remove Gaussian white noise and high-frequency vibration spectrum peak group noise from the force signal observation value at time k obtained in Step 1, to obtain the intermediate state estimate of the force signal at time k.

[0029] The state update formula for the classic Kalman filter algorithm is:

[0030]

[0031] Among them, z k The force signal observation value at time k is the current time. This is the estimated intermediate state value of the force signal at time k. The intermediate state estimate of the force signal at the previous time k-1 is given at the current time. Let Q be the intermediate estimation error variance of the previous time k-1, Q be the system process noise variance (calculated by the variance definition based on the sampling force information), and R be the observation noise variance (calculated by the variance definition based on the first observation state). initial value The force signal observation value z collected by the force sensor is set as follows. k The initial value of z0, initial value Set it to between 1 and 1.5.

[0032] The first filter uses the classic KF algorithm to correct the previous state estimate based on the current state observation and the real-time updated Kalman gain, thereby calculating the optimal state estimate for the current moment, which is the force measurement estimate obtained after the first filter.

[0033] For the measurement process, the state-space model of the classic KF algorithm is described by the following predicted state and observation equations:

[0034]

[0035] Where, x k and xk-1 These are the state values ​​at time k and time k-1, respectively; z k The value is the observation at time k (force sensor output); the process noise w at time k-1. k-1 The observation noise v at time k k The noise components are independent of each other and are all Gaussian white noise, belonging to a normal distribution; Q is the system process noise variance (calculated by the variance definition based on the sampling force information), and R is the observation noise variance (calculated by the variance definition based on the first observation state).

[0036] Calculate the intermediate prediction error variance P at time k. k :

[0037]

[0038] Intermediate Kalman gain K k for:

[0039]

[0040] Based on equations (3)-(4) and the observed value z k The state prediction at time k is corrected by weighted averaging to obtain the intermediate state estimate at time k.

[0041]

[0042] Update the intermediate estimate error variance using Kalman gain and prediction error variance. Then iteratively calculate the state at time k+1:

[0043]

[0044] Step 3: Use the improved Kalman filter algorithm to remove the intermediate state estimate of the force signal at time k obtained in Step 2. The colored noise in the signal is used to obtain the final state estimate of the force signal at time k. This is to remove strong vibration interference from the force signal.

[0045] The state update formula for the improved Kalman filter algorithm is:

[0046]

[0047] in, This is the estimated intermediate state value of the force signal at time k. This is the estimated final state value of the force signal at time k. The final state estimate of the force signal at the previous time k-1 is given by the current time. is the final estimated error variance of the previous time k-1; Q is the system process noise variance, R is the observation noise variance, and s is a manually set parameter, where 1 < s < 1.1. initial value The force signal observation value z collected by the force sensor is set as follows. k The initial value of z0, initial value Set it to between 1 and 1.5.

[0048] Considering that the first filter uses the classic KF algorithm, it can only remove white noise, but cannot effectively remove the colored noise contained in the complex noise of the sensor force signal. Therefore, this invention considers adding a second filter in series on the basis of the first filter to remove colored noise.

[0049] Given that the colored noise is specifically a first-order Gaussian-Markov model:

[0050] δ k =αδ k-1 +ε k (8)

[0051] Where: δ k and δ k-1 These are the colored noises at time k and k-1, respectively, where α is the weight and ε is the value of the noise. k The mean is 0 and the variance is σ at time k. k 2 Gaussian white noise.

[0052] As time k increases, the colored noise sequence becomes:

[0053]

[0054] Then the variance γ of colored noise k 2 =Var(δ k )for:

[0055] Var(α k-1 δ1+α k-2 ε2+α k-3 ε3+...+α 0 ε k (10)

[0056] Because α in formula (10) k-1 δ1, α k-2 ε2, α k-3 ε3、...、α 0 ε k Since they are mutually independent, δ1 = ε1. According to the variance property, equation (10) can be simplified to:

[0057]

[0058] From equation (11), it can be seen that the characteristic of the colored noise variance is that the colored noise variance at time k is the sum of the Gaussian white noise variances multiplied by the corresponding weights at the previous k times. Then, by using different numbers of KF algorithms in a cascaded structure to gradually filter out the corresponding Gaussian white noise related terms, the... Figure 2 This indicates that the filtering process for colored noise at time k is equivalent to k KF algorithms cascaded together, which respectively filter out the product of the variance of Gaussian white noise and the corresponding weights at the previous k times.

[0059] Although Gaussian white noise is a stationary random process, its variance at each time step is difficult to obtain in practical engineering applications. Let:

[0060] σ1 2 =σ2 2 =σ3 2 =...=σ k 2 =σ 2 (12)

[0061] As time k increases, the colored noise variance sequence becomes:

[0062]

[0063] From equation (13), we know that the sum of weights β must be greater than 1; regardless of the value of weight α, the variance of colored noise increases with the increase of time k, i.e., γ k 2 >γ k-1 2 .

[0064] Inspired by the above analysis, this invention considers that using multiple KF algorithms to filter colored noise at each recursive time step would greatly increase the algorithm complexity. Furthermore, as analyzed in (13), the variance gradually increases due to multiple KF filters. Therefore, an appropriate time-varying parameter is introduced to approximate the variance of the colored noise. A KF algorithm based on the modified variance is used to gradually eliminate colored noise during the recursive process, replacing the use of multiple KF algorithms to filter colored noise at once, which is easier to implement in engineering. Based on this idea, this invention proposes an improved KF algorithm.

[0065] Based on the above analysis, the expected distribution characteristics of colored noise can be described as follows:

[0066]

[0067]

[0068] In the formula, w k-1It is the process noise at time k-1. It is the transpose of the process noise at time k-1, v k It is the observation noise at time k. Let be the transpose of the observation noise at time k, Q be the system process noise variance (calculated from the variance definition based on sampling force information), and R be the observation noise variance (calculated from the variance definition based on the first observation state). s is a single design parameter. When s > 1, the noise variance gradually increases with time k, but the increase in observation noise variance is greater, thus increasing the degree of distrust in the observation noise.

[0069] From equations (14) and (15), the improved KF recursive algorithm is derived as follows:

[0070]

[0071] In the formula, and These are the final estimated values ​​of the force signal at time k-1 and time k, respectively, after the second filtering. These are the Kalman gain and estimation error variance at time k of the second filtering, respectively. The initial value is obtained from the output of the first filter, i.e.

[0072] Compared to the classic KF algorithm, the improved KF algorithm introduces a parameter s to describe the variance of colored noise. The variance of the estimation error at the previous time step is determined by introducing parameter s. Both the system process noise variance Q and the variance of the KF algorithm are reduced compared to the classic KF algorithm, thus changing the estimation error variance at the current time step. This leads to a Kalman gain Compared to the classic KF algorithm, the size is also smaller, and the state estimate of the previous time step has a greater effect on the state estimate of the current time step, thus effectively avoiding the influence of colored noise on the true power signal.

[0073] Figure 3 The figure shows the time-domain effect of filtering out strong vibration interference in the force signal using the dual Kalman filter (DKF) algorithm. As can be seen from the figure, the DKF algorithm effectively filters out white and colored noise in the force sensor output signal. Figure 4The image shows the frequency domain effect of the DKF algorithm for filtering and removing strong vibration interference from force signals. As can be seen, in the DKF filter, the first classical KF algorithm effectively filters and removes Gaussian white noise and high-frequency vibration spectrum peak noise. Based on the KF algorithm, the second improved KF algorithm effectively removes colored noise. In summary, the two filters, connected in series, form a dual Kalman filter (the first filter uses the classical KF algorithm, and the second filter uses the improved KF algorithm). This effectively filters and removes the complex noise composed of Gaussian white noise, vibration noise, and colored noise superimposed in the force measurement information during robot force-controlled grinding, effectively eliminating strong interference signals from the force sensor output during robot grinding.

[0074] It should be noted that although the embodiments described above are illustrative, they are not intended to limit the invention. Therefore, the invention is not limited to the specific embodiments described above. Any other embodiments obtained by those skilled in the art under the guidance of this invention without departing from its principles are considered to be within the protection scope of this invention.

Claims

1. A method for removing and implementing strong vibration interference in the force signal of a grinding robot, characterized by: The steps include the following: Step 1: The force sensor at the end of the robot arm collects force signals in real time, obtaining the force signal observation value z at the current time k. k ; Step 2: Use the classic Kalman filter algorithm to remove Gaussian white noise and high-frequency vibration spectrum peak group noise from the force signal observation value at time k obtained in Step 1, to obtain the intermediate state estimate of the force signal at time k. The state update formula for the classic Kalman filter algorithm is as follows: Step 3: Use the improved Kalman filter algorithm to remove the intermediate state estimate of the force signal at time k obtained in Step 2. The colored noise in the signal is used to obtain the final state estimate of the force signal at time k. To remove strong vibration interference from the force signal; The state update formula for the improved Kalman filter algorithm is: The update formula for the final estimation error variance of the improved Kalman filter algorithm is as follows: Among them, z k The force signal observation value at time k is the current time. This is the estimated intermediate state value of the force signal at time k. The intermediate state estimate of the force signal at the previous time k-1 is given at the current time. This is the estimated final state value of the force signal at time k. The final state estimate of the force signal at the previous time k-1 is given by the current time. Let Variance be the intermediate estimate error of the time step k-1 before the current time step. Let be the variance of the final estimate error at time k. Let be the variance of the final estimated error from the previous time k-1; Q be the variance of the system process noise, R be the variance of the observation noise, and s be a manually set parameter, where 1 <s<1.1。 2. The method for removing and implementing strong vibration interference in the force signal of a grinding robot according to claim 1, characterized in that, The intermediate state estimate of the force signal at the previous time k-1. initial value The final state estimate of the force signal at the previous time k-1. initial value The variance of the intermediate estimation error at the previous time k-1. initial value And the final estimate error variance of the previous time k-1 before the current time. initial value All are given in advance.

3. In the method for removing and implementing strong vibration interference in the force signal of a grinding robot according to claim 1, in step 2, the update formula for the intermediate estimation error variance of the classic Kalman filter algorithm is: in, Let Variance be the intermediate estimate error at time k. Let Q be the variance of the intermediate estimation error at the previous time k-1, Q be the variance of the system process noise, and R be the variance of the observation noise.

Images