Multi-Type Flutter Detection Method for Robot Milling

Abstract

A multi-type chatter detection method for robotic milling solves the problems that the existing robotic chatter detection methods cannot exclude the influence of the flexible vibration of the robot on chatter detection and cannot identify large-amplitude flexible vibrations during the milling process of the robot, and belongs to the field of milling processing. The present invention synchronously collects multi-channel vibration signals and signals of the internal controller of the robot, and transforms the multi-channel vibration signals into the tool feed coordinate system; based on the characterization of the milling vibration signals and the sensitivity analysis of the vibration signals to chatter, the filtered vibration displacement signal, the filtered vibration acceleration signal and the internal signal are combined to detect low-frequency chatter dominated by the structural mode of the robot, high-frequency chatter dominated by the tool mode, and large-amplitude flexible vibrations, and to determine the main chatter direction. The present invention improves the effectiveness of robotic milling chatter detection.

Description

Technical Field

[0001] The present invention relates to a method for detecting chatter in robot milling, belonging to the field of milling machining. Background Art

[0002] Chatter is extremely harmful to the machining process, which can lead to serious tool wear and robot failures, and significantly reduce the dimensional accuracy, surface quality and machining efficiency of robot milling. Online chatter detection technology can quickly detect chatter and help operators optimize the control of the machining process online or offline. However, several types of chatter occur during robot milling, including low-frequency chatter dominated by the robot structural mode and high-frequency chatter dominated by the tool mode. After detecting chatter, it is necessary to estimate the type of chatter occurrence to avoid the recurrence of the corresponding chatter.

[0003] The invention patent "An Online Monitoring Method for Milling Chatter" with the publication number CN108296881A proposes a chatter monitoring method, which decomposes the cutting force signal based on wavelet transform, establishes a multiple regression model, and realizes online monitoring of milling chatter; the invention patent "An Intelligent Monitoring System for Milling Chatter of a CNC Milling Machine" with the publication number CN113059402B proposes a milling chatter monitoring system, which includes acceleration, sound, and current sensors, a wireless data acquisition front end and an intelligent processing platform to realize the intelligent identification of milling chatter of the machine tool; the invention patent "An Online Monitoring Method for Milling Chatter Combining Comprehensive Energy Ratio and Amplitude Standard Deviation" with the publication number CN116423292A proposes an online monitoring method for milling chatter, which forms a monitoring index set through three-axis vibration acceleration signals and uses a trained detection model to monitor chatter.

[0004] The above methods are only applicable to machine tool milling. When the machine tool mills at a relatively high spindle speed, low-frequency chatter dominated by the structural mode usually does not occur, so the key detection is high-frequency chatter dominated by the tool mode. However, since the stiffness of the robot itself is usually lower than that of the spindle tool, both low-frequency and high-frequency chatter may occur at a relatively high spindle speed. In addition, the structural mode of the robot will interfere with signal acquisition and chatter detection during milling. For the detection of high-frequency chatter in the machine tool milling process, a common method for chatter detection is to find high-amplitude aperiodic frequency components near the tool mode frequency. In the high-speed milling process of the robot, there are high-amplitude low-frequency components similar to the structural natural frequency in the vibration signals of the stable and chatter states, which means that the frequency search method cannot be directly applied to the detection of low-frequency chatter. Therefore, there are significant differences between robot milling and traditional machine tool milling. During high-speed machining, both low-frequency and high-frequency chatter may occur in the robot milling process. It is extremely challenging to detect and classify various robot milling vibrations under the interference of various factors such as machining position, feed direction, and operating state.

[0005] The invention patent "A Method for Identifying Chatter in Robot Milling" with the publication number CN112529099A proposes a method for identifying chatter in robot milling. By collecting the surface topography image of the workpiece during robot milling and inputting it into the prediction model of the chatter type in milling, the chatter type of the robot is obtained. The prediction model of the chatter type in milling is trained through a publicly available dataset. The distinguishable chatter types include stable, excessive, regular chatter, and irregular chatter. It is impossible to distinguish between low-frequency and high-frequency chatter dominated by different modes, and it is difficult to realize the online detection of chatter in robot milling. The invention patent "A Method and System for Predicting the Stability of Low-Frequency Chatter in Robot Milling" with the publication number CN115890345A proposes a method for predicting the stability of low-frequency chatter in robot milling, which considers both the intermittent cutting characteristics and the modal coupling effect in milling and realizes the prediction of low-frequency chatter, but lacks in-depth research on high-frequency chatter. The invention patent "A Method for Identifying Chatter Types in Robot Milling Based on the Difference of Power Spectrum Entropy" with the publication number CN114800042A proposes a method for identifying chatter types in robot milling. By collecting the original vibration signal at the end of the robot to determine the optimal classification threshold of the power spectrum entropy difference, the identification of regenerative chatter caused by the flexibility of the tool-spindle structure and modal coupling chatter caused by the insufficient structural stiffness of the robot is realized. This method only uses the external vibration signal to calculate a chatter detection index, does not fully utilize the internal signal of the robot, cannot exclude the influence of the residual vibration of the robot on the chatter detection, and cannot identify the large-amplitude flexible vibration in the process of robot milling. Summary of the Invention

[0006] Aiming at the problem that the robot chatter detection method cannot exclude the influence of the residual vibration of the robot on the chatter detection and cannot identify the large-amplitude flexible vibration in the process of robot milling, the present invention provides a method for detecting multiple types of chatter in robot milling.

[0007] A method for detecting multiple types of chatter in robot milling according to the present invention includes:

[0008] S1. Synchronously collect the internal signal and external signal of the robot during robot milling. The internal signal is the controller signal of the robot, and the external signal is the vibration signal collected during robot milling. The vibration signal includes the acceleration and vibration displacement signals of each channel of the robot. Convert the vibration signal to the tool feed coordinate system;

[0009] S2. Process the vibration displacement signal in the external signal to eliminate the high-frequency component and the periodic component, and obtain the multi-channel displacement signal s of the robot milling process dis,k, process the multi-channel acceleration signals in the external signals to eliminate the low-frequency components, periodic components, and modulation components, and obtain the multi-channel acceleration signals during the robot machining process

[0010] S3. Use the internal signal as the input value to predict the statistical characteristics of the vibration displacement in each direction in the flange coordinate system during the robot's idle running process

[0011] S4. According to the multi-channel displacement signal s dis,k and the statistical characteristics of the vibration displacement in the corresponding direction Determine the amplitude index CSI of the vibration displacement at the robot's end R , determine the index CSI of the degree of fluctuation of the robot joint acceleration within the time window according to the internal signal θ , determine the high-frequency chatter detection index CSI according to the multi-channel acceleration signal T ;

[0012] S5. Chatter judgment:

[0013] When all the chatter detection indexes are lower than the corresponding thresholds, the robot milling process is in a stable state;

[0014] When CSI R is greater than CST R and CSI θ is less than CST θ , low-frequency chatter occurs in the robot milling process;

[0015] When CSI R and CSI θ both exceed the corresponding thresholds, large-amplitude flexible vibration occurs in the robot structure;

[0016] When CSI T is higher than the threshold CST T , and other indexes are all lower than the corresponding thresholds, it is determined that high-frequency chatter occurs in the robot milling system;

[0017] CST R 、CST θ and CST T are respectively the amplitude threshold of the vibration displacement at the robot's end, the threshold CSI θ of the degree of fluctuation of the robot joint acceleration within the time window, and the threshold of high-frequency chatter detection.

[0018] Preferably, in S1, the method of converting the vibration signal to the tool feed coordinate system is:

[0019]

[0020] s en represents the vibration signal in the tool feed coordinate system, s rfcs represents the vibration signal in the flange coordinate system, is the rotation matrix in is the rotation matrix in is the rotation matrix in represents the homogeneous transformation matrix from the base coordinate system to the workpiece coordinate system matrix, represents the homogeneous transformation matrix from the base coordinate system to the flange coordinate system matrix, represents the homogeneous transformation matrix from the flange coordinate system to the tool coordinate system matrix, L tool,x 、L tool,y and L tool,z respectively represent the position values of the TCP in the flange coordinate system; o x 、o y 、o z represent the coordinate system values of the z-axis of the tool feed coordinate system, v x 、v y 、v z represent the coordinate system values of the y-axis of the tool feed coordinate system, u x 、u y 、u z represent the coordinate system values of the x-axis of the tool feed coordinate system, represents the position of the origin of the tool feed coordinate system in the workpiece coordinate system.

[0021] Preferably, S2 includes:

[0022] The external signal is processed by the hierarchical sliding window sampling method. Among them, the main window is used to detect the external signal, and the vibration signal converted to the tool feed coordinate system is filtered to eliminate high-frequency components and periodic components, obtaining the displacement signals s dis,k of each channel in the robot milling process. The secondary window is used to detect the external signal, the cut-off frequency is obtained according to the power spectral density of the acceleration signal detected by the previous main window, and the multi-channel acceleration in the robot milling process is filtered by using the cut-off frequency to eliminate low-frequency components, obtaining the multi-channel acceleration Then, the periodic components and modulation components are eliminated to obtain the multi-channel acceleration signal The length of the main window is greater than that of the secondary window, and the value of the subscript k represents each coordinate axis of the tool feed coordinate system.

[0023] Preferably, a method for filtering the vibration signal converted to the tool feed coordinate system:

[0024] Process the vibration signal converted to the tool feed coordinate system using a Butterworth high-pass filter with a cut-off frequency of 200 Hz and a comb filter to eliminate high-frequency components and periodic components.

[0025] Preferably, obtain the cut-off frequency according to the power spectral density of the acceleration signal detected in the previous main window, and use this cut-off frequency to filter the multi-channel acceleration during the robot milling process to eliminate low-frequency components and obtain the acceleration of the corresponding channel and then eliminate periodic components and modulation components to obtain the multi-channel acceleration signal The method:

[0026] Calculate the power spectral density of the acceleration signal in the previous main window, and then use the peak search algorithm to find the highest peak value of the power spectral density and its corresponding frequency within the set frequency range Use a high-pass filter with a cut-off frequency of to eliminate low-frequency components and obtain multi-channel acceleration Use 2 comb filters to eliminate the periodic components and modulation components in

[0027] Preferably, the first comb filter is designed according to the harmonic frequency where N1 is determined according to the sampling frequency and the rotation frequency; the second comb filter is designed according to the modulation frequency range When the CSI determined in the previous main window R is lower than the corresponding threshold, N3 is set to 5, and when the CSI determined in the previous main window R is higher than the corresponding threshold, N3 is set to N1 / 2.

[0028] Preferably, determine the amplitude index CSI of the vibration displacement at the end of the robot according to the multi-channel displacement signal s dis,k and the statistical characteristics of the vibration displacement in the corresponding direction The method: R :

[0029]

[0030] Preferably, determine the CSI of the degree of fluctuation of the robot joint acceleration within the time window according to the internal signal θ The method:

[0031] The CSI of the degree of fluctuation of the robot joint acceleration within the time window θ The method:

[0032]

[0033] θ j Represents the internal signals of the first three joints of the robot.

[0034] Preferably, according to the multi-channel acceleration signals of the robot Determine the high-frequency chatter detection index CSI T :

[0035]

[0036] Preferably, in S3, the root mean square value or the half-peak value of the vibration displacement is used as the prediction variable, and the robot joint angles and joint velocities are used as the input values to train a Gaussian process regression model to obtain the statistical characteristics of the vibration displacement in the flange coordinate system.

[0037] Advantages of the present invention: Compared with a single acceleration signal, the internal and external signal fusion proposed by the present invention can achieve more effective chatter detection, and can judge the main chatter direction according to the internal signal, solve the interference of the robot milling machining feed direction on chatter detection, and eliminate the influence of the robot residual vibration on chatter detection, improving the accuracy of online chatter detection. The mapping relationship between the robot pose information, joint torque and vibration signal helps to realize position-guided process monitoring. Based on the internal signal and the robot kinematic model, the synchronized vibration signal is transformed in different coordinate systems to better perform operational modal analysis, helping the operator to more deeply analyze the machining process and conduct process monitoring, especially for the low-frequency chatter dominated by the robot structural mode. The key points of chatter detection in robot milling machining are the low-frequency and modulated components of the vibration signal caused by the low stiffness characteristics of the robot. Different types of chatter have different sensitivities to the vibration signal. Among them, the acceleration signal is sensitive to high-frequency chatter, while the displacement signal is sensitive to low-frequency chatter. The present invention distinguishes the types of chatter in robot milling machining based on this. The vibration displacement amplitude and frequency distribution in the high-speed milling process and the idle running process are similar. The present invention uses the vibration displacement in the idle running process as a reference value for low-frequency chatter detection, and trains a Gaussian process regression model to predict the root mean square value or the half-peak value of the displacement signal as a reference for detecting low-frequency chatter. The proposed combination of chatter detection indexes of the present invention has clear physical meanings and can be used to detect various types of chatter. By removing noise, modulated signal components and periodic signal components, the interference on chatter detection is reduced, and the low-frequency chatter, large-amplitude flexible vibration and high-frequency chatter in robot milling machining can be effectively and timely detected. Description of the Drawings

[0038] Figure 1 Is the flowchart of the method of the present invention.

[0039] Figure 2Vibration displacement signals and their spectrograms in the x - direction of the tool feed coordinate system under different states. Among them, (a) and (d) are the displacement signals and their spectrograms in the idle - running state, (b) and (e) are the displacement signals and their spectrograms in the stable milling state, and (c) and (f) are the displacement signals and their spectrograms in the low - frequency chatter state.

[0040] Figure 3 Vibration acceleration signals and their spectrograms collected under different chatter conditions. Among them, (a) and (d) are the acceleration signals and their spectrograms in the stable cutting state, (b) and (e) are the acceleration signals and their spectrograms in the low - frequency chatter state, and (c) and (f) are the acceleration signals and their spectrograms in the high - frequency chatter state.

[0041] Figure 4 Acceleration signals and their spectrograms in the x - direction of the tool feed coordinate system. (a) is the acceleration signal, and (b) is the acceleration spectrogram, where a1 is the spindle idling and a2 is the milling process.

[0042] Figure 5 Schematic diagram of hierarchical sliding window sampling.

[0043] Figure 6 Vibration signals and chatter detection results. (a), (b), and (c) are the displacement signals, short - time Fourier transforms, and chatter detection indexes in the x, y, and z directions respectively, (d) is the internal signal of the robot, and (e) is the machined surface of the workpiece.

[0044] Figure 7 Specimen for chatter detection verification experiment.

[0045] Figure 8 Chatter detection results and joint angular acceleration signals during rough machining. Among them, (a) is the chatter detection index, and (b) is the robot joint angular acceleration signal.

[0046] Figure 9 Chatter detection results for finish machining of diamond contour. (a) is the tool path for finish machining of diamond contour, (b) is the torque signal of joint 1, (c) is the z - direction displacement signal in the tool feed coordinate system, and (d) is the chatter detection index.

[0047] Figure 10 Vibration signals and chatter detection results of slot milling experiment. Among them, (a) and (b) are the vibration acceleration signal and displacement signal in the tool feed coordinate system, (c) is the chatter detection index, and (d), (e), and (f) are the spectrograms of the acceleration signals in regions a1, a2, and a3 respectively. Specific implementation method

[0048] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0049] It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments may be combined with each other.

[0050] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, but it is not limited to the present invention.

[0051] The multi-type chatter detection method for robot milling machining in this embodiment includes:

[0052] Step 1: Synchronously collect the internal signals and external signals of the robot during robot milling machining. The internal signals are the controller signals of the robot, and the external signals are the vibration signals collected during the robot milling process. The vibration signals include the acceleration and vibration displacement signals of each channel of the robot. The multi-channel acceleration signals are collected by a triaxial accelerometer, and the accelerometer is installed on the electric spindle. The multi-channel vibration displacement signals are calculated from the corresponding channel acceleration signals. Convert the vibration signals to the tool feed coordinate system;

[0053] A high-speed electric spindle is installed at the end of the industrial robot, which can realize operations such as milling, drilling, and grinding. The monitoring system of the robot includes a robot controller and external sensors. The internal signals refer to the controller signals of the robot, such as joint angles, TCP positions in different coordinate systems, torque, current and other signals. Signals such as joint speed and acceleration need to be obtained through joint positions and sampling periods. External sensors include triaxial accelerometers, smartphones, dynamometers, laser trackers, etc. Considering different application scenarios, the internal signals and triaxial accelerometers are used as a common sensor combination. A triaxial accelerometer with a sensitivity of 100mv / g is installed on the spindle bracket to monitor the machining process.

[0054] Table 1 Internal Signals and Acceleration Signals

[0055]

[0056]

[0057] A robotic machining system typically includes the following coordinate systems: Robot basic coordinate system (RBCS), Robot flange coordinate system (RFCS), Accelerometer coordinate system (ACS), Tool coordinate system (TCS), Workpiece coordinate system (WCS), Engagement coordinate system (ECS), and Laser tracker coordinate system (LCS), etc. When installing the accelerometer, the accelerometer coordinate system is aligned with the flange coordinate system by adjusting the installation position. The base coordinate system and the flange coordinate system are determined by the robot design parameters or calibrated by a tracker, and the origin of the tool coordinate system is set at the tip point.

[0058] In terms of signal acquisition and conversion, for the synchronously acquired three-axis acceleration signals and internal signals, the vibration signals of different chatter types collected, such as Figure 2 , 3 As shown, according to the C0 complexity and the spindle idle cutting vibration signal, a filter is designed to remove the noise related to the spindle operation. Based on the data collected inside the robot and the kinematic model, the obtained vibration signals are converted into the engagement coordinate system. The transformation matrix from the workpiece coordinate system to the tool coordinate system is calculated as follows:

[0059]

[0060] where is the homogeneous transformation matrix from the base coordinate system to the workpiece coordinate system, which is calculated from the values defined in the robot operation program; is the homogeneous transformation matrix from the base coordinate system to the flange coordinate system, which is calculated from the kinematic model and the joint angles; The homogeneous transformation matrix from the flange coordinate system to the tool coordinate system is calculated as follows:

[0061]

[0062] where L tool,x , L tool,y and L tool,z are the position values of the center position of the tool tip in the flange coordinate system.

[0063] For the i-th tool position The position of the tool feed coordinate system is determined by the current tool tip center position. The tool feed coordinate system in the workpiece coordinate system can be determined by the axis directions and feed direction of the tool coordinate system, and is calculated based on the tool position data defined in the workpiece coordinate system. The feed direction of the workpiece coordinate system is calculated based on two consecutive tool positions as follows:

[0064]

[0065] represents the position of the tool in the workpiece coordinate system;

[0066] Assuming that the Z-axis direction of the tool feed coordinate system is the same as the Z-axis direction of the tool coordinate system, then the axis directions of the tool feed coordinate system in the workpiece coordinate system are calculated as follows:

[0067]

[0068]

[0069]

[0070] Furthermore, the following variables are determined:

[0071] o x 、o y 、o z represent the coordinate values of the z-axis of the tool feed coordinate system, v x 、v y 、v z represent the coordinate values of the y-axis of the tool feed coordinate system, u x 、u y 、u z represent the coordinate values of the x-axis of the tool feed coordinate system, represents the position of the origin of the tool feed coordinate system in the workpiece coordinate system;

[0072] The rotation transformation matrix from the workpiece coordinate system to the tool feed coordinate system is calculated as follows:

[0073]

[0074] Then, the homogeneous transformation matrix from the workpiece coordinate system to the tool feed coordinate system is calculated as follows:

[0075]

[0076] represents the position of the origin of the tool feed coordinate system in the workpiece coordinate system;

[0077] The rotation transformation matrix from the flange coordinate system to the tool feed coordinate system can be obtained as follows:

[0078]

[0079] where is the rotation transformation matrix in the homogeneous transformation matrix from the flange coordinate system to the tool coordinate system, is the rotation matrix in is the rotation matrix in

[0080] Convert the vibration displacement in the flange coordinate system to the tool feed coordinate system:

[0081]

[0082] In the formula, s ecs is the vibration signal in the tool feed coordinate system; s rfcs is the vibration signal in the flange coordinate system.

[0083] Step 2: Process the vibration displacement signal in the external signal to eliminate the high-frequency components and periodic components, and obtain the multi-channel displacement signal s dis,k of the robot milling process. Process the multi-channel acceleration signal in the external signal to eliminate the low-frequency components, periodic components and modulation components, and obtain the multi-channel acceleration signal

[0084] Step 2 is a signal processing process. To ensure the accuracy of chatter identification and reduce the calculation time, mainly obtain the signals for low-frequency chatter detection, large-amplitude flexible vibration detection and robot operating frequency estimation, and the high-frequency chatter detection signal; In this embodiment, the hierarchical sliding window sampling method is used to process the external signal, as Figure 5 shown. The hierarchical sliding window sampling method means that a fixed window length is set at the beginning of the detection task. After the detection starts, a new sample with an offset length is inserted at the end of the original time window, and the signal of the same length in the start area of the original time window is deleted. The main window is used for low-frequency chatter detection, large-amplitude flexible vibration detection and robot operating frequency estimation, while the secondary window is used for high-frequency chatter detection. Since the dominant frequency of low-frequency chatter is much smaller than that of high-frequency chatter, too short a main window will make it difficult to effectively identify low-frequency chatter, so the length of the main window should be longer than that of the secondary window. To sum up, the number of samples in the main window of the acceleration signal is set to 3600, and the offset is set to 1200; the number of samples in the secondary window is set to 1200, and the offset of the secondary window is set to 600. The single sliding window sampling method is used to process the internal signal of the robot, and the number of window samples and the offset are 30 and 10 respectively. Step 2 is specifically:

[0085] Use the main window to detect the external signal to obtain the multi-channel displacement signal s dis,k of the robot milling process:

[0086] The original displacement signal is processed using a Butterworth high-pass filter with a cut-off frequency of 200 Hz and a comb filter to eliminate high-frequency components and periodic components. The frequency setting of the comb filter is determined according to the spindle rotation frequency and its harmonic frequencies

[0087] Detect the external signal using a sub-window, obtain the cut-off frequency according to the power spectral density of the acceleration signal detected in the previous main window, and use this cut-off frequency to perform filtering on the multi-channel acceleration during the robot milling process to eliminate low-frequency components and obtain multi-channel acceleration Then eliminate periodic components and modulation components to obtain a multi-channel acceleration signal Method:

[0088] Calculate the power spectral density P acc,i (f) of the acceleration signal in the previous main window

[0089] F acc,i (k) = FFT(s acc,i (t))

[0090]

[0091] Then use a peak search algorithm to find the highest peak value of the power spectral density and its corresponding frequency within the set frequency range

[0092] Use a high-pass filter with a cut-off frequency of to eliminate low-frequency components and obtain multi-channel acceleration

[0093] Use 2 comb filters to eliminate the periodic components and modulation components in to obtain a multi-channel acceleration signal

[0094] Step 3: Use the internal signal as the input value to predict the statistical characteristics of the vibration displacement in each direction under the flange coordinate system during the robot's no-load running process

[0095] In this embodiment, the joint angle and feed speed in the internal signal are used as the input values, and the statistical characteristics of the vibration displacement under the flange coordinate system are predicted based on the Gaussian process regression model, and the displacement statistical characteristics are converted to the tool feed coordinate system

[0096]

[0097] where q r(t) is the angular values of six joints and the velocities of three joints. The joint angles and the feed velocity are the average values of all data within the corresponding time window.

[0098] Step 4. According to the multi-channel displacement signal s dis,k and the statistical characteristics of the vibration displacement in the corresponding direction determine the amplitude index CSI of the vibration displacement at the end of the robot R , and according to the internal signal, determine the index CSI of the degree of fluctuation of the joint acceleration of the robot within the time window θ , and according to the multi-channel acceleration signal determine the high-frequency chatter detection index CSI T ;

[0099] The root mean square value or the half-peak peak value of the multi-channel displacement signal s dis,k is used to calculate the detection index CSI related to the low-frequency chatter of the robot R :

[0100] The root mean square calculation index CSI of the multi-channel displacement signal s dis,k : R :

[0101]

[0102] The half-peak peak value calculation index CSI of the multi-channel displacement signal s dis,k : R :

[0103]

[0104] CSI R is a real number between 0 and infinity, with a clear physical meaning, reflecting the amplitude of the vibration displacement at the end of the robot. During high spindle speed milling, it is close to the vibration displacement amplitude in the idling state. Therefore, the idling low-frequency vibration displacement is used as a reference value. When CSI R exceeds the set threshold CST R , it means that low-frequency chatter occurs in the robot milling system. Determine the threshold CST R according to experiments and simulations.

[0105] Calculate the second index CSI according to the internal signals of the first three joints of the robot θ,j to distinguish between low-frequency chatter and flexible vibration caused by inertial excitation.

[0106]

[0107] CSI θDescribes the degree of fluctuation of the robot joint acceleration within a time window, which is a real number ranging from 0 to infinity. According to the dry-running experiment signal, the threshold CST is determined based on experiments and simulations. θ When CSI R and CSI θ,j are both higher than the threshold, large-amplitude flexible vibrations with significant inertial excitation occur in the system at this time, and this flexible vibration can be suppressed by reducing the feed speed and acceleration.

[0108] Different from regenerative chatter, large-amplitude flexible vibrations tend to decay to a stable state within a time range of 3 - 10 windows. Therefore, when CSI θ,j exceeds the threshold CST θ,j for the first time, CSI R within this window and the subsequent 5 windows is set to 0.

[0109]

[0110] The transition process from the stable milling state to the chatter state reflects the energy conversion process between the periodic component and the chatter component. Therefore, the energy ratio of the filtered multi-channel acceleration signal to the original signal is defined as the third chatter detection index for detecting high-frequency chatter:

[0111]

[0112] CSI T is a real number between 0 and 1, which reflects the distribution and change of signal energy. Without considering the noise component, CSI T is close to 0 during the stable milling process, and when severe chatter occurs, CSI T is close to 1. According to the milling experiment results, the threshold CSI T is determined based on experiments and simulations.

[0113] When using two comb filters to eliminate the periodic component and the modulation component in : The first comb filter is designed according to the harmonic frequency , where N1 is determined based on the sampling frequency and the rotational frequency; the second comb filter is designed according to the modulation frequency range . When the previous CSI R is lower than the threshold, N3 is set to 5, and when CSI R is higher than the threshold, N3 is set to N1 / 2. To save calculation time, when the difference between the current window and the previous window's is less than 1 Hz, the filter parameter settings are no longer updated, and the initial comb filter can be set to the GRP model prediction value or 0.

[0114] Step 5, Flutter determination: Determine the type of flutter based on the above three flutter detection indicators. The four flutter states and their corresponding combinations of flutter detection indicators are shown in Table 2.

[0115] Table 2 Combinations of Flutter Detection Indicators and Their Corresponding Flutter States

[0116]

[0117]

[0118] When all flutter detection indicators are lower than their corresponding thresholds, the robotic milling process is in a stable state;

[0119] When CSI R is greater than CST R and CSI θ is less than CST θ the robotic milling process experiences low-frequency flutter;

[0120] When CSI R and CSI θ both exceed their corresponding thresholds, large-amplitude flexible vibrations occur in the robotic structure;

[0121] When CSI T is higher than the threshold CST T and other indicators are all lower than their corresponding thresholds, it is determined that high-frequency flutter occurs in the robotic milling system;

[0122] CST R 、CST θ and CST T are the amplitude threshold of the vibration displacement at the end of the robot, the threshold of the fluctuation degree of the joint acceleration of the robot within the time window CSI θ and the threshold for high-frequency flutter detection, respectively.

[0123] Based on the data-driven root mean square value or half-peak value prediction model of flexible vibration, higher prediction accuracy can be obtained, and complex dynamic models are not required. The vibration displacement of the robot is affected by multiple factors such as the robot pose, feed direction, feed speed, and cutting parameters. Therefore, it is unrealistic to collect vibration displacement signals under sufficient stable milling or flutter states. The vibration displacement amplitude and frequency distribution in the high-speed milling process and the no-load running process are similar. Therefore, the vibration displacement in the no-load running process is used as a reference value for low-frequency flutter detection, and a Gaussian process regression model is trained to predict the root mean square value or half-peak value of the displacement signal.

[0124] Statistical features such as the root mean square value or the peak-to-peak value of the vibration displacement are used as predictive variables, and the robot joint angles and joint velocities are used as input values to train a Gaussian process regression model. To improve the prediction accuracy, the Bayesian optimization method is used to optimize the hyperparameters. The sliding window method is adopted to process the vibration displacement data collected from the robot's no-load running experiment, and the dataset of the training samples is randomly divided into a training set and a validation set in a ratio of 3:1. Six models are trained respectively in different directions of the flange coordinate system to predict the root mean square value or the peak-to-peak value of the displacement signal for chatter detection.

[0125] In this embodiment, the chatter degree in each direction is analyzed through the chatter detection index, and the main chatter direction is judged according to the internal joint torque signal of the robot. The chatter index CSI is calculated through the vibration and external signals in the tool feed coordinate system. R and CSI T . As shown in (a), (b), and (c) of Figure 6 , the amplitude of the displacement signal and the energy of the low-frequency component increase with the occurrence of chatter. As shown in (d) of Figure 6 , chatter can be observed from the internal joint torque signal and the z-direction pose signal. The occurrence of chatter means the instability of the robot state. Therefore, the increase in the fluctuation amplitude of the internal signal can be regarded as a sign of the occurrence of chatter. Different directions of vibration displacement are affected by different joints. According to the signal similarity analysis, the vibrations in the y and z directions are mainly affected by joint 2, and the vibration in the x direction is mainly affected by joint 1.

[0126] For the three-way vibration signal, the moment when the chatter detection index CSI R exceeds the threshold is considered as the start time of chatter. For the internal signal of the robot, the moment when the waveform changes significantly is considered as the start time of chatter. The CSI R index in the x direction detects the occurrence of chatter at 3.48 seconds, which is close to the time when the waveform of the internal torque signal of joint 1 changes. According to the detection results in the x direction and the internal data, the estimated stability limit range is between 8 and 9 mm. As chatter develops, chatter is detected in the z and y directions at 3.84 s and 3.96 s respectively. Compared with the x direction, there is an obvious lag in the time when chatter is detected in the y and z directions. On the milling surface, chatter marks can be clearly observed when the axial cutting depth is 8.5 - 10 mm. Therefore, early detection of low-frequency chatter can be carried out through the calculation of multi-channel vibration signals.

[0127] To verify the effectiveness of the present invention, specific implementation cases of the present invention are provided. Using the time-frequency spectrogram and spectrogram of vibration signals, chatter detection is carried out by detecting the change of indicators over time. The short-time Fourier transform and fast Fourier transform are used to process the data. The total execution time of the three-channel data and the internal data of the sliding main window are within 40 ms, and the total execution time of the secondary window is within 7 ms. The corresponding calculation time is less than the offset of the main and secondary windows, meeting the requirements of online detection.

[0128] Regarding the identification of low-frequency chatter and large-amplitude flexible vibration, workpieces are designed according to GB / T 34880.2-2017, such as Figure 7 shown, and chatter detection is carried out during the milling process. In the rough machining stage, the chatter detection index CSI is calculated using the internal signals of the robot and multi-channel vibration signals R The curve is as Figure 8 shown. It can be obtained that no low-frequency chatter occurs during the rough machining process of this layer, but large-amplitude flexible vibration caused by inertial excitation appears in the area where the chatter detection index CSI R curve is 0. By comparing the chatter detection results and joint acceleration signals from 180 to 200 s in the figure, the internal and external fusion signals proposed by the present invention are beneficial to detecting large-amplitude flexible vibration.

[0129] Chatter detection and analysis are carried out on the finish machining process of diamond contour milling. Ignoring the process monitoring signals when the tool cuts in and out of the workpiece, only the internal and external signals during the milling process when the tool is fully cut into the workpiece are analyzed. The diamond milling trajectory is M→N→K→L, as Figure 9 shown in (a) of Figure 9 . The torque and displacement signals of robot joint 1 are as R shown in (b) and (c) of T , and it can be observed that the signal amplitude increases significantly when machining the N and L sides. According to the chatter indexes CSI Figure 9 shown in (d) of

[0130] , low-frequency chatter occurs when the tool cuts the N and L sides. Compared with chatter detection using only acceleration signals, the internal and external signal fusion can effectively distinguish low-frequency chatter from large-amplitude flexible vibration and simultaneously solve the interference of the feed direction on chatter detection during robot milling. Figure 10 shown, where a1, a2, and a3 represent the tool cutting-in, normal milling, and tool quickly lifting stages respectively. Through Figure 10The flutter detection index curve in (c) can detect high-frequency flutter, and the flutter frequency can be obtained according to the spectrogram of the acceleration filtered signal. When the tool is quickly lifted, the peak value of the vibration displacement exceeds 0.1 mm, and large-amplitude flexible vibration occurs. During this process, the tool collides with the workpiece, exacerbating the high-frequency vibration dominated by the tool mode. Therefore, only using the flutter detection index CSI T will result in false alarms of high-frequency flutter. The combination of flutter detection indexes used in the present invention can better reflect the actual situation of robot milling machining.

[0131] In this embodiment, multi-channel vibration signals and robot internal controller signals are synchronously collected, and the multi-channel vibration signals are transformed into the tool feed coordinate system; based on the characterization of milling vibration signals and the sensitivity analysis of vibration signals to flutter, the filtered vibration displacement signal, the filtered vibration acceleration signal and the internal signal are combined to detect low-frequency flutter dominated by the robot structure mode, high-frequency flutter dominated by the tool mode, and large-amplitude flexible vibration, and to judge the main flutter direction. The aim is to solve the problems of large-amplitude flexible vibration caused by robot pose, feed direction, and inertial excitation, and the interference of spindle noise on flutter detection in robot milling machining, and to improve the effectiveness of robot milling flutter detection.

[0132] Although the present invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the present invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed, as long as they do not deviate from the spirit and scope of the present invention as defined by the appended claims. It should be understood that the different dependent claims and the features described herein can be combined in a manner different from that described in the original claims. It should also be understood that the features described in connection with a single embodiment can be used in other described embodiments.

Claims

1. A multi-type chatter detection method for robotic milling, characterized in that, Including: S1. Synchronously collect the internal signals and external signals of the robot during robot milling. The internal signals are the controller signals of the robot, and the external signals are the vibration signals collected during the robot milling process. The vibration signals include the acceleration and vibration displacement signals of each channel of the robot. Convert the vibration signals to the tool feed coordinate system; S2. Process the vibration displacement signal in the external signal to eliminate the high-frequency components and periodic components, and obtain the multi-channel displacement signal s of the robot milling process dis,k , process the multi-channel acceleration signal in the external signal to eliminate the low-frequency components, periodic components and modulation components, and obtain the multi-channel acceleration signal of the robot processing process S3. Using the internal signal as the input value, predict the statistical characteristics of the vibration displacement in each direction under the flange coordinate system during the robot's dry running S4. Determine the amplitude index CSI of the vibration displacement at the end of the robot according to the multi-channel displacement signal s dis,k and the statistical characteristics of the vibration displacement in the corresponding direction ; determine the degree-of-fluctuation index CSI of the acceleration of the robot joints within the time window according to the internal signal R ; determine the high-frequency chatter detection index CSI according to the multi-channel acceleration signal θ ; T ​​ S5. Chatter determination: When all chatter detection indicators are lower than the corresponding thresholds, the robot milling process is in a stable state; When CSI R is greater than CST R and CSI θ is less than CST θ a low-frequency chatter occurs during the robotic milling process; When CSI R and CSI θ both exceed their respective thresholds, large-amplitude flexible vibrations occur in the robot structure; When CSI T is higher than the threshold CST T , and other indicators are all lower than the corresponding thresholds, it is determined that high-frequency chatter occurs in the robot milling system; CST R 、 CST θ and CST T are respectively the amplitude threshold of the vibration displacement at the end of the robot, the threshold CSI of the fluctuation degree of the joint acceleration of the robot within the time window θ and the threshold for high-frequency chatter detection.

2. The multi-type chatter detection method for robotic milling according to claim 1, wherein In S1, the method for converting the vibration signals to the tool feed coordinate system is: s ecs represents the vibration signal in the tool feed coordinate system, s rfcs represents the vibration signal in the flange coordinate system, is the rotation matrix in is the rotation matrix in is the rotation matrix in represents the homogeneous transformation matrix from the base coordinate system to the workpiece coordinate system, represents the homogeneous transformation matrix from the base coordinate system to the flange coordinate system, represents the homogeneous transformation matrix from the flange coordinate system to the tool coordinate system, L tool,x 、L tool,y and L tool,z respectively represent the position values of TCP in the flange coordinate system; o x 、o y 、o z represent the coordinate system values of the z-axis of the tool feed coordinate system, v x 、v y 、v z represent the coordinate system values of the y-axis of the tool feed coordinate system, u x 、u y 、u z represent the coordinate system values of the x-axis of the tool feed coordinate system, represents the position of the origin of the tool feed coordinate system in the workpiece coordinate system.

3. The multi-type chatter detection method for robotic milling according to claim 1, characterized in that, The S2 includes: The hierarchical sliding window sampling method is adopted to process the external signal. Among them, the main window is used to detect the external signal, and the vibration signal converted to the tool feed coordinate system is filtered to eliminate the high-frequency components and periodic components, and the displacement signals s of each channel in the robot milling process are obtained. dis,k , the secondary window is used to detect the external signal, the cut-off frequency is obtained according to the power spectral density of the acceleration signal detected by the previous main window, and the multi-channel acceleration in the robot milling process is filtered by using the cut-off frequency. The low-frequency components are eliminated to obtain the multi-channel acceleration. Then, the periodic components and modulation components are eliminated to obtain the multi-channel acceleration signal. The length of the main window is greater than that of the secondary window, and the value of the subscript k represents each coordinate axis of the tool feed coordinate system.

4. The multi-type chatter detection method for robotic milling according to claim 3, characterized in that The method for filtering the vibration signals converted to the tool feed coordinate system: Use a Butterworth high-pass filter with a cut-off frequency of 200 Hz and a comb filter to process the vibration signals converted to the tool feed coordinate system to eliminate high-frequency components and periodic components.

5. The multi-type chatter detection method for robotic milling according to claim 3, characterized in that, Obtain the cut-off frequency based on the power spectral density of the acceleration signal detected by the previous main window, and use this cut-off frequency to filter the multi-channel acceleration during the robot milling process to eliminate the low-frequency components and obtain the acceleration of the corresponding channel Then eliminate the periodic components and modulation components to obtain the multi-channel acceleration signal Method: Calculate the power spectral density of the previous main window acceleration signal, and then use the peak search algorithm to find the highest peak of the power spectral density and its corresponding frequency within the set frequency range Use a high-pass filter with a cut-off frequency of to eliminate the low-frequency components and obtain the multi-channel acceleration Use two comb filters to eliminate the periodic components and modulation components in 6. The multi-type chatter detection method for robot milling according to claim 5, wherein The first comb filter is designed according to the harmonic frequency where N1 is determined according to the sampling frequency and the rotation frequency; the second comb filter is designed according to the modulation frequency range When the CSI determined by the previous main window R is lower than the corresponding threshold, N3 is set to 5, and when the CSI determined by the previous main window R is higher than the corresponding threshold, N3 is set to N1 / 2.

7. The multi-type chatter detection method for robotic milling according to claim 1, wherein According to the multi-channel displacement signal s dis,k and the statistical characteristics of the vibration displacement in the corresponding direction determine the amplitude index CSI of the vibration displacement at the end of the robot R :

8. The multi-type chatter detection method for robotic milling according to claim 1, characterized in that Determine the CSI, an index of the fluctuation degree of the robot joint acceleration within a time window, based on the internal signal θ : Robot joint acceleration fluctuation degree index CSI within the time window θ : θ j Represents the internal signals of the first three joints of the robot.

9. The multi-type chatter detection method for robotic milling according to claim 1, wherein, Based on the multi-channel acceleration signals of the robot Determine the high-frequency flutter detection index CSI T :

10. The multi-type chatter detection method for robot milling according to claim 1, wherein, In S3, use the root mean square value or the half-peak peak value of the vibration displacement as the prediction variable, and the robot joint angles and joint speeds as the input values to train a Gaussian process regression model to obtain the statistical characteristics of the vibration displacement in the flange coordinate system.

Images