Flutter feature extraction method based on differential principal component entropy

Abstract

The invention discloses a flutter feature extraction method based on differential principal component entropy. The method belongs to the technical field of mechanical system control and fault diagnosis, and comprises the following operation steps: selecting mounting positions of a vibration sensor and a rotating speed sensor, and determining relevant parameters of data acquisition and feature extraction; acquiring synchronous vibration data Y; expanding the synchronous vibration data Y into multi-dimensional data P; performing mu-step differential processing on the multi-dimensional data P to obtain multi-dimensional data P mu; dimension reduction processing is carried out on the multi-dimensional data P [mu], and principal components are selected; calculating [mu] step differential principal component entropy; the mu is made to be equal to mu + 1, the step (4) is executed again, and differential principal component entropy calculation is continued till the difference principal component entropy of the mu max step is calculated; and taking all the calculated differential principal component entropies as extracted flutter characteristics. According to the flutter feature extraction method based on the differential principal component entropy disclosed by the invention, a plurality of flutter features can be extracted; the extracted flutter characteristics can better describe different flutter types; the flutter feature extraction method based on the differential principal component entropy disclosed by the invention can be applied to flutter control, flutter detection and diagnosis.

Description

Technical Field

[0001] The present invention belongs to the technical field of control and fault diagnosis of rotary machinery systems, and relates to a method for extracting chatter characteristics based on Difference Principal Entropy (DPE). Background Art

[0002] In the field of modern high-precision machining, chatter, as a typical non-linear self-excited vibration phenomenon, has become an important technical bottleneck restricting the machining process limit. Research shows that during the machining process of rotary machinery such as milling, when the cutting dynamics system parameters (including spindle speed, cutting depth, etc.) break through the boundary of the stability lobe diagram, the system will enter the chatter state from the non-chatter state. This non-linear instability phenomenon will not only cause problems such as abnormal increase in machining surface waviness and abnormal tool wear, but also lead to deterioration of workpiece surface integrity and sudden change in the dynamic load of the machining system, seriously affecting the machining qualification rate of high-value-added components and the overall efficiency of the production line. Therefore, it is necessary to detect and identify milling chatter.

[0003] In addition, in some non-typical milling machining scenarios, in order to improve machining efficiency and ensure a certain machining accuracy, it is sometimes allowed for the milling machining system to machine workpieces in the periodic bifurcation chatter state. At this time, it is necessary to identify the chatter type to ensure that the milling machining is in the periodic bifurcation chatter state.

[0004] Currently, the existing methods for extracting chatter characteristics mainly extract chatter characteristics from various signals in the milling process. Commonly used signals include vibration signals, electrical signals, and sound signals, etc., among which vibration signals are the commonly used signals for extracting chatter characteristics.

[0005] The methods for extracting chatter characteristics based on vibration signal analysis have made remarkable progress in recent years, among which the extraction methods based on entropy are particularly prominent. Commonly used entropies include sample entropy, approximate entropy, fuzzy entropy, diversity entropy, permutation entropy, and asynchronization entropy, etc. Although the chatter characteristics extracted by the above entropy methods can be used for chatter identification, they cannot be used for the identification of chatter types. In order to identify chatter types, Chinese Patent No. 202410194548.8 proposes a multi-resolution asynchronization entropy method. Although this method has made breakthroughs in chatter type identification, its algorithm requires a long vibration sequence. Obtaining a long vibration sequence will inevitably increase the time delay of chatter type identification, making it difficult to meet the real-time requirements of high-speed machining (spindle speed > 20,000 RPM).

[0006] The core defect of the above method lies in that traditional entropy and its multi-scale entropy features are difficult to effectively characterize different types of chatter, while the existing multi-resolution entropy is faced with the constraint of a relatively long required vibration sequence. This has severely restricted the engineering applicability of the existing technical system in advanced manufacturing scenarios such as high-speed / ultra-high-speed machining and non-traditional milling machining. Summary of the Invention

[0007] In view of the above problems, the object of the present invention is to propose a method for extracting chatter features based on differential principal component entropy, which can extract chatter features through differential principal component entropy, and the extracted chatter features can be applied to chatter control, chatter detection and identification.

[0008] The technical solution of the present invention is as follows: The method for extracting chatter features based on differential principal component entropy according to the present invention comprises the following operating steps:

[0009] Step (1): Select the installation positions of the vibration sensor and the rotational speed sensor, and determine the relevant parameters for data acquisition and feature extraction;

[0010] Step (2): Obtain the synchronous vibration data Y;

[0011] Step (3): Expand the synchronous vibration data Y into multi-dimensional data P;

[0012] Step (4): Perform μ-step differential processing on the multi-dimensional data P to obtain multi-dimensional data P μ ;

[0013] Step (5): Perform dimensionality reduction processing on the multi-dimensional data P μ and select the principal components;

[0014] Step (6): Calculate the μ-step differential principal component entropy;

[0015] Step (7): Let μ = μ + 1, and return to Step (4) to continue calculating the differential principal component entropy until the μ max -step differential principal component entropy is calculated;

[0016] Step (8): Obtain the chatter features.

[0017] Further, in Step (1), the vibration sensor is installed on the workpiece or the spindle for measuring the vibration of the workpiece or the spindle; the rotational speed sensor is installed near the spindle or the milling cutter for measuring the rotational speed of the spindle or the milling cutter; the milling cutter is installed on the spindle, and the rotational speed of the milling cutter is the same as that of the spindle.

[0018] The relevant parameters for data acquisition include: the number of teeth n of the milling cutter, the amount of synchronous sampling data M within each tooth cycle of the milling cutter, and the sampling length; the sampling length is knM, where k is the number of revolutions of the spindle;

[0019] The relevant parameters of the feature extraction include: the initial differential step number μ0, the maximum differential step number μ max , and the cumulative contribution rate threshold r; let the differential step number μ = μ0.

[0020] Further, the implementation process of step (2) is as follows: taking the spindle speed as a reference signal, obtaining the synchronous vibration data in the milling process; the obtained synchronous vibration data is synchronous vibration displacement data, synchronous vibration velocity data, or synchronous vibration acceleration data; the obtained synchronous vibration data can be expressed as:

[0021] Y = {y(1), y(2), …, y(i),..., y(knM)}.

[0022] Further, the implementation process of step (3) is as follows: according to the amount of synchronous sampling data M within each tooth cycle of the milling cutter, expanding the synchronous vibration data Y into multi-dimensional data P, and the expanded multi-dimensional data P is:

[0023]

[0024] Further, the implementation process of step (4) is as follows:

[0025] Performing μ-step difference processing on the multi-dimensional data P, and the multi-dimensional data P after μ-step difference processing μ is

[0026]

[0027] Further, the implementation process of step (5) is as follows:

[0028] Regarding the multi-dimensional data P μ as data composed of M variables and each variable containing nk - μ samples; using the principal component analysis method to perform dimensionality reduction processing on the multi-dimensional data P μ and selecting a μ principal components whose cumulative contribution rate is greater than the cumulative contribution rate threshold r for the first time

[0029]

[0030] Further, the implementation process of step (6) is as follows:

[0031] For the a μ principal components PC μ obtained in step (5), calculating their energy as

[0032]

[0033] According to the energy of the a μ principal components, calculating the energy ratio of each principal component to all energies as

[0034]

[0035] Among them, the energy ratio of the j-th principal component is

[0036]

[0037] According to a μ energy ratios of principal components, calculate the μ-step differential principal component entropy

[0038]

[0039] Furthermore, the implementation process of the step (7) is as follows:

[0040] Let μ = μ + 1, and return to step (4) to continue calculating the next differential principal component entropy until the μ max step differential principal component entropy is calculated.

[0041] Furthermore, the implementation process of the step (8) is as follows:

[0042] Regard all the calculated differential principal component entropies as the extracted flutter characteristics.

[0043] The beneficial effects of the present invention are as follows: (1) The flutter feature extraction method based on differential principal component entropy of the present invention can extract multiple flutter features and can characterize different flutter types; (2) The flutter feature extraction method based on differential principal component entropy of the present invention requires less computational cost and can quickly extract flutter features; (3) The flutter feature extraction method based on differential principal component entropy of the present invention does not require long synchronous vibration data and has a short time delay when used for flutter control and diagnosis; it overcomes the deficiencies of existing multi-resolution asynchronous entropy; (4) The flutter feature extraction method based on differential principal component entropy of the present invention is not affected by the spindle speed; (5) The flutter feature extraction method based on differential principal component entropy of the present invention requires fewer parameters and less artificial expert knowledge. Description of the Drawings

[0044] Figure 1 is the flowchart of the flutter feature extraction method based on differential principal component entropy of the present invention;

[0045] Figure 2 is the workpiece synchronous vibration displacement diagram without flutter in the embodiment of the present invention;

[0046] Figure 3 is the workpiece synchronous vibration displacement diagram during period-2 bifurcation type flutter in the embodiment of the present invention;

[0047] Figure 4It is the workpiece synchronous vibration displacement diagram during period-3 bifurcation flutter in the embodiment of the present invention;

[0048] Figure 5 It is the workpiece synchronous vibration displacement diagram during Hopf bifurcation flutter in the embodiment of the present invention;

[0049] Figure 6 It is the flutter feature extraction result diagram in the embodiment of the present invention. Specific Embodiments

[0050] The technical solution of the present invention will be further described in detail below in conjunction with specific embodiments.

[0051] As Figure 1 shown, the specific operation steps of the flutter feature extraction method based on differential principal component entropy of the present invention are as follows:

[0052] Step (1): Select the installation positions of the vibration sensor and the rotational speed sensor, and determine the relevant parameters for data acquisition and feature extraction;

[0053] Step (2): Obtain the synchronous vibration data Y;

[0054] Step (3): Expand the synchronous vibration data Y into multi-dimensional data P;

[0055] Step (4): Perform μ-step difference processing on the multi-dimensional data P to obtain the multi-dimensional data P μ ;

[0056] Step (5): Perform dimensionality reduction processing on the multi-dimensional data P μ and select the principal components;

[0057] Step (6): Calculate the μ-step differential principal component entropy;

[0058] Step (7): Let μ = μ + 1, and return to Step (4) to continue calculating the differential principal component entropy until the μ max -step differential principal component entropy is calculated;

[0059] Step (8): Obtain the flutter features.

[0060] Furthermore, in Step (1), the vibration sensor is installed on the workpiece or the spindle to measure the vibration of the workpiece or the spindle; the rotational speed sensor is installed near the spindle or the milling cutter to measure the rotational speed of the spindle or the milling cutter; the milling cutter is installed on the spindle, and the rotational speed of the milling cutter is the same as that of the spindle.

[0061] The relevant parameters for data acquisition include: the number of teeth n of the milling cutter, the amount of synchronous sampling data M within each tooth cycle of the milling cutter, and the sampling length; the sampling length is knM, where k is the spindle speed. In this embodiment, the number of teeth n of the milling cutter used is 1, the amount of synchronous sampling data M within each tooth cycle of the milling cutter is set to 400, the sampling length knM is 8000, and the spindle speed k is 20.

[0062] The relevant parameters for feature extraction include: the initial differential step μ0, the maximum differential step μ max , and the cumulative contribution rate threshold r; let the differential step μ = μ0. In this embodiment, the initial differential step μ0 is set to 1, the maximum differential step μ max is 3, and the cumulative contribution rate threshold r is 0.99.

[0063] Furthermore, the implementation process of step (2) is as follows: Taking the spindle speed as the reference signal, obtain the synchronous vibration data during the milling process; the obtained synchronous vibration data is synchronous vibration displacement data, synchronous vibration velocity data, or synchronous vibration acceleration data; the obtained synchronous vibration data can be expressed as:

[0064] Y = {y(1), y(2), y(3),..., y(i),..., y(8000)}.

[0065] Furthermore, the implementation process of step (3) is as follows: According to the amount of synchronous sampling data M within each tooth cycle of the milling cutter, expand the synchronous vibration data Y into multi-dimensional data P. The expanded multi-dimensional data P is:

[0066]

[0067] Furthermore, the implementation process of step (4) is as follows:

[0068] Perform μ-step difference processing on the multi-dimensional data P. The multi-dimensional data P after μ-step difference processing μ is

[0069]

[0070] Furthermore, the implementation process of step (5) is as follows:

[0071] Regard the multi-dimensional data P μ as data consisting of M variables and each variable containing nk - μ samples; use the principal component analysis method to perform dimensionality reduction processing on the multi-dimensional data P μ and select a μ principal components whose cumulative contribution rate is greater than the cumulative contribution rate threshold r for the first time

[0072]

[0073] Furthermore, the implementation process of step (6) is as follows:

[0074] For the a μ principal components PC μ obtained in step (5), calculate their energies as

[0075]

[0076] According to the energies of the a μ principal components, calculate the energy ratio of each principal component to the total energy as

[0077]

[0078] wherein, the energy ratio of the j-th principal component is

[0079]

[0080] According to the energy ratios of the a μ principal components, calculate the μ-step differential principal component entropy

[0081]

[0082] Furthermore, the implementation process of step (7) is as follows:

[0083] Let μ = μ + 1, and return to step (4) to continue calculating the next differential principal component entropy until the μ max -step differential principal component entropy is calculated.

[0084] Furthermore, the implementation process of step (8) is as follows:

[0085] Regard all the calculated differential principal component entropies as the extracted flutter features.

[0086] Figures 2 - 5 The synchronous vibration signals of the workpiece are respectively shown when the milling machining states are non-flutter, period-2 bifurcation type flutter, period-3 bifurcation type flutter, and Hopf type flutter. Figure 6 This is the flutter detection result of the embodiment of the present invention. It can be seen from the figure that the period-2 bifurcation type flutter, period-3 bifurcation type flutter, and Hopf type flutter are relatively concentrated; the non-flutter is distributed far from the energy entropy of the flutter.

[0087] In summary, the flutter feature extraction method based on differential principal component entropy disclosed by the method of the present invention can extract multiple flutter features without being affected by the spindle speed and can characterize different flutter types. The calculation cost of this method is relatively low, and flutter features can be extracted quickly. The flutter feature extraction method based on differential principal component entropy of the present invention requires few parameters and little artificial expert knowledge. In addition, the flutter detection method based on contribution entropy and cumulative contribution entropy disclosed by the method of the present invention does not require long synchronous vibration data and has a short time delay when used for flutter control and diagnosis; it overcomes the deficiencies of existing multi-resolution asynchronous entropy.

[0088] The above are only examples of the implementation of the present invention and are not intended to limit the present invention. All equivalent replacements made within the principles of the present invention shall be included within the protection scope of the present invention. The content not elaborated in detail in the present invention belongs to the well-known prior art in the technical field of the present profession.

Claims

1. A flutter feature extraction method based on differential principal component entropy, characterized in that The operation steps are as follows: Step (1): Select the installation positions of the vibration sensor and the rotational speed sensor, and determine the relevant parameters for data acquisition and feature extraction; Step (2): Obtain the synchronous vibration data Y; Step (3): Expand the synchronous vibration data Y into multi-dimensional data P; Step (4): Perform μ-step difference processing on the multi-dimensional data P to obtain the multi-dimensional data P μ ; Step (5): For the multi-dimensional data P μ Perform dimensionality reduction processing to select the principal components; Step (6): Calculate the μ-step differential principal component entropy; Step (7): Let μ = μ + 1, and return to Step (4) to continue the calculation of the differential principal component entropy until the μ-step differential principal component entropy is calculated; max ​ Step (8): Obtain the chatter characteristics.

2. The flutter feature extraction method based on differential principal component entropy according to claim 1, wherein In step (1), the vibration sensor is installed on the workpiece or the spindle to measure the vibration of the workpiece or the spindle; the rotational speed sensor is installed near the spindle or the milling cutter to measure the rotational speed of the spindle or the milling cutter; the milling cutter is installed on the spindle, and the rotational speed of the milling cutter is the same as that of the spindle. The relevant parameters for data acquisition include: the number of teeth n of the milling cutter, the amount of synchronous sampling data M within each tooth cycle of the milling cutter, and the sampling length; the sampling length is knM, where k is the number of revolutions of the spindle; The relevant parameters of the feature extraction include: the initial difference step number μ0, the maximum difference step number μ max , and the cumulative contribution rate threshold r; let the difference step number μ = μ0.

3. The flutter feature extraction method based on differential principal component entropy according to claim 1, wherein In step (2), using the spindle speed as a reference signal, obtain the synchronous vibration data during the milling process; the obtained synchronous vibration data is synchronous vibration displacement data, synchronous vibration velocity data, or synchronous vibration acceleration data; the obtained synchronous vibration data can be expressed as: Y = {y(1), y(2), …, y(i), …, y(knM)}.

4. The flutter feature extraction method based on differential principal component entropy according to claim 1, wherein In step (3), the specific implementation process of expanding the synchronous vibration data Y into multi-dimensional data P is: according to the amount of synchronous sampling data M within each tooth cycle of the milling cutter, expand the synchronous vibration data Y into multi-dimensional data P, and the expanded multi-dimensional data P is:

5. The flutter feature extraction method based on differential principal component entropy according to claim 1, characterized in that, The specific implementation process of step (4) is: Perform μ-step difference processing on the multi-dimensional data P, and the multi-dimensional data P after μ-step difference processing μ is 6. The flutter feature extraction method based on differential principal component entropy according to claim 1, characterized in that In step (5), for the multi-dimensional data P μ Implement dimensionality reduction processing, and the specific implementation process of selecting the principal components is as follows: Regarding the multi-dimensional data P μ as the data composed of M variables and each variable containing nk - μ samples; Using the principal component analysis method to perform dimensionality reduction processing on the multi-dimensional data P μ and selecting a μ principal components whose cumulative contribution rate is greater than the cumulative contribution rate threshold r for the first time 7. The flutter feature extraction method based on differential principal component entropy according to claim 1, characterized in that The specific implementation process of step (6) is as follows: For the a obtained in step (5) μ principal components PC μ , calculate their energies as According to a μ the energy of each principal component, calculate the energy ratio of each principal component to all energies as where the energy ratio of the j-th principal component is According to the energy ratio of a μ principal components, calculate the μ-step differential principal component entropy 8. The flutter feature extraction method based on differential principal component entropy according to claim 1, characterized in that In step (7), the specific implementation process of calculating the differential entropy and using it as a chatter index is as follows: Let μ = μ + 1, and return to step (4) to continue the calculation of the next differential principal component entropy until μ is calculated. max Up to the step of differential principal component entropy.

9. The flutter feature extraction method based on differential principal component entropy according to claim 1, characterized in that In step (8), the specific implementation process of determining whether chatter occurs is as follows: Regard all the calculated differential principal component entropies as the extracted chatter characteristics.

Images