JP2020082304A - Chattering vibration detection device, chattering vibration detection method, chattering vibration detection program and chattering vibration restraining device - Google Patents

Abstract

To accurately detect chattering vibration.SOLUTION: A chattering vibration detection device (100) is a device for determining whether there is chattering vibration or not based on sensor data (D0) output from a sensor (204) for detecting vibration of a cutting tool (201) of a processing device, and includes a first storage unit (101) for storing sensor data with a label (101a), a characteristic quantity extracting unit (103) for executing first extraction processing to extract one or more first characteristic quantities from the sensor data with a label as learning data (L2), and second extraction processing to extract determination object data (D2) including one or more second characteristic quantities from the sensor data, a chattering vibration learning unit (104) for generating a learning model (105a) from the learning data, and a chattering vibration determination unit (106) for determining whether or not the determination object data is data showing the existence of chattering vibration by referring to the learning model.SELECTED DRAWING: Figure 2

Description

The present invention relates to a chatter vibration detection device that detects chatter vibration, a chatter vibration detection method and chatter vibration detection program used to detect chatter vibration, and a chatter vibration suppression device that suppresses chatter vibration.

2. Description of the Related Art Various processing devices are used for cutting an object to be processed (that is, a workpiece) with a rotating cutting tool. In such a processing apparatus, chatter vibration may occur during cutting. Chatter vibration can be classified into regenerative chatter vibration and forced chatter vibration. Reproduction chatter vibration, for example, due to the undulations that occur on the processing surface of the processing object cut by the preceding cutting edge of the cutting tool, the thickness of the portion cut off from the processing surface of the processing object by the next cutting edge fluctuates, As a result, the cutting force fluctuates and is generated. The forced chatter vibration occurs, for example, when some forced vibration resonates with the vibration generated by the rotation of the cutting tool of the processing device. When chatter vibration occurs, a striped pattern is generated on the processing surface of the processing target object, and the quality of the processing surface is degraded or the cutting edge of the cutting tool is easily damaged (for example, see Non-Patent Document 1).

For example, in the device described in Patent Document 1, the number of blades of the rotary blade (that is, the number of blade edges) is set to the frequency [Hz] (=rotation speed [rpm]/60) of the rotation speed [rpm] of the cutting tool, which is a cutting tool. ) To calculate the frequency [Hz], and chatter vibration (reproduction chatter vibration in this case) when the peak value of the frequency other than the calculated frequency (that is, peak power) exceeds the threshold value. It is judged that there is a sign of.

Further, the device described in Patent Document 2 multiplies the number of revolutions [rpm] of the cutting tool by the number of blades of the cutting tool, and further divides by 60 to calculate a cutting fundamental frequency [Hz], and an integer of this cutting fundamental frequency. When there is a peak at twice the frequency and the peak power exceeds the threshold value, it is determined that chatter vibration (forced chatter vibration in this case) is occurring.

JP, 2010-247316, A (for example, page 7) JP, 2018-020426, A (for example, paragraph 0044, 0055).

Eiji Syamoto, "Generation Mechanism and Suppression of Chatter Vibration in Cutting", Daido Steel Co., Ltd., Research & Development Division, Electric Steelmaking/Daido Steel Co., Ltd. Vol. 82, No. 2, pp. 143-155, 2011 December 27 Platt J. C. , "Probabilistic Outputs For Support Vector Machines and Comparisons to Regularized Likelihood Methods," Advances in Largs Margin, Claspins, Clasps, Clasps, Clasps, Clasps, Clasps, Clasps, Clasps, Clasps, Largs, Mars, Margin, Claspins, Clasps, Clasps, Clasps, Clasps, Lambs, Mars. 1-11.

However, in the device described in Patent Document 1, a peak exists at a frequency [Hz] having a value different from the calculated frequency, and even if the peak power exceeds the threshold, the frequency at which the peak exists. Chatter vibration may not occur in [Hz]. Even if a peak exists at a frequency [Hz] different from the calculated frequency and the peak power does not exceed the threshold value, chatter vibration occurs at the frequency [Hz] at which the peak exists. Sometimes.

Further, in the device described in Patent Document 2, there is a peak at a frequency [Hz] that is an integral multiple of the cutting fundamental frequency, and even if the peak power exceeds a threshold, the frequency at which the peak exists [Hz] ] Chatter vibration may not occur. Even when a peak exists at a frequency [Hz] that is an integral multiple of the cutting fundamental frequency and the peak power does not exceed the threshold value, chatter vibration occurs at the frequency at which the peak exists [Hz]. Sometimes.

That is, there is a problem in that the chatter vibration detection accuracy is low only by determining the presence or absence of chatter vibration based on the result of comparison between the peak power and a predetermined threshold value.

The present invention has been made to solve the above-described conventional problems, and a chatter vibration detection device capable of detecting chatter vibration with high accuracy, and chatter vibration capable of detecting chatter vibration with high accuracy. An object of the present invention is to provide a detection method, a chatter vibration detection program, and a chatter vibration suppressor capable of suppressing chatter vibration with high accuracy.

A chatter vibration detection device according to an aspect of the present invention is a device that determines the presence or absence of chatter vibration, based on sensor data output from a sensor that detects vibration of a cutting tool of a processing device that processes a workpiece. A first storage unit for storing the labeled sensor data created by adding a label indicating the presence or absence of chatter vibration to the sensor data, and one or more first feature amounts from the labeled sensor data. A feature extraction unit that performs a first extraction process of extracting the learning data and a second extraction process of extracting determination target data including one or more second feature amounts from the sensor data; A chatter vibration learning unit that generates a learning model from learning data, and a chatter vibration determination unit that determines whether or not the determination target data is data that indicates chatter vibration are provided with reference to the learning model. It is characterized by

A chatter vibration detection method according to an aspect of the present invention is a method of determining the presence or absence of chatter vibration based on sensor data output from a sensor that detects vibration of a cutting tool of a processing device that processes a workpiece. And a step of performing a first extraction process of extracting one or more first feature amounts as learning data from the labeled sensor data created by adding a label indicating the presence or absence of chatter vibration to the sensor data. A step of performing a second extraction process of extracting determination target data including one or more second feature amounts from the sensor data; a step of generating a learning model from the learning data; And a step of determining whether or not the data indicates chatter vibration with reference to the learning model.

By using the chatter vibration detection device, chatter vibration detection method, or chatter vibration detection program according to the present invention, chatter vibration can be detected with high accuracy.

Further, by using the chatter vibration suppressing device according to the present invention, chatter vibration can be suppressed with high accuracy.

It is a figure which shows roughly the structure of the chatter vibration suppression apparatus which concerns on Embodiment 1 of this invention, and the structure of the processing apparatus which performs cutting. FIG. 3 is a block diagram schematically showing the configuration of the chatter vibration detection device according to the first embodiment. FIG. 6 is an explanatory diagram showing a learning model generation process in the chatter vibration detection apparatus according to the first embodiment. FIG. 5 is an explanatory diagram showing a process of determining the presence or absence of chatter vibration in the chatter vibration detection apparatus according to the first embodiment. FIG. 6 is an explanatory diagram showing an operation of an FFT calculation unit of the chatter vibration detection device according to the first embodiment. 5 is an explanatory diagram showing an operation of a feature amount extraction unit of the chatter vibration detection device according to the first embodiment. FIG. FIG. 4 is a diagram showing an example of learning data including a feature vector generated by a feature amount extraction unit in a table format during a learning process of the chatter vibration detection apparatus according to the first embodiment. FIG. 6 is an explanatory diagram showing an operation during determination processing of the chatter vibration detection device according to the first embodiment. FIG. 5 is a diagram showing, in a tabular format, an example of determination target data including a feature amount for one frame input to the chatter vibration determination unit of the chatter vibration detection apparatus according to the first embodiment. (A) to (I) are diagrams showing an example of the feature amount extracted by the feature amount extraction unit of the chatter vibration detection apparatus according to the second embodiment of the present invention. It is a figure which shows the example of the data for learning containing the feature vector produced|generated by the feature amount extraction part in the table format in the learning process of the chatter vibration detection apparatus which concerns on Embodiment 3 of this invention. FIG. 16 is a diagram showing an example of determination target data including a feature amount for one frame input to the chatter vibration determination unit in a table format during the determination process of the chatter vibration detection device according to the third embodiment. It is a block diagram which shows roughly the structure of the chatter vibration detection apparatus which concerns on Embodiment 4 and 5 of this invention. It is a block diagram which shows roughly the structure of the chatter vibration detection apparatus which concerns on Embodiment 6 of this invention. It is a figure which shows roughly the structure of the chatter vibration suppression apparatus which concerns on Embodiment 7 of this invention, and the structure of the processing apparatus which performs a cutting process. It is a block diagram which shows roughly the structure of the chatter vibration detection apparatus which concerns on Embodiment 7. It is a block diagram which shows roughly the structure of the chatter vibration detection apparatus which concerns on Embodiment 8 of this invention. It is a block diagram which shows roughly the structure of the chatter vibration detection apparatus which concerns on Embodiment 9 of this invention. FIG. 27 is a block diagram schematically showing the configuration of a chatter vibration detection device according to a modification of the ninth embodiment. It is a figure which shows the example of the hardware constitutions of the chatter vibration detection apparatus which concerns on the modification of Embodiments 1-9.

A chatter vibration detection device, a chatter vibration detection method, a chatter vibration detection program, and a chatter vibration suppression device according to an embodiment of the present invention will be described below with reference to the drawings. The following embodiments are merely examples, and various modifications can be made within the scope of the present invention. In the figure, x, y and z represent coordinate axes in the xyz orthogonal coordinate system.

<<1>> Embodiment 1.
FIG. 1 is a diagram schematically showing a configuration of a chatter vibration suppressing device 10 and a configuration of a processing device 200 that performs a cutting process according to the first embodiment.

The processing apparatus 200 for processing the processing target object is, for example, a device that scrapes off an unnecessary portion of the processing target object 206 that is the processing target object using the cutting tool 201. The cutting tool 201 is, for example, an end mill. In this case, the processing device 200 performs end mill processing. The processing device 200 includes a cutting tool 201 that rotates around an axis line that faces the x direction, a main shaft portion 202 that supports the cutting tool 201, and a stage 203 that supports the main shaft portion 202 so as to be movable in the +x direction and the −x direction. It has and. In addition, the processing apparatus 200 supports a sensor (for example, an acceleration sensor) 204 that detects vibration of the cutting tool 201 and a processing target object 206 so as to be movable in +y direction, −y direction, +z direction, and −z direction. A unit 205 and an NC (Numerical Control) device 207, which is a control device that controls the operation of the processing device 200, are provided. The NC device 207 controls processing conditions such as the rotation speed and the depth of cut of the cutting tool 201 and the feed speed of the processing object 206 in cutting. Further, the NC device 207 can notify the chatter vibration detection device 100 of the processing conditions. The sensor 204 outputs a detection signal indicating vibration in three axial directions orthogonal to each other, that is, the x direction, the y direction, and the z direction, or sensor data D0 that is sampling data of the detection signal.

Chatter vibration suppressor 10 includes chatter vibration detector 100 and chatter vibration suppressor 107. The chatter vibration detection device 100 determines whether chatter vibration is present in the cutting tool 201 of the processing device 200 based on the sensor data D0 output from the sensor 204, and outputs determination result information D3 indicating the presence of chatter vibration. When the detection signal is output from the sensor 204, the chatter vibration detection device 100 generates the sampling data D0. The chatter vibration suppression unit 107 sends to the processing apparatus 200 a control signal C1 for suppressing chatter vibration, that is, a control signal C1 for performing chatter vibration suppression operation based on the determination result information D3 output from the chatter vibration detection apparatus 100. Send. At this time, the control signal C1 may be transmitted to the NC device 207. The processing device 200 that receives the control signal C1 slightly changes the rotation speed of the cutting tool 201 from the rotation speed when it is determined that chatter vibration exists, for example. By changing the rotation speed, the processing apparatus 200 can perform the cutting of the processing target object 206 with the cutting tool 201 that rotates at the rotation speed at which chatter vibration does not occur.

FIG. 2 is a block diagram schematically showing the configuration of chatter vibration detection apparatus 100 according to the first embodiment. Chatter vibration detection apparatus 100 is an apparatus capable of implementing the chatter vibration detection method according to the first embodiment. The chatter vibration detection apparatus 100 performs a learning process of generating a learning model 105a for determining the presence or absence of chatter vibration by learning using the sensor data D0 output from the sensor 204, and a sensor output from the sensor 204 during cutting. For the data D0, the learning model 105a is used to perform determination processing for determining the presence or absence of chatter vibration.

As shown in FIG. 2, the chatter vibration detection apparatus 100 includes a storage unit 101 as a first storage unit, an FFT (Fast Fourier Transform) calculation unit 102 as a conversion processing unit that performs a Fourier transform, and a feature amount extraction. A unit 103, a chatter vibration learning unit 104, a storage unit 105 as a second storage unit, and a chatter vibration determination unit 106 are provided. The storage unit 101 and the storage unit 105 may be part of a storage area of an information storage device provided outside the chatter vibration detection device 100. The chatter vibration detection device 100 may be an information processing device such as a computer that executes a chatter vibration detection program. Further, chatter vibration suppression device 10 may be an information processing device such as a computer. Further, chatter vibration suppressing device 10 and NC device 207 can be configured by an information processing device such as a computer.

The storage unit 101 stores in advance the labeled sensor data 101a created by adding a label that is information indicating the presence or absence of chatter vibration to the learning sensor data. Chatter vibration detection apparatus 100 generates learning model 105a using labeled sensor data 101a stored in storage unit 101 during chatter vibration learning processing. The labeled sensor data 101a is data generated by associating the sensor data collected in advance with information indicating the presence or absence of chatter vibration when the sensor data is collected as a label and the sensor data.

The FFT operation unit 102 performs a Fourier transform on the labeled sensor data 101a during the learning process to generate data (also referred to as “first data”) L1 in the frequency space (ie, frequency domain) (“first process”). 1) conversion process) and a conversion process of generating frequency space data (also referred to as “second data”) D1 by performing Fourier transform of the time-series sensor data D0 during the determination process (“first conversion process”). 2 conversion process").

The feature amount extraction unit 103 acquires the processing condition parameter D4 including the cutting fundamental frequency from the NC device 207. The processing condition parameter D4 is, for example, the cutting rotation speed (that is, the cutting rotation frequency), the material of the processing target, the feed rate of the processing target, the cutting thickness of the processing target, the type of processing (for example, end mill processing or parting). Processing etc.), etc. The feature amount extraction unit 103 uses the processing condition parameter D4 to generate one or more feature amounts (“first feature amount” from the frequency space data L1 generated from the labeled sensor data 101a by the FFT calculation unit 102 during the learning process. (Also referred to as “first extraction process”) for extracting the learning data L2 including a feature vector including a feature vector composed of the sensor data D0. Extraction processing (also referred to as "second extraction processing") for extracting determination target data D2 including one or more characteristic quantities (also referred to as "second characteristic quantities") from the frequency space data D1 generated from. ) And do. By increasing the number of feature quantities included in the feature vector, it is possible to improve the accuracy of the determination of the presence or absence of chatter vibration. Therefore, it is desirable that the number of feature quantities included in the feature vector is plural.

The chatter vibration learning unit 104 receives the learning data L2 including the feature vector generated by the feature amount extraction unit 103, and outputs it as a feature amount included in the feature vector and a label indicating the presence or absence of chatter vibration corresponding to the feature amount. Based on this, a learning algorithm prepared in advance is used to acquire a boundary surface used for determining the presence or absence of chatter vibration, and a learning model 105a which is the result of this learning is generated. An example of the boundary surface is shown as a hyperplane in FIG. 8 described later. The generated learning model 105a is stored in the storage unit 105.

At the time of the process of determining the presence or absence of chatter vibration, the sensor data D0 output from the sensor 204 during cutting is input to the FFT calculation unit 102, and converted by the FFT calculation unit 102 into frequency space data D1. At the time of cutting, the characteristic amount extraction unit 103 extracts the characteristic amount D2 from the frequency space data D1. The feature amount extraction unit 103 extracts the feature amount D2 at the time of cutting by the same method as at the time of the learning process, but unlike the case at the time of the learning process, does not give a label indicating the presence or absence of chatter vibration to the feature amount D2. .. The feature amount extraction unit 103 outputs the extracted feature amount D2 to the chatter vibration determination unit 106.

The chatter vibration determination unit 106 uses the learning model 105a stored in the storage unit 105 to determine the presence or absence of chatter vibration for the feature amount D2, and outputs determination result information D3 indicating the result of this determination. To do. In the first embodiment, the determination result information D3 is output to the chatter vibration suppressing unit 107.

By the way, chatter vibration can be classified into forced chatter vibration and replay chatter vibration. The forced chatter vibration is generated, for example, when the blade of the cutting tool 201 contacts the object 206 to be processed. Regenerative chatter vibration causes fluctuations in the thickness of the part cut from the work surface of the work piece by the next cutting edge due to the undulations that occur on the work surface of the work piece cut by the cutting edge of the cutting tool. , The cutting force fluctuates. The forced chatter vibration is a vibration having a correlation with the cutting fundamental frequency Fc. The reproduction chatter vibration is vibration that has no correlation with the fundamental cutting frequency Fc.

FIG. 3 is an explanatory diagram showing a process of generating the learning model 105a in the chatter vibration detection apparatus 100 according to the first embodiment, that is, a process at the time of learning processing.

The storage unit 101 previously stores, in advance, sensor data D0 output from the sensor 204 that captures chatter vibration that occurs during cutting and sensor data D0 that is output from the sensor 204 during normal operation in which chatter vibration does not occur. Collect in large quantities. The storage unit 101 classifies each sensor data D0 into a file with chatter vibration and a file without chatter vibration in advance, and gives a label with chatter vibration and a label without chatter vibration to each file in advance. As a result, the labeled sensor data 101a is generated.

In the first embodiment, the presence/absence of chatter vibration for label application is determined based on the sound generated during cutting or based on the result of observing the processed surface of the processing target. The presence or absence of chatter vibration is determined by a person. However, the presence/absence of chatter vibration may be determined by a device that analyzes sound or an image of a processed surface. The sensor data in which one or more feature quantities are labeled is the labeled sensor data 101a.

The labeled sensor data 101a is collected by the FFT calculation unit 102 for each of the xs, y, and z direction sensor data collected at the sampling frequency Fs, and is collected for each Ws sample data. Is performed as one frame. The plurality of frames generated by the grouping may have overlapping portions in the time axis direction. Note that a method of determining the number Ws of sample data which configures one frame will be described later. The FFT calculation unit 102 performs a Short Time FFT that converts the frame data of the plurality of frames thus generated into the data L1 of the frequency space in time order (step S11 of FIG. 3 ). Short Time FFT is also described as SFFT.

Next, the feature amount extraction unit 103 extracts the feature amount (feature amount 1 to feature amount f in FIG. 3) from each of the plurality of frames converted into the frequency space data L1, and determines whether chatter vibration is present. Learning data L2 including the labeled feature vector is generated (step S12 in FIG. 3). The feature vector includes, for example, f (f is a positive integer) feature amount and a label indicating the presence or absence of chatter vibration as elements.

Next, the chatter vibration learning unit 104 generates a learning model 105a for determining the presence or absence of chatter vibration by performing a learning process for determining the presence or absence of chatter vibration on the learning data L2 (FIG. 3). Step S13).

FIG. 4 is an explanatory diagram showing a process of determining the presence or absence of chatter vibration in chatter vibration detection apparatus 100 according to the first embodiment, that is, a process at the time of the determination process.

In order to determine the presence or absence of chatter vibration in real time from the sensor data in the x, y, and z directions sampled during cutting, the FFT operation unit 102 uses the sensor data D0 output from the sensor 204 as Ws samples. Each data is framed and converted to frequency space data D1 by SFFT processing (step S21 in FIG. 4).

Next, the feature amount extraction unit 103 extracts the determination target data D2 including one or more feature amounts from the frequency space data D1 (step S22 in FIG. 4). The determination target data D2 is, for example, a feature vector including a plurality of feature amounts of feature amount 1 to feature amount f. However, the determination target data D2 does not include a label indicating the presence or absence of chatter vibration.

Next, the chatter vibration determination unit 106 determines the presence or absence of chatter vibration for each frame in real time, and sequentially outputs the determination result information D3 indicating the result of the determination to the chatter vibration suppression unit 107 (step S23). The chatter vibration suppression unit 107 performs control for chatter vibration suppression based on the determination result information D3 indicating the result of the determination of the presence or absence of chatter vibration.

FIG. 5 is an explanatory diagram showing the operation of the FFT calculation unit 102 of the chatter vibration detection device 100 according to the first embodiment. FIG. 5 shows the frequency spectrum of the data (L1 or D1) in the frequency space generated by performing the SFFT process on the Ws sample data forming one frame. Ws is a positive integer. Since the data sampling frequency is Fs [Hz], the frequency resolution Fstep, which is the minimum division unit on the frequency coordinate axis (that is, the horizontal axis) in FIG. 5, can be calculated by the following equation (1). ..

The basic cutting frequency Fc [Hz] of the cutting tool 201 is determined by the following formula (2) depending on the rotation speed Rc [rpm] of the cutting tool 201 and the number of blades Bc of the cutting tool 201.

In order to determine the presence or absence of chatter vibration with high accuracy, it is desirable to extract an effective feature amount and generate the learning model 105a from this effective feature amount. Chatter vibration is classified into forced chatter vibration having a correlation with the cutting fundamental frequency Fc and regenerative chatter vibration having no correlation with the cutting fundamental frequency Fc. That is, when there is a remarkable peak (that is, a peak with a large peak power) at a frequency that is an integral multiple of the cutting fundamental frequency Fc, it is considered that forced chatter vibration is likely to occur. Further, when there is a remarkable peak at a frequency that is a non-integer multiple of the cutting fundamental frequency Fc, it is considered highly possible that reproduction chatter vibration is occurring. Therefore, in order to extract a feature amount effective for chatter vibration detection, it is desirable that a peak having a frequency that is an integral multiple of the cutting fundamental frequency Fc is reliably extracted. For that purpose, it is desirable to determine the frequency resolution Fstep so that the frequency that is an integral multiple of the cutting fundamental frequency Fc is an integer multiple of the frequency resolution Fstep. In other words, it is desirable to satisfy the following expression (3).

ここで、ｍは、正の整数である。

Here, m is a positive integer.

From the equations (1) to (3), the number Ws of sample data forming one frame, that is, the window size of the FFT is obtained by the following equation (4).

As described above, the FFT calculation unit 102 determines the number Ws of sample data forming one frame based on the sampling frequency Fs and the cutting fundamental frequency Fc, as shown in Expression (4). The cutting fundamental frequency Fc is acquired from the NC device 207, for example.

FIG. 6 is an explanatory diagram showing the operation of the feature quantity extraction unit 103 of the chatter vibration detection apparatus 100 according to the first embodiment.

In the extraction process for the learning process, the feature amount extraction unit 103 performs one or more peaks (learning process) on the frequency space data (that is, the frequency spectrum) converted in frame units by the FFT operation unit 102. The peak at time is also referred to as “first peak”), and the frequency of one or more peaks and the peak power that is the power of one or more peaks are calculated. The feature amount extraction unit 103 performs the following process on the extracted peak in the extraction process for the learning process. First, the feature amount extraction unit 103 divides the entire frequency band Ba of the frequency spectrum into N, and divides the N frequency bands (that is, divided bands) B1 to B5 obtained by the division and the entire frequency band Ba. For example, a process of extracting the following first to tenth learning information (U1) to (U10) as feature quantities is performed. Note that N is a positive integer, and N=5 in FIG.

(U1) First learning information indicating the total, average, or variance of the powers of the first peaks.
(U2) Second learning information indicating the number of first peaks.
(U3) Third learning information indicating the power and frequency of the first peak having the power within the top three of the first peaks.
(U4) Fourth learning information indicating the total, average, or variance of the powers of the first peaks having a frequency that is a non-integer multiple of the cutting fundamental frequency Fc.
(U5) Fifth learning information indicating the power and frequency of the first peak having the power within the top three of the first peaks having a frequency that is a non-integer multiple of the cutting fundamental frequency Fc.
(U6) Sixth learning information indicating the number of first peaks having a frequency that is a non-integer multiple of the cutting fundamental frequency Fc.
(U7) Seventh learning information indicating the ratio of the number of first peaks having a frequency that is a non-integer multiple of the cutting fundamental frequency Fc to the number of first peaks in the entire frequency band Ba.
(U8) Eighth learning information indicating the total, average, or variance of the powers of the first peaks having a frequency that is an integral multiple of the cutting fundamental frequency Fc.
(U9) Ninth learning information indicating the power and frequency of the first peak having the power within the top three ranks of the first peak having the frequency that is an integral multiple of the cutting fundamental frequency Fc.
(U10) Tenth learning information indicating the number of first peaks having a frequency that is an integral multiple of the cutting fundamental frequency Fc.

The chatter vibration learning unit 104 can generate the learning model 105a from a combination of two or more pieces of learning information among the first to tenth learning information (U1) to (U10) in the learning process.

In the extraction process for the determination process, the feature amount extraction unit 103 performs the same process as the extraction process for the learning process. In the extraction process for the determination process, the feature amount extraction unit 103 performs one or more peaks (determination process) on the data in the frequency space (that is, the frequency spectrum) converted in frame units by the FFT operation unit 102. The peak at time is also referred to as “second peak”), and the frequency of one or more second peaks and the power of one or more second peaks are obtained. A second peak is obtained for the data converted into the frequency spectrum, and the frequency of the second peak and the power of the second peak are obtained. The feature amount extraction unit 103 performs the following process on the extracted second peak. First, the feature amount extraction unit 103 divides the entire frequency band Ba of the frequency spectrum into N, and divides the N frequency bands (that is, divided bands) B1 to B5 obtained by the division and the entire frequency band Ba. For example, a process of extracting the following first to tenth determination information (V1) to (V10) as feature quantities is performed.

(V1) First determination information indicating the total, average, or variance of the powers of the second peaks.
(V2) Second determination information indicating the number of second peaks.
(V3) Third determination information indicating the power and frequency of the second peak having the power within the top three ranks of the second peak.
(V4) Fourth determination information indicating the total, average, or variance of the powers of the second peaks having a frequency that is a non-integer multiple of the cutting fundamental frequency Fc.
(V5) Fifth determination information indicating the power and frequency of the second peak having the power within the top three of the second peaks having the non-integer multiple of the cutting fundamental frequency Fc.
(V6) Sixth determination information indicating the number of second peaks having a frequency that is a non-integer multiple of the cutting fundamental frequency Fc.
(V7) Seventh determination information indicating the ratio of the number of second peaks having a frequency that is a non-integer multiple of the cutting fundamental frequency Fc to the number of second peaks in the entire frequency band Ba.
(V8) Eighth determination information indicating the total, average, or variance of the powers of the second peaks having a frequency that is an integral multiple of the cutting fundamental frequency Fc.
(V9) Ninth determination information indicating the power and frequency of the second peak having the power within the top three of the second peaks having the frequency that is an integral multiple of the cutting fundamental frequency Fc.
(V10) Tenth determination information indicating the number of second peaks having a frequency that is an integral multiple of the cutting fundamental frequency Fc.

In the determination process, chatter vibration determination unit 106 determines that determination target data D2 is based on a combination of two or more extracted determination information items among the first to tenth determination information items (V1) to (V10). It is determined whether or not the data indicates chatter vibration by referring to the learning model 105a. The determination process performed by the chatter vibration determination unit 106 is a process corresponding to the learning process performed when the learning information stored in the learning model 105a is acquired.

FIG. 7 is a diagram showing an example of learning data L2 including a feature vector generated by the feature amount extraction unit 103 in a table format during the learning process of chatter vibration detection apparatus 100 according to the first embodiment. FIG. 7 generates one or more feature vectors composed of n (n is a positive integer) feature quantities and a label that is information indicating the presence or absence of chatter vibration. Chatter vibration learning unit 104 generates learning model 105a from learning data L2 including the generated one or more feature vectors, and stores this in storage unit 105.

FIG. 8 is an explanatory diagram showing an operation of the chatter vibration learning unit 104 of the chatter vibration detection apparatus 100 according to the first embodiment. 8 is a hyperplane (that is, a boundary surface) that classifies two feature amounts of the feature amounts of FIG. 7, that is, a multidimensional feature amount space having feature amounts 1 and n as coordinate axes into two classes. ) Is shown. In FIG. 8, for example, the points of the combination of the characteristic amounts indicated by white circles indicate the absence of chatter vibration, and the points of the combinations of the characteristic amounts indicated by the shaded circles indicate the presence of chatter vibration. The hyperplane indicated by a curve indicates the boundary of chatter vibration. Methods that can be used to obtain a hyperplane in a multidimensional feature space, that is, learning algorithms that can be used are known, and specifically, support vector machines, decision trees, discriminant analysis, and logistic regression. , A nearest neighbor classifier, an ensemble classifier that fuses the results of a plurality of classifiers for classification, or a neural network is known. It is preferable that the chatter vibration learning unit 104 selects a combination of two or more feature amounts having the highest classification performance and generates a feature vector from the selected two or more feature amounts. As described above, the chatter vibration learning unit 104 can generate the learning model 105a that enables the presence/absence of chatter vibration to be determined from the selected two or more feature amounts. The multidimensional feature amount space may be a three-dimensional space or more.

FIG. 9 is a diagram showing, in a tabular form, an example of determination target data D2 including the feature amount for one frame input to chatter vibration determination unit 106 during the determination process of chatter vibration detection apparatus 100 according to the first embodiment. is there. At the time of cutting processing, the sensor data output from the sensors 204 in the x, y, and z directions are subjected to framing by the FFT calculation unit 102 and conversion processing into frequency space data by a method similar to that at the time of learning processing. A feature amount extraction process for each frame is performed by the amount extraction unit 103. When the feature amount for one frame is input to the chatter vibration determination unit 106 in the data format shown in FIG. 9, the chatter vibration determination unit 106 uses the learning model 105a acquired in advance learning to input the input one frame. The presence or absence of chatter vibration is determined in real time for each minute data, and the determination result information D3 is output to the chatter vibration suppressing unit 107.

As described above, the chatter vibration detection apparatus 100 or the chatter vibration detection method according to the first embodiment can be used to detect chatter vibration with high accuracy. Specifically, since the presence or absence of chatter vibration is determined based on the combination of feature amounts using a multidimensional feature amount space having a plurality of feature amounts as coordinate axes, the determination accuracy of the presence or absence of chatter vibration is improved. be able to.

Further, the FFT calculation unit 102 determines the window size Ws in the Fourier transform so that the cutting fundamental frequency Fc becomes an integral multiple of the frequency resolution Fstep, and the feature amount extraction unit 103 extracts the peak in the frequency space. A peak with high kurtosis can be obtained as a peak having a frequency that is an integral multiple of the cutting fundamental frequency Fc. Therefore, a peak with high kurtosis can be determined as a peak indicating the presence of forced chatter vibration, and a peak with a relatively low kurtosis can be determined as a peak indicating the presence of regenerative chatter vibration.

In addition, the feature amount extraction unit 103 divides the entire frequency band Ba of the frequency spectrum into N parts, and extracts the feature amount for each of the divided bands and the entire frequency band Ba obtained by the division, thereby cutting Even if the fundamental frequency Fc slightly changes during cutting, or even if the material and processing conditions of the cutting tool or the object to be processed change and the resonance frequency changes, the fluctuation of each feature amount within the divided band is reduced. It is possible. That is, the same learning model 105a is highly versatile. In particular, when chatter vibration is detected, the same learning model 105a may be used without switching the learning model 105a to be used, when the cutting fundamental frequency Fc is slightly changed to perform the cutting as chatter vibration suppression control. it can.

In addition, it is very difficult for the designer to determine a plurality of thresholds used when determining the presence or absence of chatter vibration from a large number of feature quantities. However, the chatter vibration learning unit 104 can easily find a hyperplane that classifies the presence or absence of chatter vibration in a multidimensional frequency space by learning.

The chatter vibration is a forced chatter vibration that occurs when the vibration of a frequency that is an integral multiple of the cutting fundamental frequency Fc is very large, and a play chatter that occurs when the vibration of a frequency that is a non-integer multiple of the cutting fundamental frequency Fc is large. It is classified as vibration. A peak and a peak power of a frequency which is a non-integer multiple of the cutting fundamental frequency Fc, and a peak and a peak power of a frequency which is an integral multiple of the cutting fundamental frequency Fc are used as statistical information of the frequency for learning processing and determination processing. Since it is used, it is possible to find an appropriate hyperplane that classifies the presence or absence of chatter vibration.

<<2>> Embodiment 2.
The chatter vibration detection device according to the second embodiment has the same configuration as that shown in FIGS. 1 and 2. Therefore, FIGS. 1 and 2 are referred to when describing the chatter vibration detection device according to the second embodiment. The chatter vibration detection apparatus according to the second embodiment is different from the chatter vibration detection apparatus 100 according to the first embodiment with respect to the type of feature quantity extracted by the feature quantity extraction unit 103. Except for this point, the chatter vibration detection apparatus according to the second embodiment is the same as chatter vibration detection apparatus 100 according to the first embodiment.

The feature amount extraction unit 103 according to the second embodiment extracts the feature amount as follows. First, the feature amount extraction unit 103 in the second embodiment obtains the detection accuracy of the presence or absence of chatter vibration using the feature amount extracted by the feature amount extraction unit 103 in the first embodiment. Next, the feature quantity extraction unit 103 according to the second embodiment obtains the detection accuracy of the presence or absence of chatter vibration when one of the extracted feature quantities is excluded. In this way, the detection accuracy of the presence or absence of chatter vibration is compared with each other when one of the extracted feature values is excluded, and the excluded feature value is specified when the detection accuracy is the lowest. To do. The feature amount excluded when the detection accuracy is the lowest is considered to be the most effective feature amount for chatter vibration detection. In the second embodiment, the top 10 most effective feature amounts for detecting chatter vibration are determined by such a method. By generating the learning model 105a by combining these top ten feature quantities, it is possible to highly accurately determine the presence or absence of chatter vibration.

In the second embodiment, the feature amount extraction unit 103 sets 1 to the frequency space data (that is, the frequency spectrum) converted in the FFT operation unit 102 in units of frames in the extraction process for the learning process. One or more peaks (first peaks during learning processing) are obtained, and one or more first peak frequencies and one or more first peak powers are obtained. The feature amount extraction unit 103 performs the following process on the extracted first peak. First, the feature amount extraction unit 103 divides the entire frequency band Ba of the frequency spectrum into N, and for the N frequency bands (that is, divided bands) obtained by the division and the entire frequency band Ba, for example, A process of extracting the following first to tenth learning information (J1) to (J10) as feature amounts is performed.

(J1) The maximum peak power (also referred to as "first maximum peak power") that is the maximum power of the powers of the first peaks having a frequency that is a non-integer multiple of the cutting fundamental frequency Fc in the entire frequency band Ba. Learning information indicating the.
(J2) Second learning information indicating the total power of the first peak in the divided band including the cutting fundamental frequency Fc of the plurality of divided bands.
(J3) Third learning information indicating the average of the power of the first peak in the divided band including the cutting fundamental frequency Fc of the plurality of divided bands.
(J4) Fourth learning information indicating the dispersion of the power of the first peak in the divided band including the cutting fundamental frequency Fc of the plurality of divided bands.
(J5) Fifth learning information indicating the power of the first peak having the first largest power in the first peak in the divided band including the cutting fundamental frequency Fc of the plurality of divided bands.
(J6) Sixth learning information indicating the power of the first peak having the second largest power in the first peak in the divided band including the cutting fundamental frequency Fc of the plurality of divided bands.
(J7) Seventh learning information indicating the power of the first peak having the third largest power in the first peak in the divided band including the cutting fundamental frequency Fc of the plurality of divided bands.
(J8) Eighth learning information indicating the number of first peaks having a frequency that is a non-integer multiple of the cutting fundamental frequency Fc in the divided band including the cutting fundamental frequency Fc of the plurality of divided bands.
(J9) Ninth learning information indicating the number of first peaks in the divided band including the cutting fundamental frequency Fc of the plurality of divided bands.
(J10) Tenth learning information indicating the ratio of the number of first peaks in a divided band including the cutting fundamental frequency Fc of a plurality of divided bands to the total number of first peaks in all frequency bands.

The chatter vibration learning unit 104 can generate the learning model 105a from a combination of two or more pieces of learning information among the first to tenth learning information (J1) to (J10) in the learning process.

In the extraction process for the determination process, the feature amount extraction unit 103 performs the same process as the extraction process for the learning process. In the extraction process for the determination process, the feature amount extraction unit 103 performs one or more peaks (determination process) on the data in the frequency space (that is, the frequency spectrum) converted in frame units by the FFT operation unit 102. The second peak at time) is obtained, and the frequency of the one or more second peaks and the power of the one or more second peaks are obtained. A second peak is obtained for the data converted into the frequency spectrum, and the frequency of the second peak and the power of the second peak are obtained. The feature amount extraction unit 103 performs the following process on the extracted second peak. First, the feature amount extraction unit 103 divides the entire frequency band Ba of the frequency spectrum into N, and for the N frequency bands (that is, divided bands) obtained by the division and the entire frequency band Ba, for example, A process of extracting the following first to tenth determination information (K1) to (K10) as feature amounts is performed.

(K1) The maximum peak power that is the maximum power of the powers of the second peaks having a frequency that is a non-integer multiple of the cutting fundamental frequency Fc in the entire frequency band Ba (also referred to as “second maximum peak power”). Determination information indicating the.
(K2) Second determination information indicating the total power of the second peak in the divided band including the cutting fundamental frequency Fc of the plurality of divided bands.
(K3) Third determination information indicating the average of the power of the second peak in the divided band including the cutting fundamental frequency Fc of the plurality of divided bands.
(K4) Fourth determination information indicating the dispersion of the power of the second peak in the divided band including the cutting fundamental frequency Fc of the plurality of divided bands.
(K5) Fifth determination information indicating the power of the second peak having the first largest power in the second peak in the divided band including the cutting fundamental frequency Fc of the plurality of divided bands.
(K6) Sixth determination information indicating the power of the second peak having the second largest power in the second peak in the divided band including the cutting fundamental frequency Fc of the plurality of divided bands.
(K7) Seventh determination information indicating the power of the second peak having the third highest power in the second peak in the divided band including the cutting fundamental frequency Fc of the plurality of divided bands.
(K8) Eighth determination information indicating the number of second peaks having a frequency that is a non-integer multiple of the cutting fundamental frequency Fc in the divided band including the cutting fundamental frequency Fc of the plurality of divided bands.
(K9) Ninth determination information indicating the number of second peaks in the divided band including the cutting fundamental frequency Fc of the plurality of divided bands.
(K10) Tenth determination information indicating a ratio of the number of second peaks in a divided band including the cutting fundamental frequency Fc among a plurality of divided bands to the total number of second peaks in all frequency bands.

In the determination process, chatter vibration determination unit 106 determines that determination target data D2 is based on a combination of two or more extracted determination information items among the first to tenth determination information items (K1) to (K10). It is determined whether or not the data indicates chatter vibration by referring to the learning model 105a. The determination process performed by the chatter vibration determination unit 106 is a process corresponding to the learning process performed when the learning information stored in the learning model 105a is acquired.

10A to 10I are diagrams showing examples of the feature amount extracted by the feature amount extraction unit 103 of the chatter vibration detection apparatus according to the second embodiment. 10A to 10I, among the feature quantities extracted by the feature quantity extraction unit 103, the maximum of the powers of peaks having a frequency that is a non-integer multiple of the cutting fundamental frequency Fc in the entire frequency band Ba. The first learning information (J1) indicating the maximum peak power, which is the power, and any one or more of the second to tenth learning information (J2) to (J10) are combined into the actual feature amount space, The result of plotting the sensor data from the sensor 204 which detects the vibration in the x direction during cutting and visualizing the distribution of the presence or absence of chatter vibration is shown.

FIG. 10A shows a sensor 204 that detects vibration in the x direction during actual cutting in a feature amount space that is a combination of the first learning information (J1) and the eighth learning information (J8). The results of plotting the sensor data and visualizing the distribution of the presence or absence of chatter vibration are shown.

FIG. 10B shows a sensor 204 that detects vibration in the x direction during actual cutting in a feature amount space that is a combination of the first learning information (J1) and the ninth learning information (J9). The results of plotting the sensor data and visualizing the distribution of the presence or absence of chatter vibration are shown.

FIG. 10C shows a sensor 204 that detects vibration in the x direction during actual cutting in the feature amount space in which the first learning information (J1) and the tenth learning information (J10) are combined. The results of plotting the sensor data and visualizing the distribution of the presence or absence of chatter vibration are shown.

FIG. 10D shows a sensor 204 for detecting vibration in the x direction during actual cutting in the feature amount space in which the first learning information (J1) and the second learning information (J2) are combined. The results of plotting the sensor data and visualizing the distribution of the presence or absence of chatter vibration are shown.

FIG. 10E shows a sensor 204 that detects vibration in the x direction during actual cutting in a feature amount space that is a combination of the first learning information (J1) and the third learning information (J3). The results of plotting the sensor data and visualizing the distribution of the presence or absence of chatter vibration are shown.

FIG. 10F shows a sensor 204 that detects vibration in the x direction during actual cutting in the feature amount space that is a combination of the first learning information (J1) and the fourth learning information (J4). The results of plotting the sensor data and visualizing the distribution of the presence or absence of chatter vibration are shown.

FIG. 10G shows a sensor 204 that detects vibrations in the x direction during actual cutting in the feature amount space in which the first learning information (J1) and the fifth learning information (J5) are combined. The results of plotting the sensor data and visualizing the distribution of the presence or absence of chatter vibration are shown.

FIG. 10H shows a sensor 204 that detects vibration in the x direction during actual cutting in a feature amount space that is a combination of the first learning information (J1) and the sixth learning information (J6). The results of plotting the sensor data and visualizing the distribution of the presence or absence of chatter vibration are shown.

FIG. 10(I) shows a sensor 204 that detects vibrations in the x direction at the time of actual cutting in a feature amount space obtained by combining the first learning information (J1) and the seventh learning information (J7). The results of plotting the sensor data and visualizing the distribution of the presence or absence of chatter vibration are shown.

As can be understood from FIGS. 10A to 10I, when the feature amount of the “maximum peak power of the non-integer multiple frequencies of the cutting fundamental frequency Fc in the entire frequency band Ba” on the horizontal axis is large, It can be seen that the distribution with chatter vibration (that is, a white triangle mark) has a distribution with no chatter vibration (that is, a black circle) when the feature value has a small value.

Further, looking at the vertical axis of FIGS. 10A to 10I, “peak power dispersion in divided band B1 (divided band including cutting fundamental frequency Fc)” or “divided band B1 (cutting fundamental frequency) It is understood that when the “third largest peak power in the divided band including Fc” has a small value, a distribution without chatter vibration exists, and when it has a large value, a distribution with chatter vibration exists. When viewed from a combination of these two, it can be roughly understood that the lower left region is a distribution region without chatter vibration and the other regions are distribution regions with chatter vibration.

As described above, when the chatter vibration detection device or the chatter vibration detection method according to the second embodiment is used, “the maximum peak power of the non-integer multiple frequencies of the cutting fundamental frequency Fc in the entire frequency band Ba” is , Is a very effective feature amount for classifying the distribution of chatter vibrations, and further, “peak power dispersion in divided band B1 (divided band including cutting fundamental frequency Fc)” or “divided band B1”. By using an effective feature amount such as "the third largest peak power in the divided band including the cutting fundamental frequency Fc", it becomes possible to accurately and easily determine the presence or absence of chatter vibration. ..

Note that it is also possible to determine the presence or absence of chatter vibration using a multidimensional frequency space based on other combinations of a plurality of feature amounts.

In the second embodiment, from the viewpoint of how much the determination accuracy of the presence or absence of chatter vibration decreases when one of the feature quantities in the first embodiment is reduced, the feature quantity used for the determination of the presence or absence of chatter vibration is reduced. An example of selecting the type is described. That is, when the feature amount is decreased by one and the chatter vibration determination accuracy is significantly reduced, the feature amount is considered to be a combination of feature amounts that is very effective in determining the presence or absence of chatter vibration. .. By applying such processing to the generation of the learning model 105a, the presence/absence of chatter vibration can be determined with high accuracy.

<<3>> Third embodiment.
The chatter vibration detection device according to the third embodiment has the same configuration as that shown in FIGS. 1 and 2. Therefore, FIG. 1 and FIG. 2 are also referred to when describing the chatter vibration detection device according to the third embodiment. The chatter vibration detection apparatus according to the third embodiment is different from the chatter vibration detection apparatus according to the first or second embodiment with respect to the feature amount extracted by the feature amount extraction unit 103. Other than this point, the chatter vibration detection device according to the third embodiment is the same as the chatter vibration detection device according to the first or second embodiment.

FIG. 11 is a diagram showing an example of learning data L2 including a feature vector generated by the feature amount extraction unit 103 in a table format during the learning process of the chatter vibration detection device according to the third embodiment. FIG. 12 is a diagram showing an example of the determination target data D2 including the feature amount for one frame input to the chatter vibration determination unit 106 in the table format during the determination process of the chatter vibration detection device according to the third embodiment. ..

The chatter vibration detection apparatus according to the third embodiment acquires the processing condition parameter D4 from the NC apparatus 207. The processing condition parameter D4 is, for example, the cutting rotation speed (that is, the cutting rotation frequency), the material of the processing target, the feed rate of the processing target, the cutting thickness of the processing target, the type of processing (for example, end mill processing or parting). Processing etc.), etc.

During the learning process for the presence or absence of chatter vibration, the feature amount extraction unit 103 generates one or more feature vectors as shown in FIG. The chatter vibration learning unit 104 stores the learning model 105 a generated from the generated one or more feature vectors in the storage unit 105.

When determining the presence or absence of chatter vibration, the feature amount extraction unit 103 generates a feature amount as shown in FIG. 12, and the chatter vibration determination unit 106 uses the learning model for the feature amount generated by the feature amount extraction unit 103. The presence or absence of chatter vibration is determined with reference to 105a.

As described above, if the chatter vibration detection device or chatter vibration detection method according to the third embodiment is used, parameters for cutting processing are acquired from the NC device 207 and the presence or absence of chatter vibration is learned in addition to the feature vector. I do. Also, the presence/absence of chatter vibration is determined in addition to the feature vector by obtaining the parameters for cutting from the NC device 207. As described above, when the chatter vibration detection device or the chatter vibration detection method according to the third embodiment is used, it is not necessary to prepare the learning model 105a for each of various cutting conditions, and one learning model 105a can be used as a plurality of learning models. It can be used to determine the presence or absence of chatter vibration under cutting conditions.

<<4>> Fourth Embodiment
FIG. 13 is a block diagram schematically showing the configuration of chatter vibration detection device 400 according to the fourth embodiment. 13, constituent elements that are the same as or correspond to the constituent elements illustrated in FIG. 2 are denoted by the same reference numerals as those illustrated in FIG. Chatter vibration detection apparatus 400 according to the fourth embodiment temporarily stores learning label input section 301 as an operation input section used when a user inputs a learning label, and sensor data D0 during cutting. A storage unit 303 and a learning data generation unit 302 that adds a learning label input by the user to the sensor data D0 temporarily stored in the temporary storage unit 303 and adds the learning data to the labeled sensor data 101a are provided. The difference from the chatter vibration detection device according to the first to third embodiments is that it is different. Except for this point, the chatter vibration detection device according to the fourth embodiment is the same as the chatter vibration detection device according to any of the first to third embodiments.

The chatter vibration detection devices according to the first to third embodiments use the learning model 105a generated by the previous learning process to perform the process of determining whether chatter vibration is present. However, when cutting a new pattern, the learning model 105a may not be able to handle the cutting of a new pattern, and it may not be possible to accurately determine the presence or absence of chatter vibration.

Therefore, in chatter vibration detection apparatus 400 according to the fourth embodiment, sensor data D0 is temporarily stored in temporary storage unit 303 during cutting. Then, the determination result information on the presence or absence of chatter vibration determined by the chatter vibration determination unit 106 is displayed on the determination result display unit 304. The user confirms the displayed determination result information. If the determination result information is different from the state of presence or absence of chatter vibration that is directly recognized by the user, the user recognizes that an erroneous determination has occurred, and inputs a correct determination result from the learning label input unit 301, for example, by hand. To do. The learning data generation unit 302 adds the learning label L0 input by the user to the sensor data D0 output from the sensor 204 that is temporarily stored in the temporary storage unit 303, and creates new labeled sensor data. , Add new labeled sensor data (that is, new learning data) to the existing labeled sensor data. Then, the FFT calculation unit 102 and the feature amount extraction unit 103 generate new feature data as new learning data, and the chatter vibration learning unit 104 performs learning to update the learning model 105a.

As described above, when the chatter vibration detection device 400 or the chatter vibration detection method according to the fourth embodiment is used, if the chatter vibration determination unit 106 determines that the chatter vibration is determined to be an erroneous determination, the user learns it. It is possible to input a correct determination result from the label input unit 301. Therefore, additional learning can be performed on-site (for example, a service performed by a service provider by visiting the chatter vibration detection device 100 of the processing device 200), and a new chatter vibration pattern can be flexibly generated. Can respond.

<<5>> Fifth Embodiment
The chatter vibration detection device according to the fifth embodiment has the same configuration as that shown in FIG. Therefore, FIG. 13 is referred to when describing the chatter vibration detection device according to the fifth embodiment. In the chatter vibration detection apparatus according to the fifth embodiment, the determination result display unit 304 presents the chatter vibration detection certainty factor indicating the certainty factor of the chatter vibration determination result to the user as the presentation information D3a. It is different from the chatter vibration detection device related to. Except for this point, the chatter vibration detecting apparatus according to the fifth embodiment is the same as the chatter vibration detecting apparatus according to the fourth embodiment.

The chatter vibration determination unit 106 can acquire the chatter vibration detection certainty factor by a known method. For example, when the chatter vibration determination unit 106 is a support vector machine, the chatter vibration determination unit 106 determines the signed distance from the boundary surface that divides the two classes by using a sigmoid function to determine that the input belongs to a specific class. It can be converted into a posterior probability of being present and output as a certainty factor. This method is described in Non-Patent Document 2, for example.

As described above, if the chatter vibration detection device or the chatter vibration detection method according to the fifth embodiment is used, the user can grasp the certainty factor of the chatter vibration determination result, and if each progress is low, the user learns correctly. The learning model 105a can be updated on-site by entering the label for

<<6>> Sixth Embodiment
FIG. 14 is a block diagram schematically showing the configuration of chatter vibration detection device 600 according to the sixth embodiment. 14, constituent elements that are the same as or correspond to the constituent elements shown in FIG. 13 are assigned the same reference numerals as the reference numerals shown in FIG. Chatter vibration detection device 600 according to the sixth embodiment differs from chatter vibration detection device according to the fifth embodiment in that chatter vibration detection certainty D5 is provided to learning data generation unit 302. Except for this point, the chatter vibration detection device according to the sixth embodiment is the same as the chatter vibration detection device according to the fifth embodiment.

In chatter vibration detection apparatus 600 according to the sixth embodiment, learning data generation unit 302 causes temporary storage unit to determine when chatter vibration determination confidence D5 output from chatter vibration determination unit 106 is higher than a preset threshold value. The determination result determined by the chatter vibration determination unit 106 is added to the sensor data D0 temporarily stored in 303 as a learning label.

As described above, when the chatter vibration detection device 600 or the chatter vibration detection method according to the sixth embodiment is used, the chatter vibration determination unit 106 provides the chatter vibration detection certainty D5 to the learning data generation unit 302. The determination result of chatter vibration is determined to be correct, and the result of the presence or absence of chatter vibration can be added to the sensor data D0 as a learning label, and learning enhancement can be automatically performed on-site.

<<7>> Seventh Embodiment
FIG. 15 is a diagram schematically showing the configuration of the chatter vibration suppression device 70 and the configuration of a processing device 200 that performs cutting according to the seventh embodiment. 15, constituent elements that are the same as or correspond to the constituent elements shown in FIG. 1 are assigned the same reference numerals as those shown in FIG. The system including the chatter vibration detection device 700 according to the seventh embodiment obtains the processing position information L3 from the NC device 207, and further, the processing surface imaging unit 401 that images the processing surface after cutting, and the processing surface imaging unit 401. 1 is different from the system shown in FIG. 1 in that it includes an image determination unit 402 that determines the presence or absence of chatter vibration from the image of the machined surface captured by.

FIG. 16 is a block diagram schematically showing the configuration of chatter vibration detection device 700 according to the seventh embodiment. 16, components that are the same as or correspond to the components shown in FIG. 13 are assigned the same reference numerals as those shown in FIG. 16 shows the internal configuration of chatter vibration detection device 700 according to the seventh embodiment. The processing surface photographing unit 401 and the image determination unit 402 described above are added to the configuration of the first embodiment. Except for this point, chatter vibration detection apparatus 700 according to the seventh embodiment is the same as chatter vibration detection apparatus according to the fourth embodiment.

Next, the operation of the seventh embodiment will be described. The temporary storage unit 303 temporarily stores the sensor data D0 output from the sensor 204 during cutting, and also temporarily stores the processing position information L3 in association with the sensor data D0 during cutting. When the cutting process is completed, the machined surface imaging unit 401 captures an image of the machined surface. Then, the image determination unit 402 performs determination processing of the presence or absence of chatter vibration from the image data I0 of the captured image of the processed surface, and outputs determination result information I1 including coordinates in the image. Here, a determination algorithm for determining the presence or absence of chatter vibration from the image of the processed surface is known. In this determination algorithm, for example, by deep learning using a convolutional neural network suitable for image recognition, images with and without chatter vibration are learned in advance, and the learning model is used.

Then, the learning data generation unit 302 determines the determination result information I1 including the coordinates in the image corresponding to the determination result of the presence or absence of chatter vibration acquired from the image data I0 of the processing surface, and the processing position information acquired from the NC device 207. Based on L3, a new learning label for the presence or absence of chatter vibration is added to the temporarily stored sensor data D0. Specifically, the relationship between the pixel coordinates in the image and the coordinates of the actual machining coordinate system has been obtained by pre-calibration, and it is possible to perform coordinate conversion from pixel coordinates to the coordinates of the actual machining coordinate system. There is. By converting the pixel coordinates with chatter vibration or the pixel coordinates without chatter vibration detected by the image determination unit 402 into the coordinates of the actual machining coordinate system, the sensor data D0 output from the sensor 204 at the time of cutting is compared. It is possible to add a label indicating the presence or absence of chatter vibration with an accuracy of each frame. The labeled sensor data is added as new learning data to the labeled sensor data 101a. Then, the FFT calculation unit 102 and the feature amount extraction unit 103 generate a feature vector from the labeled sensor data including the added learning data. The chatter vibration learning unit 104 updates the learning model 105a by performing learning using the feature vector. The additional processing of the learning data described above is performed when the determination result of the chatter vibration determination unit 106 is different from the determination result of the image determination unit 402.

As described above, when the chatter vibration detection device 700 or the chatter vibration detection method according to the seventh embodiment is used, when the determination result of the chatter vibration determination unit 106 is different from the determination result of the image determination unit 402. A label is automatically attached to the sensing data using the determination result of the image determination unit 402. Therefore, the user does not need to perform input work for generating the learning label, and additional learning can be performed on-site.

In addition, the learning data generation unit 302 can correctly apply the learning label to the corresponding sensor data D0 even when chatter vibration occurs during cutting. Therefore, it is possible to automate the update of the learning model, and it is possible to strengthen the learning by updating the learning model.

<<8>> Eighth Embodiment.
FIG. 17 is a block diagram schematically showing the configuration of chatter vibration detection device 800 according to the eighth embodiment. 17, constituent elements that are the same as or correspond to the constituent elements shown in FIG. 16 are assigned the same reference numerals as those shown in FIG. Other than this point, chatter vibration detection apparatus 800 according to the eighth embodiment is the same as chatter vibration detection apparatus according to the seventh embodiment.

In the seventh embodiment, the learning data addition process is performed when the determination result of the chatter vibration determination unit 106 is different from the determination result of the image determination unit 402. On the other hand, in chatter vibration detection apparatus 800 according to the eighth embodiment, chatter vibration determination unit 106 outputs chatter vibration detection certainty D5 described in the fifth embodiment to learning data generation unit 302. When the output chatter vibration detection certainty factor D5 is lower than a predetermined threshold value, the learning data generation unit 302 includes the coordinates in the image corresponding to the determination result of the chatter vibration presence or absence from the image determination unit 402. The learning data is updated using the determination result information I1.

As described above, if chatter vibration detection apparatus 800 or chatter vibration detection method according to the eighth embodiment is used, chatter vibration determination unit 106 outputs chatter vibration detection certainty D5, and chatter vibration detection certainty D5. Is lower than a predetermined threshold, learning data is generated using the learning label by the image determination unit 402 and the temporarily stored sensor data D0. Therefore, new learning data effective for improving accuracy can be added. Therefore, it is possible to automate the update of the learning model, and it is possible to strengthen the learning by updating the learning model.

<<9>> Embodiment 9.
In the fifth and sixth embodiments, the chatter vibration determination unit 106 in FIG. 14 generates additional learning data by the learning data generation unit 302 according to the chatter vibration detection certainty degree D5, and uses this as the labeled sensor data 101a. In addition, additional learning is performed on-site to update the learning model 105a. Similarly, in the seventh and eighth embodiments, the chatter vibration determination unit 106 in FIG. 17 generates additional learning data by the learning data generation unit 302 according to the chatter vibration detection certainty degree D5, and the labeled sensor data 101a. The learning model 105a is updated by performing additional learning on-site.

FIG. 18 is a block diagram schematically showing the configuration of the chatter vibration detection device 900 according to the ninth embodiment. 18, constituent elements that are the same as or correspond to the constituent elements shown in FIG. 13 are assigned the same reference numerals as those shown in FIG. FIG. 19 is a block diagram schematically showing the configuration of the chatter vibration detection device 901 according to the modification of the ninth embodiment. 19, components that are the same as or correspond to the components shown in FIG. 18 are assigned the same reference numerals as those shown in FIG.

In chatter vibration detection devices 900 and 901 according to the ninth embodiment, learning data generation unit 302 newly generates learning data on-site, and as shown in FIG. 18 or FIG. When added, the server 501 is provided with the communication control unit 108 for transmitting the labeled sensor data 101a. The communication control unit 108 may transmit only the added or changed learning data to the server 501 in order to suppress the data communication amount. The server 501 is connected to the chatter vibration detection device 900 or 901, a plurality of chatter vibration detection devices (not shown) via a network, and uses learning data collected from the plurality of chatter vibration detection devices. , FFT operation unit 102, feature amount extraction unit 103, and chatter vibration learning unit 104 may be used for learning. By connecting to such a server 501, the newly generated learning model 105a can be used for updating the learning models 105a of the plurality of chatter vibration detection devices of the same type. The update process based on the transmission data from the server 501 can be performed at a timing required by each chatter vibration detection device.

Further, whether or not to update the learning model based on the transmission data from the server 501 may be determined with reference to the processing condition parameter D4. For example, when the machining condition parameter D4 of the chatter vibration detection device adopting the learning model and the processing condition parameter D4 of the chatter vibration detection device having the learning model to be updated match, or the degree of matching is predetermined. The learning model of the chatter vibration detection apparatus having the learning model to be updated may be updated when the learning model is equal to or more than the threshold value. In other words, the communication control unit 108 transmits the processing condition parameter D4 corresponding to the learning data to the server 501, and the processing has the same processing condition parameter D4 as the transmitted processing condition parameter D4 or a predetermined relationship with the processing condition parameter D4. The learning data of the condition parameter D4 may be received as a new learning model.

As described above, when the chatter vibration detection devices 900 and 901 or the chatter vibration detection method according to the ninth embodiment are used, the server 501 collects the learning data generated by each chatter vibration detection device 100 and performs new learning. The model 105a is generated and delivered to each chatter vibration detection device 100. For this reason, it is possible to more efficiently enhance the learning and to detect various chatter vibrations with higher accuracy than the on-site learning using only the chatter vibration detection device.

Further, the server 501 generates the learning model 105a for each of the same or similar processing conditions, and each chatter vibration detection device is generated with the learning data of the processing condition parameter D4 which is the same as or close to the processing condition to be implemented. The learning model 105a can be acquired from the server 501. In this case, the presence or absence of chatter vibration can be determined with high accuracy. Moreover, the learning data can be efficiently generated by generating learning data for such individual processing conditions in the server.

<<10>> Modification.
FIG. 20 is a diagram showing an example of the hardware configuration of the chatter vibration detection device according to the modifications of the first to ninth embodiments. The chatter vibration detection device according to the modifications of the first to ninth embodiments includes, for example, a program as software, that is, a memory 602 as a storage device that stores the chatter vibration detection program according to the first to ninth embodiments, and a memory. A processor 601 as an information processing unit that executes the program stored in 602 is provided. The chatter vibration detection device according to the modification is, for example, a computer. The chatter vibration detection program according to the modification is stored in the memory 602 from a storage medium that stores information via a reading device (not shown) or via a communication interface (not shown) connectable to the Internet or the like. .. In addition, the chatter vibration detection device according to the modified example includes an input device that is a user operation unit such as a mouse 603 and a keyboard 604, a display device 605 that displays an image, and a voice output unit (not shown) that outputs voice. Output device such as. Further, the chatter vibration detection device according to the modification may include an auxiliary storage device 606 that stores various information such as a database. The auxiliary storage device 606 may be a storage device existing on a cloud that can be connected via a communication interface (not shown).

The FFT calculation unit 102, the feature amount extraction unit 103, the chatter vibration learning unit 104, the chatter vibration determination unit 106, the chatter vibration suppression unit 107, the learning data generation unit 302, the image determination unit 402 in any of the first to ninth embodiments. The communication control unit 108 and the whole or part of the NC device 207 can be realized by the processor 601 that executes a program stored in the memory 602 of FIG. Further, the storage unit 101, the storage unit 105, and the temporary storage unit 303 may be a part of the auxiliary storage device 606 of FIG.

The configurations of the first to ninth embodiments and their modifications can be combined with each other as appropriate.

10, 70 Chatter vibration suppression device, 100, 400, 600, 700, 800, 900, 901 Chatter vibration detection device, 101 storage unit (first storage unit), 101a labeled sensor data, 102 FFT calculation unit (conversion processing Section), 103 feature amount extraction section, 104 chatter vibration learning section, 105 storage section (second storage section), 105a learning model, 106 chatter vibration determination section, 107 chatter vibration suppression section, 200 machining device, 201 cutting tool, 202 main shaft section, 203 stage, 204 sensor (acceleration sensor), 205 support section, 206 machining object (workpiece), 207 NC device, 301 learning label input section, 302 learning data generation section, 303 temporary storage section, 304 determination result display unit, 401 processed surface photographing unit, 402 image determination unit, 501 server.

Claims (17)

A chatter vibration detection device that determines the presence or absence of chatter vibration, based on sensor data output from a sensor that detects the vibration of a cutting tool of a processing device that processes a workpiece,
A first storage unit for storing the labeled sensor data created by adding a label indicating the presence or absence of chatter vibration to the sensor data;
A first extraction process for extracting one or more first feature amounts as learning data from the labeled sensor data, and extraction of determination target data including one or more second feature amounts from the sensor data. A feature amount extraction unit that performs a second extraction process,
A chatter vibration learning unit that generates a learning model from the learning data,
Whether or not the determination target data is data indicating chatter vibration, the chatter vibration determination unit for determining with reference to the learning model,
Chatter vibration detection device characterized by having. A first conversion process for generating first data in frequency space by Fourier-transforming the labeled sensor data, and second data in frequency space by Fourier-transforming the sensor data output from the sensor. A second conversion process for generating
The chatter vibration detection device according to claim 1, wherein the feature amount extraction unit extracts the learning data from the first data and the determination target data from the second data. The conversion processing unit, the cutting fundamental frequency corresponding to the number of blades of the cutting tool for cutting the workpiece per unit time, the Fourier transform so as to be an integer multiple of the frequency resolution determined by the sampling frequency of the sensor data, The chatter vibration detection device according to claim 2, wherein a window size in conversion is determined. The feature amount extraction unit,
The one or more first peaks based on the frequency of the one or more first peaks extracted from the first data output from the conversion processing unit and the power of the one or more first peaks. 1 feature quantity is extracted,
The one or more second peaks based on the frequency of the one or more second peaks extracted from the second data output from the conversion processing unit and the power of the one or more second peaks. The chatter vibration detection device according to claim 3, wherein the feature amount of 2 is extracted. The feature amount extraction unit, in the first extraction processing,
Dividing the entire frequency band of the frequency spectrum of the first data into a predetermined number of frequency bands,
In each of the plurality of divided bands that are the plurality of frequency bands,
First learning information indicating the sum, average, or variance of the powers of the first peaks,
Second learning information indicating the number of the first peaks,
Of the first peak, third learning information indicating the power and frequency of the first peak having power within the top three,
Fourth learning information indicating the sum, average, or variance of the powers of the first peaks having a frequency that is a non-integer multiple of the cutting fundamental frequency,
Fifth learning information indicating the power and frequency of the first peak having power within the top three of the first peaks having a frequency that is a non-integer multiple of the cutting fundamental frequency,
Sixth learning information indicating the number of the first peaks having a frequency that is a non-integer multiple of the cutting fundamental frequency,
Seventh learning information indicating a ratio of the number of the first peaks having a frequency that is a non-integer multiple of the cutting fundamental frequency with respect to the number of the first peaks in the entire frequency band,
Eighth learning information indicating a sum, an average, or a variance of the power of the first peak having a frequency that is an integral multiple of the cutting fundamental frequency,
Of the first peak having a frequency that is an integral multiple of the cutting fundamental frequency, ninth learning information indicating the power and frequency of the first peak having the power within the top three positions, and the cutting fundamental frequency Tenth learning information indicating the number of the first peaks having an integral multiple frequency,
Of two or more of the learning information,
The chatter vibration learning unit generates the learning model from a combination of two or more pieces of learning information among the first to tenth learning information,
The feature amount extraction unit, in the second extraction processing,
Dividing the entire frequency band of the frequency spectrum of the second data into the plurality of divided bands,
In each of the plurality of divided bands,
First determination information indicating the sum, average, or variance of the powers of the second peaks,
Second determination information indicating the number of the second peaks,
Of the second peak, third determination information indicating the power and frequency of the second peak having power within the top three positions,
Fourth determination information indicating the sum, average, or variance of the powers of the second peaks having a frequency that is a non-integer multiple of the cutting fundamental frequency,
Fifth determination information indicating the power and frequency of the second peak having power within the top three of the second peaks having a frequency that is a non-integer multiple of the cutting fundamental frequency,
Sixth determination information indicating the number of the second peak having a frequency that is a non-integer multiple of the cutting fundamental frequency,
Seventh determination information indicating a ratio of the number of the second peaks having a frequency that is a non-integer multiple of the cutting fundamental frequency with respect to the number of the second peaks in the entire frequency band,
Eighth determination information indicating the sum, average, or variance of the powers of the second peaks having a frequency that is an integral multiple of the cutting fundamental frequency,
Of the second peak having a frequency that is an integral multiple of the cutting fundamental frequency, ninth determination information indicating the power and frequency of the second peak having the power within the top three positions, and the cutting fundamental frequency Tenth determination information indicating the number of the second peaks having an integral multiple frequency,
Of two or more of the judgment information,
Whether the chatter vibration determination unit is data indicating that chatter vibration is present, based on a combination of the extracted two or more pieces of determination information among the first to tenth determination information. The chatter vibration detection device according to claim 4, wherein whether or not the determination is made is determined with reference to the learning model. The feature amount extraction unit, in the first extraction processing,
Dividing the entire frequency band of the frequency spectrum of the first data into a predetermined number of frequency bands,
First learning information indicating a first maximum peak power which is the maximum power among the powers of the first peaks having a frequency that is a non-integer multiple of the cutting fundamental frequency in the entire frequency band,
Second learning information indicating a total power of a first peak in a divided band including the cutting fundamental frequency among the plurality of divided bands,
Third learning information indicating an average of powers of first peaks in a divided band including the cutting fundamental frequency among the plurality of divided bands,
Fourth learning information indicating a dispersion of power of a first peak in a divided band including the cutting fundamental frequency among the plurality of divided bands,
Fifth learning information indicating the power of the first peak having the first largest power in the first peak in the divided band including the cutting fundamental frequency among the plurality of divided bands,
Sixth learning information indicating the power of the first peak having the second largest power in the first peak in the divided band including the cutting fundamental frequency among the plurality of divided bands,
Seventh learning information indicating the power of the first peak having the third largest power in the first peak in the divided band including the cutting fundamental frequency among the plurality of divided bands,
Eighth learning information indicating the number of first peaks having a frequency that is a non-integer multiple of the cutting fundamental frequency in a divided band including the cutting fundamental frequency among the plurality of divided bands,
Of the plurality of divided bands, ninth learning information indicating the number of first peaks in the divided band including the cutting fundamental frequency, and in the divided band including the cutting fundamental frequency of the plurality of divided bands, 10th learning information showing the ratio of the number of 1 peaks to the total number of 1st peaks in all the above-mentioned frequency bands,
Of two or more of the learning information,
The chatter vibration learning unit generates the learning model from a combination of two or more pieces of learning information among the first to tenth learning information,
The feature amount extraction unit, in the second extraction processing,
Dividing the entire frequency band of the frequency spectrum of the second data into the plurality of divided bands,
First determination information indicating a second maximum peak power which is the maximum power of the powers of the second peaks having a frequency that is a non-integer multiple of the cutting fundamental frequency in the entire frequency band,
Second determination information indicating a total power of second peaks in a divided band including the cutting fundamental frequency among the plurality of divided bands,
Third determination information indicating an average of powers of second peaks in a divided band including the cutting fundamental frequency among the plurality of divided bands,
Fourth determination information indicating a power distribution of a second peak in a divided band including the cutting fundamental frequency of the plurality of divided bands,
Fifth determination information indicating the power of the second peak having the first largest power in the second peak in the divided band including the cutting fundamental frequency among the plurality of divided bands,
Sixth determination information indicating the power of the second peak having the second largest power in the second peak in the divided band including the cutting fundamental frequency among the plurality of divided bands,
Seventh determination information indicating the power of the second peak having the third largest power in the second peak in the divided band including the cutting fundamental frequency among the plurality of divided bands,
Eighth determination information indicating the number of second peaks having a frequency that is a non-integer multiple of the cutting fundamental frequency in a divided band including the cutting fundamental frequency among the plurality of divided bands,
Of the plurality of divided bands, ninth determination information indicating the number of second peaks in the divided band including the cutting fundamental frequency, and in the divided band including the cutting fundamental frequency of the plurality of divided bands, 10th determination information indicating the ratio of the number of 2 peaks to the total number of 2nd peaks in all the frequency bands,
Of two or more of the judgment information,
The chatter vibration determination unit determines whether the determination target data is data indicating chatter vibration based on a combination of two or more pieces of determination information among the first to tenth determination information. The chatter vibration detection device according to claim 4, wherein the determination is performed by referring to a learning model. The learning data includes a processing condition parameter that is a condition for cutting by the processing device,
The chatter vibration learning unit generates the learning model from learning data including the processing condition parameter,
The determination target data includes a processing condition parameter by the processing device,
The chatter vibration determination unit determines whether or not the determination target data including the processing condition parameter is data indicating that there is chatter vibration, with reference to the learning model. Chatter vibration detection device. A learning label input unit that receives user input,
A temporary storage unit for temporarily storing the sensor data,
New labeled sensor data is generated from the sensor data stored in the temporary storage unit and the user input, and the labeled sensor data stored in the first storage unit is converted into the new label. A learning data generation unit that updates using the attached sensor data,
The chatter vibration detecting device according to any one of claims 1 to 7, further comprising: The chatter vibration determination unit generates chatter vibration detection certainty indicating the certainty of the result of the determination by the chatter vibration determination unit, and outputs a display signal for displaying the chatter vibration detection certainty on a display device. The chatter vibration detection device according to claim 8. A temporary storage unit for temporarily storing the sensor data,
New labeled sensor data is generated from the result of determination of chatter vibration presence or absence based on the image obtained by photographing the processed surface of the processing target, and is stored in the first storage unit. A learning data generation unit that updates the labeled sensor data using the new labeled sensor data,
The chatter vibration detecting device according to any one of claims 1 to 7, further comprising: A temporary storage unit for temporarily storing the sensor data,
An image determination unit that determines the presence or absence of chatter vibration based on the image obtained by photographing the processing surface of the processing target,
A learning data generation unit that generates new labeled sensor data from the result of the determination, and updates the labeled sensor data stored in the first storage unit using the new labeled sensor data. ,
The chatter vibration detecting device according to any one of claims 1 to 7, further comprising: The chatter vibration determination unit generates chatter vibration detection certainty indicating the certainty of the result of the determination by the chatter vibration determination unit,
When the chatter vibration determination unit outputs the chatter vibration detection certainty factor is lower than a predetermined certainty factor threshold, the learning data generation unit, the labeled sensor data based on the result of the determination using the image. The chatter vibration detection device according to claim 10 or 11, characterized by updating. A second storage unit that stores the learning model;
A communication control unit that collects learning data from a plurality of chatter vibration detection devices and enables communication with a server that performs learning using the learning data,
Further equipped with,
The chatter vibration detection device according to any one of claims 1 to 12, wherein the communication control unit updates the learning model using a new learning model generated by the server. The communication control unit transmits a processing condition parameter corresponding to the learning data to the server, and learns a processing condition parameter that is the same as the transmitted processing condition parameter or has a predetermined relationship with the processing condition parameter. The chatter vibration detecting apparatus according to claim 13, wherein the use data is received as the new learning model. Based on the sensor data output from the sensor that detects the vibration of the cutting tool of the processing device for processing the workpiece, a chatter vibration detection method for determining the presence or absence of chatter vibration,
A step of performing a first extraction process for extracting one or more first feature amounts as learning data from the labeled sensor data created by adding a label indicating the presence or absence of chatter vibration to the sensor data;
Performing a second extraction process of extracting determination target data including one or more second feature amounts from the sensor data,
Generating a learning model from the learning data,
And a step of determining whether or not the determination target data is data indicating chatter vibration, by referring to the learning model. A chatter vibration detection program that causes a computer to perform chatter vibration detection processing that determines the presence or absence of chatter vibration based on sensor data output from a sensor that detects vibration of a cutting tool of a processing device that processes a workpiece. ,
On the computer,
A first extraction process for extracting one or more first feature quantities as learning data from the labeled sensor data created by adding a label indicating the presence or absence of chatter vibration to the sensor data;
A second extraction process of extracting determination target data including one or more second feature amounts from the sensor data;
Generating a learning model from the learning data,
A chatter vibration detection program, which executes a process of determining whether or not the determination target data is data indicating chatter vibration, with reference to the learning model. The chatter vibration detecting device according to any one of claims 1 to 14,
Based on the result of the determination output from the chatter vibration determination unit of the chatter vibration detection device, chatter vibration suppressor for controlling the operation of the processing device,
A chatter vibration suppressing device characterized by being provided with.

Images