Search

KR-20260064890-A - Chatter Detection Apparatus Using Ensemble Models and method there of

KR20260064890AKR 20260064890 AKR20260064890 AKR 20260064890AKR-20260064890-A

Abstract

An automated system for creating medical device certification documents is provided. The automated system for creating medical device certification documents according to an embodiment of the present invention includes: a user management unit that collects and manages a user's personal information and medical device certification document creation history and provides a customized solution based on the collected information; a data analysis unit that collects medical device certification documents and related data, and constructs a data set by classifying, pre-processing, post-processing, and anonymizing the collected data; a document creation unit that trains an AI model with the data set, generates a medical device certification document by the trained AI model, detects errors in the generated medical device certification document, analyzes the cause, and corrects them; and a feedback unit that collects and analyzes feedback from a user terminal to optimize the AI model.

Inventors

  • 김준
  • 정용희
  • 김보현

Assignees

  • 한국생산기술연구원

Dates

Publication Date
20260508
Application Date
20241030

Claims (17)

  1. A signal processing unit that collects vibration signals generated in a milling process based on multi-signal processing and generates individual spectrograms; An image blending unit that generates a blended image by performing blending processing on the spectrogram images generated by the signal processing unit, and generates a dataset based on the generated blended image; A CNN training unit that trains a plurality of CNN models (Convolutional Neural Networks) for determining chatter based on the dataset generated by the image blending unit; and An ensemble application unit that detects chatter based on the plurality of CNN models generated in the above CNN learning unit; Chatter detection device utilizing an ensemble model including
  2. In paragraph 1, The above signal processing unit is, An STFT unit that performs a Short-Time Fourier Transform (STFT) on the vibration signal to generate a two-dimensional spectrogram image according to the amplitude and frequency of the vibration signal; A CWT unit that generates a two-dimensional spectrogram image by performing a Continuous Wavelet Transform (CWT) on the above vibration signal; and An HHT unit that generates a two-dimensional spectrogram image by performing HHT (Hilbert-Huang Transform) on the above vibration signal; Chatter detection device utilizing an ensemble model including
  3. In paragraph 2, The above image blending unit is, The spectrogram images generated by the signal processing unit are each assigned a transparency according to the signal transformation and summed to blend the images, and The above blended image is preprocessed to generate the above dataset, and A chatter detection device utilizing an ensemble model in which the transparency is a real value between 0 and 1 and the sum is 1.
  4. In paragraph 3, The above CNN learning unit is, Based on the above dataset, the plurality of CNN models are trained, and the plurality of CNN models are sorted according to validation accuracy. A chatter detection device utilizing an ensemble model that verifies performance by setting the model with the highest accuracy as BBM (Best Blending Model) and other models as SM (Supporting Model).
  5. In paragraph 4, The above ensemble application part is, An ensemble score is derived by summing (a+b) the product (a) of the prediction value, model accuracy, and confidence score of each of the plurality of CNN models and the BBM, and the sum (b) of the product (a+b) of the prediction value, model accuracy, and confidence evaluation and the SM of each CNN model, and dividing the result by the number of CNN models + 1. A chatter detection device utilizing an ensemble model that classifies into a normal class or a chatter class by deriving the average value of a feature map through Adaptive Average Precision (AAP) based on the above-derived ensemble score.
  6. In paragraph 3, The image blending unit is a chatter detection device utilizing an ensemble model that converts the blended image to a size of 128x128 pixels and preprocesses it.
  7. In paragraph 1, Each of the above multiple CNN models consists of four convolution layers, and A chatter detection device utilizing an ensemble model in which the above convolution layer increases the number of channels by applying a Leaky ReLU activation function, reduces the size of the feature map by applying a Max Pooling function, and includes dropout set to a ratio of 0.5 to prevent overfitting.
  8. In paragraph 5, A chatter detection device utilizing an ensemble model in which the above predicted value appears as -1 when the processing state is normal and 1 when there is chatter.
  9. In paragraph 5, A chatter detection device utilizing an ensemble model, wherein the above model accuracy represents the accuracy for a validation set per CNN model generated above.
  10. In paragraph 5, A chatter detection device utilizing an ensemble model in which the confidence score has a value from 0 to 1, and the more certain the prediction result of the CNN model, the closer the value is to 1.
  11. In paragraph 5, A chatter detection device utilizing an ensemble model in which the above ensemble score has a value between -1 and 1, and diagnoses the vibration signal as chatter when it has a value greater than or equal to 0.
  12. A signal processing step that collects vibration signals generated in a milling process based on multi-signal processing and generates individual spectrograms; An image blending step that generates a blended image by performing blending processing on the spectrogram images generated in the signal processing step, and generates a dataset based on the generated blended image; A CNN training step for training a plurality of CNN models to determine chatter based on the dataset generated in the above image blending step; and An ensemble application step for detecting chatter based on the plurality of CNN models generated in the above CNN training step; Chatter detection method using an ensemble model including
  13. In Paragraph 12, The above signal processing step is, An STFT step of performing STFT on the vibration signal to generate a two-dimensional spectrogram image according to the amplitude and frequency of the vibration signal; A CWT step of performing CWT on the above vibration signal to generate a two-dimensional spectrogram image; and HHT step of performing HHT on the above vibration signal to generate a two-dimensional spectrogram image; Chatter detection method using an ensemble model including
  14. In Paragraph 12, The above image blending step is, The spectrogram images generated in the above signal processing step are each assigned a transparency according to the above signal transformation, and the images are blended by summing them. The above blended image is preprocessed to generate the above dataset, and A chatter detection method utilizing an ensemble model in which the transparency is a real value between 0 and 1 and the sum is 1.
  15. In Paragraph 12, The above CNN training step is, Based on the above dataset, the plurality of CNN models are trained, and the plurality of CNN models are sorted according to validation accuracy. A chatter detection method using an ensemble model that verifies performance by setting the model with the highest accuracy as BBM and other models as SM.
  16. In paragraph 15, The above ensemble application step is, An ensemble score is derived by summing (a+b) the product (a) of the prediction value, accuracy, and reliability evaluation of each of the plurality of CNN models and the BBM, and the sum (b) of the product (a+b) of the prediction value, accuracy, and reliability evaluation and the SM of each CNN model, and dividing the result by the number of CNN models + 1. A chatter detection method using an ensemble model that classifies into a normal class or a chatter class by deriving the average value of a feature map through AAP based on the ensemble score derived above.
  17. A computer-readable recording medium having a program recorded thereon for performing the method of any one of paragraphs 12 through 16.

Description

Chatter Detection Apparatus Using Ensemble Models and Method Thereof The present invention relates to a chatter detection device and method utilizing an ensemble model, and in particular, to a chatter detection device and method based on a convolutional neural network model learned through an image generated by multi-signal processing of vibration signals generated in a milling process. Milling is a machine tool-based machining technology utilized in various manufacturing fields, primarily for the production of precision parts. In the milling process, chatter is an abnormal vibration phenomenon occurring between the material surface and the tool outside the spindle's rotational frequency. It significantly impacts the performance and productivity of machine tools by causing reduced process precision, shortened tool life, and defects. Therefore, if chatter is detected early in the milling process, it can be eliminated by modifying machining conditions. This prevents tool damage, thereby reducing maintenance costs and equipment downtime while maintaining the quality of the machined products. Conventional studies on chatter detection have collected vibration signals from accelerometer sensors, preprocessed them using single-signal processing techniques, and detected the occurrence of chatter by utilizing features extracted from each signal. However, due to the limitations of single-signal processing techniques, this method faces difficulties in detecting various types of chatter. Therefore, there is a need for devices and methods that can improve chatter detection accuracy and ensure detection performance even under changes in process conditions. FIG. 1 is a schematic diagram of a chatter detection system utilizing an ensemble model according to one embodiment of the present invention. FIG. 2 is a cutting process photograph (a), a surface comparison photograph (b), a diagram of chatter occurrence (c), and a tool wear photograph (d) to explain a chatter detection device utilizing an ensemble model according to an embodiment of the present invention. FIG. 3 is a block diagram of a chatter detection device utilizing an ensemble model according to an embodiment of the present invention. FIG. 4 is a detailed block diagram of the signal processing unit of a chatter detection device utilizing an ensemble model according to one embodiment of the present invention. FIG. 5 is a spectrogram image for explaining a signal processing technique of a chatter detection device using an ensemble model according to an embodiment of the present invention. FIG. 6 is an STFT spectrogram (a), SWT spectrogram (b), HHT spectrogram (c), blending image (d), and transparency calculation formula (e) for explaining the image blending section of a chatter detection device utilizing an ensemble model according to an embodiment of the present invention. FIG. 7 is a diagram illustrating the CNN learning unit of a chatter detection device utilizing an ensemble model according to an embodiment of the present invention. FIG. 8 is a diagram illustrating the CNN learning unit of a chatter detection device utilizing an ensemble model according to an embodiment of the present invention. FIG. 9 is a diagram illustrating an ensemble model of a chatter detection device utilizing an ensemble model according to an embodiment of the present invention. FIG. 10 is a single-signal processing comparison graph (a) and other image merging comparison graph (b) of a chatter detection device utilizing an ensemble model according to one embodiment of the present invention. FIG. 11 is a graph showing the performance test results of a chatter detection device using an ensemble model according to one embodiment of the present invention. FIG. 12 is a flowchart of a chatter detection method using an ensemble model according to an embodiment of the present invention. Hereinafter, embodiments of the present invention are described in detail with reference to the attached drawings so that those skilled in the art can easily implement the present invention. The present invention may be embodied in various different forms and is not limited to the embodiments described herein. In the drawings, parts unrelated to the explanation have been omitted to clearly explain the present invention, and the same reference numerals have been used for identical or similar components throughout the specification. Hereinafter, a chatter detection device and method utilizing an ensemble model according to an embodiment of the present invention will be described in more detail with reference to the drawings. FIG. 1 is a schematic diagram of a chatter detection system utilizing an ensemble model according to one embodiment of the present invention. Referring to FIG. 1, a chatter detection system (1) utilizing an ensemble model may include a chatter detection device (100) utilizing an ensemble model, a user terminal (10), a metal milling machine (20), and an accelerometer sensor (30). The chatter d