Micro-vibration source detection and separation method and system based on machine learning

Abstract

The invention discloses a micro-vibration source detection and separation method and system based on machine learning, and the method comprises the steps: fusing the actually measured micro-vibration data generated by the construction of a newly-built factory building in the microelectronic industry with simulation micro-vibration data, assembling the single-vibration-source micro-vibration data in the fused micro-vibration data to form multi-vibration-source micro-vibration data; enhancing and expanding the single-vibration-source micro-vibration data and the multi-vibration-source micro-vibration data through an adversarial generation algorithm; the single-vibration-source micro-vibration data, the single-vibration-source micro-vibration data enhanced and expanded by the single-vibration-source micro-vibration data, the multi-vibration-source micro-vibration data and the multi-vibration-source micro-vibration data enhanced and expanded by the multi-vibration-source micro-vibration data serve as training and testing sample data, and model training is conducted on a vibration source data separation model through the multi-vibration-source micro-vibration data and the single-vibration-source micro-vibration data; obtaining a trained vibration source data separation model; and testing and verifying the trained vibration source data separation model. According to the invention, multi-vibration-source micro-vibration data can be separated into single-vibration micro-vibration data.

Description

Technical Field

[0001] The present invention relates to the field of civil engineering, and in particular to a method and system for detecting and separating micro-vibration sources based on machine learning. Background Art

[0002] With the development of the microelectronics industry, more and more high-precision factory buildings are being built. At the same time, due to the agglomeration effect of the microelectronics industry, many microelectronics companies currently choose to build new factories next to existing high-precision IC factories. This may lead to excessive micro-vibrations in nearby sites caused by construction vibrations during the construction process, which may cause the equipment inside the existing factory to alarm and stop working. Due to the complex environment and numerous equipment at the construction site, there is little research on the characteristics of the construction vibration source, and due to the complexity of the construction vibration data, a set of vibration data often includes multiple vibrations with different characteristics, such as environmental vibration, traffic vibration, and impact vibration. At present, there is no relevant method that can detect and separate the complex measured micro-vibration data one by one according to the classification of the vibration source. Summary of the invention

[0003] The purpose of the present invention is to provide a method and system for detecting and separating micro-vibration sources based on machine learning, so as to solve the problem of being unable to detect micro-vibration data generated by the construction of new plant buildings in the microelectronics industry and to separate multi-vibration source micro-vibration data into single-source micro-vibration data.

[0004] In order to solve the above technical problems, the present invention provides a method for detecting and separating micro-vibration sources based on machine learning, comprising:

[0005] The measured micro-vibration data containing interference noise generated by the construction of a new plant in the microelectronics industry is integrated with the simulated micro-vibration data without interference noise, and the single-vibration source micro-vibration data in the integrated micro-vibration data are assembled to form multi-vibration source micro-vibration data;

[0006] Enhance and expand single-source micro-vibration data and multi-source micro-vibration data through adversarial generation algorithm;

[0007] Single-source micro-vibration data and its enhanced and expanded single-source micro-vibration data, multi-source micro-vibration data and its enhanced and expanded multi-source micro-vibration data are used as test training and sample data, and a vibration source data separation model is trained through the multi-source micro-vibration data and single-source micro-vibration data to obtain a trained vibration source data separation model;

[0008] The trained vibration source data separation model is tested and verified.

[0009] Furthermore, in the method for detecting and separating micro-vibration sources based on machine learning provided by the present invention, the source data separation model is composed of three deep neural network models, namely, a time domain model, a frequency domain model and a time-frequency analysis model. The loss function of the source data separation model is composed of a weighted average of the loss functions of the time domain model, the frequency domain model and the time-frequency analysis model. During the model training process, the goal of minimizing the loss function value is achieved by continuously adjusting the parameters in the time domain model, the frequency domain model and the time-frequency analysis model. When the loss function value is lower than the set threshold, it is considered that the trained source data separation model can separate multi-source micro-vibration data of micro-vibrations with various characteristics into single-source micro-vibration data.

[0010]

[0011] is the final loss function, is the loss function of the time domain model, is the loss function of the frequency domain model, is the loss function of the time-frequency analysis model, y is the real single-source micro-vibration data in the data set, This is the single source micro-vibration data predicted by the vibration source data separation model.

[0012] Furthermore, the micro-vibration source detection and separation method based on machine learning provided by the present invention tests and verifies the trained source data separation model from three aspects: time domain, frequency domain and time-frequency analysis, and ensures the prediction accuracy and generalization ability of the source data separation model trained with test sample data through three-aspect cross-validation.

[0013] Furthermore, the present invention provides a method for detecting and separating micro-vibration sources based on machine learning, wherein the source data separation model is a convolutional neural network, a recurrent neural network or a Transformer network.

[0014] In order to solve the above technical problems, the present invention also provides a micro-vibration source detection and separation system based on machine learning, comprising:

[0015] The data fusion and assembly module is used to fuse the measured micro-vibration data containing interference noise generated by the construction of a new plant in the microelectronics industry with the simulated micro-vibration data without interference noise, and assemble the single-vibration source micro-vibration data in the fused micro-vibration data to form multi-vibration source micro-vibration data;

[0016] The data enhancement and expansion module enhances and expands the single-source micro-vibration data and multi-source micro-vibration data of the data fusion and assembly module through the adversarial generation algorithm;

[0017] The vibration source data separation model training module uses the single vibration source micro-vibration data and multi-vibration source micro-vibration data of the data fusion and assembly module; the single vibration source micro-vibration data and multi-vibration source micro-vibration data enhanced and expanded by the data enhancement and expansion module are used as test sample data, and the multi-vibration source micro-vibration data are separated into single vibration source micro-vibration data through the vibration source data separation model for model training, and the trained vibration source data separation model is obtained;

[0018] The model testing and verification module tests and verifies the trained vibration source data separation model.

[0019] Furthermore, in the micro-vibration source detection and separation system based on machine learning provided by the present invention, the source data separation model is composed of three deep neural network models, namely, a time domain model, a frequency domain model and a time-frequency analysis model. The loss function of the source data separation model is composed of a weighted average of the loss functions of the time domain model, the frequency domain model and the time-frequency analysis model. During the model training process, the goal of minimizing the loss function value is achieved by continuously adjusting the parameters in the time domain model, the frequency domain model and the time-frequency analysis model. When the loss function value is lower than the set threshold, it is considered that the trained source data separation model can separate multi-source micro-vibration data of micro-vibrations with various characteristics into single-source micro-vibration data.

[0020]

[0021] is the final loss function, is the loss function of the time domain model, is the loss function of the frequency domain model, is the loss function of the time-frequency analysis model, y is the real single-source micro-vibration data in the data set, This is the single source micro-vibration data predicted by the vibration source data separation model.

[0022] Furthermore, the micro-vibration source detection and separation system based on machine learning provided by the present invention tests and verifies the trained source data separation model from three aspects: time domain, frequency domain and time-frequency analysis in the model testing and verification module, and ensures the prediction accuracy and generalization ability of the source data separation model trained with test sample data through cross-validation in three aspects.

[0023] Furthermore, in the micro-vibration source detection and separation system based on machine learning provided by the present invention, the source data separation model is a convolutional neural network, a recurrent neural network or a Transformer network.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] The present invention provides a method and system for detecting and separating micro-vibration sources based on machine learning. The method and system fuse the measured micro-vibration data generated during the construction of a new microelectronic plant in a microelectronic industry cluster area or near the site of an existing microelectronic enterprise with the simulated micro-vibration data, and enhance and expand the fused single-source micro-vibration data and the multi-source micro-vibration data formed by their assembly as training data for a source data separation model. The multi-source micro-vibration data is separated into single-source micro-vibration data to obtain a trained source data separation model, thereby increasing the data volume and production quality of the training data and improving the output accuracy of the trained source data separation model. After testing and verification, the trained source data separation model can separate the multi-source micro-vibration data generated during the construction of a new microelectronic plant in a microelectronic industry cluster area or near the site of an existing microelectronic enterprise into single-source micro-vibration data, thereby overcoming technical barriers. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a flow chart of the method for detecting and separating micro-vibration sources based on machine learning;

[0027] Figure 2 This is a structural composition relationship diagram of the micro-vibration source detection and separation system based on machine learning;

[0028] As shown in the figure:

[0029] 200. Micro-vibration source detection and separation system based on machine learning, 210. Data fusion and assembly module, 220. Data enhancement and expansion module, 230. Vibration source data separation model training module, 240. Model testing and verification module. DETAILED DESCRIPTION

[0030] The present invention is described in detail below in conjunction with the accompanying drawings: The advantages and features of the present invention will become more apparent from the following description. It should be noted that the accompanying drawings are in very simplified form and in non-precise proportions, and are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention.

[0031] Please refer to Figure 1 , an embodiment of the present invention provides a method for detecting and separating micro-vibration sources based on machine learning, which may include:

[0032] Step S1, the measured micro-vibration data containing interference noise generated by the construction of a new plant in the microelectronics industry is merged with the simulated micro-vibration data without interference noise, and the single-source micro-vibration data in the merged micro-vibration data is assembled to form multi-source micro-vibration data. The merged micro-vibration data can meet the characteristics of far-field construction micro-vibration and improve the accuracy of the data.

[0033] Step S2, enhance and expand the single-source micro-vibration data and the multi-source micro-vibration data through the adversarial generation algorithm. By enhancing and expanding the data, the amount of training data in the following step S3 can be increased, and the training accuracy can be improved.

[0034] Step S3, using single-source micro-vibration data and its enhanced and expanded single-source micro-vibration data, multi-source micro-vibration data and its enhanced and expanded multi-source micro-vibration data as training and testing sample data, and training the vibration source data separation model through the multi-source micro-vibration data and single-source micro-vibration data to obtain the trained vibration source data separation model. In order to separate multi-source micro-vibration data into single-source micro-vibration data, the source data separation model can be composed of three deep neural network models, namely, time domain model, frequency domain model and time-frequency analysis model. The loss function of the source data separation model is composed of the weighted average of the loss functions of the time domain model, frequency domain model and time-frequency analysis model. In the model training process, the goal of minimizing the loss function value is achieved by continuously adjusting the parameters in the time domain model, frequency domain model and time-frequency analysis model. When the loss function value is lower than the set threshold, it is considered that the trained source data separation model can separate multi-source micro-vibration data with various characteristics of micro-vibration into single-source micro-vibration data; wherein the various characteristics of micro-vibration include but are not limited to the persistence and randomness of environmental vibration, the instantaneousness and impact characteristics of traffic vibration, etc.

[0035]

[0036] is the final loss function, is the loss function of the time domain model, is the loss function of the frequency domain model, is the loss function of the time-frequency analysis model, y is the real single-source micro-vibration data in the data set, This is the single source micro-vibration data predicted by the vibration source data separation model.

[0037] Step S4, testing and verifying the trained vibration source data separation model. Through testing and verification, the accuracy of the detection of the trained vibration source data separation model can be determined, and it can be ensured that the trained vibration source data separation model has sufficient prediction accuracy and sufficient accuracy for the prediction of all test sample data. In combination with step S3, the trained vibration source data separation model can be tested and verified from three aspects: time domain, frequency domain and time-frequency analysis. Through cross-validation in three aspects, the prediction accuracy and generalization ability of the vibration source data separation model trained with test sample data are ensured. Generalization ability refers to the model's ability to perform on data that has not been seen during the training process, and whether it can be universal. The vibration source data separation model includes but is not limited to convolutional neural networks, recurrent neural networks, and Transformer networks.

[0038] Please refer to Figure 2 The present invention also provides a micro-vibration source detection and separation system 100 based on machine learning, which adopts the above method, including a data fusion and assembly module 110, a data enhancement and expansion module 120, a vibration source data separation model training module 130 and a model testing and verification module 140, wherein:

[0039] The data fusion and assembly module 110 is used to fuse the measured micro-vibration data containing interference noise generated by the construction of a new plant in the microelectronics industry with the simulated micro-vibration data without interference noise, and assemble the single-vibration source micro-vibration data in the fused micro-vibration data to form multi-vibration source micro-vibration data.

[0040] The data enhancement and expansion module 120 enhances and expands the single-source micro-vibration data and multi-source micro-vibration data of the data fusion and assembly module 110 through an adversarial generation algorithm.

[0041] The vibration source data separation model training module 130 uses the single-source micro-vibration data and multi-source micro-vibration data of the data fusion and assembly module 110; the single-source micro-vibration data and multi-source micro-vibration data enhanced and expanded by the data enhancement and expansion module 120 as test sample data, separates the multi-source micro-vibration data into single-source micro-vibration data through the vibration source data separation model for model training, and obtains the trained vibration source data separation model.

[0042] The model testing and verification module 140 tests and verifies the trained vibration source data separation model.

[0043] The machine learning-based micro-vibration source detection and separation method and system 100 provided in the embodiment of the present invention fuses the measured micro-vibration data generated during the construction of a new microelectronics plant in a microelectronics industry cluster area or near the site of an existing microelectronics enterprise with the simulated micro-vibration data, and enhances and expands the fused single-source micro-vibration data and the multi-source micro-vibration data formed by their assembly as training data for a source data separation model, and separates the multi-source micro-vibration data into single-source micro-vibration data to obtain a trained source data separation model, thereby increasing the data volume and production quality of the training data and improving the output accuracy of the trained source data separation model. After testing and verification, the trained source data separation model can separate the multi-source micro-vibration data generated during the construction of a new microelectronics plant in a microelectronics industry cluster area or near the site of an existing microelectronics enterprise into single-source micro-vibration data, thereby overcoming technical barriers.

[0044] The present invention is not limited to the above-mentioned specific implementation modes. Obviously, the above-mentioned embodiments are only some embodiments of the embodiments of the present invention, but not all embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of the present invention. Those skilled in the art can make other levels of modifications and changes to the present invention. In this way, if these modifications and changes of the present invention fall within the scope of the claims of the present invention, the present invention is also intended to include these changes and changes.

Claims

1. A method for detecting and separating micro-vibration sources based on machine learning, characterized in that: include: The measured micro-vibration data containing interference noise generated by the construction of a new plant in the microelectronics industry is integrated with the simulated micro-vibration data without interference noise, and the single-vibration source micro-vibration data in the integrated micro-vibration data are assembled to form multi-vibration source micro-vibration data; Enhance and expand single-source micro-vibration data and multi-source micro-vibration data through adversarial generation algorithm; Single-source micro-vibration data and its enhanced and expanded single-source micro-vibration data, multi-source micro-vibration data and its enhanced and expanded multi-source micro-vibration data are used as training and testing sample data, and a vibration source data separation model is trained through the multi-source micro-vibration data and the single-source micro-vibration data to obtain a trained vibration source data separation model; The trained vibration source data separation model is tested and verified.

2. The method for detecting and separating micro-vibration sources based on machine learning according to claim 1, characterized in that: The vibration source data separation model is composed of three deep neural network models, namely the time domain model, the frequency domain model and the time-frequency analysis model. The loss function of the vibration source data separation model is composed of the weighted average of the loss functions of the time domain model, the frequency domain model and the time-frequency analysis model. During the model training process, the goal of minimizing the loss function value is achieved by continuously adjusting the parameters in the time domain model, the frequency domain model and the time-frequency analysis model. When the loss function value is lower than the set threshold, it is considered that the trained vibration source data separation model can separate the multi-vibration source micro-vibration data with various characteristics into single-vibration source micro-vibration data. is the final loss function, is the loss function of the time domain model, is the loss function of the frequency domain model, is the loss function of the time-frequency analysis model, y is the real single-source micro-vibration data in the data set, This is the single source micro-vibration data predicted by the vibration source data separation model.

3. The method for detecting and separating micro-vibration sources based on machine learning according to claim 2, characterized in that: The trained vibration source data separation model is tested and verified from three aspects: time domain, frequency domain and time-frequency analysis. The prediction accuracy and generalization ability of the vibration source data separation model trained with test sample data are ensured through cross-validation in three aspects.

4. The method for detecting and separating micro-vibration sources based on machine learning according to claim 1, characterized in that: The vibration source data separation model is a convolutional neural network, a recurrent neural network or a Transformer network.

5. A micro-vibration source detection and separation system based on machine learning, characterized in that: include: The data fusion and assembly module is used to fuse the measured micro-vibration data containing interference noise generated by the construction of a new plant in the microelectronics industry with the simulated micro-vibration data without interference noise, and assemble the single-vibration source micro-vibration data in the fused micro-vibration data to form multi-vibration source micro-vibration data; The data enhancement and expansion module enhances and expands the single-source micro-vibration data and multi-source micro-vibration data of the data fusion and assembly module through the adversarial generation algorithm; The vibration source data separation model training module combines the single-source micro-vibration data and multi-source micro-vibration data of the data fusion and assembly module; The data enhancement and expansion module enhances and expands the single-source micro-vibration data and the multi-source micro-vibration data. The single-source micro-vibration data and the multi-source micro-vibration data are used as training and testing sample data. The vibration source data separation model is trained through the multi-source micro-vibration data and the single-source micro-vibration data to obtain the trained vibration source data separation model. The model testing and verification module tests and verifies the trained vibration source data separation model.

6. The micro-vibration source detection and separation system based on machine learning according to claim 5 is characterized in that: The vibration source data separation model is composed of three deep neural network models, namely the time domain model, the frequency domain model and the time-frequency analysis model. The loss function of the vibration source data separation model is composed of the weighted average of the loss functions of the time domain model, the frequency domain model and the time-frequency analysis model. During the model training process, the goal of minimizing the loss function value is achieved by continuously adjusting the parameters in the time domain model, the frequency domain model and the time-frequency analysis model. When the loss function value is lower than the set threshold, it is considered that the trained vibration source data separation model can separate the multi-vibration source micro-vibration data with various characteristics into single-vibration source micro-vibration data. is the final loss function, is the loss function of the time domain model, is the loss function of the frequency domain model, is the loss function of the time-frequency analysis model, y is the real single-source micro-vibration data in the data set, This is the single source micro-vibration data predicted by the vibration source data separation model.

7. The micro-vibration source detection and separation system based on machine learning according to claim 6 is characterized in that: In the model testing and verification module, the trained vibration source data separation model is tested and verified from three aspects: time domain, frequency domain and time-frequency analysis. The prediction accuracy and generalization ability of the vibration source data separation model trained with test sample data are ensured through cross-validation in three aspects.

8. The micro-vibration source detection and separation system based on machine learning according to claim 5, characterized in that: The vibration source data separation model is a convolutional neural network, a recurrent neural network or a Transformer network.

Images