A reactor mechanical defect early warning method and system based on deep learning

Abstract

The present invention discloses a method and system for warning of mechanical defects of reactors based on deep learning. The method comprises: extracting the original sound signal and original vibration signal of the reactor; filtering out interference signals in the original sound signal and the original vibration signal, and determining an estimated sound source signal and an estimated vibration source signal; calculating the cepstral coefficients of the estimated sound source signal and the estimated vibration source signal; and performing deep learning on the cepstral coefficients of the sound signal and the vibration signal based on the SRU neural network, and issuing a warning for mechanical defects of the reactor. Thus, after filtering the interference signal and reducing the time-frequency spectrum, a data-driven warning model for mechanical defects of the reactor is established through a deep learning algorithm, accurately grasping the operating status of the reactor and improving the safe operation level of the reactor.

Description

Technical Field

[0001] The present invention relates to the technical field of reactor defect early warning technology, and more specifically, to a reactor mechanical defect early warning method and system based on deep learning. Background Art

[0002] Conventional reactor mechanical defect warning mainly adopts a defect diagnosis method based on signal analysis. Signal analysis theory is used to obtain multiple eigenvectors at a deeper level in the system time domain and frequency domain. The relationship between these eigenvectors and system defects is used to warn of defects. However, it is very difficult to establish an analytical model of characteristic quantities and defects. Summary of the Invention

[0003] According to the present invention, a method and system for warning mechanical defects of reactors based on deep learning are provided to solve the technical problem in the prior art that the defect diagnosis method based on signal analysis needs to obtain multiple feature vectors, but it is very difficult to establish an analytical model of feature quantities and defects.

[0004] According to a first aspect of the present invention, a method for early warning of mechanical defects of a reactor based on deep learning is provided, comprising:

[0005] Extracting the original sound signal and original vibration signal of the reactor;

[0006] filtering out interference signals in the original sound signal and the original vibration signal, and determining an estimated sound source signal and an estimated vibration source signal;

[0007] Calculating cepstral coefficients of the estimated sound source signal and the estimated vibration source signal;

[0008] Based on the SRU neural network, deep learning is performed on the cepstral coefficients of sound signals and vibration signals, and early warning of mechanical defects in reactors is issued.

[0009] Optionally, filtering out interference signals in the original sound signal and the original vibration signal to determine the estimated sound source signal and the estimated vibration source signal includes:

[0010] Decomposing the original sound signal and the original vibration signal by using wavelet packets to obtain sound subband signals and vibration subband signals;

[0011] The mutual information value of the sound sub-band signal and the vibration sub-band signal is calculated.

[0012] Optionally, filtering out interference signals in the original sound signal and the original vibration signal to determine the estimated sound source signal and the estimated vibration source signal further includes:

[0013] Selecting a subspace signal from the mutual information value, reconstructing the subspace signal, and determining a reconstructed signal;

[0014] Perform fastICA decomposition on the reconstructed signal to obtain a separation matrix;

[0015] Blind source separation is performed on the original sound signal and the original vibration signal according to the separation matrix to determine an estimated sound source signal and an estimated vibration source signal.

[0016] Optionally, calculating cepstral coefficients of the estimated sound source signal and the estimated vibration source signal includes:

[0017] framing the estimated sound source signal and the estimated vibration source signal to determine frame data;

[0018] Applying a Hamming window to each frame of data to perform windowing processing, and determining each frame of data after windowing;

[0019] The frequency spectrum of each frame of data is obtained by performing a fast Fourier transform on each frame of windowed data, and the frequency spectrum of each frame of data is modulo-squared to obtain the power spectrum.

[0020] Optionally, calculating the cepstral coefficients of the estimated sound source signal and the estimated vibration source signal further includes:

[0021] smoothing the power spectrum using a triangular bandpass filter to filter harmonics;

[0022] The logarithm of the output of each triangle band filter group is calculated and discrete cosine transform is performed to obtain the cepstral coefficient of each frame of data.

[0023] According to another aspect of the present invention, a reactor mechanical defect early warning system based on deep learning is provided, comprising:

[0024] Extracting original signal module, used to extract original sound signal and original vibration signal of reactor;

[0025] an interference signal filtering module, configured to filter out interference signals in the original sound signal and the original vibration signal, and determine an estimated sound source signal and an estimated vibration source signal;

[0026] A cepstral coefficient calculation module, used to calculate the cepstral coefficients of the estimated sound source signal and the estimated vibration source signal;

[0027] The early warning mechanical defect module is used to perform deep learning on the cepstral coefficients of sound signals and vibration signals based on the SRU neural network, and to issue early warnings for mechanical defects of the reactor.

[0028] Optionally, the interference signal filtering module includes:

[0029] The original signal decomposition submodule is used to decompose the original sound signal and the original vibration signal respectively using wavelet packets to obtain sound subband signals and vibration subband signals;

[0030] The mutual information value calculation submodule is used to calculate the mutual information value of the sound subband signal and the vibration subband signal.

[0031] Optionally, the interference signal filtering module further includes:

[0032] a determination and reconstruction signal submodule, configured to select a subspace signal from the mutual information value, reconstruct the subspace signal, and determine a reconstructed signal;

[0033] A separation matrix submodule is used to perform fastICA decomposition on the reconstructed signal to obtain a separation matrix;

[0034] The estimated signal determination submodule is used to perform blind source separation on the original sound signal and the original vibration signal according to the separation matrix to determine an estimated sound source signal and an estimated vibration source signal.

[0035] Optionally, a module for calculating cepstral coefficients includes:

[0036] A framing submodule, configured to frame the estimated sound source signal and the estimated vibration source signal, and determine frame data;

[0037] A windowing submodule, configured to perform windowing processing on each frame of data by applying a Hamming window to determine each frame of data after windowing;

[0038] The power spectrum submodule is used to perform fast Fourier transform on each frame of windowed data to obtain the spectrum of each frame of data, and to square the spectrum of each frame of data to obtain the power spectrum.

[0039] Optionally, the cepstral coefficient calculation module further includes:

[0040] a power spectrum smoothing submodule, configured to smooth the power spectrum and filter harmonics using a triangular bandpass filter;

[0041] The cepstral coefficient submodule is used to obtain the logarithm of the output of each triangle band filter group, perform discrete cosine transform, and obtain the cepstral coefficient of each frame data.

[0042] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and the computer program is used to execute the method described in any of the above embodiments of the present invention.

[0043] Therefore, there is no need to understand the physical mechanism of the mechanical defects of the reactor, nor is there any need to establish an analytical model of the relationship between the sound signal and vibration signal characteristics and the mechanical defects of the reactor. After interference signal filtering and time-spectrum dimensionality reduction, a data-driven reactor mechanical defect early warning model is established through a deep learning algorithm to accurately grasp the operating status of the reactor and improve the safe operation level of the reactor. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] A more complete understanding of exemplary embodiments of the present invention may be obtained by referring to the following drawings:

[0045] Figure 1 Schematic diagram of a process flow of a reactor mechanical defect early warning method based on deep learning according to this embodiment;

[0046] Figure 2 Schematic diagram of the process of the reactor mechanical defect early warning method according to this embodiment;

[0047] Figure 3 Schematic diagram of a deep learning-based reactor mechanical defect warning system according to this embodiment. DETAILED DESCRIPTION

[0048] Exemplary embodiments of the present invention will now be described with reference to the accompanying drawings. However, the present invention may be embodied in many different forms and is not limited to the embodiments described herein. These embodiments are provided to provide a thorough and complete disclosure of the present invention and to fully convey the scope of the present invention to those skilled in the art. The terminology used in the exemplary embodiments shown in the accompanying drawings is not intended to limit the present invention. In the accompanying drawings, identical elements are denoted by the same reference numerals.

[0049] Unless otherwise specified, the terms used herein (including technical terms) have the meanings commonly understood by those skilled in the art. In addition, it is understood that terms defined in commonly used dictionaries should be understood to have the same meanings as those in the context of the relevant fields, and should not be understood as idealized or overly formal meanings.

[0050] According to a first aspect of the present invention, a method 100 is provided, referring to Figure 1 As shown, the method 100 includes:

[0051] S101: extracting the original sound signal and the original vibration signal of the reactor;

[0052] S102: filtering out interference signals in the original sound signal and the original vibration signal, and determining an estimated sound source signal and an estimated vibration source signal;

[0053] S103: Calculating cepstral coefficients of the estimated sound source signal and the estimated vibration source signal;

[0054] S104: Based on the SRU neural network, deep learning is performed on the cepstral coefficients of the sound signal and the vibration signal, and an early warning is issued for mechanical defects of the reactor.

[0055] Specifically, in step S101, the reactor sound signal and vibration signal are extracted;

[0056] In step S102, the SDICA blind source separation algorithm is applied to filter out interference signals in the original sound signal and vibration signal;

[0057] In step S103, the Mel time spectrum is applied to calculate the Mel cepstral coefficients of the sound signal and the vibration signal, and perform dimensionality reduction processing;

[0058] In step S104, the SRU neural network is applied to perform deep learning on the cepstral coefficients of the sound signal and the vibration signal to achieve early warning of mechanical defects of the reactor.

[0059] The steps of the SDICA blind source separation algorithm include:

[0060] Use wavelet packets to decompose the reactor sound signal and vibration signal to obtain sub-band signals;

[0061] Calculate the mutual information value (MI) of each subspace of the signal after wavelet decomposition;

[0062] Select one or several subspaces with the smallest MI value, which are relatively independent subspaces, and reconstruct these subspaces;

[0063] The reconstructed signal is decomposed by fastICA to obtain the separation matrix;

[0064] The separation matrix is ​​used to perform blind source separation on the original sound signal and the vibration signal.

[0065] The Mel-time spectrum calculation process includes:

[0066] The signal after blind source separation is framed, with a 1-second signal as a sample. Each frame has a length of 0.04 seconds and a frame shift of 0.01 seconds.

[0067] Apply Hamming window to each frame of data for windowing processing;

[0068] Perform fast Fourier transform on each frame of windowed data to obtain the spectrum of each frame of data, and square the spectrum to obtain the power spectrum;

[0069] A 50Hz frequency-multiplied triangular bandpass filter is applied to smooth the power spectrum and filter out harmonics;

[0070] The logarithm of each filter bank output is calculated, and then discrete cosine transform (DCT) is performed to obtain the frequency cepstral coefficient (FMCC) of each frame data.

[0071] refer to Figure 2 As shown in the figure, the extracted sound signal and vibration signal of the reactor are decomposed by wavelet packets to extract the self-contained signal, and then the mutual information value is calculated by extracting the self-contained signal. One or more subspaces with the smallest mutual information value are selected to reconstruct the signal. The fastICA decomposition algorithm is applied to the reconstructed signal to calculate the separation matrix, and then the sound and vibration signals are calculated to obtain the estimated source signal; the estimated source signal is framed, windowed and fast Fourier transformed to obtain the power spectrum of each frame of data, and then filtered by a 50Hz multiplied triangle bandpass filter, and then the logarithm is calculated and discrete cosine transform is performed to obtain the cepstral coefficients of each frame of data; the cepstral coefficients of the sound and vibration signals are used as the input of the SRU neural network to provide early warning of mechanical defects of the reactor.

[0072] Optionally, filtering out interference signals in the original sound signal and the original vibration signal to determine the estimated sound source signal and the estimated vibration source signal includes:

[0073] Decomposing the original sound signal and the original vibration signal by using wavelet packets to obtain sound subband signals and vibration subband signals;

[0074] The mutual information value of the sound sub-band signal and the vibration sub-band signal is calculated.

[0075] Optionally, filtering out interference signals in the original sound signal and the original vibration signal to determine the estimated sound source signal and the estimated vibration source signal further includes:

[0076] Selecting a subspace signal from the mutual information value, reconstructing the subspace signal, and determining a reconstructed signal;

[0077] Perform fastICA decomposition on the reconstructed signal to obtain a separation matrix;

[0078] Blind source separation is performed on the original sound signal and the original vibration signal according to the separation matrix to determine an estimated sound source signal and an estimated vibration source signal.

[0079] Optionally, calculating cepstral coefficients of the estimated sound source signal and the estimated vibration source signal includes:

[0080] framing the estimated sound source signal and the estimated vibration source signal to determine frame data;

[0081] Applying a Hamming window to each frame of data to perform windowing processing, and determining each frame of data after windowing;

[0082] The frequency spectrum of each frame of data is obtained by performing a fast Fourier transform on each frame of windowed data, and the frequency spectrum of each frame of data is modulo-squared to obtain the power spectrum.

[0083] Optionally, calculating the cepstral coefficients of the estimated sound source signal and the estimated vibration source signal further includes:

[0084] smoothing the power spectrum using a triangular bandpass filter to filter harmonics;

[0085] The logarithm of the output of each triangle band filter group is calculated and discrete cosine transform is performed to obtain the cepstral coefficient of each frame of data.

[0086] Therefore, there is no need to understand the physical mechanism of the mechanical defects of the reactor, nor is there any need to establish an analytical model of the relationship between the sound signal and vibration signal characteristics and the mechanical defects of the reactor. After interference signal filtering and time-spectrum dimensionality reduction, a data-driven reactor mechanical defect early warning model is established through a deep learning algorithm to accurately grasp the operating status of the reactor and improve the safe operation level of the reactor.

[0087] According to another aspect of the present invention, a reactor mechanical defect early warning system 300 based on deep learning is also provided. Figure 3 As shown, the system 300 includes:

[0088] The original signal extraction module 310 is used to extract the original sound signal and the original vibration signal of the reactor;

[0089] an interference signal filtering module 320, configured to filter out interference signals from the original sound signal and the original vibration signal, and determine an estimated sound source signal and an estimated vibration source signal;

[0090] A cepstral coefficient calculation module 330 is used to calculate the cepstral coefficients of the estimated sound source signal and the estimated vibration source signal;

[0091] The mechanical defect warning module 340 is used to perform deep learning on the cepstral coefficients of the sound signal and the vibration signal based on the SRU neural network, and to issue an early warning for the mechanical defects of the reactor.

[0092] Optionally, the interference signal filtering module 320 includes:

[0093] The original signal decomposition submodule is used to decompose the original sound signal and the original vibration signal respectively using wavelet packets to obtain sound subband signals and vibration subband signals;

[0094] The mutual information value calculation submodule is used to calculate the mutual information value of the sound subband signal and the vibration subband signal.

[0095] Optionally, the interference signal filtering module 320 further includes:

[0096] a determination and reconstruction signal submodule, configured to select a subspace signal from the mutual information value, reconstruct the subspace signal, and determine a reconstructed signal;

[0097] A separation matrix submodule is used to perform fastICA decomposition on the reconstructed signal to obtain a separation matrix;

[0098] The estimated signal determination submodule is used to perform blind source separation on the original sound signal and the original vibration signal according to the separation matrix to determine an estimated sound source signal and an estimated vibration source signal.

[0099] Optionally, the cepstral coefficient calculation module 340 includes:

[0100] A framing submodule, configured to frame the estimated sound source signal and the estimated vibration source signal, and determine frame data;

[0101] A windowing submodule, configured to perform windowing processing on each frame of data by applying a Hamming window to determine each frame of data after windowing;

[0102] The power spectrum submodule is used to perform fast Fourier transform on each frame of windowed data to obtain the spectrum of each frame of data, and to square the spectrum of each frame of data to obtain the power spectrum.

[0103] Optionally, the cepstral coefficient calculation module 340 further includes:

[0104] a power spectrum smoothing submodule, configured to smooth the power spectrum and filter harmonics using a triangular bandpass filter;

[0105] The cepstral coefficient submodule is used to obtain the logarithm of the output of each triangle band filter group, perform discrete cosine transform, and obtain the cepstral coefficient of each frame data.

[0106] A reactor mechanical defect early warning system 300 based on deep learning in an embodiment of the present invention corresponds to a reactor mechanical defect early warning method 100 based on deep learning in another embodiment of the present invention, and will not be repeated here.

[0107] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and the computer program is used to execute the method described in any of the above embodiments of the present invention.

[0108] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiment of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal translation scripting language JavaScript, etc.

[0109] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0110] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0111] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0112] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application.

[0113] Obviously, those skilled in the art may make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is intended to include these modifications and variations.

Claims

1. A reactor mechanical defect early warning method based on deep learning, characterized in that: include: Extracting the original sound signal and original vibration signal of the reactor; filtering out interference signals in the original sound signal and the original vibration signal, and determining an estimated sound source signal and an estimated vibration source signal; Calculating cepstral coefficients of the estimated sound source signal and the estimated vibration source signal; Based on the SRU neural network, deep learning is performed on the cepstral coefficients of sound and vibration signals, and early warning of mechanical defects in reactors is provided; Filtering out interference signals in the original sound signal and the original vibration signal to determine an estimated sound source signal and an estimated vibration source signal, comprising: Decomposing the original sound signal and the original vibration signal by using wavelet packets to obtain sound sub-band signals and vibration sub-band signals; Calculating the mutual information value of the sound sub-band signal and the vibration sub-band signal; Filtering out interference signals in the original sound signal and the original vibration signal to determine an estimated sound source signal and an estimated vibration source signal, further comprising: Selecting a subspace signal from the mutual information value, reconstructing the subspace signal, and determining a reconstructed signal; Perform fastICA decomposition on the reconstructed signal to obtain a separation matrix; Performing blind source separation on the original sound signal and the original vibration signal according to the separation matrix to determine an estimated sound source signal and an estimated vibration source signal; Calculating cepstral coefficients of the estimated sound source signal and the estimated vibration source signal, comprising: framing the estimated sound source signal and the estimated vibration source signal to determine frame data; Applying a Hamming window to each frame of data to perform windowing processing, and determining each frame of data after windowing; Perform fast Fourier transform on each frame of windowed data to obtain the spectrum of each frame of data, and square the spectrum of each frame of data to obtain the power spectrum; Calculating the cepstral coefficients of the estimated sound source signal and the estimated vibration source signal further includes: smoothing the power spectrum using a triangular bandpass filter to filter harmonics; The logarithm of the output of each triangle band filter group is calculated and discrete cosine transform is performed to obtain the cepstral coefficient of each frame of data.

2. A reactor mechanical defect early warning system based on deep learning, characterized in that: include: Extracting original signal module, used to extract original sound signal and original vibration signal of reactor; an interference signal filtering module, configured to filter out interference signals in the original sound signal and the original vibration signal, and determine an estimated sound source signal and an estimated vibration source signal; A cepstral coefficient calculation module, used to calculate the cepstral coefficients of the estimated sound source signal and the estimated vibration source signal; The mechanical defect warning module is used to perform deep learning on the cepstral coefficients of sound and vibration signals based on the SRU neural network and issue early warnings for mechanical defects in reactors. Interference signal filtering module, including: The original signal decomposition submodule is used to decompose the original sound signal and the original vibration signal respectively using wavelet packets to obtain sound subband signals and vibration subband signals; A mutual information value calculation submodule, configured to calculate the mutual information value of the sound subband signal and the vibration subband signal; The interference signal filtering module also includes: a determination and reconstruction signal submodule, configured to select a subspace signal from the mutual information value, reconstruct the subspace signal, and determine a reconstructed signal; A separation matrix submodule is used to perform fastICA decomposition on the reconstructed signal to obtain a separation matrix; a determination estimation signal submodule, configured to perform blind source separation on the original sound signal and the original vibration signal according to the separation matrix to determine an estimated sound source signal and an estimated vibration source signal; Calculate the cepstral coefficient module, including: A framing submodule, configured to frame the estimated sound source signal and the estimated vibration source signal, and determine frame data; A windowing submodule is used to apply a Hamming window to each frame of data for windowing processing, and determine each frame of data after windowing; A power spectrum submodule is used to perform a fast Fourier transform on each frame of windowed data to obtain the spectrum of each frame of data, and to square the spectrum of each frame of data to obtain the power spectrum; The module for calculating cepstral coefficients also includes: a power spectrum smoothing submodule, configured to smooth the power spectrum and filter harmonics using a triangular bandpass filter; The cepstral coefficient submodule is used to obtain the logarithm of the output of each triangle band filter group, perform discrete cosine transform, and obtain the cepstral coefficient of each frame data.

Images