An ICEEMDAN-STFT-based voiceprint map-based state evaluation system and method for a hydraulic turbine unit

Abstract

The application discloses a kind of based on ICEEMDAN-STFT voiceprint map hydro-turbine unit state evaluation system and method, belong to digital processing technical field.System includes acoustic sensor, signal conditioning circuit, data acquisition unit, core processing unit, typical working condition voiceprint database, state evaluation unit and fault early warning terminal.Acoustic sensor acquires unit operation sound signal, after conditioning, acquisition, core processing unit uses ICEEMDAN decomposition, according to correlation coefficient and variance contribution rate Effective mode component is screened and reconstructs noise reduction, again by STFT generates voiceprint time-frequency map and feature matrix;State evaluation unit matches real-time features with typical working condition voiceprint database, identifies operating state and outputs fault early warning.The application can weaken complex background noise and modal aliasing influence, improve the discernibility of blade damage, bolt loosening and bearing fault.

Description

Technical Field

[0001] This invention belongs to the field of digital processing technology, specifically relating to a turbine unit condition assessment system and method based on ICEEMDAN-STFT acoustic signature spectrum. Background Technology

[0002] Hydropower units operate in environments characterized by high humidity, strong vibration, and complex hydraulic disturbances. Their operating acoustic signals are non-stationary, non-linear, and have strong background noise. The impact or modulation components generated in the early stages of blade damage, loose connecting bolts, and bearing failures are relatively weak and easily masked by water flow noise, shaft fundamental frequency, and structural resonance components, making it difficult to identify them in a timely manner using only conventional time-domain indicators.

[0003] Existing empirical mode decomposition (EMD) and ensemble EMD methods are prone to mode aliasing or residual noise when processing strong noise signals, and direct time-frequency transformation may result in unclear energy distribution of weak faults. Therefore, a condition assessment system is needed that can adaptively decompose the original acoustic signal, effectively screen modes, and perform high-resolution time-frequency mapping, and can be matched with acoustic patterns under typical operating conditions. Summary of the Invention

[0004] The purpose of this invention is to provide a turbine unit condition assessment system and method based on ICEEMDAN-STFT acoustic signature spectrum, so as to reduce the impact of complex background noise and mode aliasing on the extraction of weak fault features, and realize turbine unit operating status identification and fault early warning.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows: a turbine generator condition assessment system based on acoustic signature spectrum, comprising a turbine generator, acoustic sensors, signal conditioning circuits, a data acquisition unit, a core processing unit, a typical operating condition acoustic signature database, a condition assessment unit, and a fault early warning terminal;

[0006] The acoustic sensor is installed at the monitoring location of the turbine unit and is used to collect the raw acoustic signals generated when the turbine unit is running.

[0007] The signal conditioning circuit is connected to the acoustic sensor and is used to perform anti-aliasing filtering, amplification, and impedance matching on the original acoustic signal; the data acquisition unit is connected to the signal conditioning circuit and is used to convert the conditioned analog acoustic signal into a digital acoustic signal.

[0008] The core processing unit is connected to the data acquisition unit and is used to perform noise separation, effective mode screening, signal reconstruction and voiceprint feature matrix generation on the digital acoustic signal; the typical operating condition voiceprint database is used to store standard voiceprint features corresponding to different operating conditions of the turbine unit.

[0009] The status assessment unit is connected to the core processing unit and the typical operating condition voiceprint database, respectively, and is used to match the real-time voiceprint feature matrix with the standard voiceprint features and identify the operating status; the fault early warning terminal is connected to the status assessment unit and is used to output the fault location, fault type and early warning information when the identification result is abnormal.

[0010] Preferably, the acoustic sensor is a broadband acoustic sensor; the signal conditioning circuit includes an anti-aliasing filter circuit, an amplification circuit, and an impedance matching circuit; the sampling frequency of the data acquisition unit is not less than twice the highest fault characteristic frequency to be identified.

[0011] Preferably, the core processing unit includes a signal denoising preprocessing subunit and a speaker graph generation subunit; the signal denoising preprocessing subunit uses an improved fully adaptive noise set empirical mode decomposition algorithm to obtain multiple intrinsic mode components, and selects effective mode components based on the correlation coefficient and variance contribution rate of each intrinsic mode component with the original acoustic signal; the speaker graph generation subunit uses short-time Fourier transform to convert the reconstructed signal into a speaker graph feature matrix containing time, frequency and energy information.

[0012] Preferably, the typical operating condition acoustic signature database includes at least standard acoustic signature features for normal operating conditions, blade damage operating conditions, bolt loosening operating conditions, and bearing failure operating conditions; the status assessment unit determines the operating status based on the matching results of real-time acoustic signature features and standard acoustic signature features.

[0013] The method for performing condition assessment of a hydro-turbine unit using the aforementioned system includes the following steps:

[0014] Step S1: Acoustic sensors are used to collect the raw acoustic signals of the turbine unit, and digital acoustic signals are obtained through signal conditioning circuits and data acquisition units.

[0015] Step S2 involves sequentially executing the following steps: original acoustic signal input, decomposition parameter initialization, adding adaptive white noise, calculating residuals and intrinsic modal components, decomposition termination judgment, calculating correlation coefficient and variance contribution rate, filtering effective modal components, and superposition reconstruction, to obtain the denoised reconstructed signal.

[0016] Step S3: The following steps are executed sequentially: reconstructed signal input step, framing and windowing step, short-time Fourier transform step, voiceprint power spectral density calculation step, and voiceprint feature matrix generation step, to obtain the real-time voiceprint feature matrix.

[0017] Step S4: Call the typical operating condition voiceprint database, and perform the feature matching and correlation analysis steps and the operating status identification steps;

[0018] Step S5: Perform the anomaly judgment step. If the identification result is normal, perform the normal status recording step. If the identification result is abnormal, perform the fault location and early warning step.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] (1) By adding adaptive white noise step by step and calculating residuals and intrinsic mode components, the mode aliasing in traditional empirical mode decomposition can be reduced; by combining correlation coefficient and variance contribution rate to screen effective modes, environmental noise and trend terms can be suppressed while retaining the main fault components.

[0021] (2) By framing, windowing and short-time Fourier transform, the one-dimensional reconstructed acoustic signal is converted into an acoustic feature matrix containing time, frequency and energy information, which is beneficial to present the energy distribution of blade damage, bolt loosening and bearing failure in different frequency bands.

[0022] (3) By matching real-time voiceprint features with a typical operating condition voiceprint database, a complete processing link can be formed from signal acquisition, noise separation, voiceprint generation to condition assessment and fault early warning, which facilitates remote monitoring and condition maintenance of the turbine unit. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments are briefly described below.

[0024] Figure 1 This is a structural block diagram of the turbine unit condition assessment system of the present invention;

[0025] Figure 2 This is a flowchart of the signal denoising preprocessing of the present invention;

[0026] Figure 3 This is a flowchart of the voiceprint spectrum generation and status assessment process of the present invention.

[0027] Among them: 101, turbine unit; 102, acoustic sensor; 103, signal conditioning circuit; 104, data acquisition unit; 105, core processing unit; 106, typical operating condition acoustic database; 107, condition assessment unit; 108, fault early warning terminal. Detailed Implementation

[0028] The technical solution of the present invention will be further described below with reference to the accompanying drawings. It should be understood that the embodiments described are only for illustrating the present invention and are not intended to limit the scope of protection of the present invention.

[0029] like Figure 1As shown, the turbine unit 101 is the object being monitored. Acoustic sensors 102 are positioned at monitoring locations such as the generator side, bearing housing, or outer wall of the volute of the turbine unit 101, and are used to convert changes in operating sound pressure into analog electrical signals. The signal conditioning circuit 103 receives the analog electrical signals output from the acoustic sensors 102 and performs anti-aliasing filtering, amplification, and impedance matching on them. The data acquisition unit 104 converts the conditioned analog electrical signals into digital acoustic signals according to a preset sampling frequency and transmits the digital acoustic signals to the core processing unit 105.

[0030] The core processing unit 105 is used to execute Figure 2 The signal denoising preprocessing shown and Figure 3 The process of generating the acoustic signature map is shown. The typical operating condition acoustic signature database 106 stores standard acoustic signature features corresponding to normal operating conditions, blade damage conditions, bolt loosening conditions, and bearing failure conditions. The condition assessment unit 107 receives the real-time acoustic signature feature matrix output by the core processing unit 105 and calls the typical operating condition acoustic signature database 106 for feature matching and condition recognition. The fault early warning terminal 108 receives the recognition results from the condition assessment unit 107 and displays the fault location, fault type, abnormal frequency band, and early warning level when the recognition result is abnormal.

[0031] like Figure 2 As shown, in the original acoustic signal input step, the digital acoustic signal output by the data acquisition unit 104 is input to the core processing unit 105; in the decomposition parameter initialization step, the noise amplitude coefficient, set number of sets, maximum decomposition order, and stopping condition are set; in the adaptive white noise addition step, white noise is added according to the current residual scale; in the residual and intrinsic modal component calculation step, the noisy signal is decomposed by set, and the current residual and intrinsic modal components are calculated; in the decomposition termination judgment step, it is determined whether the current residual meets the stopping condition. If it does not meet the condition, the process returns to the adaptive white noise addition step to continue decomposition. If it meets the condition, the process proceeds to the subsequent filtering process.

[0032] In the steps of calculating the correlation coefficient and variance contribution rate, the correlation coefficient between each intrinsic mode component and the original acoustic signal and the contribution rate of each intrinsic mode component to the variance of the original acoustic signal are calculated respectively. In the step of screening effective mode components, intrinsic mode components that meet the preset conditions for correlation coefficient and variance contribution rate and contain the target fault frequency band energy are retained, while high-frequency random noise components and low-frequency trend components are removed. In the step of superimposing and reconstructing the noise-reduced signal, the retained effective mode components are superimposed to obtain the reconstructed noise-reduced signal.

[0033] like Figure 3As shown, in the reconstructed signal input step, the reconstructed signal output from the previous step is sent to the acoustic signature generation process; in the framing and windowing step, the frame length, frame shift, and window function are set according to the sampling frequency and the rated rotation frequency of the turbine unit; in the short-time Fourier transform step, the windowed signal of each frame is subjected to discrete Fourier transform to obtain a two-dimensional time-frequency complex matrix; in the acoustic signature power spectral density calculation step, the amplitude square operation is performed on the time-frequency complex matrix; in the acoustic signature feature matrix generation step, the time, frequency, and energy information are normalized to form a real-time acoustic signature feature matrix.

[0034] The typical operating condition voiceprint database provides standard voiceprint features to the feature matching and association analysis steps. In the feature matching and association analysis steps, the similarity between real-time voiceprint features and each standard voiceprint feature is calculated, and the frequency band of abnormal energy and its variation over time are analyzed. In the operating status identification step, the current operating status is determined based on the highest matching degree and preset criteria. In the anomaly judgment step, it is determined whether the identification result is abnormal. If it is normal, the normal status recording step is executed to store the current voiceprint features and operating parameters. If it is abnormal, the fault location and early warning step is executed to send the fault location, fault type, and early warning information to the fault early warning terminal 108.

[0035] During system operation, acoustic sensors 102 continuously or at preset intervals collect the operating acoustic signals of the turbine unit 101. The condition assessment unit 107 can write the current identification results, the corresponding acoustic signature feature matrix, and the acquisition time into the typical operating condition acoustic signature database 106 for historical trend comparison. For blade damage, the focus can be on analyzing the blade's passing frequency and its sideband energy; for loose bolts, the focus can be on analyzing low-frequency broadband energy and modulation components; for bearing failures, the focus can be on analyzing mid-to-high frequency periodic impact components.

[0036] The monitoring location, sampling parameters, screening thresholds, and matching criteria described in the above embodiments can be adjusted according to different types of turbine units and on-site noise conditions; any equivalent substitutions or modifications made without departing from the technical concept of this invention shall fall within the protection scope of this invention.

Claims

1. A turbine generator condition assessment system based on ICEEMDAN-STFT acoustic signature spectrum, characterized in that, It includes a turbine unit (101), an acoustic sensor (102), a signal conditioning circuit (103), a data acquisition unit (104), a core processing unit (105), a typical operating condition acoustic database (106), a condition assessment unit (107), and a fault early warning terminal (108). The acoustic sensor (102) is installed at the monitoring position of the turbine unit (101) and is used to collect the original acoustic signals generated when the turbine unit (101) is running. The signal conditioning circuit (103) is connected to the acoustic sensor (102) and is used to perform anti-aliasing filtering, amplification and impedance matching on the original acoustic signal; the data acquisition unit (104) is connected to the signal conditioning circuit (103) and is used to convert the conditioned analog acoustic signal into a digital acoustic signal. The core processing unit (105) is connected to the data acquisition unit (104) and is used to perform noise separation, effective mode screening, signal reconstruction and voiceprint feature matrix generation on the digital acoustic signal. The typical operating condition voiceprint database (106) is used to store standard voiceprint features corresponding to different operating conditions of the turbine unit; The status assessment unit (107) is connected to the core processing unit (105) and the typical operating condition voiceprint database (106) respectively, and is used to match the real-time voiceprint feature matrix with the standard voiceprint features and identify the operating status; the fault early warning terminal (108) is connected to the status assessment unit (107) and is used to output the fault location, fault type and early warning information when the identification result is abnormal.

2. The turbine unit condition assessment system based on ICEEMDAN-STFT acoustic signature spectrum according to claim 1, characterized in that, The acoustic sensor (102) is a broadband acoustic sensor; the signal conditioning circuit (103) includes an anti-aliasing filter circuit, an amplification circuit and an impedance matching circuit; the sampling frequency of the data acquisition unit (104) is not less than twice the highest fault characteristic frequency to be identified.

3. The turbine unit condition assessment system based on ICEEMDAN-STFT acoustic signature spectrum according to claim 1, characterized in that, The core processing unit (105) includes a signal denoising preprocessing subunit and a speaker graph generation subunit. The signal denoising preprocessing subunit uses an improved fully adaptive noise set empirical mode decomposition algorithm to obtain multiple intrinsic mode components, and selects effective mode components based on the correlation coefficient and variance contribution rate of each intrinsic mode component with the original acoustic signal. The speaker graph generation subunit uses short-time Fourier transform to convert the reconstructed signal into a speaker graph feature matrix containing time, frequency and energy information.

4. The turbine unit condition assessment system based on ICEEMDAN-STFT acoustic signature spectrum according to claim 1, characterized in that, The typical operating condition acoustic signature database (106) includes at least standard acoustic signature features for normal operating conditions, blade damage operating conditions, bolt loosening operating conditions, and bearing failure operating conditions; the status assessment unit (107) determines the operating status based on the matching results of real-time acoustic signature features and standard acoustic signature features.

5. A method for assessing the condition of a turbine unit using the ICEEMDAN-STFT acoustic signature system as described in any one of claims 1 to 4, characterized in that, Includes the following steps: Step S1: The original acoustic signal of the turbine unit (101) is collected by the acoustic sensor (102), and the digital acoustic signal is obtained by the signal conditioning circuit (103) and the data acquisition unit (104); Step S2 involves sequentially executing the following steps: original acoustic signal input, decomposition parameter initialization, adding adaptive white noise, calculating residuals and intrinsic modal components, decomposition termination judgment, calculating correlation coefficient and variance contribution rate, filtering effective modal components, and superposition reconstruction, to obtain the denoised reconstructed signal. Step S3: The following steps are executed sequentially: reconstructed signal input step, framing and windowing step, short-time Fourier transform step, voiceprint power spectral density calculation step, and voiceprint feature matrix generation step, to obtain the real-time voiceprint feature matrix. Step S4: Call the typical operating condition voiceprint database (106) and perform the feature matching and correlation analysis steps and the operating status identification steps; Step S5: Perform the anomaly judgment step. If the identification result is normal, perform the normal status recording step. If the identification result is abnormal, perform the fault location and early warning step.

Images