Thermal power plant pipeline abnormal vibration diagnosis method based on voiceprint recognition and related equipment

Abstract

The invention relates to the technical field of pipeline anomaly detection, and discloses a thermal power plant pipeline abnormal vibration diagnosis method based on voiceprint recognition and related equipment.The method comprises the steps that a pickup sensor is arranged at the monitoring position of a thermal power plant pipeline to collect acoustic signals, the acoustic signals are preprocessed, and preprocessed signals are obtained; performing multi-dimensional acoustic feature extraction and fusion according to the pre-processed signal to generate a voiceprint feature vector, establishing a voiceprint baseline library corresponding to different operation conditions of the pipeline, and performing dynamic updating; and outputting a pipeline abnormal vibration diagnosis result through working condition matching, voiceprint difference comparison and sound source positioning based on the voiceprint feature vector and the updated voiceprint baseline library. The technical problem that abnormal signals and background noise are difficult to distinguish through traditional acoustic monitoring in complex operation environments such as high temperature, high pressure and pump valve noise is solved.

Description

Technical Field

[0001] This invention relates to the field of pipeline anomaly detection technology, specifically to a method and related equipment for diagnosing abnormal vibrations in thermal power plant pipelines based on voiceprint recognition. Background Technology

[0002] The abnormal vibration diagnosis method for thermal power plant pipelines utilizes acoustic monitoring, signal processing, and feature recognition technologies to collect and analyze acoustic and vibration information generated during pipeline operation. Typically, sound pickup sensors or accelerometers are deployed at key locations along the pipeline to collect acoustic or vibration signals generated during operation in real time. These signals are then processed through filtering, feature extraction, and pattern recognition to obtain the pipeline's operational characteristics. By comparing the acoustic or vibration characteristics under different operating conditions, the pipeline's operational status can be monitored and diagnosed, assisting in the detection of potential anomalies during pipeline operation and providing technical support for the safe and stable operation of thermal power units.

[0003] In actual thermal power plant pipeline operation, a large amount of background noise is generated due to the flow of media inside the pipeline, the start-up of pumps and valves, and external environmental factors. Traditional acoustic monitoring methods are difficult to effectively distinguish these normal operating noises from abnormal pipeline vibration signals, resulting in low accuracy of anomaly diagnosis and the risk of false alarms or missed alarms. Summary of the Invention

[0004] In order to overcome the shortcomings of the existing technology, the purpose of this invention is to provide a method and related equipment for diagnosing abnormal vibrations in thermal power plant pipelines based on voiceprint recognition, so as to solve the technical problem that traditional acoustic monitoring is unable to distinguish abnormal signals from background noise in complex operating environments such as high temperature, high pressure and pump and valve noise.

[0005] This invention is achieved through the following technical solution: In a first aspect, the present invention provides a method for diagnosing abnormal vibrations in thermal power plant pipelines based on voiceprint recognition, comprising: Acoustic sensors are deployed at monitoring locations on pipelines in thermal power plants to collect acoustic signals. The acoustic signals are then preprocessed to obtain preprocessed signals. Based on the preprocessed signal, multi-dimensional acoustic features are extracted and fused to generate acoustic feature vectors. At the same time, an acoustic baseline library corresponding to different operating conditions of the pipeline is established and dynamically updated. Based on the aforementioned acoustic signature feature vector and the updated acoustic signature baseline library, the abnormal vibration diagnosis results of the pipeline are output through operating condition matching, acoustic signature differential comparison, and sound source localization.

[0006] Preferably, the arrangement of the sound pickup sensor includes: Based on the pipeline's geometric parameters and operating noise distribution, candidate sensor placement points are determined using sound field propagation simulation results. Select locations with high acoustic signature differences from the candidate locations to form a sensor layout scheme covering the main pipeline and branch pipelines; According to the layout plan, sound pickup sensors are installed on the surface of the pipe, and the spatial coordinates of each sensor are recorded for subsequent sound source localization.

[0007] Preferably, the preprocessing includes bandpass filtering and noise suppression processing, specifically including: The frequency range of the acquired acoustic signal is limited by a bandpass filter, retaining only the target frequency band signal within the set passband; A noise estimation model is constructed using a reference noise signal, and noise reduction is performed on the bandpass filtered signal based on this noise estimation model to obtain a preprocessed signal.

[0008] Preferred multi-dimensional acoustic feature extraction includes: The short-time energy, zero-crossing rate, and autocorrelation function values ​​of the preprocessed signal are calculated in the time domain to obtain the time-domain characteristic parameters; The preprocessed signal is subjected to Fast Fourier Transform to extract the amplitude spectrum and power spectrum, and the spectral centroid and spectral flatness are calculated to obtain the frequency domain characteristic parameters. Mel frequency cepstral analysis is performed on the preprocessed signal to calculate the cepstral coefficient sequence and obtain the acoustic cepstral characteristic parameters.

[0009] Preferably, multi-dimensional acoustic feature fusion includes: The time-domain feature parameters, frequency-domain feature parameters, and acoustic cepstral feature parameters are normalized according to a unified dimension to obtain a normalized feature set. Principal component analysis is performed on the normalized feature set to extract the eigenvectors of the principal components in order to reduce feature redundancy. The main component feature vectors are input into a convolutional neural network to extract deep feature representations and obtain deep acoustic feature vectors. The depth acoustic feature vector is fused with the main component feature vector to form the voiceprint feature vector.

[0010] Preferably, the establishment and dynamic updating of the voiceprint baseline library includes: Acoustic signals were collected under different steady-state operating conditions of the pipeline, and acoustic fingerprint sample sets were generated based on acoustic fingerprint feature vectors. Clustering calculations are performed on the voiceprint sample set for each working condition to obtain and store the voiceprint center vector for each working condition, thus forming a mapping relationship between the working condition and the baseline voiceprint feature vector. During operation, a new normal voiceprint feature vector is acquired and its similarity to the baseline voiceprint feature vector is calculated. When the similarity meets the preset threshold, the corresponding baseline voiceprint feature vector is updated and stored using a weighted average method.

[0011] Preferably, based on the voiceprint feature vector and the updated voiceprint baseline library, the output pipeline abnormal vibration diagnosis result is obtained through operating condition matching, voiceprint differential comparison and sound source localization. The operating condition matching includes collecting real-time pressure, temperature and flow data of the pipeline, inputting the operating condition classification model based on support vector machine, outputting the current operating condition identifier and calling the corresponding baseline voiceprint feature vector. The voiceprint differential comparison includes normalizing the real-time voiceprint feature vector and the baseline voiceprint feature vector, calculating the Euclidean distance dimension by dimension to generate the differential voiceprint vector, and outputting the differential voiceprint result after threshold discrimination. The sound source separation and localization includes performing independent component analysis and decomposition on the acoustic signal to screen out the target sound source components that match the differential acoustic signature results; calculating the time delay data of the target acoustic signature signal at each sensor through cross-correlation; combining the sensor spatial coordinate input with the sound source localization model based on the time difference of arrival algorithm; and outputting the spatial coordinates of the abnormal sound source and the diagnostic results.

[0012] Secondly, the present invention also provides a diagnostic system for abnormal vibration of pipelines in thermal power plants based on voiceprint recognition, comprising: The data acquisition and preprocessing module is used to deploy sound pickup sensors at the monitoring location of the pipeline in the thermal power plant to collect acoustic signals, and to preprocess the acoustic signals to obtain preprocessed signals. The feature vector fusion processing module is used to extract and fuse multi-dimensional acoustic features based on the preprocessed signal to generate acoustic feature vectors, and at the same time to establish and dynamically update the acoustic baseline library corresponding to different operating conditions of the pipeline. The abnormal vibration diagnosis and processing module is used to output the abnormal vibration diagnosis results of the pipeline based on the acoustic feature vector and the updated acoustic baseline library, through working condition matching, acoustic differential comparison and sound source localization.

[0013] Thirdly, the present invention also provides a mobile terminal, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method for diagnosing abnormal vibrations in thermal power plant pipelines based on voiceprint recognition.

[0014] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for diagnosing abnormal vibrations in thermal power plant pipelines based on voiceprint recognition.

[0015] Compared with the prior art, the present invention has the following beneficial technical effects: This invention provides a method for diagnosing abnormal vibrations in thermal power plant pipelines based on acoustic signature recognition. By normalizing the real-time acquired pipeline acoustic signature feature vectors and baseline acoustic signature feature vectors, and then calculating the Euclidean distance dimensionally, it achieves the ability to capture abnormal acoustic features during pipeline operation in real time. This solves the problem that traditional acoustic monitoring methods struggle to distinguish abnormal signals from background noise under complex operating environments such as high temperature, high pressure, and pump / valve noise. By performing independent component analysis on the acquired acoustic signals and comparing similarity with the differential acoustic signature results, it automatically extracts specific sound source components related to pipeline anomalies from complex sound fields, solving the problem that traditional filtering or energy discrimination methods struggle to accurately identify abnormal sound sources under interference from multiple superimposed sound sources. This invention inputs the target sound source signal obtained from differential acoustic signature analysis, the spatial arrangement information of each pickup sensor, and the time delay data obtained from cross-correlation calculations into a sound source localization model based on the time difference of arrival, achieving precise spatial localization of abnormal sound sources in the pipeline. This solves the problem that traditional localization methods are easily affected by background noise and cannot confirm the attribution of abnormal sound sources. Attached Figure Description

[0016] Figure 1 This is a flowchart of the abnormal vibration diagnosis method for thermal power plant pipelines based on voiceprint recognition in an embodiment of the present invention; Figure 2 This is a schematic diagram of the abnormal vibration diagnosis system for thermal power plant pipelines based on voiceprint recognition in an embodiment of the present invention. In the diagram: 1. Data acquisition and preprocessing module; 2. Feature vector fusion processing module; 3. Abnormal vibration diagnosis and processing module. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0018] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0019] The present invention will now be described in further detail with reference to the accompanying drawings: The purpose of this invention is to provide a method and related equipment for diagnosing abnormal vibrations in thermal power plant pipelines based on voiceprint recognition, so as to solve the technical problem that traditional acoustic monitoring is unable to distinguish abnormal signals from background noise under complex operating environments such as high temperature, high pressure and pump and valve noise.

[0020] Example 1 See Figure 1 This embodiment provides a method for diagnosing abnormal vibrations in thermal power plant pipelines based on voiceprint recognition, including the following process: Step 1: Deploy acoustic sensors at monitoring locations on the power plant pipelines to collect acoustic signals, and preprocess the acoustic signals to obtain preprocessed signals; Specifically, the arrangement of the sound pickup sensor includes: Based on the pipeline's geometric parameters and operating noise distribution, candidate sensor placement points are determined using sound field propagation simulation results. Select locations with high acoustic signature differences from the candidate locations to form a sensor layout scheme covering the main pipeline and branch pipelines; According to the layout plan, sound pickup sensors are installed on the surface of the pipe, and the spatial coordinates of each sensor are recorded for subsequent sound source localization.

[0021] Specifically, the preprocessing includes bandpass filtering and noise suppression processing, specifically including: The frequency range of the acquired acoustic signal is limited by a bandpass filter, retaining only the target frequency band signal within the set passband; A noise estimation model is constructed using a reference noise signal, and noise reduction is performed on the bandpass filtered signal based on this noise estimation model to obtain a preprocessed signal.

[0022] In this embodiment, based on the geometric parameters of the pipeline and the noise distribution under operating conditions, a sound field propagation simulation model is established to calculate the propagation path and attenuation characteristics of sound waves in the pipeline and its surrounding space, thereby obtaining multiple candidate placement points. By simulating and analyzing acoustic signals collected from different candidate locations, the corresponding acoustic signature difference index is extracted. The acoustic signature difference index can be calculated by calculating the cross-correlation coefficient between the signals of each candidate point or by using a difference function based on spectral entropy. When the difference index is high, it indicates that the placement point is more conducive to distinguishing different sound source characteristics. Based on this, a set of placement points with high difference index is selected to form a sound pickup sensor placement scheme covering the main pipeline and branch pipelines. After determining the placement scheme, the sound pickup sensors are fixedly installed on the pipeline surface, and the spatial position parameters of each sensor in the three-dimensional coordinate system are recorded by measuring or reading the construction drawings. This spatial position information will be used as the calculation input in the subsequent sound source localization process. After acoustic signal acquisition is completed, the signal preprocessing stage begins. First, the acquired raw signal is frequency-limited using a bandpass filter. The passband range of the filter is determined based on the natural frequency of the pipe material and the expected abnormal vibration frequency range. For example, selecting... to The passband, of which Indicates the lower cutoff frequency. Represents the upper limit cutoff frequency, satisfying The frequency components will be preserved; in the filtering operation, the filter transfer function It can be represented as: ; in The frequency variable is used; the signal after bandpass filtering can effectively remove low-frequency environmental disturbances and high-frequency electromagnetic noise, and obtain the signal component of the target frequency band. After bandpass filtering, noise suppression is also required. One possible approach is to introduce a reference noise signal, which could originate from an auxiliary channel in the surrounding environment unaffected by pipe vibration. Based on the reference noise signal, a noise estimation model is constructed, for example, by estimating the noise power spectrum using the minimum mean square error (MMSE) method. Let the target signal be... The reference noise signal is The power spectra obtained by Fourier transform are as follows: and Then the noise estimate can be expressed as: ; in This represents the expectation operation. The power spectrum of the reference noise is then subtracted from the noise estimate in the target signal spectrum to obtain the noise-suppressed spectral components. The signal is then restored to the time domain through inverse Fourier transform, and the resulting signal is the preprocessed signal. This preprocessed signal has a higher signal-to-noise ratio and clearer voiceprint features, which facilitates subsequent feature extraction and anomaly diagnosis. Through the above implementation methods, high-quality acquisition and effective preprocessing of pipeline acoustic signals can be achieved. The reasonable arrangement of the pickup sensors ensures the spatial coverage and sound source differentiation of the acoustic signals, while the combination of bandpass filtering and noise suppression can reduce background noise interference while preserving the characteristics of key frequency bands, thereby providing reliable input data for subsequent acoustic feature extraction and abnormal vibration diagnosis.

[0023] Multi-dimensional acoustic feature extraction is performed on the preprocessed signal, and the extracted results are fused to generate a voiceprint feature vector.

[0024] Step 2: Based on the preprocessed signal, perform multi-dimensional acoustic feature extraction and fusion to generate acoustic feature vectors. At the same time, establish an acoustic baseline library corresponding to different operating conditions of the pipeline and update it dynamically. Specifically, multi-dimensional acoustic feature extraction includes: The short-time energy, zero-crossing rate, and autocorrelation function values ​​of the preprocessed signal are calculated in the time domain to obtain the time-domain characteristic parameters; The preprocessed signal is subjected to Fast Fourier Transform to extract the amplitude spectrum and power spectrum, and the spectral centroid and spectral flatness are calculated to obtain the frequency domain characteristic parameters. Mel frequency cepstral analysis is performed on the preprocessed signal to calculate the cepstral coefficient sequence and obtain the acoustic cepstral characteristic parameters.

[0025] The multi-dimensional acoustic feature fusion includes: The time-domain feature parameters, frequency-domain feature parameters, and acoustic cepstral feature parameters are normalized according to a unified dimension to obtain a normalized feature set. Principal component analysis is performed on the normalized feature set to extract the eigenvectors of the principal components in order to reduce feature redundancy. The main component feature vectors are input into a convolutional neural network to extract deep feature representations and obtain deep acoustic feature vectors. The depth acoustic feature vector is fused with the main component feature vector to form the voiceprint feature vector.

[0026] In this embodiment, a preprocessed signal is received, and the preprocessed signal is segmented into several frames to obtain signal samples, which are used as inputs for feature extraction. For each frame of signal, time-domain feature calculation, frequency-domain feature calculation, and cepstral feature calculation are performed in sequence to obtain the corresponding time-domain feature parameter set, frequency-domain feature parameter set, and cepstral feature parameter set. The obtained feature parameters are subjected to a unified scale transformation and stored as inputs for subsequent feature fusion. For temporal feature extraction, the energy value of each frame is calculated based on short-time energy. Its expression is ,in Represents discrete-time signals. Represents the window function. Indicates window length; calculates the zero-crossing rate for each frame based on the zero-crossing rate. Its expression is ,in The sign function is represented; the autocorrelation sequence of each frame is calculated based on the autocorrelation function. ,in The lag order is given; the set of time-domain characteristic parameters is composed of... , It is combined with the autocorrelation values ​​under several lag orders and output as time-domain characteristic parameters; For frequency domain feature extraction, a long-window fast Fourier transform is performed on each frame of signal. ,in To change the number of points, For frequency index; based on amplitude spectrum Calculate the power spectrum Calculate the spectral centroid by frequency index ,in Indicates index Corresponding frequency values; calculate spectral flatness ,in This represents the number of frequency points within the effective frequency band; the set of frequency domain characteristic parameters consists of the amplitude spectrum, power spectrum, and spectral centroid. With spectral flatness The parameters are composed and output as frequency domain feature parameters; For cepstral feature extraction, a Mel filter bank is constructed for each frame of signal and the energy of each filter band is calculated. (in , (Indicates the number of filter banks), take the logarithm of the energy and perform a discrete cosine transform on the logarithmic energy sequence to obtain the cepstral coefficient sequence. The expression for calculating the cepstral coefficients is: ,in This serves as an index for the cepstral coefficient sequence; the cepstral coefficient sequence is output as an acoustic cepstral feature parameter. After obtaining the time-domain feature parameter set, frequency-domain feature parameter set, and acoustic cepstral feature parameters, a unified dimension normalization process is performed. The normalization operation adopts z-score transform. ,in and These represent the mean and standard deviation of the feature dimension within the training or statistical window, respectively; after normalization, they form a feature matrix with uniform dimensions. The data is then input into the principal component analysis module; within the principal component analysis module, the covariance matrix is ​​calculated. ,right Perform eigenvalue decomposition Before choosing The matrix is ​​composed of principal component vectors. And through matrix multiplication Obtain the principal component eigenvectors The principal component eigenvectors are output as the structured representation after dimensionality reduction. The principal component feature vectors are input into a one-dimensional convolutional neural network for deep feature extraction. The network structure includes one or more one-dimensional convolutional layers, each followed by batch normalization and ReLU activation functions. Pooling layers are inserted between the convolutional layers as designed to reduce temporal dimensionality. The output of the convolutional network is flattened and mapped to a deep acoustic feature vector through several fully connected layers. The parameters of the convolutional neural network are updated iteratively through supervised training on the sample set. The training loss function can be either cross-entropy or mean squared error, and the network weights are updated using the backpropagation algorithm. Depth acoustic feature vectors With principal component eigenvectors The original fusion vector is obtained by performing concatenation and fusion. Subsequently implement Normalization ,in Represents the 2-norm; normalized vector The final speakerprint feature vector is output and written to the online feature cache for use in baseline construction, differential alignment and subsequent sound source separation modules; Time-domain features, frequency-domain features, and cepstral features respectively provide representations of different dimensions of signal information. The combination of principal component analysis and convolutional neural networks reduces redundant dimensions while maintaining feature discriminative power. (Further details on cascading fusion and...) Normalization ensures the numerical stability of the generated voiceprint feature vectors in subsequent comparison and matching processes; it can provide standardized and consistent input for baseline library construction, differential analysis, and sound source separation.

[0027] Acoustic signals are collected under different operating conditions of the pipeline, and an acoustic signature baseline library corresponding to the operating conditions is established. During operation, the acoustic signature baseline library is dynamically updated according to the new normal acoustic signature feature vector.

[0028] Step 3: Based on the acoustic signature feature vector and the updated acoustic signature baseline library, output the diagnosis result of abnormal pipeline vibration through operating condition matching, acoustic signature differential comparison and sound source localization.

[0029] Specifically, the establishment and dynamic updating of the voiceprint baseline database includes: Acoustic signals were collected under different steady-state operating conditions of the pipeline, and acoustic fingerprint sample sets were generated based on acoustic fingerprint feature vectors. Clustering calculations are performed on the voiceprint sample set for each working condition to obtain and store the voiceprint center vector for each working condition, thus forming a mapping relationship between the working condition and the baseline voiceprint feature vector. During operation, a new normal voiceprint feature vector is acquired and its similarity to the baseline voiceprint feature vector is calculated. When the similarity meets the preset threshold, the corresponding baseline voiceprint feature vector is updated and stored using a weighted average method.

[0030] In this embodiment, acoustic signals are first collected under different operating conditions of the pipeline in steady-state operation, and an acoustic signature sample set for each operating condition is generated. Steady-state operation is achieved by monitoring operating parameters within a time window. The fluctuation range within is used to determine this, among which Indicates the length of the time window. These represent the stability thresholds for pressure, temperature, and flow rate, respectively, within a time window. The change in internal pressure is less than or equal to Temperature change range less than or equal to The change in flow rate is less than or equal to The system is considered to be in a steady state and sample acquisition is triggered. The sample acquisition process includes preprocessing the acquired signal and extracting the voiceprint feature vector using the weighted 4 method. Output operating condition voiceprint sample set ; Clustering is performed on the voiceprint sample set for each operating condition to obtain the voiceprint center vector. The clustering algorithm can be implemented using k-means or Gaussian mixture model. k-means clustering obtains the center vector by solving the following minimization problem. : ; in The vector L2 norm is used to solve for the set of center vectors of the voiceprint under operating conditions. And establish a mapping relationship between working condition labels and center vectors; Each center vector is written to the baseline database while simultaneously maintaining statistical information. Each record in the baseline database contains a center vector. Sample count Covariance matrix With the most recent update timestamp The covariance matrix is ​​defined as: ; The above statistical information is used for subsequent dynamic updates and similarity calculations and is stored as metadata for baseline entries; Acquire new normal voiceprint feature vectors during online operation. The similarity with the center vector of the corresponding working condition in the baseline library is calculated, and cosine similarity is used as the similarity measure, which is defined as: ; in This represents the dot product of vectors, and the calculated similarity is used to determine whether a preset similarity threshold is met. ; When similarity At that time, the center vector and statistical information are incrementally updated according to the weighted average strategy. The center vector update can be performed using the counting weighted formula: ; And update the sample count. The covariance matrix is ​​updated using the incremental formula. ; After the update is complete, With update timestamp Write back to the baseline library; When new samples The similarity with all center vectors in the baseline library is less than the threshold. At that time, Write the candidate sample to the cache and count it. When the cumulative number of samples in the cache reaches a threshold... When a new steady-state operating condition distribution is formed, cluster analysis is performed on the cached sample set to determine whether a new steady-state operating condition distribution is formed. If a new center vector is generated by clustering, it is inserted into the baseline library as a new baseline entry and the corresponding statistical information is initialized. Otherwise, the cached samples are transferred to the manual annotation or offline analysis process. To improve the efficiency of similarity calculation and baseline retrieval, the baseline library establishes a vector index structure for the center vector. The index implements optional local sensitive hashing or nearest neighbor search tree. During online retrieval, the candidate center vector set is first returned through the index, and the precise similarity calculation is performed on the candidate set to determine the target center vector. The baseline library update and retrieval module is deployed in a service-oriented manner. The baseline management module provides an atomic write interface and version management. Asynchronous update tasks are triggered during low-load periods to avoid affecting the real-time diagnostic process. The baseline management ensures the consistency of concurrent updates through a transaction mechanism. The above steps form a closed-loop implementation process from sample collection to baseline update, which can be used for online maintenance of the baseline library and expansion of new operating condition entries. The baseline library can provide a structured representation of typical voiceprint distributions under different steady-state operating conditions and track long-term operating characteristics through statistical information and dynamic update strategies, thereby providing consistent and searchable reference data for subsequent differential comparison and anomaly judgment.

[0031] The system obtains the real-time operating parameters of the pipeline, determines the current operating condition based on the operating parameters, and calls the baseline acoustic signature feature vector corresponding to the current operating condition from the acoustic signature baseline library.

[0032] Specifically, the operating condition matching includes collecting real-time pressure, temperature, and flow data of the pipeline, inputting the data into a support vector machine-based operating condition classification model, outputting the current operating condition identifier, and calling the corresponding baseline voiceprint feature vector.

[0033] In this embodiment, a sensor array that collects real-time pipeline operating parameters continuously monitors pressure, temperature, and flow rate, and buffers the raw measurements according to a preset time window. Time synchronization and missing value interpolation are performed on the buffered raw measurements. The interpolation method can be linear interpolation or forward imputation based on the most recent valid sample. Subsequently, a low-pass filter is applied to the interpolated signal to suppress measurement noise, and sliding statistical features are calculated. The sliding statistical features include the time window length... mean within ,variance and the trend slope based on linear regression The above statistical features are concatenated into a vector of operating parameters in a fixed order. ,in Indicates the vector dimension; For the runtime parameter vector To eliminate the influence of dimensions, a standardization transformation is performed using z-score standardization. ,in and These represent the mean vector and standard deviation vector obtained from the training set statistics, respectively, and are read from the configuration store during online runtime; the standardized vectors are... As input to the working condition classification model; During the training phase, the operating condition classification model constructs a training set based on historical operating data and manually labeled operating conditions. The training employs a support vector machine-based classifier, which uses a one-to-one strategy to handle multiple work condition identifiers. The discriminant function for binary classification is expressed as follows: ,in Represents the training sample vector. This indicates a second-class label. Represents the Lagrange multipliers. Indicates the bias term. The kernel function is represented by a radial basis function, defined in this embodiment. ,in For kernel width parameter, The vector's 2-norm is represented; the training process obtains this by solving the dual problem. and Cross-validation was used to determine the hyperparameters. With penalty factor After training, the support vectors, corresponding coefficients, and normalization parameters are stored as model files and deployed on the online inference service. During the online inference phase, the standardized vectors will be... The input is fed into a deployed multi-class support vector machine classification network, and the binary discriminant function is calculated for each class. The final working condition label is obtained by combining the judgments according to the one-to-one voting rule. To assess the confidence level of the discrimination, the absolute value of the distance to the discrimination surface can be calculated. The normalized confidence score is obtained by using max-min normalization. ,when Less than the configured threshold In such cases, the decision will be marked as low confidence and a manual review will be triggered, or the default baseline entry will be used. As a temporary mapping; According to the working condition label Retrieve the corresponding baseline voiceprint feature vector from the baseline management table. The retrieval process utilizes a mapping table to quickly locate mapping relationships. This table records the correspondence between operating condition identifiers and baseline entry indexes and supports index caching to accelerate queries. The retrieved baseline voiceprint feature vectors are then used to... Used as reference input for differential alignment and passed to subsequent modules; In terms of model maintenance, during operation, samples of operating parameters with manual annotations or automatically confirmed by trusted rules, along with corresponding working condition labels, are periodically aggregated into the retraining data pool. Retraining is periodically triggered to update the support vector machine model and recalculate the normalized parameters. and Model updates are version controlled and distributed to the online inference service via canary releases to ensure operational continuity. It can structure physical measurement parameters into stable input vectors and accurately identify the current operating conditions through a pre-trained classification model, thereby providing clear criteria for the accurate retrieval of baseline voiceprint vectors and supporting subsequent differential comparison and anomaly diagnosis processes.

[0034] The real-time voiceprint feature vector is compared with the baseline voiceprint feature vector to obtain the differential voiceprint result.

[0035] The voiceprint differential comparison includes normalizing the real-time voiceprint feature vector and the baseline voiceprint feature vector, calculating the Euclidean distance dimension by dimension to generate the differential voiceprint vector, and outputting the differential voiceprint result after threshold discrimination.

[0036] In this embodiment, the real-time acoustic signature feature vector generated by the pipeline during operation is first obtained. Simultaneously, the baseline acoustic signature feature vector corresponding to the current operating condition is retrieved from the acoustic signature baseline library. Since the two sets of vectors differ in data source and generation mechanism, normalization is required before differential calculation. This normalization maps feature parameters of different dimensions to a unified numerical range to avoid a single feature dimension having excessive weight in the calculation. The normalization process can be accomplished through a minimum-maximum linear transformation, the mathematical expression of which is: ; in, Indicates the first The eigenvector at the th eigenvector in the th th eigenvector Values ​​in each dimension; and These represent the minimum and maximum values ​​of this dimension in the sample set, respectively. This represents the normalized eigenvalues.

[0037] After normalization, the system calculates the Euclidean distance between the normalized real-time voiceprint feature vector and the normalized baseline voiceprint feature vector dimension by dimension to construct the differential voiceprint vector; the formula for calculating the Euclidean distance is: ; in, The normalized real-time voiceprint feature vector represents the first... Values ​​of each dimension The normalized baseline voiceprint feature vector is represented at the th... Values ​​of each dimension This represents the difference value corresponding to that dimension; the difference voiceprint vector is formed by the set of difference results calculated on all feature dimensions.

[0038] After the differential voiceprint vector is generated, anomaly identification needs to be performed by a threshold discrimination module; this module presets a threshold. The output is completed based on the following criteria: ; in, This represents the number of dimensions of the voiceprint feature vector. This represents the average of all differences. This represents the differential voiceprint result. A value of 1 indicates that the difference exceeds the threshold, i.e., there is an anomaly, while a value of 0 indicates that the difference is below the threshold, i.e., the normal state is maintained.

[0039] Through the above process, the system can dynamically compare real-time operating features with baseline reference features in the voiceprint feature space, output differential voiceprint results, and realize timely judgment of operating status; this process can convert complex acoustic signals into structured differential information, which is convenient for subsequent call and application of sound source separation and anomaly localization modules.

[0040] Sound source separation is performed based on differential acoustic signature results, the acoustic signature signal of the target pipeline is extracted, and the location of abnormal sound sources is calculated by combining the sensor placement position, and the pipeline anomaly diagnosis result is output.

[0041] The sound source separation and localization includes performing independent component analysis and decomposition on the acoustic signal to screen out the target sound source components that match the differential acoustic signature results; calculating the time delay data of the target acoustic signature signal at each sensor through cross-correlation; combining the sensor spatial coordinate input with the sound source localization model based on the time difference of arrival algorithm; and outputting the spatial coordinates of the abnormal sound source and the diagnostic results.

[0042] In this embodiment, differential acoustic signature results and corresponding multi-channel acoustic signals at monitoring locations are obtained, and time synchronization and window framing processing are performed on each channel signal to form an aligned frame set; DC component removal processing is performed on each frame signal and normalization is performed according to frame energy to eliminate sensor gain differences and improve blind source separation stability; the preprocessed multi-channel frame set is used as input to the blind source separation module for subsequent decomposition. Within the blind source separation module, a linear hybrid model is established. ,in Indicates time The observation vector, Indicates the number of monitoring channels. Represents the independent sound source component vector. The matrix is ​​a mixture matrix. To perform independent component analysis, the observed vectors are first centered to achieve a mean of zero, and then the centered data is whitened to obtain whitened vectors. The whitening process is completed by solving the covariance matrix of the observed data and performing eigenvalue decomposition. After whitening, the Fast Independent Component Analysis (FastICA) algorithm is one possible implementation method, which solves the separation matrix iteratively through fixed points. To maximize the non-Gaussianity, the iterative update rule can be approximated by the following formula: ; in This is the currently estimated separation vector. For comparison functions, such as Its derivative, the expectation operation can be approximated by the mean of intra-frame samples; an orthogonalization step is applied to all separation vectors to ensure their mutual independence, and the separation matrix is ​​obtained after iterative convergence. Then, the sequence of isolated independent sound source components is calculated. It also outputs several independent sound source component signals; For each independent sound source component, the corresponding acoustic feature vector is extracted according to the multi-dimensional feature extraction process. ,in For component indexing; use the differential voiceprint vector With each component eigenvector Similarity comparison is performed, and the similarity measure is expressed as cosine similarity: ; Where “·” represents the dot product of vectors, Represents the L2 norm; based on the similarity threshold Filter out the set of target sound source components that match the difference vector If the set contains more than one component, the positioning process is entered in parallel to support multi-source concurrent positioning. For the set of sound source components selected as the target, the corresponding time-domain waveform is extracted from each sensor channel, and the arrival time difference is estimated using a cross-correlation function, where the cross-correlation is defined in discrete form: ; in and They represent the first With the Discrete time-series samples of target components from each sensor channel, indexed This represents the number of delayed samples, with a sampling rate of [missing information]. The position of cross-correlation peak Obtain delay estimate To improve the accuracy of time delay estimation, parabolic fitting interpolation can be used at the cross-correlation peak to obtain subsampling time offset estimation. A positioning equation is established based on the Time Difference of Arrival (TDOA), assuming the sensor's spatial coordinates are... The coordinates of the target sound source are The speed of sound is Then the sensor The theoretical arrival time difference with reference sensor 1 is expressed as: ; The estimated time delay vector With model function By performing nonlinear least squares fitting and solving the following optimization problem, the target position estimate can be obtained. : ; This optimization can be solved iteratively using the Gauss-Newton or Levenberg-Marquardt algorithm, and convergence can be accelerated by estimating the sensor coordinate matrix and initial values. To reduce the impact of sound velocity uncertainty on the positioning results, ambient temperature can be considered. Correct the speed of sound, for example, by using an approximation. (Units: meters per second and degrees Celsius), and the corrected Substitute into the positioning model; After the location solution is completed, the target space coordinates are output. The confidence index is calculated by combining the component similarity score and the positioning residual, and a record of the abnormal pipeline location is generated. This record includes a timestamp, target coordinates, confidence index, corresponding sensor index, and triggering differential feature vector reference identifier. This location data is transmitted to the diagnostic summary module for visualization and maintenance work order generation.

[0043] In summary, the method for diagnosing abnormal vibrations in thermal power plant pipelines based on acoustic signature recognition provided in this embodiment achieves the ability to capture abnormal acoustic features in pipeline operation in real time by normalizing the real-time acquired pipeline acoustic signature feature vector and the baseline acoustic signature feature vector and then calculating the Euclidean distance dimensionally. This solves the problem that traditional acoustic monitoring methods are unable to distinguish abnormal signals from background noise in complex operating environments such as high temperature, high pressure, and pump and valve noise. By performing independent component analysis and decomposition on the acquired acoustic signals and comparing similarity with the differential acoustic signature results, the method automatically extracts specific sound source components related to pipeline anomalies from complex sound fields, solving the problem that traditional filtering or energy discrimination methods are unable to accurately identify abnormal sound sources under interference from multiple superimposed sound sources. This invention achieves accurate spatial positioning of abnormal sound sources in pipelines by inputting the target sound source signal obtained from differential acoustic signature analysis, the spatial arrangement information of each pickup sensor, and the time delay data obtained from cross-correlation calculation into a sound source localization model based on the time difference of arrival. This solves the problem that traditional localization methods are easily interfered with by background noise and cannot confirm the attribution of abnormal sound sources.

[0044] Example 2 according to Figure 2 As shown, this embodiment provides a diagnostic system for abnormal vibration of pipelines in thermal power plants based on voiceprint recognition, including: The data acquisition and preprocessing module 1 is used to deploy a pickup sensor at the monitoring location of the pipeline in the thermal power plant to acquire acoustic signals, and to preprocess the acoustic signals to obtain a preprocessed signal. Feature vector fusion processing module 2 is used to extract and fuse multi-dimensional acoustic features based on the preprocessed signal to generate acoustic feature vectors, and at the same time to establish an acoustic baseline library corresponding to different operating conditions of the pipeline and dynamically update it. The abnormal vibration diagnosis and processing module 3 is used to output the abnormal vibration diagnosis results of the pipeline based on the acoustic feature vector and the updated acoustic baseline library, through working condition matching, acoustic differential comparison and sound source localization.

[0045] Example 3 The present invention also provides a mobile terminal, including a memory, a processor, and a computer program stored in the memory and executable on the processor, such as a voiceprint recognition-based diagnostic program for abnormal vibration of pipelines in thermal power plants.

[0046] When the processor executes the computer program, it implements the steps of the above-described method for diagnosing abnormal vibrations in thermal power plant pipelines based on voiceprint recognition, for example: Acoustic sensors are deployed at monitoring locations on pipelines in thermal power plants to collect acoustic signals. The acoustic signals are then preprocessed to obtain preprocessed signals. Based on the preprocessed signal, multi-dimensional acoustic features are extracted and fused to generate acoustic feature vectors. At the same time, an acoustic baseline library corresponding to different operating conditions of the pipeline is established and dynamically updated. Based on the aforementioned acoustic signature feature vector and the updated acoustic signature baseline library, the abnormal vibration diagnosis results of the pipeline are output through operating condition matching, acoustic signature differential comparison, and sound source localization.

[0047] Alternatively, when the processor executes the computer program, it implements the functions of each module in the above system, for example: The data acquisition and preprocessing module 1 is used to deploy a pickup sensor at the monitoring location of the pipeline in the thermal power plant to acquire acoustic signals, and to preprocess the acoustic signals to obtain a preprocessed signal. Feature vector fusion processing module 2 is used to extract and fuse multi-dimensional acoustic features based on the preprocessed signal to generate acoustic feature vectors, and at the same time to establish an acoustic baseline library corresponding to different operating conditions of the pipeline and dynamically update it. The abnormal vibration diagnosis and processing module 3 is used to output the abnormal vibration diagnosis results of the pipeline based on the acoustic feature vector and the updated acoustic baseline library, through working condition matching, acoustic differential comparison and sound source localization.

[0048] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the mobile terminal.

[0049] For example, the computer program can be divided into a data acquisition and preprocessing module 1, a feature vector fusion processing module 2, and an abnormal vibration diagnosis processing module 3; The specific functions of each module are as follows: The data acquisition and preprocessing module 1 is used to deploy a pickup sensor at the monitoring location of the pipeline in the thermal power plant to acquire acoustic signals, and to preprocess the acoustic signals to obtain a preprocessed signal. Feature vector fusion processing module 2 is used to extract and fuse multi-dimensional acoustic features based on the preprocessed signal to generate acoustic feature vectors, and at the same time to establish an acoustic baseline library corresponding to different operating conditions of the pipeline and dynamically update it. The abnormal vibration diagnosis and processing module 3 is used to output the abnormal vibration diagnosis results of the pipeline based on the acoustic feature vector and the updated acoustic baseline library, through working condition matching, acoustic differential comparison and sound source localization.

[0050] The mobile terminal can be a computing device such as a desktop computer, laptop, handheld computer, or cloud server. The mobile terminal may include, but is not limited to, a processor and memory.

[0051] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the mobile terminal, connecting various parts of the mobile terminal via various interfaces and lines.

[0052] The memory can be used to store the computer program and / or module. The processor implements various functions of the mobile terminal by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory.

[0053] The memory may primarily include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a function (such as sound playback, image playback, etc.); the data storage area may store data created based on the use of the mobile phone (such as audio data, phonebook, etc.). Furthermore, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disks, RAM, plug-in hard disks, SmartMediaCards (SMC), Secure Digital (SD) cards, FlashCards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0054] Example 4 The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method for diagnosing abnormal vibrations in thermal power plant pipelines based on voiceprint recognition.

[0055] If the modules / units integrated in the mobile terminal are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.

[0056] Based on this understanding, all or part of the processes in the above method can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of the above-described aggregated reinforcement learning resource scheduling method. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate form.

[0057] The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0058] It should be noted that the content contained in the computer-readable medium may be appropriately added to or subtracted from the content as required by the legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium may not include electrical carrier signals and telecommunication signals.

[0059] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for diagnosing abnormal vibrations in thermal power plant pipelines based on voiceprint recognition, characterized in that, include: Acoustic sensors are deployed at monitoring locations on pipelines in thermal power plants to collect acoustic signals. The acoustic signals are then preprocessed to obtain preprocessed signals. Based on the preprocessed signal, multi-dimensional acoustic features are extracted and fused to generate acoustic feature vectors. At the same time, an acoustic baseline library corresponding to different operating conditions of the pipeline is established and dynamically updated. Based on the aforementioned acoustic signature feature vector and the updated acoustic signature baseline library, the abnormal vibration diagnosis results of the pipeline are output through operating condition matching, acoustic signature differential comparison, and sound source localization.

2. The method for diagnosing abnormal vibrations in thermal power plant pipelines based on voiceprint recognition according to claim 1, characterized in that, The arrangement of the pickup sensor includes: Based on the pipeline's geometric parameters and operating noise distribution, candidate sensor placement points are determined using sound field propagation simulation results. Select locations with high acoustic signature differences from the candidate locations to form a sensor layout scheme covering the main pipeline and branch pipelines; According to the layout plan, sound pickup sensors are installed on the surface of the pipe, and the spatial coordinates of each sensor are recorded for subsequent sound source localization.

3. The method for diagnosing abnormal vibrations in thermal power plant pipelines based on voiceprint recognition according to claim 1, characterized in that, The preprocessing includes bandpass filtering and noise suppression, specifically including: The frequency range of the acquired acoustic signal is limited by a bandpass filter, retaining only the target frequency band signal within the set passband; A noise estimation model is constructed using a reference noise signal, and noise reduction is performed on the bandpass filtered signal based on this noise estimation model to obtain a preprocessed signal.

4. The method for diagnosing abnormal vibrations in thermal power plant pipelines based on voiceprint recognition according to claim 1, characterized in that, The multi-dimensional acoustic feature extraction includes: The short-time energy, zero-crossing rate, and autocorrelation function values ​​of the preprocessed signal are calculated in the time domain to obtain the time-domain characteristic parameters; The preprocessed signal is subjected to Fast Fourier Transform to extract the amplitude spectrum and power spectrum, and the spectral centroid and spectral flatness are calculated to obtain the frequency domain characteristic parameters. Mel frequency cepstral analysis is performed on the preprocessed signal to calculate the cepstral coefficient sequence and obtain the acoustic cepstral characteristic parameters.

5. The method for diagnosing abnormal vibrations in thermal power plant pipelines based on voiceprint recognition according to claim 1, characterized in that, The multi-dimensional acoustic feature fusion includes: The time-domain feature parameters, frequency-domain feature parameters, and acoustic cepstral feature parameters are normalized according to a unified dimension to obtain a normalized feature set. Principal component analysis is performed on the normalized feature set to extract the eigenvectors of the principal components in order to reduce feature redundancy. The main component feature vectors are input into a convolutional neural network to extract deep feature representations and obtain deep acoustic feature vectors. The depth acoustic feature vector is fused with the main component feature vector to form the voiceprint feature vector.

6. The method for diagnosing abnormal vibrations in thermal power plant pipelines based on voiceprint recognition according to claim 1, characterized in that, The establishment and dynamic updating of the voiceprint baseline library includes: Acoustic signals were collected under different steady-state operating conditions of the pipeline, and acoustic fingerprint sample sets were generated based on acoustic fingerprint feature vectors. Clustering calculations are performed on the voiceprint sample set for each working condition to obtain and store the voiceprint center vector for each working condition, thus forming a mapping relationship between the working condition and the baseline voiceprint feature vector. During operation, a new normal voiceprint feature vector is acquired and its similarity to the baseline voiceprint feature vector is calculated. When the similarity meets the preset threshold, the corresponding baseline voiceprint feature vector is updated and stored using a weighted average method.

7. The method for diagnosing abnormal vibrations in thermal power plant pipelines based on voiceprint recognition according to claim 1, characterized in that, The process of outputting the abnormal vibration diagnosis result of the pipeline based on the voiceprint feature vector and the updated voiceprint baseline library, through operating condition matching, voiceprint differential comparison and sound source localization, includes collecting real-time pressure, temperature and flow data of the pipeline, inputting the operating condition classification model based on support vector machine, outputting the current operating condition identifier and calling the corresponding baseline voiceprint feature vector. The voiceprint differential comparison includes normalizing the real-time voiceprint feature vector and the baseline voiceprint feature vector, calculating the Euclidean distance dimension by dimension to generate the differential voiceprint vector, and outputting the differential voiceprint result after threshold discrimination. The sound source separation and localization includes performing independent component analysis and decomposition on the acoustic signal to screen out the target sound source components that match the differential acoustic signature results; calculating the time delay data of the target acoustic signature signal at each sensor through cross-correlation; combining the sensor spatial coordinate input with the sound source localization model based on the time difference of arrival algorithm; and outputting the spatial coordinates of the abnormal sound source and the diagnostic results.

8. A diagnostic system for abnormal vibration of pipelines in thermal power plants based on voiceprint recognition, characterized in that, include: The data acquisition and preprocessing module is used to deploy sound pickup sensors at the monitoring locations of pipelines in thermal power plants to collect acoustic signals, and to preprocess the acoustic signals to obtain preprocessed signals. The feature vector fusion processing module is used to extract and fuse multi-dimensional acoustic features based on the preprocessed signal to generate acoustic feature vectors, and at the same time to establish and dynamically update the acoustic baseline library corresponding to different operating conditions of the pipeline. The abnormal vibration diagnosis and processing module is used to output the abnormal vibration diagnosis results of the pipeline based on the acoustic feature vector and the updated acoustic baseline library, through working condition matching, acoustic differential comparison and sound source localization.

9. A mobile terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the abnormal vibration diagnosis method for thermal power plant pipelines based on voiceprint recognition as described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the abnormal vibration diagnosis method for thermal power plant pipelines based on voiceprint recognition as described in any one of claims 1-7.

Images