Rotating machinery vibration signal multi-channel acquisition and adaptive noise reduction system and method
By using multi-channel synchronous acquisition and adaptive noise reduction methods, the vibration signals of rotating machinery are spatiotemporally aligned and adaptively denoised, which solves the problem of strong noise drowning out weak fault features and improves the fault identification capability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- XIAN THERMAL POWER RES INST CO LTD
- Filing Date
- 2026-03-18
- Publication Date
- 2026-07-31
AI Technical Summary
In complex excitation environments, traditional single-point or quasi-synchronous acquisition methods and fixed-parameter filtering techniques are insufficient to extract weak fault features of rotating machinery, resulting in a high rate of missed diagnoses. Single-channel data cannot provide sufficient spatial resolution, and strong background noise overwhelms weak fault features.
A multi-channel synchronous acquisition array is used to perform spatiotemporal alignment processing on the vibration signal of rotating machinery. The rotational speed phase information is extracted based on the cross-correlation analysis method. The noise reduction strategy is optimized by combining the ant colony algorithm. The fault indication index is extracted by using a pre-constructed vibration assessment model to achieve adaptive noise reduction.
It improves the response to vibration fault identification in rotating machinery, ensures that the noise-reduced signal is clean and useful, reduces the false negative rate, and improves the accuracy of fault diagnosis.
Smart Images
Figure CN122493814A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of mechanical vibration signal processing technology, specifically relating to a multi-channel acquisition and adaptive noise reduction system and method for rotating machinery vibration signals. Background Technology
[0002] Rotating machinery is widely used in key equipment such as steam turbines, generators, compressors, fans, and pumps in power generation, petrochemicals, metallurgy, aerospace, and high-end manufacturing. Failures in these machines can lead to unplanned shutdowns and major safety accidents. Vibration signals, as the most direct and sensitive indicator of the health status of rotating machinery, have become the recognized primary diagnostic source. However, in the complex excitation environments of industrial sites, traditional single-point or quasi-synchronous acquisition methods and fixed-parameter filtering techniques are insufficient to meet the needs of extracting subtle early fault features.
[0003] Considering that modern rotating machinery often has the structural characteristics of multiple supports, multiple rotors, and multiple gear couplings, using a single or sparse measurement point can only obtain local and partial vibration responses. It is very easy to have limited measurement accuracy due to modal nodes and signal cancellation. Especially in coupled fault scenarios, strong background noise will drown out weak fault features, and single-channel data cannot provide sufficient spatial resolution, resulting in a high rate of misdiagnosis and missed diagnosis. To address the above problems, we propose a multi-channel acquisition and adaptive noise reduction system and method for rotating machinery vibration signals. Summary of the Invention
[0004] The purpose of this invention is to address the shortcomings of existing technologies by providing a multi-channel acquisition and adaptive noise reduction system and method for rotating machinery vibration signals. This solves the problem that in coupled fault scenarios, strong background noise can overwhelm weak fault features, and single-channel data cannot provide sufficient spatial resolution, resulting in a high rate of missed diagnoses of rotating machinery vibration.
[0005] This invention is implemented as follows: a method for multi-channel acquisition and adaptive noise reduction of vibration signals from rotating machinery, the method comprising: Based on a multi-channel synchronous acquisition array, multi-channel vibration signals are acquired in real time. The multi-channel vibration signals are spatiotemporally aligned and processed to output spatiotemporally stamped multi-channel vibration signals. The system acquires multi-channel vibration signals after time-space stamp alignment, extracts real-time rotational speed and phase information from the multi-channel vibration signals based on cross-correlation analysis, and determines the equipment operating conditions based on the real-time rotational speed and phase information. Based on the equipment operating conditions, typical noise reduction strategies corresponding to real-time speed phase information are retrieved from the noise reduction strategy library. A noise reduction optimization function for real-time speed phase information is constructed. The ant colony algorithm is used to solve the noise reduction optimization function and output an adaptive noise reduction strategy. Based on the adaptive noise reduction strategy, the real-time speed phase information is adaptively denoised and the adaptive noise reduction information is output. The adaptive noise reduction signal is acquired, and the fault indication index is extracted from the adaptive noise reduction signal based on the pre-constructed vibration assessment model. The multi-channel fusion characterization result is then output.
[0006] Preferably, the method for spatiotemporal alignment processing of multi-channel vibration signals includes: The channel weights in the multi-channel system are determined based on principal component analysis. The channel with the maximum weight value is taken as the main channel. The vibration signal of the main channel is subjected to spectral analysis to identify the phase corresponding to the vibration signal. The phase of the main channel is used as the reference phase. Spectral analysis is performed on the vibration signals of non-main channels in multi-channel vibration signals to identify the corresponding phases of the reference phase in the non-main channel vibration signals, calculate the phase difference between the corresponding phase and the reference phase, and determine the time delay between the non-corresponding phase and the reference phase based on the phase difference. Using time delay as prior information, the vibration signals of non-main channels are timestamped to obtain multi-channel signals with timestamped alignment. The system identifies the acceleration information of the measurement points in the multi-channel vibration signal, maps the acceleration information of the measurement points to the same rotating cross section, identifies the optimal coordinate transformation matrix based on the modal guarantee criterion, and uses the optimal coordinate transformation matrix as a constraint to spatially align the multi-channel vibration signal, outputting the spatiotemporally aligned multi-channel vibration signal.
[0007] Preferably, the method for determining equipment operating conditions based on real-time rotational speed and phase information includes: Real-time rotational speed phase information is obtained, and the rotational frequency and harmonic components of the associated equipment are extracted based on the real-time rotational speed phase information. Statistical analysis is performed on the multi-channel vibration signals to determine the peak factor, spectral centroid, and background noise of the multi-channel vibration signals. The rotational frequency, harmonic components, peak factor, spectral centroid, and background noise of the associated equipment are used as the operating condition information set. Obtain a set of operating condition information, determine the speed operating condition index, load operating condition index, and environmental noise operating condition index based on the set of operating condition information, and determine the speed weighting coefficient, load weighting coefficient, and noise weighting coefficient based on the combined weighting method. The operating condition indicators of speed, load, and environmental noise are linearly weighted based on the speed weighting coefficient, load weighting coefficient, and noise weighting coefficient to determine the operating condition information value. The equipment operating condition corresponding to the operating condition information value is determined based on the support vector machine, and the equipment operating condition label is added to the corresponding multi-channel vibration signal.
[0008] Preferably, the method for retrieving typical noise reduction strategies corresponding to real-time rotational speed and phase information from the noise reduction strategy library based on equipment operating conditions includes: Real-time speed phase information is obtained, and the operating conditions of the equipment are identified based on the real-time speed phase information. Using the operating conditions of the equipment as an index, typical noise reduction strategies in the noise reduction strategy library are traversed, and the strategy similarity between the operating conditions of the equipment and the typical noise reduction strategies is determined based on the cosine similarity method. Determine whether the strategy similarity exceeds a preset similarity threshold; If the similarity between the equipment's operating conditions and typical noise reduction strategies exceeds a preset similarity threshold, a mapping relationship with the typical noise reduction strategy is established, and a mapping relationship queue is output. Load the mapping relationship queue, identify the typical denoising strategies associated with the mapping relationship queue, fuse the typical denoising strategies, and use the strategy similarity of the typical denoising strategies in the mapping relationship queue as prior information to assign weights to the typical denoising strategies, thereby obtaining the weighted denoising strategy fusion set.
[0009] Preferably, the method for solving the noise reduction optimization function using the ant colony algorithm includes: Real-time speed phase information is obtained, and multiple noise reduction objectives are defined for the noise reduction optimization function based on the real-time speed phase information. The noise reduction objectives are to maximize fault fidelity, minimize noise residue, and minimize signal distortion. The set of denoising strategies after weight labeling is used as the strategy weight constraint. The noise reduction optimization function is defined by combining multiple noise reduction objectives and strategy weight constraints. The noise reduction optimization function is expressed as:
[0010]
[0011] in, These are fault fidelity, noise residue, and signal distortion. These are the weighting coefficients for the fault fidelity item, noise residue item, and signal distortion item, respectively. Traverse the denoising strategy fusion set after weight labeling, encode the typical denoising strategy in the denoising strategy fusion set as ant search path, and use the typical denoising strategy encoding after weight labeling as heuristic information to initialize the pheromone concentration of the ant search path and generate the initial denoising strategy path. Obtain the initial noise reduction strategy path, and determine the transition probability from the current typical noise reduction strategy to the next typical noise reduction strategy based on the initial noise reduction strategy path. Construct the solution transition path based on the transition probability. The transition probability formula is expressed as:
[0012] in, Indicates time From the current typical noise reduction strategies Shift to the next typical noise reduction strategy The transition probability, Indicates the initial pheromone concentration. These are the pheromone weight parameters and the heuristic factor weight parameters, respectively. The number of candidate typical noise reduction strategies; Substitute the solution transfer path into the denoising optimization function to calculate the updated pheromone after iteration. Evaluate the quality of the updated pheromone solution after iteration, and take the optimal solution as the optimal solution for the current iteration. After the iteration ends, select the global optimal solution as the final adaptive denoising strategy.
[0013] Preferably, the vibration assessment model is based on a convolutional neural network (CNN) architecture and further includes an input layer and an output layer. An adaptive denoising layer and a time-frequency domain transformation layer are provided between the CNN and the input layer. A multi-channel fusion layer is provided between the CNN and the output layer. The CNN includes an embedding layer, a convolutional layer, and a fully connected layer. The multi-channel fusion layer is a Bayesian network model based on decision-level fusion. The time-frequency domain transformation layer is a long short-term memory network model based on an attention mechanism. The adaptive denoising layer is used to acquire an adaptive denoising signal, perform variational mode decomposition on the adaptive denoising signal to obtain a denoised reconstructed signal based on variational mode decomposition, and then pass the denoised reconstructed signal to the time-frequency domain transformation layer. The layer-switching method extracts time-domain, frequency-domain, and time-frequency-domain features from the denoised and reconstructed signal, and constructs a high-dimensional representation feature matrix based on these features. The embedding layer is used to vectorize the high-dimensional representation feature matrix, and performs convolution operations on the vectorized high-dimensional representation feature matrix based on multiple convolution kernels of the convolutional layer to extract fault indication features from the high-dimensional representation feature matrix. The fault indication features are then fully connected with the time-domain, frequency-domain, and time-frequency-domain features through a fully connected layer to obtain a fully connected feature set. The multi-channel fusion layer is used to obtain the fully connected feature set, and performs decision-level fusion on the fully connected feature set based on a Bayesian network to obtain the multi-channel fusion representation result and the corresponding global health status.
[0014] Preferably, the method for extracting fault indication indicators from adaptive noise reduction signals based on a pre-built vibration assessment model includes: The adaptive denoising signal is acquired, and variational mode decomposition is performed on the adaptive denoising signal to obtain the denoised reconstructed signal based on variational mode decomposition. The denoised reconstructed signal is then passed to the time-frequency domain transform layer. The constrained variational model of the denoised and reconstructed signal is expressed as:
[0015] in, These are the noise-reconstructed signal, the adaptive noise-reduced signal, and the center frequency of the adaptive noise-reduced signal, respectively. Indicates the number of modal components. These are the time derivative and the Dirac function, respectively. The time-frequency domain transform layer extracts time-domain features, frequency-domain features, and time-frequency domain features from the denoised and reconstructed signal, and constructs a high-dimensional representation feature matrix based on these features. The high-dimensional representation feature matrix is expressed as follows:
[0016] in, These are respectively high-dimensional representation feature matrix, time-domain feature, frequency-domain feature, and time-frequency-domain feature; The embedding layer vectorizes the high-dimensional representation feature matrix, and performs convolution operations on the vectorized high-dimensional representation feature matrix based on multiple convolution kernels of the convolution layer to extract the fault indication features of the high-dimensional representation feature matrix. The fault indication features are then fully connected with the time domain features, frequency domain features, and time-frequency domain features through a fully connected layer to obtain a fully connected feature set. The fault indication features are represented as follows:
[0017] in, express Fault indication features of each convolutional kernel output, These are the high-dimensional representation feature weight matrix, the high-dimensional representation feature vector matrix, and the bias vector, respectively. A multi-channel fusion layer acquires a fully connected feature set, and a decision-level fusion of the fully connected feature set is performed based on a Bayesian network to obtain the multi-channel fusion representation result and the corresponding global health status.
[0018] On the other hand, the present invention also provides a multi-channel acquisition and adaptive noise reduction system for rotating machinery vibration signals, the multi-channel acquisition and adaptive noise reduction system for rotating machinery vibration signals comprising: The spatiotemporal alignment module is based on a multi-channel synchronous acquisition array to acquire multi-channel vibration signals in real time, performs spatiotemporal alignment processing on the multi-channel vibration signals, and outputs spatiotemporally aligned multi-channel vibration signals. The operating condition determination module is used to acquire multi-channel vibration signals after time and space stamp alignment, extract real-time rotational speed and phase information from the multi-channel vibration signals based on cross-correlation analysis, and determine the operating condition of the equipment based on the real-time rotational speed and phase information. The adaptive noise reduction module retrieves typical noise reduction strategies corresponding to real-time speed phase information from the noise reduction strategy library based on the equipment's operating conditions, constructs a noise reduction optimization function for real-time speed phase information, solves the noise reduction optimization function using the ant colony algorithm, outputs an adaptive noise reduction strategy, and adaptively reduces noise for real-time speed phase information based on the adaptive noise reduction strategy, outputting adaptive noise reduction information. The feature fusion module is used to acquire adaptive noise reduction signals, extract fault indication indicators from the adaptive noise reduction signals based on a pre-built vibration assessment model, and output multi-channel fusion characterization results.
[0019] Preferably, the spatiotemporal alignment module includes: The reference signal determination unit determines the channel weights in the multi-channel based on the principal component analysis method, takes the channel corresponding to the maximum weight value as the main channel, performs spectral analysis on the vibration signal of the main channel, identifies the phase corresponding to the vibration signal, and takes the phase of the main channel as the reference phase. The time delay calculation unit is used to perform spectral analysis on the vibration signals of non-main channels in multi-channel vibration signals, identify the corresponding phase of the corresponding reference phase in the non-main channel vibration signals, calculate the phase difference between the corresponding phase and the reference phase, and determine the time delay between the non-corresponding phase and the reference phase based on the phase difference. The spatiotemporal alignment processing unit uses time delay as prior information to perform time stamp alignment on the vibration signals of non-main channels, obtaining multi-channel signals after time stamp alignment. It identifies the acceleration information of the measurement points of the multi-channel vibration signals, maps the acceleration information of the measurement points to the same rotating cross section, identifies the optimal coordinate transformation matrix based on the modal guarantee criterion, and uses the optimal coordinate transformation matrix as a constraint to perform spatial alignment on the multi-channel vibration signals, outputting multi-channel vibration signals after spatiotemporal alignment.
[0020] Compared with the prior art, the embodiments of this application have the following main advantages: In this embodiment of the invention, typical noise reduction strategies corresponding to real-time rotational speed and phase information in the noise reduction strategy library are retrieved based on the equipment operating conditions. The ant colony algorithm is used to find the optimal solution in the search space composed of typical noise reduction strategies. This enables adaptive noise reduction of multi-channel vibration signals collected under different operating conditions, ensuring that the noise-reduced signal is clean and useful. It solves the problem of strong noise drowning out weak fault features and improves the vibration assessment model's response to vibration fault identification of rotating machinery.
[0021] In this embodiment of the invention, when processing the spatiotemporal alignment of multi-channel vibration signals, Principal Component Analysis (PCA) is introduced to automatically select the main channel based on the contribution of vibration energy, avoiding the subjectivity of manually specifying references. Furthermore, through spectrum analysis and phase difference calculation, the time delay between each channel is accurately quantified, achieving sub-sampling point-level synchronization of signals on timestamps. This eliminates timing errors caused by differences in sensor wiring or trigger jitter. Simultaneously, the Modal Assurance Criterion (MAC) is introduced to measure the orthogonality of modal vectors to solve for the optimal coordinate transformation matrix. This projects each channel signal from its inherent sensor coordinate system and unifies it into the objective physical coordinate system of the device. The final output is not only a simple time-aligned signal, but also spatiotemporally unified vibration field data with clear physical meaning, enabling the vibration assessment model to fully utilize the temporal and spatial correlation of the signal.
[0022] In this embodiment of the invention, when determining the operating condition of the equipment based on real-time rotational speed and phase information, a comprehensive and representative set of operating condition information is constructed. The set of operating condition information not only includes the rotational frequency and harmonic components that directly reflect the mechanical motion state, but also incorporates multi-dimensional features such as peak factor, spectral centroid, and background noise that reflect the statistical characteristics of the signal. Thus, the operating condition description is no longer a single rotational speed reading, but a three-dimensional portrait that can comprehensively reflect the internal mechanical state of the equipment and the external working environment. Finally, through the decision model constructed by linear weighting and support vector machine, the complex multi-dimensional operating condition information can be condensed into a clear and tagged equipment operating condition.
[0023] In this embodiment of the invention, when retrieving typical noise reduction strategies corresponding to real-time rotational speed and phase information in the noise reduction strategy library based on the equipment operating conditions, the real-time identified equipment operating conditions are used as an index to accurately perform targeted searches in the noise reduction strategy library. Furthermore, the cosine similarity method can quantify the similarity between the current complex operating conditions and historical typical operating conditions, thereby quickly and accurately identifying the most likely effective candidate strategies. The introduction of a similarity-based weighted prior mechanism provides guidance for the optimization process of typical noise reduction strategies and provides an initial solution set for the subsequent ant colony algorithm optimization process. This avoids the optimization algorithm performing inefficient random searches in a vast and clueless solution space, enabling it to converge to a high-quality solution more quickly.
[0024] In this embodiment of the invention, when using the ant colony algorithm to solve the noise reduction optimization function, a noise reduction optimization function is constructed with the goal of maximizing fault fidelity, minimizing noise residue, and minimizing signal distortion. This ensures that the entire optimization process is centered on protecting weak fault features. Furthermore, by utilizing the powerful global optimization capability of the ant colony algorithm, a noise reduction strategy with the strongest overall performance under the given conditions is calculated and output in real time, rather than a fixed template. This improves the online adaptive capability for filtering and denoising multi-channel vibration signals, ensuring that a dynamic and optimal combination of noise reduction strategies can be obtained.
[0025] In this embodiment of the invention, when extracting fault indication indicators from the adaptive noise-reduced signal based on the pre-built vibration assessment model, the adaptive noise-reduced signal is processed by VMD to further remove residual noise and obtain a purer noise-reconstructed signal. Convolutional Neural Networks (CNN) can automatically learn complex feature patterns in the signal. Through deep learning, CNN can extract higher-level fault features, significantly improving fault diagnosis accuracy. Finally, through the fusion of Bayesian networks, the global health status of the equipment can be evaluated, providing not only diagnostic results of local fault features but also an overall assessment of the equipment's operating status. Attached Figure Description
[0026] Figure 1 A schematic diagram of the implementation process of multi-channel acquisition and adaptive noise reduction method for vibration signals of rotating machinery is shown.
[0027] Figure 2 A schematic diagram of the implementation process of the spatiotemporal alignment processing method for multi-channel vibration signals is shown.
[0028] Figure 3 A schematic diagram of the implementation process of a method for determining equipment operating conditions based on real-time rotational speed and phase information is shown.
[0029] Figure 4 This paper presents a schematic diagram illustrating the implementation process of a typical noise reduction strategy method that retrieves real-time speed and phase information from a noise reduction strategy library based on the equipment's operating conditions.
[0030] Figure 5 The diagram illustrates the implementation process of the ant colony algorithm for solving the noise reduction optimization function.
[0031] Figure 6 A schematic diagram of the implementation process of a method for extracting fault indication indicators from adaptive noise reduction signals based on a pre-built vibration assessment model is shown.
[0032] Figure 7 A schematic diagram of a multi-channel acquisition and adaptive noise reduction system for vibration signals of rotating machinery is shown. Detailed Implementation
[0033] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.
[0034] Currently, in coupled fault scenarios, strong background noise can overwhelm weak fault features, and single-channel data cannot provide sufficient spatial resolution, leading to a high rate of missed diagnoses of rotating machinery vibration. To address this issue, we propose a multi-channel acquisition and adaptive denoising system and method for rotating machinery vibration signals. In short, the method first performs spatiotemporal alignment processing on the multi-channel vibration signals, determines the equipment operating conditions based on real-time rotational speed and phase information, solves the denoising optimization function using an ant colony algorithm, and outputs an adaptive denoising strategy. Then, based on the adaptive denoising strategy, the real-time rotational speed and phase information is adaptively denoised. Finally, based on a pre-built vibration assessment model, fault indication indicators are extracted from the adaptively denoised signal, and the multi-channel fusion representation result is output. In this embodiment, typical denoising strategies corresponding to real-time rotational speed and phase information are retrieved from the denoising strategy library based on the equipment operating conditions, and the optimal solution is found in the search space composed of typical denoising strategies using an ant colony algorithm. This enables adaptive denoising of multi-channel vibration signals acquired under different operating conditions, ensuring that the denoised signal is clean and useful, solving the problem of strong noise overwhelming weak fault features, and improving the vibration assessment model's response to rotating machinery vibration fault identification.
[0035] This invention provides a method for multi-channel acquisition and adaptive noise reduction of vibration signals from rotating machinery. Figure 1 This diagram illustrates the implementation flow of a multi-channel acquisition and adaptive noise reduction method for rotating machinery vibration signals. The method specifically includes: S10, based on a multi-channel synchronous acquisition array, acquires multi-channel vibration signals in real time, performs spatiotemporal alignment processing on the multi-channel vibration signals, and outputs spatiotemporally aligned multi-channel vibration signals; In this embodiment of the invention, the multi-channel synchronous acquisition array is used to acquire vibration information of rotating machinery, including but not limited to IEPE accelerometers and integrated multi-channel data acquisition units, while rotating machinery includes but is not limited to motors, generators, gearboxes, and bearings.
[0036] S20: Obtain multi-channel vibration signals after time-space stamp alignment, extract real-time rotational speed phase information from the multi-channel vibration signals based on cross-correlation analysis, and determine the equipment operating conditions based on the real-time rotational speed phase information; S30: Based on the equipment operating conditions, retrieve the typical noise reduction strategies corresponding to the real-time speed phase information in the noise reduction strategy library, construct the noise reduction optimization function of the real-time speed phase information, use the ant colony algorithm to solve the noise reduction optimization function, output the adaptive noise reduction strategy, and adaptively reduce noise for the real-time speed phase information based on the adaptive noise reduction strategy, output the adaptive noise reduction information. S40: Acquire the adaptive noise reduction signal, extract fault indication indicators from the adaptive noise reduction signal based on the pre-built vibration assessment model, and output multi-channel fusion characterization results.
[0037] In this embodiment of the invention, typical noise reduction strategies corresponding to real-time rotational speed and phase information in the noise reduction strategy library are retrieved based on the equipment operating conditions. The ant colony algorithm is used to find the optimal solution in the search space composed of typical noise reduction strategies. This enables adaptive noise reduction of multi-channel vibration signals collected under different operating conditions, ensuring that the noise-reduced signal is clean and useful. It solves the problem of strong noise drowning out weak fault features and improves the vibration assessment model's response to vibration fault identification of rotating machinery.
[0038] This invention provides a method for spatiotemporal alignment processing of multi-channel vibration signals. Figure 2 This diagram illustrates the implementation flow of a spatiotemporal alignment processing method for multi-channel vibration signals. The method specifically includes: S101, based on principal component analysis, the channel weights in the multi-channel array are determined. The channel corresponding to the maximum weight value is selected as the master channel. Spectral analysis is performed on the vibration signal of the master channel to identify the corresponding phase, and the phase of the master channel is used as the reference phase. In this embodiment of the invention, the channel weights are determined by principal component analysis, and the channel corresponding to the maximum weight value is selected as the master channel, ensuring the stability and reliability of the reference phase. This allows the vibration signals of non-master channels to be accurately aligned with the master channel, reducing signal distortion caused by differences between channels.
[0039] S102, perform spectrum analysis on the vibration signals of non-main channels in the multi-channel vibration signals, identify the corresponding phase of the corresponding reference phase in the non-main channel vibration signals, calculate the phase difference between the corresponding phase and the reference phase, determine the time delay between the non-corresponding phase and the reference phase based on the phase difference, and the phase difference calculation and timestamp alignment can accurately compensate for the time delay between each channel, ensuring the consistency of the multi-channel signals in time. S103, using time delay as prior information, timestamps are aligned on the vibration signals of non-main channels to obtain timestamp-aligned multi-channel signals; S104 identifies the acceleration information of the measurement points of the multi-channel vibration signal, maps the acceleration information of the measurement points to the same rotating cross section, identifies the optimal coordinate transformation matrix based on the modal guarantee criterion, and uses the optimal coordinate transformation matrix as a constraint to perform spatial alignment of the multi-channel vibration signal, outputting the spatiotemporally aligned multi-channel vibration signal. By mapping the acceleration information of the measurement points and identifying the optimal coordinate transformation matrix based on the modal guarantee criterion, the spatial alignment of the multi-channel signal is realized.
[0040] In this embodiment of the invention, when processing the spatiotemporal alignment of multi-channel vibration signals, principal component analysis (PCA) is introduced to automatically select the main channel based on the vibration energy contribution, avoiding the subjectivity of manually specifying references. Furthermore, through spectrum analysis and phase difference calculation, the time delay between each channel is accurately quantified, achieving subsampling-level synchronization of signals at timestamps. This eliminates timing errors caused by differences in sensor wiring or trigger jitter. Simultaneously, the Modal Assurance Criterion (MAC) is introduced to measure the orthogonality of modal vectors to solve for the optimal coordinate transformation matrix. This projects each channel signal from its inherent sensor coordinate system and unifies it into the objective physical coordinate system of the device. The final output is not only a simple time-aligned signal, but also spatiotemporally unified vibration field data with clear physical meaning, enabling the vibration assessment model to fully utilize the temporal and spatial correlation of the signal.
[0041] This invention provides a method for determining equipment operating conditions based on real-time rotational speed and phase information. Figure 3 This diagram illustrates the implementation flow of a method for determining equipment operating conditions based on real-time speed and phase information. The method specifically includes: S201: Obtain real-time rotational speed and phase information; extract the rotational frequency and harmonic components of the associated equipment based on the real-time rotational speed and phase information; perform statistical analysis on the multi-channel vibration signals to determine the peak factor, spectral centroid, and background noise of the multi-channel vibration signals; use the rotational frequency, harmonic components, peak factor, spectral centroid, and background noise of the associated equipment as the operating condition information set; by extracting the operating condition information set, the characteristic changes of the equipment under different operating conditions can be fully reflected, providing sufficient basis for subsequent operating condition identification. S202, acquire the operating condition information set, determine the speed operating condition index, load operating condition index, and environmental noise operating condition index based on the operating condition information set, and determine the speed weighting coefficient, load weighting coefficient, and noise weighting coefficient based on the combined weighting method. S203 linearly weights the speed operating condition index, load operating condition index, and environmental noise operating condition index based on the speed weighting coefficient, load weighting coefficient, and noise weighting coefficient to determine the operating condition information value. It then determines the equipment operating condition corresponding to the operating condition information value based on the support vector machine. In particular, the use of support vector machine (SVM) to classify the linearly weighted operating condition information value can accurately identify the operating condition of the equipment and add the equipment operating condition label to the corresponding multi-channel vibration signal.
[0042] In this embodiment of the invention, when determining the operating condition of the equipment based on real-time rotational speed and phase information, a comprehensive and representative set of operating condition information is constructed. The set of operating condition information not only includes the rotational frequency and harmonic components that directly reflect the mechanical motion state, but also incorporates multi-dimensional features such as peak factor, spectral centroid, and background noise that reflect the statistical characteristics of the signal. Thus, the operating condition description is no longer a single rotational speed reading, but a three-dimensional portrait that can comprehensively reflect the internal mechanical state of the equipment and the external working environment. Finally, through the decision model constructed by linear weighting and support vector machine, the complex multi-dimensional operating condition information can be condensed into a clear and tagged equipment operating condition.
[0043] This invention provides a method for retrieving typical noise reduction strategies corresponding to real-time speed and phase information from a noise reduction strategy library based on equipment operating conditions. Figure 4 This document illustrates a flowchart illustrating the implementation of a typical noise reduction strategy method that retrieves real-time speed and phase information from a noise reduction strategy library based on equipment operating conditions. Specifically, this method includes: S301: Obtain real-time speed phase information, identify equipment operating conditions based on real-time speed phase information, traverse typical noise reduction strategies in the noise reduction strategy library using the equipment operating conditions as an index, and determine the strategy similarity between the equipment operating conditions and typical noise reduction strategies based on the cosine similarity method. By obtaining speed phase information in real time and identifying equipment operating conditions, the most suitable noise reduction strategy for the current operating conditions can be dynamically selected from the noise reduction strategy library, thereby avoiding the inapplicability of fixed noise reduction strategies under different operating conditions and improving the targeting of noise reduction strategies. S302, Determine whether the strategy similarity exceeds the preset similarity threshold; S303, If the similarity between the equipment operating conditions and typical noise reduction strategies exceeds a preset similarity threshold, establish a mapping relationship with the typical noise reduction strategies and output the mapping relationship queue. S304, Load the mapping relationship queue, identify the typical denoising strategies associated with the mapping relationship queue, fuse the typical denoising strategies, and use the strategy similarity of the typical denoising strategies in the mapping relationship queue as prior information to assign weights to the typical denoising strategies, thereby obtaining the weighted denoising strategy fusion set.
[0044] In this embodiment of the invention, when retrieving typical noise reduction strategies corresponding to real-time rotational speed and phase information in the noise reduction strategy library based on the equipment operating conditions, the real-time identified equipment operating conditions are used as an index to accurately perform targeted searches in the noise reduction strategy library. Furthermore, the cosine similarity method can quantify the similarity between the current complex operating conditions and historical typical operating conditions, thereby quickly and accurately identifying the most likely effective candidate strategies. The introduction of a similarity-based weighted prior mechanism provides guidance for the optimization process of typical noise reduction strategies and provides an initial solution set for the subsequent ant colony algorithm optimization process. This avoids the optimization algorithm performing inefficient random searches in a vast and clueless solution space, enabling it to converge to a high-quality solution more quickly.
[0045] This invention provides a method for solving the noise reduction optimization function using the ant colony algorithm. Figure 5 This diagram illustrates the implementation flow of a method for solving the noise reduction optimization function using the ant colony algorithm. The method specifically includes: S401, Obtain real-time speed phase information, define multiple noise reduction objectives for the noise reduction optimization function based on the real-time speed phase information, wherein the noise reduction objectives are to maximize fault fidelity, minimize noise residue, minimize signal distortion, and use the weighted noise reduction strategy fusion set as the strategy weight constraint, and define the noise reduction optimization function by combining the multiple noise reduction objectives and the strategy weight constraint. The noise reduction optimization function is expressed as:
[0046]
[0047] in, These are fault fidelity, noise residue, and signal distortion. These are the weighting coefficients for the fault fidelity item, noise residue item, and signal distortion item, respectively.
[0048] S402, traverse the denoising strategy fusion set after weight labeling, encode the typical denoising strategy in the denoising strategy fusion set as ant search path, and use the typical denoising strategy encoding after weight labeling as heuristic information to initialize the pheromone concentration of the ant search path and generate the initial denoising strategy path. S403, obtain the initial noise reduction strategy path, and determine the transition probability from the current typical noise reduction strategy to the next typical noise reduction strategy based on the initial noise reduction strategy path, and construct the solution transition path based on the transition probability; The transition probability formula is expressed as:
[0049] in, Indicates time From the current typical noise reduction strategies Shift to the next typical noise reduction strategy The transition probability, Indicates the initial pheromone concentration. These are the pheromone weight parameters and the heuristic factor weight parameters, respectively. The number of candidate typical noise reduction strategies; S404: Substitute the solution transfer path into the denoising optimization function to calculate the updated pheromone after iteration. Evaluate the quality of the updated pheromone solution after iteration, and take the optimal solution as the optimal solution for the current iteration. After the iteration ends, select the global optimal solution as the final adaptive denoising strategy.
[0050] In this embodiment of the invention, when using the ant colony algorithm to solve the noise reduction optimization function, a noise reduction optimization function is constructed with the goal of maximizing fault fidelity, minimizing noise residue, and minimizing signal distortion. This ensures that the entire optimization process is centered on protecting weak fault features. Furthermore, by utilizing the powerful global optimization capability of the ant colony algorithm, a noise reduction strategy with the strongest overall performance under the given conditions is calculated and output in real time, rather than a fixed template. This improves the online adaptive capability for filtering and denoising multi-channel vibration signals, ensuring that a dynamic and optimal combination of noise reduction strategies can be obtained.
[0051] In this embodiment of the invention, the vibration assessment model is based on a convolutional neural network (CNN) architecture and includes an input layer and an output layer. An adaptive denoising layer and a time-frequency domain transformation layer are provided between the CNN and the input layer. A multi-channel fusion layer is provided between the CNN and the output layer. The CNN includes an embedding layer, a convolutional layer, and a fully connected layer. The multi-channel fusion layer is a Bayesian network model based on decision-level fusion, and the time-frequency domain transformation layer is a long short-term memory network model based on an attention mechanism. The adaptive denoising layer is used to acquire an adaptive denoising signal, perform variational mode decomposition on the adaptive denoising signal to obtain a denoised reconstructed signal based on variational mode decomposition, and then pass the denoised reconstructed signal to the time-frequency domain transformation layer. The domain transformation layer extracts time-domain, frequency-domain, and time-frequency-domain features based on the denoised and reconstructed signal, and constructs a high-dimensional representation feature matrix based on these features. The embedding layer is used to vectorize the high-dimensional representation feature matrix, and performs convolution operations on the vectorized high-dimensional representation feature matrix based on multiple convolution kernels of the convolutional layer to extract fault indication features from the high-dimensional representation feature matrix. The fault indication features are then fully connected with the time-domain, frequency-domain, and time-frequency-domain features through a fully connected layer to obtain a fully connected feature set. The multi-channel fusion layer is used to obtain the fully connected feature set, and performs decision-level fusion on the fully connected feature set based on a Bayesian network to obtain the multi-channel fusion representation result and the corresponding global health status.
[0052] This invention provides a method for extracting fault indication indicators from adaptive noise reduction signals based on a pre-built vibration assessment model. Figure 6 The diagram illustrates the implementation flow of a method for extracting fault indication indicators from adaptive noise-reduced signals based on a pre-built vibration assessment model. Specifically, this method includes: S501: Acquire the adaptive denoising signal, perform variational mode decomposition on the adaptive denoising signal to obtain the denoised reconstructed signal based on variational mode decomposition, and then pass the denoised reconstructed signal to the time-frequency domain transform layer. Variational Mode Decomposition (VMD) can adaptively decompose complex vibration signals into several modal components (IMFs) with specific center frequencies and finite bandwidths. This achieves more refined denoising than traditional filtering, resulting in a denoised reconstructed signal with an extremely high signal-to-noise ratio. The constrained variational model of the denoised and reconstructed signal is expressed as:
[0053] in, These are the noise-reconstructed signal, the adaptive noise-reduced signal, and the center frequency of the adaptive noise-reduced signal, respectively. Indicates the number of modal components. These are the time derivative and the Dirac function, respectively. S502, the time-frequency domain transform layer extracts time-domain features, frequency-domain features, and time-frequency domain features based on the denoised and reconstructed signal, and constructs a high-dimensional representation feature matrix based on the time-domain features, frequency-domain features, and time-frequency domain features. In this process, the heterogeneous time-domain features, frequency-domain features, and time-frequency domain features are concatenated into a high-dimensional representation feature matrix, providing a structured and information-dense input for the convolutional neural network (CNN), enabling the CNN to effectively mine deep-level, cross-dimensional correlation patterns from it; The high-dimensional representation feature matrix is expressed as follows:
[0054] in, These are respectively high-dimensional representation feature matrix, time-domain feature, frequency-domain feature, and time-frequency-domain feature; S503, the embedding layer vectorizes the high-dimensional representation feature matrix, and performs convolution operation on the vectorized high-dimensional representation feature matrix based on multiple convolution kernels of the convolution layer to extract the fault indication features of the high-dimensional representation feature matrix. The fault indication features are then fully connected with the time domain features, frequency domain features, and time-frequency domain features through a fully connected layer to obtain a fully connected feature set. The fault indication features are represented as follows:
[0055] in, express Fault indication features of each convolutional kernel output, These are the high-dimensional representation feature weight matrix, the high-dimensional representation feature vector matrix, and the bias vector, respectively. S504: The multi-channel fusion layer obtains the fully connected feature set, and performs decision-level fusion on the fully connected feature set based on a Bayesian network to obtain the multi-channel fusion representation result and the corresponding global health status.
[0056] In this embodiment of the invention, Bayesian networks can explicitly model the probabilistic dependencies between different features and between features and health states. It can not only provide a hard classification label, but also output a probabilistic global health state.
[0057] In this embodiment of the invention, when extracting fault indication indicators from the adaptive noise reduction signal based on the pre-built vibration assessment model, the adaptive noise reduction signal is processed by VMD to further remove residual noise and obtain a purer noise reduction and reconstruction signal. The convolutional neural network (CNN) can automatically learn complex feature patterns in the signal. Through deep learning, CNN can extract higher-level fault features, significantly improving fault diagnosis accuracy. Finally, through the fusion of Bayesian networks, the global health status of the equipment can be evaluated, providing not only diagnostic results of local fault features but also an overall assessment of the equipment's operating status.
[0058] On the other hand, embodiments of the present invention also provide a multi-channel acquisition and adaptive noise reduction system for rotating machinery vibration signals. Figure 7 A schematic diagram of a multi-channel acquisition and adaptive noise reduction system for vibration signals of rotating machinery is shown. The system specifically includes: The spatiotemporal alignment module 100 is based on a multi-channel synchronous acquisition array to acquire multi-channel vibration signals in real time, performs spatiotemporal alignment processing on the multi-channel vibration signals, and outputs spatiotemporally aligned multi-channel vibration signals. The spatiotemporal alignment module 100 includes: The reference signal determination unit 110 determines the channel weights in the multi-channel based on the principal component analysis method, takes the channel corresponding to the maximum weight value as the main channel, performs spectral analysis on the vibration signal of the main channel, identifies the phase corresponding to the vibration signal, and takes the phase of the main channel as the reference phase. The time delay calculation unit 120 is used to perform spectral analysis on the vibration signal of the non-main channel in the multi-channel vibration signal, identify the corresponding phase of the corresponding reference phase in the non-main channel vibration signal, calculate the phase difference between the corresponding phase and the reference phase, and determine the time delay between the non-corresponding phase and the reference phase based on the phase difference. The spatiotemporal alignment processing unit 130 uses time delay as prior information to perform time stamp alignment on the vibration signals of non-main channels, obtaining multi-channel signals after time stamp alignment. It identifies the acceleration information of the measurement points of the multi-channel vibration signals, maps the acceleration information of the measurement points to the same rotating cross section, identifies the optimal coordinate transformation matrix based on the modal guarantee criterion, and uses the optimal coordinate transformation matrix as a constraint to perform spatial alignment on the multi-channel vibration signals, outputting the spatiotemporally aligned multi-channel vibration signals.
[0059] The operating condition determination module 200 is used to acquire multi-channel vibration signals after time and space stamp alignment, extract real-time rotational speed and phase information from the multi-channel vibration signals based on cross-correlation analysis, and determine the operating condition of the equipment based on the real-time rotational speed and phase information. The adaptive noise reduction module 300 retrieves typical noise reduction strategies corresponding to real-time speed phase information from the noise reduction strategy library based on the equipment's operating conditions, constructs a noise reduction optimization function for real-time speed phase information, solves the noise reduction optimization function using the ant colony algorithm, outputs an adaptive noise reduction strategy, and adaptively reduces noise for real-time speed phase information based on the adaptive noise reduction strategy, outputting adaptive noise reduction information. The feature fusion module 400 is used to acquire the adaptive noise reduction signal, extract fault indication indicators from the adaptive noise reduction signal based on the pre-built vibration assessment model, and output multi-channel fusion characterization results.
[0060] In summary, this invention provides a multi-channel acquisition and adaptive noise reduction system and method for rotating machinery vibration signals. In the embodiments of this invention, typical noise reduction strategies corresponding to real-time rotational speed and phase information in the noise reduction strategy library are retrieved based on the equipment operating conditions. The optimal solution is found in the search space composed of typical noise reduction strategies using the ant colony algorithm. This enables adaptive noise reduction of multi-channel vibration signals acquired under different operating conditions, ensuring that the noise-reduced signal is clean and useful. It solves the problem of strong noise drowning out weak fault features and improves the vibration assessment model's response to rotating machinery vibration fault identification.
[0061] It should be noted that, for the sake of simplicity, the foregoing embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to the present invention. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.
[0062] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on these embodiments, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art can still combine, add, delete, or otherwise adjust the features of the various embodiments of the present invention according to the circumstances without conflict or creative effort, thereby obtaining different technical solutions that do not fundamentally depart from the concept of the present invention. These technical solutions also fall within the scope of protection of the present invention.
Claims
1. A method for multi-channel acquisition and adaptive noise reduction of vibration signals from rotating machinery, characterized in that, The method includes: Based on a multi-channel synchronous acquisition array, multi-channel vibration signals are acquired in real time. The multi-channel vibration signals are spatiotemporally aligned and processed to output spatiotemporally stamped multi-channel vibration signals. The system acquires multi-channel vibration signals after time-space stamp alignment, extracts real-time rotational speed and phase information from the multi-channel vibration signals based on cross-correlation analysis, and determines the equipment operating conditions based on the real-time rotational speed and phase information. Based on the equipment operating conditions, typical noise reduction strategies corresponding to real-time speed phase information are retrieved from the noise reduction strategy library. A noise reduction optimization function for real-time speed phase information is constructed. The ant colony algorithm is used to solve the noise reduction optimization function and output an adaptive noise reduction strategy. Based on the adaptive noise reduction strategy, the real-time speed phase information is adaptively denoised and the adaptive noise reduction information is output. The adaptive noise reduction signal is acquired, and the fault indication index is extracted from the adaptive noise reduction signal based on the pre-constructed vibration assessment model. The multi-channel fusion characterization result is then output.
2. The method for multi-channel acquisition and adaptive noise reduction of rotating machinery vibration signals as described in claim 1, characterized in that: The method for spatiotemporal alignment processing of multi-channel vibration signals includes: The channel weights in the multi-channel system are determined based on principal component analysis. The channel with the maximum weight value is taken as the main channel. The vibration signal of the main channel is subjected to spectral analysis to identify the phase corresponding to the vibration signal. The phase of the main channel is used as the reference phase. Spectral analysis is performed on the vibration signals of non-main channels in multi-channel vibration signals to identify the corresponding phases of the reference phase in the non-main channel vibration signals, calculate the phase difference between the corresponding phase and the reference phase, and determine the time delay between the non-corresponding phase and the reference phase based on the phase difference. Using time delay as prior information, the vibration signals of non-main channels are timestamped to obtain multi-channel signals with timestamped alignment. The system identifies the acceleration information of the measurement points in the multi-channel vibration signal, maps the acceleration information of the measurement points to the same rotating cross section, identifies the optimal coordinate transformation matrix based on the modal guarantee criterion, and uses the optimal coordinate transformation matrix as a constraint to spatially align the multi-channel vibration signal, outputting the spatiotemporally aligned multi-channel vibration signal.
3. The method for multi-channel acquisition and adaptive noise reduction of rotating machinery vibration signals as described in claim 1, characterized in that: The method for determining equipment operating conditions based on real-time rotational speed and phase information includes: Real-time rotational speed phase information is obtained, and the rotational frequency and harmonic components of the associated equipment are extracted based on the real-time rotational speed phase information. Statistical analysis is performed on the multi-channel vibration signals to determine the peak factor, spectral centroid, and background noise of the multi-channel vibration signals. The rotational frequency, harmonic components, peak factor, spectral centroid, and background noise of the associated equipment are used as the operating condition information set. Obtain a set of operating condition information, determine the speed operating condition index, load operating condition index, and environmental noise operating condition index based on the set of operating condition information, and determine the speed weighting coefficient, load weighting coefficient, and noise weighting coefficient based on the combined weighting method. The operating condition indicators of speed, load, and environmental noise are linearly weighted based on the speed weighting coefficient, load weighting coefficient, and noise weighting coefficient to determine the operating condition information value. The equipment operating condition corresponding to the operating condition information value is determined based on the support vector machine, and the equipment operating condition label is added to the corresponding multi-channel vibration signal.
4. The method for multi-channel acquisition and adaptive noise reduction of rotating machinery vibration signals as described in claim 1, characterized in that: The method for retrieving typical noise reduction strategies corresponding to real-time speed and phase information from the noise reduction strategy library based on equipment operating conditions includes: Real-time speed phase information is obtained, and the operating conditions of the equipment are identified based on the real-time speed phase information. Using the operating conditions of the equipment as an index, typical noise reduction strategies in the noise reduction strategy library are traversed, and the strategy similarity between the operating conditions of the equipment and the typical noise reduction strategies is determined based on the cosine similarity method. Determine whether the strategy similarity exceeds a preset similarity threshold; If the similarity between the equipment's operating conditions and typical noise reduction strategies exceeds a preset similarity threshold, a mapping relationship with the typical noise reduction strategy is established, and a mapping relationship queue is output. Load the mapping relationship queue, identify the typical denoising strategies associated with the mapping relationship queue, fuse the typical denoising strategies, and use the strategy similarity of the typical denoising strategies in the mapping relationship queue as prior information to assign weights to the typical denoising strategies, thereby obtaining the weighted denoising strategy fusion set.
5. The method for multi-channel acquisition and adaptive noise reduction of rotating machinery vibration signals as described in claim 4, characterized in that: The method for solving the noise reduction optimization function using the ant colony algorithm includes: Real-time speed phase information is obtained, and multiple noise reduction objectives are defined for the noise reduction optimization function based on the real-time speed phase information. The noise reduction objectives are to maximize fault fidelity, minimize noise residue, and minimize signal distortion. The denoising strategy fusion set after weight labeling is used as the strategy weight constraint. The noise reduction optimization function is defined by combining the multiple noise reduction objectives and the strategy weight constraint.
6. The method for multi-channel acquisition and adaptive noise reduction of rotating machinery vibration signals as described in claim 5, characterized in that: The method for solving the noise reduction optimization function using the ant colony algorithm also includes: Traverse the denoising strategy fusion set after weight labeling, encode the typical denoising strategy in the denoising strategy fusion set as ant search path, and use the typical denoising strategy encoding after weight labeling as heuristic information to initialize the pheromone concentration of the ant search path and generate the initial denoising strategy path. Obtain the initial noise reduction strategy path, and determine the transition probability from the current typical noise reduction strategy to the next typical noise reduction strategy based on the initial noise reduction strategy path. Construct the solution transition path based on the transition probability. Substitute the solution transfer path into the denoising optimization function to calculate the updated pheromone after iteration. Evaluate the quality of the updated pheromone solution after iteration, and take the optimal solution as the optimal solution for the current iteration. After the iteration ends, select the global optimal solution as the final adaptive denoising strategy.
7. The method for multi-channel acquisition and adaptive noise reduction of rotating machinery vibration signals as described in claim 1, characterized in that: The vibration assessment model is based on a convolutional neural network architecture and includes an input layer and an output layer. An adaptive noise reduction layer and a time-frequency domain transformation layer are set between the convolutional neural network and the input layer. A multi-channel fusion layer is set between the convolutional neural network and the output layer. The convolutional neural network includes an embedding layer, a convolutional layer, and a fully connected layer. The multi-channel fusion layer is a Bayesian network model based on decision-level fusion, and the time-frequency domain transformation layer is a long short-term memory network model based on an attention mechanism.
8. The method for multi-channel acquisition and adaptive noise reduction of rotating machinery vibration signals as described in claim 7, characterized in that: The method for extracting fault indication indicators from adaptive noise reduction signals based on a pre-built vibration assessment model includes: The adaptive denoising signal is acquired, and variational mode decomposition is performed on the adaptive denoising signal to obtain the denoised reconstructed signal based on variational mode decomposition. The denoised reconstructed signal is then passed to the time-frequency domain transform layer. The time-frequency domain transform layer extracts time-domain features, frequency-domain features, and time-frequency domain features from the denoised and reconstructed signal, and constructs a high-dimensional representation feature matrix based on these features. The embedding layer vectorizes the high-dimensional representation feature matrix, and performs convolution operations on the vectorized high-dimensional representation feature matrix based on multiple convolution kernels of the convolution layer to extract the fault indication features of the high-dimensional representation feature matrix. The fault indication features are then fully connected with the time domain features, frequency domain features, and time-frequency domain features through a fully connected layer to obtain a fully connected feature set. A multi-channel fusion layer acquires a fully connected feature set, and a decision-level fusion of the fully connected feature set is performed based on a Bayesian network to obtain the multi-channel fusion representation result and the corresponding global health status.
9. A multi-channel acquisition and adaptive noise reduction system for rotating machinery vibration signals, used to implement the multi-channel acquisition and adaptive noise reduction method for rotating machinery vibration signals as described in any one of claims 1-8, characterized in that: The multi-channel acquisition and adaptive noise reduction system for rotating machinery vibration signals includes: The spatiotemporal alignment module is based on a multi-channel synchronous acquisition array to acquire multi-channel vibration signals in real time, performs spatiotemporal alignment processing on the multi-channel vibration signals, and outputs spatiotemporally aligned multi-channel vibration signals. The operating condition determination module is used to acquire multi-channel vibration signals after time and space stamp alignment, extract real-time rotational speed and phase information from the multi-channel vibration signals based on cross-correlation analysis, and determine the operating condition of the equipment based on the real-time rotational speed and phase information. The adaptive noise reduction module retrieves typical noise reduction strategies corresponding to real-time speed phase information from the noise reduction strategy library based on the equipment's operating conditions, constructs a noise reduction optimization function for real-time speed phase information, solves the noise reduction optimization function using the ant colony algorithm, outputs an adaptive noise reduction strategy, and adaptively reduces noise for real-time speed phase information based on the adaptive noise reduction strategy, outputting adaptive noise reduction information. The feature fusion module is used to acquire adaptive noise reduction signals, extract fault indication indicators from the adaptive noise reduction signals based on a pre-built vibration assessment model, and output multi-channel fusion characterization results.
10. The multi-channel acquisition and adaptive noise reduction system for rotating machinery vibration signals as described in claim 9, characterized in that: The spatiotemporal alignment module includes: The reference signal determination unit determines the channel weights in the multi-channel based on the principal component analysis method, takes the channel corresponding to the maximum weight value as the main channel, performs spectral analysis on the vibration signal of the main channel, identifies the phase corresponding to the vibration signal, and takes the phase of the main channel as the reference phase. The time delay calculation unit is used to perform spectral analysis on the vibration signals of non-main channels in multi-channel vibration signals, identify the corresponding phase of the corresponding reference phase in the non-main channel vibration signals, calculate the phase difference between the corresponding phase and the reference phase, and determine the time delay between the non-corresponding phase and the reference phase based on the phase difference. The spatiotemporal alignment processing unit uses time delay as prior information to perform time stamp alignment on the vibration signals of non-main channels, obtaining multi-channel signals after time stamp alignment. It identifies the acceleration information of the measurement points of the multi-channel vibration signals, maps the acceleration information of the measurement points to the same rotating cross section, identifies the optimal coordinate transformation matrix based on the modal guarantee criterion, and uses the optimal coordinate transformation matrix as a constraint to perform spatial alignment on the multi-channel vibration signals, outputting multi-channel vibration signals after spatiotemporal alignment.