Methods, systems, media, and equipment for noise reduction of weak fault signals under strong background noise in aero-engines

Abstract

This invention relates to the field of acoustic emission signal feature enhancement for aero-engines, and discloses a method, system, medium, and device for denoising weak fault signals under strong background noise in aero-engines. The method includes: acquiring strong background noise signals under normal operating conditions of the aero-engine; performing wavelet packet decomposition and reconstruction to obtain multiple frequency band component signals; constructing a corresponding autoencoder for each frequency band component signal, and using the frequency band component signal as input to perform unsupervised training on the autoencoder; performing wavelet packet decomposition and reconstruction on the acoustic emission signal to be processed to obtain multiple frequency band component signals to be processed; inputting each frequency band component signal to be processed into the corresponding trained autoencoder for reconstruction; calculating the reconstruction error between the input and output signals of each autoencoder; determining whether each frequency band component signal contains outliers based on the reconstruction error; reconstructing the frequency band component signals containing outliers to obtain the denoised fault signal.

Description

Technical Field

[0001] This invention relates to the field of enhancing the acoustic emission signal characteristics of air engines, and in particular to a method, system, medium, and device for noise reduction of weak fault signals under strong background noise of air engines. Background Technology

[0002] In modern aircraft, the energy and power system, centered on the aero-engine, is the "heart" of the aircraft. The operating status of the engine directly affects the safety and economic efficiency of the entire aircraft. As the most critical power component of an aircraft, the aero-engine, due to its complex mechanical structure and harsh operating environment such as high temperature and high pressure, is more prone to mechanical damage to its internal components after a period of use. This will reduce operating efficiency and even lead to the failure of the entire unit, causing huge economic losses. Real-time monitoring of its operating status and timely and accurate diagnosis and prediction of fault information are of great research significance for ensuring flight safety. Traditional fault diagnosis techniques, such as vibration monitoring, gas path analysis and monitoring, lubricating oil monitoring, and electrostatic monitoring, cannot detect the early and subtle characteristics of engine faults in the initial stage in real time. However, engine structural health monitoring methods based on acoustic emission technology have significant advantages in dealing with such problems.

[0003] However, due to the complex structure and harsh operating conditions of aero engines, the acoustic emission signals from various components are coupled with each other. The acquired signals are often under strong background noise, making it impossible to effectively extract fault characteristic information, ultimately leading to difficulty in identifying engine faults. Strong background noise refers to the superposition of noise from all external environments and systems except the target sound source; it is characterized by high absolute amplitude and high energy. A large amount of information carried by the fault signal will be significantly distorted under the influence of strong background noise. Currently, the most commonly used turbofan engines in civil aircraft have their main noise sources from four main components: fan / compressor noise, combustion noise, turbine noise, and jet noise. During engine operation, the rotor and stator interfere with each other, forming unsteady, periodically changing aerodynamic forces; this noise constitutes discrete single-tone noise. Furthermore, during the flow of air and fuel gas inside the engine, due to viscosity and the intense combustion process, the airflow exhibits strong turbulent pulsations. Therefore, engine noise also contains strong random broadband noise. It is evident that the composition of strong background noise in aero engines is extremely complex. Achieving intelligent diagnosis of aero-engine faults under conditions of strong background noise is a challenging problem.

[0004] To address this challenge, numerous studies have discussed the interference of background noise on the original acoustic emission signal. Traditional signal analysis methods mainly include adaptive noise cancellation, digital filtering, and discrete statistical averaging, generally used for signal preprocessing. However, in practical applications, the acquired signals are mostly non-stationary and time-varying, making it difficult to effectively identify high-frequency details using only traditional methods. With the development of computer technology, various advanced signal processing techniques have been applied to acoustic emission signal processing. These include kurtosis spectrum analysis, wavelet transform, wavelet entropy, independent component analysis, empirical mode decomposition, wavelet packet decomposition, and deep learning. However, traditional methods still require complex signal processing procedures, demanding extensive signal processing knowledge from personnel. Furthermore, they necessitate analysis of various aero-engine fault mechanisms and prior knowledge of the corresponding time-spectrum characteristics of each fault. This makes it difficult to extract early fault features from aero-engine acoustic emission signals subjected to multiple internal and external excitation interferences. Simultaneously, the generalization of deep learning-based feature extraction methods remains a challenging problem. For noise types frequently encountered in real-world environments but invisible in training data, deep learning-based methods offer no advantage over traditional methods. Summary of the Invention

[0005] To address the aforementioned problems, the present invention aims to provide a method, system, medium, and device for denoising weak fault signals under strong background noise in aero-engines. It utilizes the powerful feature extraction and outlier identification capabilities of wavelet packets and autoencoders to solve the problem of acoustic emission signal denoising for early-stage faults in aero-engines under strong background noise.

[0006] To achieve the above objectives, in a first aspect, the present invention adopts the following technical solution: a method for denoising weak fault signals under strong background noise in an aero-engine, comprising: collecting acoustic emission signals under normal operating conditions of the aero-engine as strong background noise signals; performing wavelet packet decomposition and reconstruction on the strong background noise signals to obtain multiple frequency band component signals; constructing a corresponding autoencoder for each frequency band component signal, and using the frequency band component signal as input to perform unsupervised training on the autoencoder; introducing a shared feature extraction mechanism during the training process for each frequency band autoencoder, enabling the autoencoder to learn and reconstruct the strong background noise signal under that frequency band; wherein, the shared feature extraction mechanism inputs different frequency band component signals into the shared encoder. Feature extraction is performed in the device structure, and each frequency band is reconstructed through the corresponding decoder, so that the model can learn cross-frequency band correlation information while preserving the feature differences of each frequency band. Acoustic emission signals containing fault information are collected, and wavelet packet decomposition and reconstruction are performed on the acoustic emission signals to be processed to obtain multiple frequency band component signals to be processed. Each frequency band component signal to be processed is input into the corresponding trained autoencoder, which reconstructs it and calculates the reconstruction error between the input signal and the output signal of each autoencoder. Based on the reconstruction error, it is determined whether each frequency band component signal contains outliers, and all frequency band component signals determined to contain outliers are reconstructed to obtain the noise-reduced fault signal.

[0007] Furthermore, each autoencoder includes both an encoder and a decoder;

[0008] Among them, the multilayer encoder is used to progressively compress the input data to extract features. Its encoding process takes the output of the (i-1)th layer encoder as the input of the i-th layer encoder, and outputs the output of the i-th layer encoder after encoding operation. The decoder is used to reconstruct the compressed features. Its decoding process takes the output of the i-th layer decoder as the input of the (i-1)-th layer decoder and outputs the output of the (i-1)-th layer decoder after decoding.

[0009] Furthermore, unsupervised training of the autoencoder is performed, including: The frequency band component signal is used as the input of the autoencoder, and the mean square error is used as the loss function. The loss function is minimized through repeated iterative training. A threshold value is set based on the calculation result when the loss function is minimized, which is used to determine whether the frequency band component signal contains outliers; the threshold value is not a fixed constant, but is adaptively determined based on the statistical distribution of the reconstruction error during the training phase.

[0010] Furthermore, based on the reconstruction error, it is determined whether each frequency band component signal contains outliers. The specific determination process is as follows: The reconstruction error between the input and output signals of each autoencoder is used as the criterion. When the reconstruction error is greater than a preset threshold, it is determined that the frequency band component signal corresponding to the input signal contains an anomaly, which carries fault information. The reconstruction error includes not only the overall error but also the error change characteristics within a local time window, which is used to enhance the detection capability of transient impact-type abnormal signals. When the reconstruction error is not greater than the preset threshold, the component signal of the frequency band to be processed corresponding to the input signal is determined to be the background noise component.

[0011] Furthermore, all frequency band components identified as containing outliers are reconstructed to obtain the noise-reduced fault signal, including: All frequency band component signals identified as containing outliers are taken as input, and their signals are reconstructed. The reconstructed signals are output as the effective signals after feature enhancement. These effective signals contain fault information and suppress noise information.

[0012] Furthermore, wavelet packet decomposition and reconstruction are performed on the strong background noise signal to obtain multiple frequency band component signals, including: The strong background noise signal was collected as input and subjected to three-level wavelet packet decomposition to obtain the decomposition coefficients of eight frequency components from low frequency to high frequency in the third level. The eight decomposition coefficients are reconstructed to output eight frequency band component signals.

[0013] Furthermore, the acoustic emission signal to be processed, which contains fault information, is obtained by superimposing the broken lead signal with the acquired strong background noise signal at a preset signal-to-noise ratio.

[0014] Secondly, the technical solution adopted by this invention is: a noise reduction system for weak fault signals under strong background noise in aero-engines, comprising: a strong background noise signal acquisition module, which acquires acoustic emission signals under normal operating conditions of the aero-engine as strong background noise signals, performs wavelet packet decomposition and reconstruction on the strong background noise signals to obtain multiple frequency band component signals; an autoencoder training module, which constructs a corresponding autoencoder for each frequency band component signal, and performs unsupervised training on the autoencoder using the frequency band component signal as input, wherein a shared feature extraction mechanism is introduced into each frequency band autoencoder during the training process, enabling the autoencoder to learn and reconstruct the strong background noise signal under that frequency band; wherein, the shared feature extraction mechanism inputs the different frequency band component signals into the shared encoder structure. Feature extraction is performed during the modeling process, and each frequency band is reconstructed using the corresponding decoder. This allows the model to learn cross-band correlation information while preserving the differences in features between frequency bands. The acoustic emission signal processing module collects the acoustic emission signal containing fault information, performs wavelet packet decomposition and reconstruction on the signal, and obtains multiple frequency band component signals. Each frequency band component signal is then input into the corresponding trained autoencoder for reconstruction, and the reconstruction error between the input and output signals of each autoencoder is calculated. The noise reduction module determines whether each frequency band component signal contains outliers based on the reconstruction error, and reconstructs all frequency band component signals identified as containing outliers to obtain the noise-reduced fault signal.

[0015] Thirdly, the technical solution adopted by the present invention is: a computer-readable storage medium for storing one or more programs, wherein the one or more programs include instructions that, when executed by a computing device, cause the computing device to perform any of the methods described above.

[0016] Fourthly, the technical solution adopted by the present invention is: a computing device comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for performing any of the methods described above.

[0017] The present invention has the following advantages due to the adoption of the above technical solutions: 1. This invention employs an acoustic emission signal denoising method based on a combination of autoencoder and wavelet packet decomposition. First, the original acoustic emission signal is subdivided into multiple full-band subdivisions through wavelet packet decomposition. Strong background noise and fault-related frequency components are separated in different sub-bands, thus structurally realizing a layered representation of non-stationary broadband noise. Compared with the traditional overall filtering method, this method avoids the problem of effective feature attenuation caused by band overlap and provides a data foundation for subsequent model input after frequency domain decoupling.

[0018] 2. This invention constructs datasets for each frequency band component after decomposition, defines the real-collected engine background noise samples as normal data, and defines the noisy fault signals as abnormal data. It uses an autoencoder to reconstruct and learn the normal data, and realizes the identification and reconstruction of abnormal components through reconstruction error. This mechanism does not rely on the prior frequency information of the fault mechanism, but rather performs adaptive modeling based on the differences in data distribution, thereby improving the adaptability to unknown or complex coupled noise environments.

[0019] 3. This invention enhances the autoencoder's ability to analyze local frequency band features through wavelet packet decomposition, reduces the training difficulty when the model directly processes strong noise raw signals, improves convergence stability and generalization ability, and overcomes the problem of single deep learning methods being easily affected by dominant noise under non-stationary strong background noise conditions from the model structure level. Attached Figure Description

[0020] Figure 1 This is an overall flowchart of the method for denoising weak fault signals under strong background noise of an aero-engine in this embodiment of the invention; Figure 2 This is a detailed flowchart of the method for denoising weak fault signals under strong background noise of an aero-engine in an embodiment of the present invention; Figure 3 This is a schematic diagram of wavelet packet decomposition in an embodiment of the present invention; Figure 4 This is a schematic diagram of the self-encoder in an embodiment of the present invention; Figure 5 This is a learning curve diagram of AE0, AE1, AE2, AE3, AE4, AE5, AE6, and AE7 in an embodiment of the present invention. Detailed Implementation

[0021] Existing noise reduction methods for acoustic emission signals in aero-engines operating under strong background noise environments are mostly based on traditional signal processing techniques such as wavelet analysis, time-frequency analysis, higher-order statistical analysis, or empirical mode decomposition. While these methods can reduce noise interference to some extent, they generally rely on complex signal processing procedures and extensive prior knowledge of mechanisms. They require prior knowledge of the time-frequency characteristics corresponding to different faults and are difficult to effectively extract early, weak fault features under complex operating conditions involving multi-source coupling, strong random broadband noise, and discrete single-tone noise. On the other hand, existing deep learning noise reduction methods mostly focus on Gaussian white noise and lack specific research on non-stationary, high-energy background noise in actual engineering. They also have insufficient generalization ability for noise types not covered by training data and are difficult to adapt to the complex and variable noise conditions in the real operating environment of aero-engines. Therefore, this invention provides a method, system, medium, and device for denoising weak fault signals under strong background noise in aero-engines. It can combine the advantages of traditional time-frequency decomposition with the adaptive modeling capabilities of deep learning for joint denoising, effectively separating and reconstructing weak fault acoustic emission signals under real strong background noise conditions, improving the signal-to-noise ratio and the robustness of the method, thereby providing a reliable data foundation for engine structural condition monitoring and subsequent fault identification and location.

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.

[0023] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0024] In one embodiment of the present invention, a method for denoising weak fault signals under strong background noise in aero-engines is provided. This method utilizes the powerful feature extraction and outlier identification capabilities of wavelet packets and autoencoders to solve the problem of acoustic emission signal denoising for early-stage aero-engine faults under strong background noise. In this embodiment, as shown... Figure 1 , Figure 2 As shown, the method includes the following steps: 1) Acquire acoustic emission signals from the aero-engine under normal operating conditions as strong background noise signals, perform wavelet packet decomposition and reconstruction on the strong background noise signals to obtain multiple frequency band component signals.

[0025] 2) For each frequency band component signal, a corresponding autoencoder is constructed, and the autoencoder is trained unsupervised using the frequency band component signal as input. During the training process, a shared feature extraction mechanism is introduced for each frequency band autoencoder, enabling the autoencoder to learn and reconstruct strong background noise signals in that frequency band. Specifically, a multi-band collaborative modeling mechanism (i.e., a shared feature extraction mechanism) is introduced, where different frequency band component signals are input into the shared encoder structure for feature extraction, and each frequency band is reconstructed through the corresponding decoder. This allows the model to learn cross-band correlation information while preserving the feature differences between frequency bands, thereby improving the modeling ability for complex coupled noise.

[0026] 3) Acquire acoustic emission signals containing fault information, perform wavelet packet decomposition and reconstruction on the signals to obtain multiple frequency band components. Input each component into a pre-trained autoencoder for reconstruction, and calculate the reconstruction error between the input and output signals of each autoencoder. Furthermore, local temporal reconstruction error and residual signal analysis are introduced to finely characterize abnormal features, thereby improving the ability to identify weak impact-type fault signals.

[0027] 4) Determine whether each frequency band component signal contains outliers based on the reconstruction error, and reconstruct all frequency band component signals that are determined to contain outliers to obtain the noise-reduced fault signal.

[0028] In step 1) above, the strong background noise signal is decomposed and reconstructed using wavelet packets to obtain multiple frequency band component signals, including the following steps: 1.1) The strong background noise signal collected is used as input and subjected to three-level wavelet packet decomposition to obtain the decomposition coefficients of the eight frequency components from low frequency to high frequency in the third level.

[0029] Specifically, strong background noise signals under normal engine operation are collected, and then subjected to three-level wavelet packet decomposition. The signal features of eight frequency components from low to high frequencies in the third level are extracted to obtain the decomposition coefficients. , Indicates the first m The first layer n Each node.

[0030] In this embodiment, the Daubechies (db) wavelet, which is well-suited for analyzing short and fast transient signals, is used, as it can effectively extract signal abrupt change features. Daubechies-8 (db8) has a vanishing moment order of 8 and a filter length of 16, exhibiting fine bandpass performance and strong signal detail extraction capabilities. Therefore, this invention uses db8 as the fundamental wavelet for wavelet packet decomposition.

[0031] 1.2) Reconstruct the 8 decomposition coefficients respectively and output 8 frequency band component signals.

[0032] Specifically, an acoustic emission signal It is represented by formula (1). This represents the corresponding decomposition coefficient. express The reconstructed signal, with 8 frequency band components, is represented as follows: . Exploded view diagram as follows Figure 3 As shown.

[0033] (1) In step 2) above, each autoencoder includes an encoder and a decoder. The decomposed acoustic emission signal components are used as the input to the autoencoder. A corresponding autoencoder (AE) is designed and built for each frequency band. n (n=0, 1, ..., 7). Using an autoencoder, the feature enhancement problem is transformed into an outlier detection problem. In general prediction problems, the model is usually a way of representing the overall sample data structure. This representation typically captures the general properties of the overall sample, while points that behave completely differently from the overall sample in these properties are outliers (outliers). Strong background noise is unavoidable in aero-engines during normal operation. This invention treats strong background noise as "normal." When a fault occurs, "outliers" will appear in the stable strong background noise; these outliers are data carrying fault information. This invention uses an autoencoder for real-time outlier detection, accurately picking outliers, thereby achieving the effect of feature enhancement.

[0034] In this embodiment, a multi-layer encoder is used to progressively compress the input data to extract features. The encoding process uses the output of the (i-1)th layer encoder as the input of the ith layer encoder, and outputs the output of the ith layer encoder after encoding operations. A shared feature extraction layer is introduced during the encoding process to extract common noise features from multi-frequency signals, achieving multi-scale information fusion and enhancing the model's ability to jointly model broadband noise and discrete single-tone noise.

[0035] Specifically, autoencoder models such as Figure 4 As shown. Let For the first The input of the layer encoder, For the first The output of the layer encoder; the encoding process for this part is as follows: (2) in, The weights are those of the encoder. This represents the encoder bias. Both values ​​are determined during unsupervised training. This is the activation function for the encoder. This invention uses the ReLU activation function.

[0036] In this embodiment, the decoder is used to reconstruct the compressed features. Its decoding process uses the output of the i-th layer decoder as the input of the (i-1)-th layer decoder, and outputs the output of the (i-1)-th layer decoder after decoding.

[0037] Specifically, the corresponding decoding process is as follows: (3) in, For the first The output of the layer decoder is... The refactored version For the first The input to the layer decoder, and These are the weights and biases for the decoder. The loss function for the autoencoder uses mean squared variance.

[0038] In step 2) above, unsupervised training of the autoencoder includes the following steps: 2.1) The frequency band component signal is used as the input of the autoencoder, and the mean square error is used as the loss function. The loss function is minimized through repeated iterative training.

[0039] 2.2) A threshold value is set based on the calculation result when the loss function is minimized, which is used to determine whether the frequency band component signal contains outliers. In this embodiment, the threshold value is not a fixed constant, but is adaptively determined based on the statistical distribution of the reconstruction error during the training phase, so as to improve the model's adaptability under different noise intensity conditions.

[0040] In this embodiment, the loss function is minimized through repeated iterative training. Simultaneously, a threshold value is set based on the loss function calculation results for outlier detection. This yields a set of trained autoencoder models, i.e. These correspond to eight different frequency bands. During training, each frequency band autoencoder incorporates a shared feature extraction mechanism to enhance the correlation modeling capability between different frequency bands, thereby improving the representation capability of multi-source coupled noise.

[0041] In step 3) above, the acoustic emission signal to be processed containing fault information is obtained by superimposing the broken lead signal with the acquired strong background noise signal at a preset signal-to-noise ratio.

[0042] In step 4) above, the determination of whether each frequency band component signal contains outliers is based on the reconstruction error. The specific determination process includes the following steps: 4.1) The reconstruction error between the input and output signals of each autoencoder is used as the criterion. When the reconstruction error exceeds a preset threshold, it is determined that the component signal of the frequency band to be processed corresponding to the input signal contains an anomaly, which carries fault information. The reconstruction error includes not only the overall error but also the error variation characteristics within a local time window, which is used to enhance the detection capability of transient impact-type abnormal signals.

[0043] 4.2) When the reconstruction error is not greater than a preset threshold, the component signal of the frequency band to be processed corresponding to the input signal is determined to be a background noise component. By jointly analyzing the discrimination results of each frequency band, the stability of anomaly detection can be improved, and the impact of misjudgment of a single frequency band can be avoided.

[0044] In step 4) above, all frequency band component signals that are judged to contain outliers are reconstructed to obtain the noise-reduced fault signal. Specifically, all frequency band component signals that are judged to contain outliers are taken as input, and their signals are reconstructed. The reconstructed signal is output as the effective signal after feature enhancement. The effective signal contains fault information and noise information is suppressed.

[0045] In this embodiment, the obtained different frequency band components Input the corresponding autoencoder respectively After encoding and decoding, the output signal is compared with the input signal. Outliers are identified by the overall and local errors between the two. If any local information in the output signal is inconsistent with the original signal, there is reason to believe that the original signal is abnormal. After detection by the autoencoder, if Q frequency bands out of the eight frequency bands of the input signal show abnormalities, then these Q frequency bands are considered to contain a large amount of fault information. The remaining 8-Q frequency bands are considered background noise or contain a small amount of fault information. Finally, the Q frequency bands are reconstructed and treated as a feature-enhanced, effective signal containing a large amount of fault information and a small amount of noise information.

[0046] In one embodiment of the present invention, a noise reduction system for weak fault signals under strong background noise in an aero-engine is provided, comprising: The strong background noise signal acquisition module acquires the acoustic emission signal of the aero-engine under normal operating conditions as a strong background noise signal, performs wavelet packet decomposition and reconstruction on the strong background noise signal to obtain multiple frequency band component signals; The autoencoder training module constructs a corresponding autoencoder for each frequency band component signal and uses the frequency band component signal as input to perform unsupervised training on the autoencoder. During the training process, a shared feature extraction mechanism is introduced for each frequency band autoencoder, enabling the autoencoder to learn and reconstruct strong background noise signals in that frequency band. Specifically, the shared feature extraction mechanism inputs different frequency band component signals into the shared encoder structure for feature extraction, and the corresponding decoder reconstructs each frequency band, so that the model can learn cross-frequency band correlation information while preserving the feature differences of each frequency band. The acoustic emission signal processing module collects the acoustic emission signal containing fault information, performs wavelet packet decomposition and reconstruction on the acoustic emission signal to be processed, and obtains multiple frequency band component signals to be processed. Each frequency band component signal to be processed is input to the corresponding trained autoencoder, which reconstructs it and calculates the reconstruction error between the input signal and the output signal of each autoencoder. The noise reduction module determines whether each frequency band component signal contains outliers based on the reconstruction error, and reconstructs all frequency band component signals that are determined to contain outliers to obtain the noise-reduced fault signal.

[0047] In the above embodiments, each autoencoder includes an encoder and a decoder. The multi-layer encoder is used to progressively compress the input data to extract features. Its encoding process uses the output of the (i-1)th layer encoder as the input of the ith layer encoder, and outputs the output of the ith layer encoder after encoding operations. The decoder is used to reconstruct the compressed features. Its decoding process takes the output of the i-th layer decoder as the input of the (i-1)-th layer decoder and outputs the output of the (i-1)-th layer decoder after decoding.

[0048] In the above embodiments, unsupervised training of the autoencoder includes: The frequency band component signal is used as the input of the autoencoder, and the mean square error is used as the loss function. The loss function is minimized through repeated iterative training. A threshold value is set based on the calculation result when the loss function is minimized, which is used to determine whether the frequency band component signal contains outliers.

[0049] In the above embodiments, the determination of whether each frequency band component signal contains outliers is based on the reconstruction error. The specific determination process is as follows: The reconstruction error between the input signal and the output signal of each autoencoder is used as the judgment criterion. When the reconstruction error is greater than the preset threshold value, it is determined that the frequency band component signal to be processed corresponding to the input signal contains an abnormal value, and the abnormal value carries fault information. When the reconstruction error is not greater than the preset threshold, the component signal of the frequency band to be processed corresponding to the input signal is determined to be the background noise component.

[0050] In the above embodiments, all frequency band component signals identified as containing outliers are reconstructed to obtain the noise-reduced fault signal, including: All frequency band component signals identified as containing outliers are taken as input, and their signals are reconstructed. The reconstructed signals are output as the effective signals after feature enhancement. These effective signals contain fault information and suppress noise information.

[0051] In the above embodiments, wavelet packet decomposition and reconstruction are performed on the strong background noise signal to obtain multiple frequency band component signals, including: The strong background noise signal was collected as input and subjected to three-level wavelet packet decomposition to obtain the decomposition coefficients of eight frequency components from low frequency to high frequency in the third level. The eight decomposition coefficients are reconstructed to output eight frequency band component signals.

[0052] In the above embodiments, the acoustic emission signal to be processed containing fault information is obtained by superimposing the lead breakage signal and the acquired strong background noise signal at a preset signal-to-noise ratio.

[0053] The system provided in this embodiment is used to execute the above-described method embodiments. For specific processes and details, please refer to the above embodiments, which will not be repeated here.

[0054] In this embodiment, verification was conducted using an aero-engine horizontal rotor test bench and a casing simulation test bench. Specifically, the noise in the test originated from the aero-engine horizontal rotor test bench, while the casing simulation test bench was simplified compared to the actual structure. Aluminum alloy was used as the material, and the variable diameter structure was omitted, keeping the casing diameter constant. Its outer diameter is 1000mm, height is 300mm, wall thickness is 3mm, and the mounting edge height on one side of the casing is 8mm. The test piece includes 12 support structures and bearing sleeves, such as... Figure 5 As shown. This invention implements the proposed architecture using the TensorFlow framework, and all experiments were conducted on an Intel Core i7 3.6GHz CPU.

[0055] Regarding the dataset, the experiment used acoustic emission signals collected under normal operating conditions on an aero-engine horizontal rotor test bench as strong background noise, totaling 500 acoustic emission events. This was considered the normal value dataset. To simulate fault signals, 200 repeated lead-breaking tests were conducted in a casing simulation test bench. The lead-breaking signals were combined with random background noise at a certain signal-to-noise ratio and considered as outliers. The dataset contained a total of 700 sets of acoustic emission events. Furthermore, 60% of the normal value dataset was used as the training set, 20% of the normal value dataset and 50% of the outlier dataset were used as the test set, and 20% of the normal value dataset and 50% of the outlier dataset were used as the validation set.

[0056] Experimental results show that after feature enhancement using noisy signals with initial signal-to-noise ratios (SNRs) of -5dB, -2.5dB, 0dB, 2.5dB, and 5dB, the SNR can be improved to -0.13dB, 2.23dB, 6.12dB, 8.66dB, and 11.04dB, respectively. This result demonstrates that the present invention exhibits excellent feature enhancement performance.

[0057] In summary, this invention collected real background noise data in an aero-engine test bench environment and compared it with various traditional methods such as wavelet threshold denoising and EMD threshold denoising under different signal-to-noise ratio (SNR) conditions (multiple sets of comparative experiments were constructed by superimposing real background noise of different intensities). Four evaluation indicators—SNR, mean square error, mean absolute error, and signal gain parameter—were introduced for quantitative evaluation. The results show that under different SNR conditions, this invention achieves a higher SNR improvement, and the signal gain parameter is superior to the comparative methods, indicating that this invention effectively suppresses strong background noise while enhancing the characteristics of weak impacts. Furthermore, through repeated experiments under different noise intensities, the results show that this invention maintains stable noise reduction performance even under extreme conditions of low SNR, demonstrating good robustness and engineering applicability. This provides a more reliable signal basis for engine structural condition monitoring and subsequent fault identification and location.

[0058] In one embodiment of the present invention, a computing device is provided. This computing device can be a terminal and may include a processor, a communication interface, memory, a display screen, and an input device. The processor, communication interface, and memory communicate with each other via a communication bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. When the computer programs are executed by the processor, they implement the methods described in the above embodiments. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface is used for wired or wireless communication with external terminals. Wireless communication can be achieved through Wi-Fi, a network management system, NFC (Near Field Communication), or other technologies. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input device can be a touch layer covering the display screen, or buttons, a trackball, or a touchpad mounted on the casing of the computing device, or an external keyboard, touchpad, or mouse. The processor can call logical instructions stored in the memory to execute the above methods.

[0059] In one embodiment of the present invention, a computer program product is provided, the computer program product including a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, and when the program instructions are executed by a computer, the computer is able to perform the methods provided in the above-described method embodiments.

[0060] In one embodiment of the present invention, a non-transitory computer-readable storage medium is provided, which stores server instructions that cause a computer to perform the methods provided in the above embodiments.

[0061] The computer-readable storage medium provided in the above embodiments has a similar implementation principle and technical effect to the above method embodiments, and will not be described again here.

[0062] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1A device that provides the functions specified in one or more boxes.

[0063] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0064] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0065] 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 them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for denoising weak fault signals under strong background noise in an aero-engine, characterized in that, include: Acoustic emission signals from the normal operating state of the aero-engine are collected as strong background noise signals. Wavelet packet decomposition and reconstruction are performed on the strong background noise signals to obtain multiple frequency band component signals. For each frequency band component signal, a corresponding autoencoder is constructed, and the autoencoder is trained unsupervised using the frequency band component signal as input. During the training process, a shared feature extraction mechanism is introduced for each frequency band autoencoder, enabling the autoencoder to learn and reconstruct the strong background noise signal under that frequency band. Specifically, the shared feature extraction mechanism inputs the component signals of different frequency bands into the shared encoder structure for feature extraction, and the corresponding decoder reconstructs each frequency band, so that the model can learn cross-frequency band correlation information while preserving the feature differences of each frequency band. Acquire acoustic emission signals containing fault information, perform wavelet packet decomposition and reconstruction on the acoustic emission signals to be processed to obtain multiple frequency band component signals to be processed, input each frequency band component signal to the corresponding trained autoencoder, reconstruct it by the autoencoder, and calculate the reconstruction error between the input signal and the output signal of each autoencoder. Based on the reconstruction error, determine whether each frequency band component signal contains outliers, and reconstruct all frequency band component signals that are determined to contain outliers to obtain the noise-reduced fault signal.

2. The method for denoising weak fault signals under strong background noise in aero-engines as described in claim 1, characterized in that, Each autoencoder includes an encoder and a decoder; Among them, the multilayer encoder is used to progressively compress the input data to extract features. Its encoding process takes the output of the (i-1)th layer encoder as the input of the i-th layer encoder, and outputs the output of the i-th layer encoder after encoding operation. The decoder is used to reconstruct the compressed features. Its decoding process takes the output of the i-th layer decoder as the input of the (i-1)-th layer decoder and outputs the output of the (i-1)-th layer decoder after decoding.

3. The method for denoising weak fault signals under strong background noise in aero-engines as described in claim 1, characterized in that, Unsupervised training of the autoencoder includes: The frequency band component signal is used as the input of the autoencoder, and the mean square error is used as the loss function. The loss function is minimized through repeated iterative training. A threshold value is set based on the calculation result when the loss function is minimized, which is used to determine whether the frequency band component signal contains outliers; the threshold value is not a fixed constant, but is adaptively determined based on the statistical distribution of the reconstruction error during the training phase.

4. The method for denoising weak fault signals under strong background noise in aero-engines as described in claim 1, characterized in that, Based on the reconstruction error, it is determined whether each frequency band component signal contains outliers. The specific determination process is as follows: The reconstruction error between the input and output signals of each autoencoder is used as the criterion. When the reconstruction error is greater than a preset threshold, it is determined that the frequency band component signal corresponding to the input signal contains an anomaly, which carries fault information. The reconstruction error includes not only the overall error but also the error change characteristics within a local time window, which is used to enhance the detection capability of transient impact-type abnormal signals. When the reconstruction error is not greater than the preset threshold, the component signal of the frequency band to be processed corresponding to the input signal is determined to be the background noise component.

5. The method for denoising weak fault signals under strong background noise in aero-engines as described in claim 4, characterized in that, All frequency band components identified as containing outliers are reconstructed to obtain the noise-reduced fault signal, including: All frequency band component signals identified as containing outliers are taken as input, and their signals are reconstructed. The reconstructed signals are output as the effective signals after feature enhancement. These effective signals contain fault information and suppress noise information.

6. The method for denoising weak fault signals under strong background noise in aero-engines as described in claim 1, characterized in that, Wavelet packet decomposition and reconstruction are performed on the strong background noise signal to obtain multiple frequency band component signals, including: The strong background noise signal was collected as input and subjected to three-level wavelet packet decomposition to obtain the decomposition coefficients of eight frequency components from low frequency to high frequency in the third level. The eight decomposition coefficients are reconstructed to output eight frequency band component signals.

7. The method for denoising weak fault signals under strong background noise in aero-engines as described in claim 1, characterized in that, The acoustic emission signal containing fault information is collected by superimposing the broken lead signal with the collected strong background noise signal at a preset signal-to-noise ratio.

8. A noise reduction system for weak fault signals under strong background noise in an aero-engine, characterized in that, include: The strong background noise signal acquisition module acquires the acoustic emission signal of the aero-engine under normal operating conditions as a strong background noise signal, performs wavelet packet decomposition and reconstruction on the strong background noise signal to obtain multiple frequency band component signals; The autoencoder training module constructs a corresponding autoencoder for each frequency band component signal and uses the frequency band component signal as input to perform unsupervised training on the autoencoder. During the training process, a shared feature extraction mechanism is introduced for each frequency band autoencoder, enabling the autoencoder to learn and reconstruct strong background noise signals in that frequency band. Specifically, the shared feature extraction mechanism inputs different frequency band component signals into the shared encoder structure for feature extraction, and the corresponding decoder reconstructs each frequency band, so that the model can learn cross-frequency band correlation information while preserving the feature differences of each frequency band. The acoustic emission signal processing module collects the acoustic emission signal containing fault information, performs wavelet packet decomposition and reconstruction on the acoustic emission signal to be processed, and obtains multiple frequency band component signals to be processed. Each frequency band component signal to be processed is input to the corresponding trained autoencoder, which reconstructs it and calculates the reconstruction error between the input signal and the output signal of each autoencoder. The noise reduction module determines whether each frequency band component signal contains outliers based on the reconstruction error, and reconstructs all frequency band component signals that are determined to contain outliers to obtain the noise-reduced fault signal.

9. A computer-readable storage medium for storing one or more programs, characterized in that, The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform any of the methods described in claims 1 to 7.

10. A computing device, characterized in that, include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for performing any of the methods described in claims 1 to 7.

Images