Compressor knocking fault diagnosis method, device and equipment

Abstract

The invention relates to the field of compressor fault detection, and discloses a compressor knocking fault diagnosis method, device and equipment, and the method comprises the steps: obtaining a to-be-detected sound signal in the operation process of a compressor, and carrying out the filtering processing of the to-be-detected sound signal; variational mode decomposition and reconstruction are carried out on the sound signal to be detected after filtering processing so as to generate a corresponding enhanced signal, and analysis processing is carried out on the enhanced signal so as to obtain a corresponding analysis signal; time-frequency distribution transformation based on a self-adaptive weighting window is carried out on the analytic signal to generate a time-frequency image of the analytic signal, and a window width parameter of the self-adaptive weighting window is dynamically adjusted based on an instantaneous frequency gradient of the analytic signal; and inputting the time-frequency image into a pre-trained fault diagnosis model, and outputting a fault diagnosis result of the compressor by the fault diagnosis model. According to the technical scheme provided by one or more implementation modes, the accuracy of compressor knocking type fault recognition can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of compressor fault detection, and in particular to a compressor knocking fault diagnosis method, device and equipment. BACKGROUND

[0002] As a core component of the new energy vehicle thermal management system, the operation stability of the electric compressor is directly related to the comfort and safety of the vehicle. In the actual operation process, the compressor may produce abnormal knocking sound due to mechanical part loosening, valve fracture or piston impact and other faults. If not timely identified and warned, it may cause serious damage and even safety accidents.

[0003] In the related art, the fault diagnosis method for the compressor mainly relies on the time-frequency analysis technology based on image recognition and neural network. However, the traditional time-frequency analysis method is limited by the fixed time-frequency resolution, and the feature representation ability of the knocking type transient and impact type non-stationary signal is insufficient. The fault features in the generated time-frequency image are fuzzy, and there is a lot of noise interference, resulting in low accuracy of fault identification.

[0004] Therefore, how to improve the accuracy of the compressor knocking fault identification has become the focus of the current research in the field of compressor fault detection. SUMMARY

[0005] The present application provides a compressor knocking fault diagnosis method, device and equipment, which can improve the accuracy of the compressor knocking fault identification.

[0006] The first aspect of the present application provides a compressor knocking fault diagnosis method, which comprises: acquiring a to-be-tested sound signal in the operation process of a compressor, and performing filtering processing on the to-be-tested sound signal; performing variational mode decomposition and reconstruction on the to-be-tested sound signal after filtering processing to generate a corresponding enhanced signal, performing analysis processing on the enhanced signal to obtain a corresponding analysis signal; performing time-frequency distribution transformation on the analysis signal based on an adaptive weighted window to generate a time-frequency image of the analysis signal, wherein the window width parameter of the adaptive weighted window is dynamically adjusted based on the instantaneous frequency gradient of the analysis signal; inputting the time-frequency image into a pre-trained fault diagnosis model, and outputting a fault diagnosis result of the compressor by the fault diagnosis model.

[0007] In one embodiment, acquiring a to-be-tested sound signal of a compressor and performing filtering processing on the to-be-tested sound signal comprises: removing the direct current component and outliers of the to-be-tested sound signal to obtain a preliminary processed signal of the to-be-tested sound signal; performing frequency domain conversion on the preliminary processed signal to obtain a frequency spectrum of the preliminary processed signal; performing band-pass filtering on the frequency spectrum, and performing time domain conversion on the frequency spectrum after band-pass filtering to complete the filtering processing of the to-be-tested sound signal.

[0008] In one embodiment, bandpass filtering of the spectrum graph includes: obtaining a target frequency component and a noise frequency component in the spectrum graph, wherein the target frequency component is in a specific frequency range and the noise frequency component is in a non-specific frequency range; and filtering out the noise frequency component in the spectrum graph to retain the target frequency component in the spectrum graph.

[0009] In one embodiment, performing variational mode decomposition and reconstruction on the filtered acoustic signal to generate a corresponding enhanced signal includes: performing variational mode decomposition on the filtered acoustic signal to obtain multiple intrinsic mode components; selecting the percussion characteristic mode component of the acoustic signal to be tested from the multiple intrinsic mode components, and reconstructing the signal based on the percussion characteristic mode component to obtain the enhanced signal corresponding to the acoustic signal to be tested.

[0010] In one embodiment, performing a time-frequency distribution transformation based on an adaptive weighted window on the analyzed signal to generate a time-frequency image of the analyzed signal includes: acquiring the instantaneous frequency gradient of the analyzed signal at various times; determining multiple tapping frequencies based on the instantaneous frequency gradient; for any tapping frequency, determining an adaptive weighted window for the tapping frequency based on the instantaneous frequency gradient corresponding to the tapping frequency, and determining the output intensity corresponding to the tapping frequency based on the adaptive weighted window and a target kernel function; and determining the time-frequency image of the analyzed signal based on the output intensity corresponding to each tapping frequency.

[0011] In one embodiment, the adaptive weighted window includes a target time window and a target frequency window; determining the adaptive weighted window for the striking frequency based on the instantaneous frequency gradient corresponding to the striking frequency includes: obtaining an initial time window and an initial frequency window, and using the target time window as the target time window for the striking frequency; adaptively adjusting the frequency window width parameter of the initial frequency window based on the instantaneous frequency gradient corresponding to the striking frequency to obtain the target frequency window width for the striking frequency; and updating the initial frequency window based on the target frequency window width to obtain the target frequency window for the striking frequency.

[0012] In one embodiment, the fault diagnosis model is pre-trained as follows: multiple image training sets and image validation sets are acquired, and an initial convolutional neural network model is constructed. The initial convolutional neural network model is iteratively trained using the image training sets. During any iteration of the iterative training, the parameters of the convolutional neural network model in the current iteration are updated using any image training set, and the updated convolutional neural network model is evaluated using the image validation set. If the evaluation result indicates that the updated convolutional neural network model meets preset conditions, the iterative training is stopped, and the trained convolutional network model is used as the fault diagnosis model.

[0013] In one implementation, updating the parameters of a convolutional neural network model in the current iteration using any image training set includes: inputting any image training set into the convolutional neural network model in the current iteration to obtain the prediction result of the image training set; obtaining the true label of the image training set, determining the loss error between the prediction result and the true label, and updating the parameters of the current convolutional neural network model based on the loss error.

[0014] A second aspect of this application provides a compressor knocking fault diagnosis device, the device comprising: a filtering unit for acquiring a test acoustic signal of the compressor in operation and filtering the test acoustic signal; a noise processing unit for performing variational mode decomposition and reconstruction on the filtered test acoustic signal to generate a corresponding enhanced signal, and performing analytical processing on the enhanced signal to obtain a corresponding analytical signal; an image generation unit for performing time-frequency distribution transformation based on an adaptive weighted window on the analytical signal to generate a time-frequency image of the analytical signal, wherein the window width parameter of the adaptive weighted window is dynamically adjusted based on the instantaneous frequency gradient of the analytical signal; and a fault diagnosis unit for inputting the time-frequency image into a pre-trained fault diagnosis model, and having the fault diagnosis model output the fault diagnosis result of the compressor.

[0015] A third aspect of this application provides a computer device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions, the computer device being used to implement the compressor knocking fault diagnosis method described in the first aspect.

[0016] The technical solution provided in one or more embodiments of this application generates a time-frequency image of the acoustic signal under test through variational mode processing and time-frequency distribution transformation, and performs fault diagnosis on the time-frequency image to achieve accurate monitoring and diagnosis of the compressor's operating status. Specifically, the acoustic signal during compressor operation is collected and filtered to eliminate noise interference and highlight fault characteristics. Variational mode decomposition and reconstruction technology is used to extract the enhanced signal from the filtered signal, and analytical processing is performed to obtain an analytical signal containing amplitude and phase information, further avoiding noise frequency interference. Based on the time-frequency distribution transformation using an adaptive weighted window, the window width is dynamically adjusted according to the instantaneous frequency gradient of the analytical signal to generate a clear time-frequency image to highlight fault characteristics. The time-frequency image is input into a pre-trained fault diagnosis model, which automatically extracts deep features and outputs fault diagnosis results, achieving accurate identification and classification of compressor knocking faults. This technical solution dynamically adjusts the time-frequency analysis resolution through an adaptive weighted window and utilizes the automatic feature extraction capability of the fault diagnosis model, significantly improving the accuracy and reliability of fault diagnosis, achieving automatic and accurate identification of compressor knocking faults, and is suitable for monitoring the operating status of compressors under different operating conditions and noise environments.

[0017] It is evident that the technical solution provided in this application can improve the accuracy of identifying compressor knocking faults. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 A schematic diagram illustrating the steps of a compressor knocking fault diagnosis method provided in this application embodiment; Figure 2 A schematic diagram illustrating the processing of a fault diagnosis model provided in one embodiment of this application; Figure 3(a) is a schematic diagram of a time-frequency image obtained by processing with the WK-SPWVD algorithm according to an embodiment of this application; Figure 3(b) is a schematic diagram of a time-frequency image obtained by processing with the SPWVD algorithm according to an embodiment of this application; Figure 3(c) is a schematic diagram of an unprocessed time-frequency image provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of a compressor knocking fault diagnosis device provided in one embodiment of this application; Figure 5This is a schematic diagram of the structure of a computer device provided in one embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] Furthermore, the use of terms such as "first," "second," etc., in this application is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of embodiments in this application, unless otherwise stated, "multiple" means two or more. Additionally, the use of "based on" or "according to" implies openness and inclusiveness, because processes, steps, calculations, or other actions "based on" or "according to" one or more of the stated conditions or values ​​may in practice be based on additional conditions or beyond the stated values.

[0022] With the rapid development of the new energy vehicle industry, thermal management systems have been widely applied in new energy vehicles. Among them, the electric compressor, as the core power component of the thermal management system, undertakes the crucial function of refrigerant compression and circulation, and its operational stability is paramount. During long-term, high-frequency operation, electric compressors are susceptible to various factors such as road roughness, component aging, and assembly precision deviations, which can easily lead to knocking faults such as loose mechanical parts, valve plate fatigue fracture, and abnormal piston-cylinder impact. The most obvious symptom of these faults is an abnormal knocking sound. Initially, this abnormal noise is often quite weak. If knocking faults are not accurately identified and warned of in time, they may lead to serious damage or even safety accidents.

[0023] In related technologies, the diagnosis of impact-related faults largely relies on acoustic feature extraction and vibration signal analysis techniques using machine learning. This involves converting the original impact signal into a time-frequency image for fault feature analysis. However, limited by a fixed time-frequency resolution, traditional time-frequency analysis methods struggle to simultaneously achieve precision in both time and frequency directions. This results in insufficient feature representation capabilities for transient, impact-type, non-stationary signals like impacts, leading to blurred fault features in the generated time-frequency image and low accuracy in fault analysis results based on the image. Furthermore, because acoustic signals are highly susceptible to interference from environmental noise and equipment harmonics, the generated time-frequency image typically contains significant noise, making it prone to missed or false alarms in complex operating conditions, further reducing the accuracy of impact fault detection.

[0024] In view of this, existing technologies for diagnosing compressor knocking faults generally suffer from problems such as unclear representation of transient impact signals, weak resistance to noise interference, and poor generalization ability of models under actual variable operating conditions, resulting in low fault identification accuracy. This application provides one or more embodiments of a compressor knocking fault diagnosis method, apparatus, and device that can solve the above problems, accurately capture transient knocking characteristics, effectively suppress interference, and improve the accuracy of compressor knocking fault identification.

[0025] Please see Figure 1 One embodiment of this application provides a method, apparatus, and device for diagnosing compressor knocking faults. The method may include the following steps: S1: Acquire the sound signal to be tested during the operation of the compressor, and filter the sound signal to be tested.

[0026] The aforementioned acoustic signals to be tested are sound signals collected during compressor operation, which may include normal operating sounds as well as knocking sounds that may occur during malfunctions, serving as the raw data for fault diagnosis. Due to the complex operating conditions of the compressor, the acoustic signals to be tested typically contain a large amount of noise interference, including ambient noise and irrelevant signals. Filtering the acoustic signals to be tested to eliminate noise interference and filter out irrelevant low-frequency and high-frequency noise makes fault characteristics more apparent. The filtered signals are more suitable for subsequent time-frequency analysis and feature extraction, providing high-quality data support for subsequent time-frequency analysis and fault diagnosis, thus improving the accuracy and reliability of fault diagnosis.

[0027] S3: Perform variational mode decomposition and reconstruction on the filtered acoustic signal to be tested to generate the corresponding enhanced signal, and perform analytical processing on the enhanced signal to obtain the corresponding analytical signal.

[0028] The variational mode decomposition described above can be understood as an adaptive signal decomposition of the acoustic signal under test, breaking down the complex mixed signal into a series of essential mode components. Each essential mode component has a specific center frequency and a finite bandwidth, effectively capturing the local time-frequency characteristics of the signal. The reconstruction operation described above can be understood as superimposing multiple selected essential mode components to reconstruct a new time-domain signal as an enhancement signal.

[0029] Since the acoustic signal of a compressor during operation typically contains multiple frequency components, including background noise from normal operation and potential knocking characteristics from faults, the aforementioned filtering processes are insufficient to completely remove the noise influence. By performing variational mode decomposition (VMD) on the acoustic signal under test, different frequency components in the signal can be separated, allowing the acoustic signal to exhibit more distinct characteristic components. The essential mode components related to knocking faults are filtered out during the VMD process. Furthermore, by performing VMD and reconstruction on the filtered acoustic signal under test, the characteristic components related to knocking faults can be effectively extracted and enhanced. This results in an enhanced signal after reconstruction that more prominently highlights the fault characteristics, while noise and other irrelevant components are suppressed or removed. This provides a clearer input signal for subsequent time-frequency analysis, significantly improving the accuracy and reliability of fault diagnosis.

[0030] In this embodiment, the enhanced signal undergoes analytical processing to further avoid noise frequency interference. This analytical processing can be understood as converting the enhanced signal from a real signal to a complex signal to avoid negative frequency interference, thereby obtaining an analytical signal corresponding to the enhanced signal. This analytical signal contains the signal's amplitude and phase information, allowing the instantaneous frequency and amplitude of the generated analytical signal to more accurately describe the signal's dynamic characteristics. For example, a Hilbert transform is used to convert the enhanced signal into an analytical signal, resulting in an analytical signal... ,in, This represents the Hilbert transform. By performing analytical processing on the enhanced signal, a cleaner analytic signal is obtained, facilitating subsequent time-frequency distribution transformation.

[0031] S5: Perform a time-frequency distribution transformation based on an adaptive weighted window on the analyzed signal to generate a time-frequency image of the analyzed signal, wherein the window width parameter of the adaptive weighted window is dynamically adjusted based on the instantaneous frequency gradient of the analyzed signal.

[0032] The aforementioned time-frequency distribution transformation, as a signal analysis method, is used to represent the characteristics of a signal simultaneously in both time and frequency dimensions. This transformation can employ time-frequency analysis methods such as Short-Time Fourier Transform (STFT), Wavelet Transform (WT), and Wigner-Ville Distribution (WVD), enabling the analyzed signal to display its energy distribution in time and frequency as a two-dimensional image. The resulting time-frequency image uses time on the horizontal axis and frequency on the vertical axis, with color or grayscale representing energy intensity. This allows the image to intuitively display the time-frequency characteristics of the signal, facilitating subsequent analysis and classification.

[0033] In this embodiment, an adaptive weighted window is introduced into the time-frequency distribution transformation to dynamically adjust the time-frequency analysis resolution of fault characteristics. Since the compressor knocking fault signal is a non-stationary signal, its frequency components change rapidly over time. Traditional fixed-window-width time-frequency analysis struggles to simultaneously achieve both time and frequency resolution, failing to accurately capture transient characteristics. Specifically, a wider time window results in high frequency resolution but low time resolution, making it difficult to accurately capture transient changes; conversely, a narrower time window provides high time resolution but low frequency resolution, making it impossible to clearly distinguish frequency components. Therefore, by dynamically adjusting the width of the adaptive weighted window, time-frequency distribution transformation is performed based on the adaptive weighted window, allowing the window width to be adjusted according to the dynamic characteristics of the signal, thereby optimizing the time-frequency resolution at different time points.

[0034] In this embodiment, the window width of the adaptive weighted window is dynamically adjusted by analyzing the instantaneous frequency gradient of the signal. Specifically, the instantaneous frequency gradient is the rate of change of the instantaneous frequency of the analyzed signal over time. When the signal undergoes a transient change (such as a knocking fault), the instantaneous frequency gradient will increase significantly. Using an adaptive weighted window adjusted based on the instantaneous frequency gradient for time-frequency distribution transformation can automatically increase the resolution in areas where the signal frequency changes drastically (such as the knocking fault point), thereby capturing transient changes more clearly and making fault features (such as the energy bright spot of the knocking signal) more prominent in the time-frequency image. The generated time-frequency image can clearly show the time point and frequency components of the fault occurrence, while effectively suppressing cross-term interference in the time-frequency distribution, reducing misjudgments, and improving the accuracy and robustness of fault diagnosis.

[0035] In one embodiment, the SPWVD algorithm is used to perform time-frequency analysis on the analytic signal to maintain good time-frequency concentration and anti-interference capability even under low signal-to-noise ratio conditions. The SPWVD algorithm effectively suppresses cross-terms by convolving a frequency smoothing window with a time smoothing window in the frequency and time domains. However, the smoothing window used is fixed, and this fixed smoothing may not be optimal for different regions of the signal. For example, it may over-smooth in stable regions (resulting in excessive resolution loss) or under-smooth in transient regions (resulting in residual cross-terms). Therefore, an adaptive weighted window can be introduced into the SPWVD algorithm for algorithm updates, resulting in the WK-SPWVD algorithm. The WK-SPWVD algorithm is then used to perform time-frequency analysis on the analytic signal, optimizing the trade-off between time-frequency resolution and cross-term suppression, thereby clearly displaying the frequency components and their intensity at different time points.

[0036] S7: Input the time-frequency image into the pre-trained fault diagnosis model, and the fault diagnosis model outputs the fault diagnosis result of the compressor.

[0037] The aforementioned fault diagnosis model is used to identify and classify fault features in signals. This model can be constructed using a convolutional neural network to automatically learn and classify deep features from time-frequency images. The model is pre-trained to adapt to signals under different operating conditions and noise environments, exhibiting high reliability. By inputting time-frequency images, the model accurately identifies fault features in compressor operation, such as energy bright spots in knocking faults. Based on the identification results, it determines whether the compressor has experienced a knocking fault. The output is typically a classification label, such as "normal operation" or "knocking fault."

[0038] In one embodiment, a pre-trained CNN model is used as the fault diagnosis model. Specifically, a time-frequency image is input into the CNN model, which automatically extracts features from the time-frequency image through multiple convolutional and pooling operations and performs classification. The CNN model may include convolutional layers, pooling layers, normalization layers, fully connected layers, and an output layer. The convolutional layers use learnable convolutional kernels to slide across the time-frequency image to extract local features. The normalization layers normalize the features extracted by the convolutional layers, making the feature distribution more stable. The pooling layers reduce the dimensionality of the feature map through pooling operations. The fully connected layers flatten the multi-dimensional features extracted by the convolutional and pooling layers into a one-dimensional vector and further integrate and classify them. The output layer outputs the classification result as the fault diagnosis result based on the integrated feature vector and the learned fault features. For example, by calculating the similarity between the input time-frequency image and the learned fault features, a probability value is output. If the probability value exceeds a preset threshold, it is determined that a knocking fault has occurred.

[0039] Based on the above ideas, the technical solution provided in this embodiment of the application generates a time-frequency image of the sound signal under test through variational mode processing and time-frequency distribution transformation, so as to perform fault diagnosis on the time-frequency image and realize accurate monitoring and diagnosis of the compressor's operating status. Specifically, the sound signal during the compressor's operation is collected and filtered to eliminate noise interference and highlight fault characteristics. The enhanced signal is extracted from the filtered signal using variational mode decomposition and reconstruction technology, and the analytical signal containing amplitude and phase information is obtained through analytical processing to further avoid noise frequency interference. Based on the time-frequency distribution transformation of the adaptive weighted window, the window width is dynamically adjusted according to the instantaneous frequency gradient of the analytical signal to generate a clear time-frequency image to highlight fault characteristics. The time-frequency image is input into a pre-trained fault diagnosis model, which automatically extracts deep features and outputs fault diagnosis results, realizing accurate identification and classification of compressor knocking faults. This technical solution dynamically adjusts the time-frequency analysis resolution through adaptive weighted window and utilizes the automatic feature extraction capability of the fault diagnosis model, which significantly improves the accuracy and reliability of fault diagnosis, realizes the automatic and accurate identification of compressor knocking faults, and is suitable for monitoring the operating status of compressors under different working conditions and noise environments.

[0040] In one embodiment, acquiring the acoustic signal to be tested from the compressor and filtering the acoustic signal to be tested includes: removing the DC component and outliers of the acoustic signal to be tested to obtain a preliminary processed signal of the acoustic signal to be tested; performing frequency domain transformation on the preliminary processed signal to obtain a spectrum of the preliminary processed signal; performing bandpass filtering on the spectrum; and performing time domain transformation on the spectrum after bandpass filtering to complete the filtering process of the acoustic signal to be tested.

[0041] The DC component mentioned above can be understood as a constant offset in the signal that does not change with time, typically exhibiting a non-zero mean. Removing the DC component from the measured acoustic signal avoids the generation of zero-frequency spikes in the frequency domain, which could interfere with subsequent frequency domain analysis. The outliers mentioned above can be understood as significant deviations from the normal range caused by measurement interference, transient noise, or equipment malfunction. These outliers can be identified and removed through signal filtering. Removing outliers from the measured acoustic signal improves data quality and prevents them from misleading subsequent transformations and feature extraction.

[0042] In this embodiment, the pre-processed signal undergoes frequency domain transformation to convert its time-domain representation into a frequency-domain representation, thereby obtaining the signal's spectrum. Specifically, the FFT (Fast Fourier Transform) algorithm is applied to the discrete time-domain signal sequence to obtain a complex-form spectrum, facilitating signal component analysis in the frequency dimension and providing a basis for frequency-domain filtering. Further, bandpass filtering is applied to the spectrum, and the bandpass-filtered spectrum is then transformed into a time-domain representation.

[0043] In this embodiment, bandpass filtering is performed on the spectrum to remove noise. Specifically, the target frequency component and noise frequency component are obtained from the spectrum. The noise frequency component is filtered out to retain the target frequency component in the spectrum. The target frequency component is within a specific frequency range, while the noise frequency component is within a non-specific frequency range. For example, the specific frequency range is 300Hz-8000Hz. In the frequency domain, components below 300Hz and above 8000Hz are set to zero, and then the signal is transformed back to the time domain using an inverse FFT to obtain the filtered time-domain signal.

[0044] In one embodiment, the acoustic signal to be measured Filtering is performed. Specifically, the acoustic signal to be measured is determined. mean The DC component is removed by referring to the mean value, based on... The signal after removing the DC component can be obtained. Furthermore, based on the threshold method of 5 times the standard deviation, the standard deviation of the signal is calculated. Identify more than The data points are treated as outliers for outlier detection and removal. Typically, the median of the signal is used to replace outliers to obtain the time-domain signal after DC component and outlier removal. For time-domain signals Perform a windowed FFT operation to obtain the frequency domain representation. , , frequency domain signal With bandpass filter Multiply to obtain the filtered frequency domain signal. For the filtered frequency domain signal Perform an inverse FFT to obtain the filtered time-domain signal. , .

[0045] The technical solution provided in this embodiment completes the filtering processing of the acoustic signal under test through operations such as DC component and outlier removal, and bandpass filtering. Specifically, by removing the DC component, peak interference generated by the DC component at zero frequency in the frequency domain is avoided. By removing outliers, data quality is effectively improved, preventing outliers from misleading subsequent signal transformation and feature extraction. Furthermore, through frequency domain bandpass filtering and time domain conversion, the target frequency components in the compressor acoustic signal are effectively preserved and noise is filtered out, providing a high-quality signal data foundation for subsequent analysis of the compressor acoustic signal. This solution is suitable for monitoring the operating status of compressors under different operating conditions and noise environments.

[0046] In one implementation, variational mode decomposition and reconstruction are performed on the filtered acoustic signal to be tested to generate a corresponding enhanced signal, specifically including the following steps: S41: Perform variational mode decomposition on the filtered acoustic signal to be tested to obtain multiple intrinsic mode components; S43: Among the multiple intrinsic mode components, the striking feature mode component of the sound signal to be tested is selected, and the signal is reconstructed based on the striking feature mode component to obtain the enhanced signal corresponding to the sound signal to be tested.

[0047] In this embodiment, the VMD algorithm can be used to perform variational mode decomposition on the acoustic signal under test to obtain multiple intrinsic mode components (IMCs), and the spectral entropy of each IMC is determined as the basis for component selection. Spectral entropy measures the degree of disorder in the spectral energy distribution of each IMC. IMCs with lower spectral entropy typically correspond to impact components (including transient characteristics of impact faults), while IMCs with higher spectral entropy usually contain noise. Removing and reconstructing IMCs with higher spectral entropy can improve the signal-to-noise ratio of the acoustic signal.

[0048] In this embodiment, intrinsic mode components (IMCs) are screened and reconstructed based on a preset threshold. Specifically, IMCs are screened based on the preset threshold to select IMCs with low spectral entropy for signal reconstruction. The screened IMCs are then superimposed to reconstruct a new time-domain signal as the enhanced signal corresponding to the acoustic signal under test. By removing most of the high spectral entropy noise and irrelevant vibration components, the weak impact characteristics are relatively amplified in the reconstructed signal, and the energy bright spots corresponding to the impact fault are clearer in the generated time-frequency diagram.

[0049] In one embodiment, for the sound signal to be measured ,Will Divided into Each intrinsic mode component That is, each intrinsic mode component is represented as , representing the characteristic components of the original signal at different frequencies and time scales. For each intrinsic mode component Perform a Fourier transform to obtain the spectrum. The spectral entropy of the intrinsic mode component is calculated based on the spectrum. ,in, The power spectral density represents the intrinsic modal components. Furthermore, spectral entropy was selected. Below the preset threshold of Each intrinsic mode component The signal is reconstructed based on the selected intrinsic mode components to obtain the enhanced signal. , ,in, , This indicates that it includes all spectral entropies below a threshold. The set of indices of the intrinsic mode components.

[0050] The technical solution provided in this embodiment decomposes the filtered acoustic signal under test using a variational mode decomposition algorithm to obtain multiple intrinsic mode components. Using spectral entropy as a screening criterion, high spectral entropy components containing noise are eliminated, and transient feature components with low spectral entropy and related to the impact fault are selected for signal reconstruction, thereby generating an enhanced signal. This process improves the signal-to-noise ratio of the acoustic signal, making the impact characteristics more prominent in the reconstructed signal. The generated time-frequency diagram shows clearer energy bright spots, which helps the subsequent fault diagnosis model to more accurately identify and classify fault signals, significantly improving the accuracy and reliability of fault diagnosis.

[0051] In one implementation, the analytic signal is subjected to a time-frequency distribution transformation based on an adaptive weighted window to generate a time-frequency image of the analytic signal, specifically including the following steps: S51: Obtain the instantaneous frequency gradient of the analytical signal at each moment, and determine multiple striking frequencies based on the instantaneous frequency gradient.

[0052] S53: For any of the tapping frequencies, based on the instantaneous frequency gradient corresponding to the tapping frequency, determine an adaptive weighted window for the tapping frequency, and based on the adaptive weighted window and the target kernel function, determine the output intensity corresponding to the tapping frequency.

[0053] S55: Determine the time-frequency image of the analytical signal based on the output intensity corresponding to each of the said tapping frequencies.

[0054] The analytical signal described above is a complex signal obtained through Hilbert transform, containing the amplitude and phase information of the original signal. The instantaneous frequency gradient described above characterizes the degree to which the instantaneous frequency of the signal changes over time. The impact frequency described above refers to the frequency component in the analytical signal related to the impact fault, such as the transient impact frequency generated by mechanical impact. The adaptive weighted window described above is a window function dynamically adjusted according to the instantaneous frequency gradient, used to optimize resolution at the impact frequency during time-frequency distribution transformation. The time-frequency image described above is a two-dimensional image showing the energy distribution of the analytical signal in both time and frequency dimensions, used for fault diagnosis and feature recognition. Specifically, by integrating the output intensities of all impact frequencies, a complete time-frequency image can be generated. The output intensity described above reflects the salience of the impact frequency in the time-frequency image, which can be achieved through the color brightness of the energy bright spots.

[0055] In one embodiment, for the analyzed signal This determines the adaptive weighting window for the analytic signal. Specifically, its instantaneous frequency... In the discrete domain, it can be approximated as ,in, This represents the phase angle when taken as a complex number. express The conjugate complex number of , further, the instantaneous frequency gradient of the analytic signal is determined based on the change of instantaneous frequency with time, expressed as . Instantaneous frequency gradient A relatively large value indicates that there is an instantaneous frequency jump caused by the impact in the vicinity of that moment. Furthermore, based on the instantaneous frequency gradient... To adjust the window width of the adaptive weighted window so that it can detect drastic changes in frequency ( When the window width is large, immediately adjust the window width parameter to capture the details of that instant at a higher resolution.

[0056] In this embodiment, the adaptive weighted window includes a target time window and a target frequency window, used in time-frequency analysis to optimize time and frequency resolution. Based on step S53 above, the window width parameter (e.g., frequency window width) of the adaptive weighted window is dynamically adjusted based on the instantaneous frequency gradient corresponding to the striking frequency. This window width parameter represents the range of the window function in the time and frequency dimensions; a narrower window width results in higher time and frequency resolution, while a wider window width improves the smoothness of the time-frequency analysis. Specifically, a pre-set initial time window and initial frequency window are obtained. The initial time window is used as the target time window for the striking frequency. Based on the instantaneous frequency gradient corresponding to the striking frequency, the window width parameter of the initial frequency window is adaptively adjusted to obtain the target frequency window width for the striking frequency. The initial frequency window is then updated based on the target frequency window width to obtain the target frequency window for the striking frequency.

[0057] In this embodiment, the target time window is directly set as the target time window of the impact frequency, and the frequency window width parameter of the initial frequency window is adaptively adjusted. Since the size of the time window may be sufficient to accurately locate the time point of the fault occurrence, keeping the time window unchanged ensures the temporal resolution at the moment of the fault occurrence while reducing computational complexity. Specifically, based on the instantaneous frequency gradient corresponding to the impact frequency, the frequency window width parameter of the initial frequency window is adaptively adjusted to obtain the target frequency window width of the impact frequency. The initial frequency window is then updated based on the target frequency window width to obtain the target frequency window of the impact frequency. For regions with large instantaneous frequency gradients, indicating rapid signal changes, a narrower frequency window is needed to capture these transient changes, enhancing the frequency components related to the impact fault and making these frequency components more prominent in the time-frequency image, facilitating subsequent fault identification and classification. By precisely adjusting the window width to optimize frequency resolution, the accuracy and efficiency of fault diagnosis are improved.

[0058] In one embodiment, this time-frequency distribution transformation based on adaptive weighted window and SPWVD algorithm is used as the WK-SPWVD algorithm. The target time window and target frequency window of the striking frequency are constructed according to the WK-SPWVD algorithm: First, the time window width parameter of the initial time window is defined. and the frequency window width parameter of the initial frequency window , , ,in, , These are adaptive factors used to adjust the time window width and frequency window width parameters, respectively, to obtain the adaptively updated target frequency window width, while maintaining the initial time window width. An initial time window is then established based on the initial time window width as the target time window. Establish the target frequency window based on the target frequency window width. , , .

[0059] In this embodiment, the target kernel function of the SPWVD algorithm is updated based on the target time window and target frequency window to serve as the target kernel function of the WK-SPWVD algorithm. The updated target kernel function can be expressed as follows: ,in, For the target frequency window width, For the target time window width, Indicates a point in time The first of the target frequency windows One element, Indicates a point in time At the point where the time smoothing window is located... One element, The first analytic signal represents the... One sample, The term is a complex exponential term used to transform the signal from the time domain to the frequency domain. The output intensity at each frequency point is determined based on the target kernel function, and a time-frequency matrix is ​​constructed based on the output intensity at each frequency point. Each point in the time-frequency matrix represents its own time and frequency, and the element value of that point is the output intensity corresponding to that frequency point.

[0060] Furthermore, the aforementioned time-frequency matrix is ​​converted into a time-frequency image (e.g., through an image library), such that the horizontal axis of the time-frequency image represents the number of time points, and the vertical axis represents the number of frequency points, mapping the matrix element values ​​to color brightness. For example, the aforementioned time-frequency matrix... , indicating signal The time-frequency energy density distribution will The matrix is ​​converted into a pseudo-color image, that is... The values ​​of each matrix element are mapped to a color value (usually an RGB triple) using a color mapping function to form a time-frequency image. The time-frequency image features a frequency line that suddenly appears or brightens significantly at frequency transition points.

[0061] Compared to the traditional SPWVD algorithm, the WK-SPWVD algorithm, which performs time-frequency distribution transformation, offers several advantages. In regions where signal changes are gradual and frequency components are stable, the target kernel function can provide stronger smoothing (using a wider window) and more thoroughly suppress cross-terms. In regions where signal frequencies change rapidly (such as transients and frequency abrupt changes), the target kernel function can provide weaker smoothing (using a narrower window), resulting in more accurate instantaneous frequency localization and clearer capture of rapid changes. Through this adaptive adjustment, the WK-SPWVD algorithm, compared to the fixed-kernel SPWVD algorithm, can provide sharper frequency localization in regions requiring high resolution, while offering better cross-term suppression in regions requiring a clean spectrum.

[0062] The technical solution provided in this embodiment generates a time-frequency image from an analytical signal through time-frequency distribution transformation based on an adaptive weighted window, thereby improving the accuracy and efficiency of compressor knocking fault diagnosis. This solution first calculates the instantaneous frequency gradient of the analytical signal to determine the frequency components related to the knocking fault. Then, it dynamically adjusts the window width parameters of the adaptive weighted window for each knocking frequency, including the target time window and the target frequency window, to optimize the time and frequency resolution. Using the updated window function and target kernel function, the output intensity of each knocking frequency is calculated, and the output intensities of all knocking frequencies are integrated to generate a complete time-frequency image. This time-frequency image displays the energy distribution of the knocking fault through color and brightness, providing an intuitive basis for fault diagnosis. Specifically, by defining an initial window width parameter and adaptively adjusting it according to the instantaneous frequency gradient, the transient changes of the signal are accurately captured, enhancing the frequency components related to the knocking fault, making the fault characteristics more prominent in the time-frequency image, facilitating subsequent fault identification and classification, and thus improving the accuracy and efficiency of fault diagnosis.

[0063] In one implementation, the above-mentioned fault diagnosis model is pre-trained according to the following steps: S71: Obtain multiple image training sets and image verification sets, construct an initial convolutional neural network model, and iteratively train the initial convolutional neural network model using the image training sets; S73: In any iteration of the iterative training, the parameters of the convolutional neural network model in the current iteration are updated using any image training set, and the updated convolutional neural network model is evaluated using the image validation set. S75: If the evaluation results indicate that the updated convolutional neural network model meets the preset conditions, stop iterative training and use the trained convolutional network model as the fault diagnosis model.

[0064] The aforementioned image training set and image validation set consist of time-frequency image data labeled with correct classification tags, including time-frequency images under normal and fault conditions. First, an initial convolutional neural network model containing a convolutional neural network is established, and the model parameters are initialized. Model training begins using the image training set, allowing the model to learn the relationship between data features and labels (normal or fault). Forward propagation is used to calculate prediction results, and backpropagation is used to update model parameters, enabling the model to learn fault-related features in the time-frequency images. During iterative training, the model parameters are adjusted based on the gradient information of the loss function to improve the model's prediction performance.

[0065] Furthermore, the updated convolutional neural network model is updated using an image validation set. Model performance is validated based on preset conditions, and a decision is made whether to stop training. Specifically, these preset conditions can be that the accuracy of the image validation set reaches a specified value. Model parameters that meet these preset conditions are saved, and the current convolutional neural network model is used as the final fault diagnosis model. By pre-training and evaluating the fault diagnosis model through the above steps, it can be ensured that the model accurately classifies time-frequency images in practical applications, identifying whether the compressor's operating status is normal or faulty.

[0066] In this embodiment, any image training set is used to update the parameters of the convolutional neural network model in the current iteration process. The model parameters are optimized through multiple iterations to minimize the error between the predicted result and the true label. Specifically, any image training set is input into the convolutional neural network model in the current iteration process to obtain the prediction result for the image training set. Further, the true label of the image training set is obtained, and the loss error between the predicted result and the true label is determined. Based on the loss error, the parameters of the current convolutional neural network model are updated. The true label can be the actual classification or regression value of each sample in the dataset. The loss error can be understood as the difference between the predicted result and the true label, usually quantified by a loss function. The smaller the loss error, the closer the model's prediction result is to the true label. Adjusting the model parameters according to the gradient information of the loss function can reduce the prediction error and improve the model's prediction accuracy.

[0067] In this embodiment, the convolutional neural network model includes an input layer, multiple convolutional layer combinations, a fully connected layer, and an output layer. Each convolutional layer combination includes a convolutional layer, a pooling layer, and a normalization layer. In adjacent convolutional layer combinations, the output of the pooling layer of the previous combination is used as the input of the convolutional layer of the next combination. During training, the input layer receives a training set of images containing multiple time-frequency images and inputs these images into a convolutional layer array. In each array, each convolutional layer extracts features from the images, including knocking fault features and normal features, to generate a feature map for each time-frequency image. These feature maps are then normalized and dimensionality-reduced using normalization and pooling layers. The processed feature maps are then input into the next convolutional layer array until the output reaches the fully connected layer. The fully connected layer flattens the extracted multidimensional features (knocking fault features and normal features) into one-dimensional vectors (fault feature vector and normal feature vector), mapping these vectors to the output space to form vector representations of each feature. This output space is used in subsequent model applications. The output layer performs classification based on a loss function using the vector representations from the training process to determine whether the time-frequency image of the test sound signal possesses knocking fault features. The loss function can be cross-entropy loss, which measures the difference between the model's predicted probability distribution and the true label.

[0068] For example, the aforementioned fault feature vector and normal feature vector are determined based on the RGB values ​​of the time-frequency image. The time-frequency image is divided into an N×N color block matrix, where each color block represents a matrix element value. The RGB value of each color block can be viewed as a three-dimensional feature vector, where the RGB value represents the energy distribution of the signal at different times and frequencies. For details, please refer to... Figure 2 Any time-frequency image is input into a convolutional neural network model (the fault diagnosis model to be trained). This model consists of four convolutional layers. The RGB features of the image are input into the convolutional layer combination for feature recognition and classification. The identified knocking fault features and normal features are mapped to the output space in vector form through fully connected layers. The output layer outputs the probability of each category (including fault and normal) through an activation function (such as the softmax activation function). As the number of convolutional layers increases, the model can learn features from simple to complex. Early layers may learn basic edges and textures, while deeper layers may learn more complex patterns and object features. Each layer further extracts and abstracts features based on the previous layer, forming a hierarchical feature representation, which improves the automation and accuracy of fault diagnosis.

[0069] In this embodiment, the aforementioned image training set and image validation set are obtained as follows: Acoustic signal data of the compressor during operation are collected, including 300 sets of normal data and 20 sets of impact data, for subsequent production of the image training set and image validation set. The collected acoustic signals are processed by removing DC components and outliers, and then transformed from the time domain to the frequency domain using an FFT algorithm to obtain the signal's spectrum. The spectrum is then bandpass filtered (retaining signals within the 300~8000Hz frequency range) to obtain the filtered signal. The filtered acoustic signal is then subjected to VMD decomposition to obtain several intrinsic mode components. The spectral entropy of each intrinsic mode component is calculated, and low spectral entropy (high impact) components are selected and reconstructed to enhance the impact-type transient characteristics. The processed signal is then converted using the WK-SPWVD algorithm, and corresponding time-frequency images are created from the converted data. The completed image dataset is then divided into the image training set and image validation set at a 9:1 ratio.

[0070] For example, performing WK-SPWVD algorithm (SPWVD algorithm based on adaptive weighted window) transformation on the processed signal and creating a corresponding time-frequency image of the transformed data includes: for the processed signal, obtaining the instantaneous frequency gradient of the analytical signal at each time step; determining multiple striking frequencies based on the instantaneous frequency gradient; for any striking frequency, determining an adaptive weighted window for the striking frequency based on the instantaneous frequency gradient corresponding to the striking frequency; determining the output intensity corresponding to the striking frequency based on the adaptive weighted window and the target kernel function; and determining the time-frequency image of the analytical signal based on the corresponding output intensity of each striking frequency. The target kernel function is expressed as follows: ,in, For the target frequency window width, For the target time window width, Indicates a point in time The first of the target frequency windows One element, Indicates a point in time At the point where the time smoothing window is located... One element, The first analytic signal represents the... One sample, This is a complex exponential term used to convert a signal from the time domain to the frequency domain.

[0071] In this embodiment, the hyperparameters to be adjusted during training include the adaptive factor for the frequency window width and the model parameters. The target frequency window width is expressed as... ,in, For the target frequency window width, For instantaneous frequency gradient, For adaptive factors, adaptive factors The value of can be dynamically adjusted based on the validation results of the fault diagnosis model during training. The optimal value is selected through the cross-validation results output by the loss function. Simultaneously, during training, model parameters are dynamically adjusted based on the cross-validation results. Taking CNN as an example, this includes kernel size, number of convolutional layers, etc., and the optimal combination is found by traversing different hyperparameter combinations.

[0072] In one embodiment, please refer to Figures 3(a), 3(b), and 3(c). Figure 3(a) shows the time-frequency image obtained after processing by the WK-SPWVD algorithm, Figure 3(b) shows the time-frequency image obtained after processing by the SPWVD algorithm, and Figure 3(c) shows the time-frequency image obtained from a normal signal without processing. In the figures, colors represent the energy distribution of the signal at different times and frequencies; generally, brighter colors indicate higher energy, and areas of concentrated energy (bright areas) may correspond to fault features in the signal. The image in Figure 3(a) shows clear frequency components and time variations, indicating that the WK-SPWVD algorithm can effectively capture the transient characteristics of the signal. This high-resolution time-frequency image helps to more accurately identify and locate faults. Compared to Figure 3(a), Figure 3(b) has lower resolution and less obvious fault features. Figure 3(c) is a time-frequency image obtained directly from a normal signal without any time-frequency analysis algorithm processing, showing a relatively uniform energy distribution and no obvious fault features. The advantages of the WK-SPWVD algorithm in improving the resolution and accuracy of time-frequency analysis are evident. It can more clearly capture the transient characteristics of signals, thus aiding in fault diagnosis. Experimental results show that the accuracy rate of knocking fault identification after processing with the WK-SPWVD algorithm reaches 96%, significantly higher than the 91.7% accuracy rate of knocking fault identification after processing with the SPWVD algorithm.

[0073] The technical solution provided in this embodiment diagnoses compressor knocking faults by pre-training a fault diagnosis model. The fault diagnosis model can automatically extract and classify features from time-frequency images. By dynamically adjusting the window width parameter of the adaptive weighted window, the model can optimize the resolution of time-frequency analysis, more clearly capture the transient features of the signal, and thus improve the accuracy of fault diagnosis. Specifically, the fault diagnosis model divides the time-frequency image into a color block matrix and determines the feature vector based on its RGB values. Through multiple iterations to optimize the model parameters, the error between the predicted result and the true label is minimized, improving the model's prediction accuracy and ensuring that the model can effectively classify time-frequency images and accurately identify whether the compressor's operating status is normal or faulty, thereby achieving automated and highly accurate fault diagnosis.

[0074] Please see Figure 4 This application also provides a compressor knocking fault diagnosis device, the device comprising: The filtering processing unit 100 is used to acquire the sound signal to be tested during compressor operation and to filter the sound signal to be tested. The noise processing unit 200 is used to perform variational mode decomposition and reconstruction on the filtered acoustic signal to be tested to generate a corresponding enhanced signal, and to perform analytical processing on the enhanced signal to obtain a corresponding analytical signal. The image generation unit 300 is used to perform a time-frequency distribution transformation based on an adaptive weighted window on the analyzed signal to generate a time-frequency image of the analyzed signal, wherein the window width parameter of the adaptive weighted window is dynamically adjusted based on the instantaneous frequency gradient of the analyzed signal. The fault diagnosis unit 400 is used to input the time-frequency image into a pre-trained fault diagnosis model, and the fault diagnosis model outputs the fault diagnosis result of the compressor.

[0075] in, In one embodiment, the filtering processing unit 100 is specifically used to acquire the sound signal to be tested during the operation of the compressor, remove the DC component and outliers of the sound signal to be tested to obtain a preliminary processed signal of the sound signal to be tested, perform frequency domain transformation on the preliminary processed signal to obtain a spectrum of the preliminary processed signal, perform bandpass filtering on the spectrum, and perform time domain transformation on the spectrum after bandpass filtering to complete the filtering processing of the sound signal to be tested.

[0076] In one embodiment, the noise processing unit 200 is specifically used to perform variational mode decomposition on the filtered sound signal to be tested to obtain multiple intrinsic mode components, select the percussion characteristic mode component of the sound signal to be tested from the multiple intrinsic mode components, and reconstruct the signal based on the percussion characteristic mode component to obtain the enhanced signal corresponding to the sound signal to be tested, and perform analytical processing on the enhanced signal to obtain the corresponding analytical signal.

[0077] In one embodiment, the image generation unit 300 is specifically used to acquire the instantaneous frequency gradient of the analytical signal at each moment, determine multiple tapping frequencies based on the instantaneous frequency gradient, determine an adaptive weighted window for any tapping frequency based on the instantaneous frequency gradient corresponding to the tapping frequency, determine the output intensity corresponding to the tapping frequency based on the adaptive weighted window and the target kernel function, and determine the time-frequency image of the analytical signal based on the output intensity corresponding to each tapping frequency.

[0078] In one embodiment, the device further includes a model training unit, which is specifically used to acquire multiple image training sets and image verification sets, construct an initial convolutional neural network model, iteratively train the initial convolutional neural network model using the image training sets, update the parameters of the convolutional neural network model in the current iteration using any image training set during any iteration of the iterative training, evaluate the updated convolutional neural network model using the image verification set, and stop iterative training if the evaluation result indicates that the updated convolutional neural network model meets preset conditions, and use the trained convolutional network model as the fault diagnosis model.

[0079] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0080] The compressor knocking fault diagnosis device in this application embodiment is presented in the form of a functional unit. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, or other devices that can provide the above functions.

[0081] Please see Figure 5 , Figure 5 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application, such as... Figure 5 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 5 Take a processor 10 as an example.

[0082] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0083] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0084] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0085] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0086] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0087] This application also provides a computer-readable storage medium. The methods described in this application can be implemented in hardware or firmware, or implemented as recordable on a storage medium, or implemented as computer code downloaded over a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and subsequently stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the methods shown in the above embodiments are implemented.

[0088] This application provides a computer program product including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the method of any embodiment of this application.

[0089] The apparatus, module, or unit described in the above embodiments can be implemented by a computer chip or entity, or by a product having a certain function. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0090] For ease of description, the above devices are described separately by function as various units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0091] Those skilled in the art will understand that embodiments of this application can be provided as methods, apparatus, or computer devices. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0092] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus, and computer devices 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 1 A device that provides the functions specified in one or more boxes.

[0093] 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 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0094] 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.

[0095] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0096] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the apparatus embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0097] The above description is merely an embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of this application should be included within the scope of the claims of this application.

[0098] Although embodiments of this application have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of this application, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for diagnosing compressor knocking faults, characterized in that, The method includes: Acquire the acoustic signal to be measured during the operation of the compressor, and filter the acoustic signal to be measured; The filtered acoustic signal under test is subjected to variational mode decomposition and reconstruction to generate a corresponding enhanced signal. The enhanced signal is then analyzed to obtain a corresponding analytical signal. The analyzed signal is subjected to a time-frequency distribution transformation based on an adaptive weighted window to generate a time-frequency image of the analyzed signal, wherein the window width parameter of the adaptive weighted window is dynamically adjusted based on the instantaneous frequency gradient of the analyzed signal; The time-frequency image is input into a pre-trained fault diagnosis model, which then outputs the fault diagnosis result of the compressor.

2. The method according to claim 1, characterized in that, Acquiring the acoustic signal to be measured from the compressor and filtering the acoustic signal to be measured includes: The DC component and outliers of the sound signal under test are removed to obtain a preliminary processed signal of the sound signal under test. The preliminary processed signal is frequency domain transformed to obtain the spectrum of the preliminary processed signal; The spectrum is bandpass filtered, and the bandpass filtered spectrum is then time-domain transformed to complete the filtering process of the acoustic signal under test.

3. The method according to claim 2, characterized in that, Bandpass filtering of the spectrum includes: Obtain the target frequency component and noise frequency component in the spectrum, wherein the target frequency component is in a specific frequency range and the noise frequency component is in a non-specific frequency range; Noise frequency components in the spectrum are filtered out in order to retain the target frequency components in the spectrum.

4. The method according to claim 1, characterized in that, Variational mode decomposition and reconstruction are performed on the filtered acoustic signal to generate the corresponding enhanced signal, including: Variational mode decomposition is performed on the filtered acoustic signal to obtain multiple intrinsic mode components; Among the multiple intrinsic mode components, the impact feature mode component of the sound signal under test is selected, and the signal is reconstructed based on the impact feature mode component to obtain the enhanced signal corresponding to the sound signal under test.

5. The method according to claim 1, characterized in that, Performing an adaptive weighted window-based time-frequency distribution transformation on the analyzed signal to generate a time-frequency image of the analyzed signal includes: The instantaneous frequency gradient of the analytical signal at each moment is obtained, and multiple striking frequencies are determined based on the instantaneous frequency gradient; For any of the aforementioned tapping frequencies, an adaptive weighted window for the tapping frequency is determined based on the instantaneous frequency gradient corresponding to the tapping frequency, and the output intensity corresponding to the tapping frequency is determined based on the adaptive weighted window and the target kernel function. The time-frequency image of the analytical signal is determined based on the output intensity corresponding to each of the said tapping frequencies.

6. The method according to claim 5, characterized in that, The adaptive weighted window includes a target time window and a target frequency window; the adaptive weighted window for determining the striking frequency based on the instantaneous frequency gradient corresponding to the striking frequency includes: Obtain an initial time window and an initial frequency window, and use the target time window as the target time window for the tapping frequency; Based on the instantaneous frequency gradient corresponding to the striking frequency, the frequency window width parameter of the initial frequency window is adaptively adjusted to obtain the target frequency window width of the striking frequency. The initial frequency window is updated based on the target frequency window width to obtain the target frequency window for the tapping frequency.

7. The method according to claim 1, characterized in that, The fault diagnosis model is pre-trained in the following manner: Multiple image training sets and image verification sets are obtained, and an initial convolutional neural network model is constructed. The initial convolutional neural network model is then iteratively trained using the image training sets. In any iteration of the iterative training, the parameters of the convolutional neural network model in the current iteration are updated using any image training set, and the updated convolutional neural network model is evaluated using the image validation set. If the evaluation results indicate that the updated convolutional neural network model meets the preset conditions, the iterative training is stopped, and the trained convolutional network model is used as the fault diagnosis model.

8. The method according to claim 7, characterized in that, Updating the parameters of the convolutional neural network model in the current iteration using any image training set includes: Input any image training set into the convolutional neural network model in the current iteration process to obtain the prediction result of the image training set; Obtain the true labels of the image training set, determine the loss error between the prediction result and the true labels, and update the parameters of the current convolutional neural network model based on the loss error.

9. A compressor knocking fault diagnosis device, characterized in that, The device includes: A filtering processing unit is used to acquire the sound signal to be tested during compressor operation and to filter the sound signal to be tested. The noise processing unit is used to perform variational mode decomposition and reconstruction on the filtered acoustic signal to be tested to generate a corresponding enhanced signal, and to perform analytical processing on the enhanced signal to obtain a corresponding analytical signal. An image generation unit is used to perform a time-frequency distribution transformation based on an adaptive weighted window on the analyzed signal to generate a time-frequency image of the analyzed signal, wherein the window width parameter of the adaptive weighted window is dynamically adjusted based on the instantaneous frequency gradient of the analyzed signal; The fault diagnosis unit is used to input the time-frequency image into a pre-trained fault diagnosis model, and the fault diagnosis model outputs the fault diagnosis result of the compressor.

10. A computer device, characterized in that, include: A memory and a processor are communicatively connected, the memory stores computer instructions, and the processor executes the computer instructions to perform the compressor knocking fault diagnosis method according to any one of claims 1 to 7.

Images