Large compressor fault diagnosis method, medium and computing equipment

Abstract

The invention discloses a large compressor fault diagnosis method, a medium and computing equipment. The method comprises the following steps: selecting a multi-parameter corrosion monitoring sensor; obtaining sequential signals of a rotating machine of the compressor under different working conditions based on a sensor; the time sequence signals are preprocessed; virtual samples of different fault states and noise levels are generated based on a self-adaptive multi-feature generation model; constructing a multi-modal fusion convolutional network model, and performing training and verification; and based on the trained multi-modal fusion convolutional network model, carrying out fault diagnosis on the large compressor. According to the method, fault diagnosis is carried out on the large compressor by constructing a multi-modal fusion convolutional network model and based on the trained multi-modal fusion convolutional network model, high-resolution time-frequency features are dynamically extracted, and the method can adapt to complex non-stationary signals; according to the method, virtual samples of different fault states and noise levels are generated through a self-adaptive multi-feature generation model; according to the method, time domain, frequency domain and envelope spectrum features in an original sample are extracted through a multi-modal feature extraction method, and the diagnosis precision and robustness are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology for large compressors, and in particular to a fault diagnosis method, medium, and computing device for large compressors. Background Technology

[0002] Large compressors refer to compressors with a displacement exceeding 100 cubic meters per minute (m³ / min). As core equipment in industrial systems, they are widely used in petrochemical, power, aerospace, and manufacturing industries. The stability of their operation directly affects the system's safety, production efficiency, and economic benefits. However, due to prolonged exposure to high loads, variable environments, and complex operating conditions, compressors often experience various types of failures, such as bearing wear, blade cracks, and seal failure. These failures can lead to decreased equipment performance, increased energy consumption, and even serious safety accidents. Therefore, how to efficiently and accurately monitor the condition and diagnose faults in compressors is a major technical challenge in the industrial field.

[0003] In the development of fault diagnosis technology, traditional methods have primarily relied on signal processing and machine learning, depending mainly on expert experience in specific fields and hand-designed feature extraction algorithms. For example, methods such as Fourier Transform (FT), Short-Time Fourier Transform (STFT), and Wavelet Transform (WT) are used to transform the original signal from the time domain to the frequency domain or time-frequency domain, and fault-related features are extracted for classification. However, these methods have significant limitations in many aspects.

[0004] To address the limitations of traditional methods, researchers have attempted to combine the physical interpretability of traditional signal processing with the automated feature extraction capabilities of deep learning, integrating time-frequency analysis with deep learning techniques—a novel research direction. For example, by using time-frequency plots (such as wavelet time-frequency plots and short-time Fourier time-frequency plots) as input to CNNs, time-frequency domain information can be effectively utilized to improve the accuracy of fault classification. However, the aforementioned methods still suffer from the following problems: Insufficient feature adaptability: Traditional time-frequency plot generation methods are mostly designed with fixed parameters and lack the ability to dynamically adjust according to data, resulting in insufficient representation of non-stationary signals by the model.

[0005] Limitations of cross-domain transfer: Existing deep learning models based on time-frequency graphs lack the ability to handle the differences in distribution between the source and target domains, resulting in poor performance in cross-domain diagnosis.

[0006] Computational efficiency and real-time bottlenecks: Using time-frequency graphs as input to deep learning models increases computational overhead and makes it difficult to meet the real-time requirements of industrial sites. Summary of the Invention

[0007] The main objective of this invention is to provide a method, medium, and computing device for diagnosing faults in large compressors, in order to solve the problems existing in the prior art.

[0008] To achieve the above objectives, the present invention provides a method for fault diagnosis of a large compressor, the method comprising: S10, acquiring temporal signals of the rotating machinery of the compressor under different operating conditions based on sensors; S20, preprocessing the temporal signals; S30, generating virtual samples with different fault states and noise levels based on an adaptive multi-feature generation model and the preprocessed temporal signals; S40, constructing a multimodal fusion convolutional network model based on the virtual samples, and training and validating it; S50, performing fault diagnosis of the large compressor based on the trained multimodal fusion convolutional network model.

[0009] Optionally, step S20 includes the following steps: S210, performing noise reduction processing on the time-series signal; S220, performing segmentation processing on the continuous time signal in the time-series signal based on sliding window technology; S230, performing standardization processing on the time-series signal.

[0010] Optionally, step S210 includes the following steps: S2110, dynamically adjusting the cutoff frequency based on an adaptive low-pass filter to remove high-frequency components while retaining fault characteristic frequencies, wherein the high-frequency components refer to time-series signals with frequencies from f1 to ∞; S2120, performing discrete wavelet decomposition on the time-series signal to decompose it into different frequency bands, removing low-energy noise components using a soft thresholding method, and reconstructing the denoised time-series signal; S2130, decomposing the time-series signal into a series of intrinsic mode functions based on EMD technology; S2140, performing frequency domain analysis on each intrinsic mode function to remove high-frequency noise modes and reconstruct the time-series signal, wherein the high-frequency noise refers to sound with a frequency range of 1000Hz-20000Hz.

[0011] Optionally, after step S20, the following steps are also included: S2001, performing feature enhancement on the preprocessed time-series signal; S2002, extracting the signal envelope based on short-time Fourier transform or Hilbert transform to enhance the separability of fault features; S2003, performing dimensionality reduction on the time-series signal based on principal component analysis.

[0012] Optionally, step S30 includes the following steps: S310, extracting time-domain, frequency-domain, and envelope spectrum features from the original sample based on a multimodal feature extraction method; S320, constructing an adaptive multi-feature generation model; S330, inputting the extracted time-domain, frequency-domain, and envelope spectrum features into the adaptive multi-feature generation model for training; S340, generating virtual samples with different fault states and noise levels based on the trained adaptive multi-feature generation model.

[0013] Optionally, step S40 includes the following steps: S410, constructing a multimodal fusion convolutional network model, the multimodal fusion convolutional network model including an input layer, a convolutional layer, a pooling layer, an attention mechanism layer, a fusion layer, and a fully connected layer; S420, performing channel fusion, spatial fusion, and final fusion on different modal features of virtual samples based on a spatial attention mechanism to complete adaptive weighting of features, perform compressor fault state diagnosis, and verify the judgment results; S430, training the multimodal fusion convolutional network model.

[0014] Optionally, step S50 includes the following steps: S510, performing multimodal fusion convolutional network model verification based on the fault data in the original sample; S520, performing large compressor fault diagnosis based on the trained multimodal fusion convolutional network model.

[0015] In addition, to achieve the above objectives, embodiments of this application also provide a computer-readable storage medium including instructions that, when executed on a computer, cause the computer to perform the large compressor fault diagnosis method described in any embodiment of this application.

[0016] Furthermore, to achieve the above objectives, embodiments of this application also provide a computing device, the computing device comprising: at least one processor, a memory, and an input / output unit; wherein, the memory is used to store a computer program, and the processor is used to call the computer program stored in the memory to execute the large compressor fault diagnosis method described in any embodiment of this application.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: The large compressor fault diagnosis method provided in this application constructs a multimodal fusion convolutional network model and performs fault diagnosis based on the trained multimodal fusion convolutional network model. It dynamically extracts high-resolution time-frequency features and can adapt to complex non-stationary signals. The method generates virtual samples with different fault states and noise levels through an adaptive multi-feature generation model. The method extracts time-domain, frequency-domain, and envelope spectrum features from the original samples through a multimodal feature extraction method, thereby realizing dynamic time-frequency feature extraction and cross-domain transfer of compressor multi-condition signals, significantly improving diagnostic accuracy and robustness. Attached Figure Description

[0018] Figure 1 A flowchart of a large compressor fault diagnosis method provided in the embodiments of this application; Figure 2 This is a flowchart of the adaptive multi-feature generation process provided in the embodiments of this application; Figure 3A flowchart of a multimodal fusion convolutional network provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of the medium provided in the embodiments of this application; Figure 5 A schematic diagram of the structure of a computing device provided in an embodiment of this application.

[0019] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of the application. Rather, these embodiments are provided to make the disclosure more thorough and complete, and to fully convey the scope of the disclosure to those skilled in the art.

[0021] To address the aforementioned technical problems, embodiments of this application provide a method for diagnosing faults in large compressors. This method can be executed by a computer, such as... Figure 1 As shown, the method may include the following steps: S10, based on sensors, acquires the timing signals of the compressor's rotating machinery under different operating conditions.

[0022] Specifically, this embodiment employs multimodal signal acquisition technology to acquire real-time compressor operating data through vibration, temperature, and pressure sensors. These sensors are deployed in critical locations, including bearings, blades, compression chambers, and air passage interfaces, to ensure that the acquired signals cover the core parameters of equipment operation and potential fault characteristics.

[0023] Furthermore, the different operating conditions include normal operation, abnormal conditions, and fault conditions, and the sensors include vibration, temperature, and pressure sensors.

[0024] In this exemplary embodiment, the locations of the measurement points for collecting operational data are shown in Table 1: Table 1

[0025] S20, preprocess the timing signal.

[0026] In an exemplary embodiment, step S20 may include the following steps: S210, Denoise the timing signal; S220, The continuous time signal in the time sequence signal is segmented based on the sliding window technique; S230, the timing signal is standardized.

[0027] Specifically, after preprocessing the time-series signals, raw samples can be obtained, which include time-series signals of normal operation, abnormal state, and fault state.

[0028] Sliding window technology is a commonly used technique in computer vision, primarily for tasks such as object detection, image or signal classification. Its core idea is to gradually slide a fixed-size rectangular window (which can be resized) across the image, covering every region. For each window region, a classifier (such as HOG + SVM, convolutional neural networks, etc.) is used to determine whether the region contains a target object, or to segment the signal.

[0029] In an exemplary embodiment, step S210 may include the following steps: S2110, based on an adaptive low-pass filter, dynamically adjusts the cutoff frequency to remove high-frequency components while retaining fault characteristic frequencies. The high-frequency components refer to time-series signals with frequencies from f1 to ∞. S2120, Perform discrete wavelet decomposition on the time-series signal to decompose the time-series signal into different frequency bands, combine with the soft thresholding method to remove low-energy noise components, and reconstruct the denoised time-series signal. S2130, the time-series signal is decomposed into a series of intrinsic mode functions (IMFs) based on EMD technology. S2140, perform frequency domain analysis on each intrinsic mode function (IMF), remove high-frequency noise modes, and reconstruct the time-series signal.

[0030] High-frequency noise refers to sound with a frequency range of 1000Hz-20000Hz (such as electromagnetic interference and environmental vibration).

[0031] Specifically, in step S220, the continuous-time signal is segmented using a sliding window technique. The length of each window is calculated based on the target frequency range and sampling rate (e.g., 200ms to 500ms). The segmentation strategy is designed with a fixed window length and dynamic overlap (e.g., 50%), which ensures the integrity of the segmented signal while capturing the continuous characteristics between time-series signals.

[0032] In step S230, the acquired time-series signals are subjected to data standardization processing, including mean removal, normalization, synchronization alignment, and interpolation completion.

[0033] Furthermore, step S230 normalizes the time-series signal to unify the feature scale and avoid model training instability caused by differences in units. Zero-mean unit variance (Z-score) or min-max normalization methods are used to normalize the signal amplitude to the [-1,1] or [0,1] interval.

[0034] An adaptive filter is a filter that uses an adaptive algorithm to change its parameters and structure in response to changes in the environment. Generally, the structure of an adaptive filter remains unchanged. The coefficients of an adaptive filter are time-varying coefficients updated by the adaptive algorithm.

[0035] Empirical Mode Decomposition (EMD) is an adaptive time-frequency analysis method for analyzing nonlinear and non-stationary signals. The core idea of ​​EMD is to decompose a complex signal layer by layer into several intrinsic mode functions (IMFs) with specific characteristics. Each IMF component represents a vibrational mode at a different scale in the signal. Specifically, EMD identifies and extracts local features in the signal, gradually stripping away high-frequency to low-frequency components to achieve multi-component decomposition. This process is based on the signal's local time-domain characteristics and does not depend on any external basis functions.

[0036] In this exemplary embodiment, preprocessing the timing signal may further include the following steps: Use a bandpass filter to remove high-frequency and low-frequency noise from the normalized signal; By decomposing the signal with noise removed, the signal is decomposed into resonant and non-resonant components, thereby enhancing the fault characteristics. The non-resonant components are removed, and the resonant components are retained for reconstruction to generate a denoised signal, thus completing the signal preprocessing.

[0037] Furthermore, key features (such as time-domain and frequency-domain features) are extracted from the standardized and denoised signals to provide a foundation for the generation of virtual samples. Time-domain and frequency-domain features are extracted, and the mean, variance, peak value, and kurtosis of the signal data are calculated. Frequency-domain features are extracted, and the spectral density, dominant frequency, and sideband ratio are calculated. Envelope detection is performed on the signal by extracting envelope spectral features, and the frequency components of the envelope are extracted.

[0038] In this embodiment, in order to accurately simulate the dynamic behavior of a large compressor under different fault conditions, a new loss function is derived. This loss function combines reconstruction error and KL divergence to optimize the performance of the variational autoencoder (VAE).

[0039] The extracted features are used to generate diverse virtual fault samples to simulate feature combinations of different fault states and noise levels.

[0040] In an exemplary embodiment, after step S20, the following steps may also be included: S2001, Perform feature enhancement on the preprocessed temporal signal; S2002 extracts the signal envelope based on short-time Fourier transform or Hilbert transform, enhancing the separability of fault features; S2003, Dimensionality reduction of the time-series signal is performed based on principal component analysis to reduce data redundancy and improve processing efficiency.

[0041] S30 generates virtual samples with different fault states and noise levels based on an adaptive multi-feature generation model.

[0042] In an exemplary embodiment, reference is made to Figure 2 Step S30 may include the following steps: S310, based on a multimodal feature extraction method, extracts time-domain, frequency-domain, and envelope spectrum features from the original sample; S320, constructing an adaptive multi-feature generation model, specifically as follows: Figure 2 As shown; S330, The extracted time-domain, frequency-domain, and envelope spectrum features are input into the adaptive multi-feature generation model for training; S340, based on the trained adaptive multi-feature generation model, generates virtual samples with different fault states and noise levels.

[0043] The loss function of the adaptive multi-feature generation model is: 𝐿=𝐿recon+𝛽⋅𝐿 KL Where Lrecon is the mean squared error between the input features and the generated features; 𝐿 KL is the KL divergence between the latent variables and the standard normal distribution; is a hyperparameter used to control the trade-off between the reconstruction error Lrecon and the KL divergence.

[0044] A further technical solution of the present invention is: the method for extracting time-domain, frequency-domain, and envelope spectrum features from the original sample is as follows: By calculating the mean, variance, peak value, and kurtosis of the time-domain signal, as well as the spectral density, dominant frequency, and sideband ratio of the frequency-domain signal, and combining this with envelope spectrum analysis, a high-dimensional feature vector is generated.

[0045] Mean removal: Eliminate the DC component and center the signal so that the mean of the signal is 0.

[0046] Normalization: Scaling data to a standard range (such as 0 to 1 or -1 to 1) to reduce the impact of differences in numerical ranges.

[0047] Noise reduction: Use filters (such as bandpass filters) to remove high-frequency and low-frequency noise while preserving intermediate-frequency information.

[0048] Envelope spectroscopy is used to analyze the characteristics of non-stationary signals (such as shock and wear signals). Envelope detection: First, perform a Hilbert transform on the signal to obtain the signal envelope.

[0049]

[0050] in, It is a signal Hilbert transform, It is an envelope signal.

[0051] The extracted time-domain, frequency-domain, and envelope spectral features are combined to form a high-dimensional feature vector:

[0052] Where μ represents the mean; σ² represents the mean; Peak represents the peak value; Kurtosis represents the kurtosis; and PSD represents the power spectral density, which indicates the energy distribution of the signal at different frequencies.

[0053] These features can be used for subsequent model training and fault diagnosis. Through these steps, the vibration signal features of large compressors can be comprehensively extracted, providing sufficient data information for fault detection.

[0054] To further enhance diagnostic robustness, this invention introduces dynamic feature focusing and few-sample optimization strategies. The dynamic feature focusing module uses a multi-head attention mechanism to weight the extracted time-frequency features, automatically focusing on key fault frequency bands and important feature regions while filtering out low-value information, significantly improving the model's adaptability to complex signals. When there are insufficient samples in the target domain, this invention employs a meta-learning-based optimization method. Leveraging the rapid learning capabilities of the source domain task, it helps the model quickly adapt to small sample data in the target domain. This meta-learning strategy not only solves the problem of scarce labeled data in industrial settings but also enables the model to quickly complete online optimization as data is updated.

[0055] In the multimodal time-frequency feature extraction stage, this invention designs an adaptive time-frequency convolutional layer (TFconv). This module combines complex convolution kernel technology with dynamic time-frequency feature extraction, extracting high-resolution time-frequency features from the signal by embedding short-time Fourier transform (STFT), chirplet transform, and Morlet wavelet kernel function. The convolution kernel parameters (including center frequency, bandwidth, etc.) are dynamically optimized through backpropagation during model training to ensure that the extracted features are adapted to the characteristics of signals under different operating conditions. Furthermore, to integrate multimodal features from multiple sensors, this invention proposes a feature fusion method based on an attention mechanism. Through dynamic weight allocation, features of key fault modes are weighted to eliminate interference from sensor noise and redundant signals, thereby enhancing the discriminability of fault modes. Extract time-domain, frequency-domain, and envelope spectrum features from the original samples; then, construct an adaptive multi-feature generation model for feature extraction, and input the extracted time-domain, frequency-domain, and envelope spectrum features into the adaptive multi-feature generation model for training. Use the trained adaptive multi-feature generation model to generate virtual samples with different fault states and noise levels; the loss function of the adaptive multi-feature generation model is: 𝐿=𝐿recon+𝛽⋅𝐿 KL Where Lrecon is the mean squared error between the input features and the generated features, 𝐿 KL It is the KL divergence between the latent variables and the standard normal distribution; in the loss function of the adaptive multi-feature generation model, 𝛽 is a hyperparameter used to control the trade-off between the reconstruction error Lrecon and the KL divergence.

[0056] The adaptive time-frequency convolutional layer's convolution kernel consists of real and imaginary parts, embedding short-time Fourier transform (STFT), chirplet, and Morlet wavelet kernel functions, enabling it to model the dynamic changes of signals at different frequencies and time scales. It dynamically optimizes convolution kernel parameters (such as center frequency and bandwidth) to adapt to the characteristics of non-stationary signals.

[0057] The output features a high-resolution time-frequency plot containing the signal's temporal and frequency distribution characteristics for subsequent analysis. Time-frequency features from different modes, such as vibration, temperature, and pressure, are integrated through a mapping network to form a unified feature representation. A modal attention mechanism is used to assign weights to each mode, enhancing the feature contribution of key modes (such as vibration signals) while suppressing the influence of interfering modes.

[0058] Furthermore, the generated virtual samples include a variational autoencoder (VAE) and a generative adversarial network (GAN). The VAE encodes features as latent variables and decodes the latent variables into new samples. The GAN improves the realism and diversity of the virtual samples by introducing a multi-layer convolutional discriminator. The virtual sample generation process is as follows: Generating virtual samples: Input the perturbed latent variables into the decoder and generator to generate virtual fault samples; Distinguishing between real and virtual samples: By introducing different combinations of noise levels and fault characteristics, real and virtual samples can be distinguished.

[0059] A further technical solution of the present invention is: the generative adversarial network includes a generator and a discriminator; The generator receives random noise σ or latent variables from a variational autoencoder to generate virtual fault samples: 𝑥′=Generator(𝜖) Where 𝜖 is a random noise variable that follows a specific distribution; 𝑥′ is a data sample (such as an image or text) synthesized by the generator based on noise, with the goal of approximating the real data distribution; Generator represents the generator, which is a special function or object.

[0060] The discriminator is used to distinguish between real samples and virtual samples, and outputs the real probability: D(x)→[0,1]; By generating realistic virtual samples through adversarial training, the discriminator is unable to distinguish between virtual and real samples. Discriminator objective: To maximize the ability to distinguish between real and virtual samples.

[0061] S40, Reference Figure 3 Based on the virtual samples, a multimodal fusion convolutional network model is constructed, and then trained and validated.

[0062] In an exemplary embodiment, step S40 may include the following steps: S410 is used for network architecture design, which includes an input layer, convolutional layer, pooling layer, attention mechanism layer, fusion layer, and fully connected layer. S420 uses a spatial attention mechanism to perform feature fusion on different modal features of virtual samples. Feature fusion includes channel fusion, spatial fusion, and final fusion. It completes adaptive weighting of features, performs compressor fault state diagnosis, and verifies the judgment results.

[0063] S430, Train the multimodal fusion convolutional network model.

[0064] Specifically, refer to Figure 3 The training process is as follows: (1) Prepare the data; (2) Develop training strategies; (3) Perform the optimization process; (4) Perform model validation.

[0065] S50, based on the trained multimodal fusion convolutional network model, performs fault diagnosis on large compressors.

[0066] In an exemplary embodiment, step S50 may include the following steps: S510, Verify the multimodal fusion convolutional network model based on the fault data in the original sample; S520, based on the trained multimodal fusion convolutional network model, performs fault diagnosis on large compressors to ensure the model's generalization ability in different application scenarios.

[0067] Furthermore, this embodiment utilizes adversarial training technology to achieve knowledge transfer between the source domain (laboratory data) and the target domain (field data). Addressing the issue of distribution differences between the source and target domains, this invention designs a domain-adaptive loss function, combining maximum mean difference (MMD) and adversarial discriminator techniques to significantly reduce the feature distribution differences between the two domains, thereby ensuring the model's generalization performance in the target domain. In addition, by using a shared feature representation network, this invention trains the model in the source domain using labeled data, while simultaneously transferring the extracted general features to the target domain, improving fault identification capabilities in real-world working conditions.

[0068] This method constructs a multimodal fusion convolutional network model and uses the trained model to diagnose faults in large compressors, dynamically extracting high-resolution time-frequency features and adapting to complex non-stationary signals. It generates virtual samples with different fault states and noise levels through an adaptive multi-feature generation model. Furthermore, it extracts time-domain, frequency-domain, and envelope spectrum features from the original samples using a multimodal feature extraction method, thereby achieving dynamic time-frequency feature extraction and cross-domain transfer of compressor signals under multiple operating conditions, significantly improving diagnostic accuracy and robustness.

[0069] Based on the above embodiments, this application also provides a computer-readable storage medium, see reference. Figure 4The computer-readable storage medium shown is an optical disc 50, on which a computer program (i.e., a program product) is stored. When the computer program is run by a processor, it implements the steps described in the above-described method implementation, such as: selecting a multi-parameter corrosion monitoring sensor; acquiring the temporal signals of the compressor's rotating machinery under different operating conditions based on the sensor; preprocessing the temporal signals; generating virtual samples with different fault states and noise levels based on an adaptive multi-feature generation model; constructing a multimodal fusion convolutional network model and training and validating it; and performing fault diagnosis of a large compressor based on the trained multimodal fusion convolutional network model. The specific implementation methods of each step will not be repeated here.

[0070] It should be noted that examples of the computer-readable storage medium may also include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical and magnetic storage media, which will not be elaborated here.

[0071] Furthermore, based on the above embodiments, this application also provides a computing device. Figure 5 A block diagram is shown of an exemplary computing device 60 suitable for implementing embodiments of the present application. The computing device 60 may be a computer system or a server. Figure 5 The computing device 60 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0072] like Figure 5 As shown, the components of computing device 60 may include, but are not limited to: one or more processors or processing units 601, system memory 602, and bus 603 connecting different system components (including system memory 602 and processing unit 601).

[0073] The computing device 60 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the computing device 60, including volatile and non-volatile media, removable and non-removable media.

[0074] System memory 602 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 6021 and / or cache memory 6022. Computing device 60 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, ROM 6023 may be used to read and write non-removable, non-volatile magnetic media (…). Figure 5Not shown in the image (usually referred to as a "hard drive"). Although not shown in Figure 5 The diagram illustrates that disk drives for reading and writing to removable non-volatile disks (e.g., "floppy disks") and optical disc drives for reading and writing to removable non-volatile optical discs (e.g., CD-ROMs, DVD-ROMs, or other optical media) can be provided. In these cases, each drive can be connected to a bus 603 that connects different system components via one or more data media interfaces. The system memory 602 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of this application.

[0075] A program / utility 6025 having a set (at least one) of program modules 6024 may be stored, for example, in system memory 602, and such program modules 6024 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment. Program modules 6024 typically perform the functions and / or methods described in the embodiments of this application.

[0076] The computing device 60 can also communicate with one or more external devices 604 (such as a keyboard, pointing device, display, etc.). This communication can be performed via the input / output (I / O) interface 605. Furthermore, the computing device 60 can also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 606. Figure 5 As shown, network adapter 606 communicates with other modules of computing device 60 (such as processing unit 601, etc.) via bus 603, which connects different system components. It should be understood that, although... Figure 5 Other hardware and / or software modules may be used in conjunction with computing device 60, as not shown in the diagram.

[0077] The processing unit 601 executes various functional applications and data processing by running programs stored in the system memory 602. For example, it selects a multi-parameter corrosion monitoring sensor; acquires temporal signals of the compressor's rotating machinery under different operating conditions based on the sensor; preprocesses the temporal signals; generates virtual samples with different fault states and noise levels based on an adaptive multi-feature generation model; constructs a multimodal fusion convolutional network model and performs training and validation; and performs large compressor fault diagnosis based on the trained multimodal fusion convolutional network model. It should be noted that although several units / modules or sub-units / sub-modules of the large compressor fault diagnosis device are mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of this application, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.

[0078] In the description of this application, it should be noted that the terms "first", "second", and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0079] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0080] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the shown or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0081] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0082] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0083] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0084] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The protection scope of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the technical scope disclosed in this application. Such modifications, changes, 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 this application, and should all be covered within the protection scope of this application. Therefore, the protection scope of this application should be determined by the protection scope of the claims.

[0085] Furthermore, although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

Claims

1. A method for diagnosing faults in a large compressor, characterized in that, The method includes the following steps: S10, based on sensors, acquires the timing signals of the compressor's rotating machinery under different operating conditions; S20, preprocess the timing signal; S30 generates virtual samples with different fault states and noise levels based on an adaptive multi-feature generation model and preprocessed temporal signals. S40, Construct a multimodal fusion convolutional network model based on the virtual samples, and train and validate it; S50, based on the trained multimodal fusion convolutional network model, performs fault diagnosis on large compressors.

2. The method for diagnosing faults in a large compressor according to claim 1, characterized in that, Step S20 includes the following steps: S210, Denoise the timing signal; S220, The continuous time signal in the time sequence signal is segmented based on the sliding window technique; S230, the timing signal is standardized.

3. The method for diagnosing faults in a large compressor according to claim 2, characterized in that, Step S210 includes the following steps: S2110, based on an adaptive low-pass filter, dynamically adjusts the cutoff frequency to remove high-frequency components while retaining fault characteristic frequencies. The high-frequency components refer to time-series signals with frequencies from f1 to ∞. S2120, Perform discrete wavelet decomposition on the time-series signal to decompose the time-series signal into different frequency bands, combine with the soft thresholding method to remove low-energy noise components, and reconstruct the denoised time-series signal. S2130, Based on EMD technology, the time-series signal is decomposed into a series of intrinsic mode functions; S2140, Perform frequency domain analysis on each intrinsic mode function, remove high-frequency noise modes and reconstruct the timing signal, wherein the high-frequency noise refers to sound with a frequency range of 1000Hz to 20000Hz.

4. The method for diagnosing faults in a large compressor according to claim 1, characterized in that, Following step S20, the following steps are also included: S2001, Perform feature enhancement on the preprocessed temporal signal; S2002 extracts the signal envelope based on short-time Fourier transform or Hilbert transform, enhancing the separability of fault features; S2003, Dimensionality reduction of the time-series signal is performed based on principal component analysis.

5. The method for diagnosing faults in a large compressor according to claim 1, characterized in that, Step S30 includes the following steps: S310, based on a multimodal feature extraction method, extracts time-domain, frequency-domain, and envelope spectrum features from the original sample; S320, constructing an adaptive multi-feature generation model; S330, The extracted time-domain, frequency-domain, and envelope spectrum features are input into the adaptive multi-feature generation model for training; S340, based on the trained adaptive multi-feature generation model, generates virtual samples with different fault states and noise levels.

6. The method for diagnosing faults in a large compressor according to claim 1, characterized in that, Step S40 includes the following steps: S410, Construct a multimodal fusion convolutional network model, which includes an input layer, a convolutional layer, a pooling layer, an attention mechanism layer, a fusion layer, and a fully connected layer; S420, based on the spatial attention mechanism, performs channel fusion, spatial fusion and final fusion on different modal features of virtual samples, completes adaptive weighting of features, performs compressor fault state diagnosis, and verifies the judgment results; S430, Train the multimodal fusion convolutional network model.

7. The method for diagnosing faults in a large compressor according to claim 1, characterized in that, Step S50 includes the following steps: S510, Verify the multimodal fusion convolutional network model based on the fault data in the original sample; S520 performs fault diagnosis on large compressors based on the trained multimodal fusion convolutional network model.

8. A computer-readable storage medium, characterized in that, It includes instructions that, when run on a computer, cause the computer to perform the method for calculating pyrolysis parameters of shale oil reservoirs as described in any one of claims 1-7.

9. A computing device, characterized in that, The computing device includes: At least one processor, memory, and input / output unit; The memory is used to store computer programs, and the processor is used to call the computer programs stored in the memory to execute the large compressor fault diagnosis method according to any one of claims 1-7.

Images