Hoist machinery fault sample generation method based on dual-domain idempotent loss generation network

Abstract

The application discloses a hoisting machinery fault sample generation method based on a dual-domain idempotent loss generation network, and the method comprises the following steps: collecting original vibration signals of a hoisting machinery transmission system; pre-processing the original vibration signals to obtain corresponding time domain signals and frequency domain signals; constructing a dual-domain idempotent loss generation network model, wherein the model comprises a generator and a discriminator, and an idempotent loss and a dual-domain similarity loss are introduced into a loss function of the generator; wherein the idempotent loss is used to constrain the output consistency of the generator at adjacent network layers, the dual-domain similarity loss comprises a time domain loss and a frequency domain loss, the time domain loss is a mean square error of the time domain signals corresponding to generated data and real data in the time domain, and the frequency domain loss is a Euclidean distance between the frequency spectra of the frequency domain signals corresponding to the generated data and the real data; and the model is trained, and a fault sample is generated by using the trained model. The dual-domain idempotent loss mechanism solves the problem of the scarcity of fault samples.

Description

Technical Field

[0001] This invention belongs to the field of fault data generation technology for lifting machinery transmission systems, specifically relating to a method for generating fault samples for lifting machinery based on a dual-domain idempotent loss generation network. Background Technology

[0002] Lifting machinery is widely used in modern industrial fields (such as port loading and unloading, construction engineering, and manufacturing), significantly improving logistics efficiency and production automation levels. However, due to wear, fatigue, corrosion, and environmental factors such as high temperature and humidity, its core transmission components (such as gears and bearings) are prone to performance degradation after long-term high-intensity operation, and may even cause sudden failures. Such failures not only lead to serious equipment damage, high maintenance costs, and downtime losses, but may also cause major safety accidents such as personal injury and environmental pollution. Therefore, efficient diagnosis of faults in the transmission system of lifting machinery has become a key technical challenge that urgently needs to be solved to ensure the safe and stable operation of modern production processes.

[0003] Existing diagnostic technologies (such as vibration monitoring and acoustic analysis) assess the health status of equipment by acquiring operational status signals in real time. However, in actual engineering, fault signal samples are extremely scarce (mainly due to the low frequency of fault occurrence and high data acquisition costs), resulting in traditional data-driven models (especially deep learning models) performing poorly in terms of diagnostic accuracy, robustness, and cross-condition generalization ability.

[0004] While data augmentation techniques (such as SMOTE or GANs) can alleviate class imbalance to some extent by generating synthetic samples, they often introduce overfitting risks due to biased distribution of generated samples, reducing the model's generalization performance. Furthermore, while generative models such as Generative Adversarial Networks can effectively expand the sample library, they often focus on time-domain signal features, neglecting the complex information contained in the frequency domain and time-frequency domain. They cannot fully capture the nonlinearity and multi-scale details of signals, making it difficult to accurately characterize the true dynamic characteristics of various fault modes such as gear tooth breakage and bearing spalling.

[0005] Therefore, obtaining high-quality, multi-source fusion fault samples under varying operating conditions such as load fluctuations and speed changes is of decisive significance for achieving accurate prediction of faults and intelligent health management of lifting machinery transmission systems. This will significantly extend equipment service life, reduce maintenance costs, and ensure industrial safety. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for generating fault samples of lifting machinery based on a dual-domain idempotent loss generation network. This method proposes a dual-domain idempotent loss generation network model, which generates fault data of the transmission system of lifting machinery based on a dual-domain idempotent loss mechanism composed of idempotent loss and dual-domain similarity loss, so as to solve the problem of insufficient diagnostic accuracy caused by the scarcity of mechanical fault diagnosis samples under variable working conditions.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] The first aspect of the present invention provides a method for generating fault samples of lifting machinery based on a dual-domain idempotent loss generation network, the method comprising:

[0009] S1. Collect the original vibration signals of the lifting machinery transmission system under different working conditions and different fault modes;

[0010] S2. Preprocess the original vibration signal, including:

[0011] The original vibration signal is segmented and normalized to obtain the original time-domain signal;

[0012] Perform a Fast Fourier Transform on the original time-domain signal to obtain the corresponding frequency-domain signal;

[0013] The original time-domain signal is denoised by multi-scale wavelet packet transform to obtain the corresponding time-domain signal.

[0014] S3. Construct a dual-domain idempotent loss generative network model based on a generative adversarial network framework. This model includes a generator and a discriminator, and introduces idempotent loss and dual-domain similarity loss into the generator's loss function. The idempotent loss is used to constrain the consistency of the generator's output in adjacent network layers. The dual-domain similarity loss includes time-domain loss and frequency-domain loss. The time-domain loss is the mean square error of the time-domain signal corresponding to the generated data and the real data in the time domain, and the frequency-domain loss is the Euclidean distance between the frequency-domain signal spectra corresponding to the generated data and the real data.

[0015] S4. Train a dual-domain idempotent loss generator network model based on a dual-domain idempotent loss mechanism consisting of idempotent loss and dual-domain similarity loss, and use the trained dual-domain idempotent loss generator network model to generate fault samples of crane transmission systems.

[0016] Furthermore, the total loss of the dual-domain idempotent loss generative network model includes generative adversarial loss, idempotent loss, and dual-domain similarity loss.

[0017] Furthermore, the generative adversarial loss includes the generator loss and the discriminator loss, where:

[0018] The generator loss is:

[0019] ;

[0020] In the formula, For generator loss; To conform to the noise vector distribution A random noise vector; For random noise vectors Expectations; To generate data; This represents the output of the discriminator;

[0021] The discriminator loss is:

[0022] ;

[0023] In the formula, For discriminator loss; To conform to the actual data distribution Real data; To verify the actual data Expectations; This represents the discriminator's output on the real data.

[0024] Furthermore, the idempotent loss is:

[0025] ;

[0026] In the formula, For idempotent loss; To conform to the noise vector distribution A random noise vector; For random noise vectors Expectations; and The generator is in the th Layer and first Data output generated by the layer; For Euclidean distance.

[0027] Furthermore, the dual-domain similarity loss is:

[0028] ;

[0029] In the formula, For dual-domain similarity loss; For time domain loss; For frequency domain loss; and These are the weighting coefficients;

[0030] The time domain loss is:

[0031] ;

[0032] In the formula, This represents the number of time-domain data points. To generate the time-domain signal corresponding to the data; This is the time-domain signal corresponding to the actual data;

[0033] The frequency domain loss is:

[0034] ;

[0035] In the formula, This represents the number of frequency domain data points. Represents the Fast Fourier Transform. To generate the frequency domain signal corresponding to the data, The frequency domain signal corresponding to the actual data; It is the square of the Euclidean distance.

[0036] Furthermore, the multi-scale wavelet packet transform is as follows:

[0037] Select the Daubechies wavelet pair signal Perform wavelet packet transform to obtain coefficients at different scales and frequency bands, using the following formula:

[0038] ;

[0039] In the formula, For the first Layer The coefficients of the subband; These are wavelet packet basis functions;

[0040] For sub-bands with high noise levels, a threshold denoising method is used, as shown in the following formula:

[0041] ;

[0042] In the formula, The noise reduction threshold; These are the denoised coefficients;

[0043] The denoised signal is then:

[0044] ;

[0045] In the formula, This is the signal after noise reduction.

[0046] Furthermore, an experimental platform was built to simulate the operating state of the lifting machinery transmission system under different working conditions, and various mechanical failure modes were simulated by prefabricating faulty parts, so as to collect the original vibration signals using vibration acceleration sensors.

[0047] During training, a random noise vector is input into the generator to obtain the generated data in the time domain, and the time domain loss is calculated directly. Then, the generated data in the time domain is subjected to a fast Fourier transform to obtain the frequency domain signal corresponding to the generated data, and the frequency domain loss is calculated again.

[0048] According to a second aspect of the present invention, a crane machinery fault sample generation system based on a dual-domain idempotent loss generation network is provided, applying the crane machinery fault sample generation method based on a dual-domain idempotent loss generation network as described in any one of the first aspects. The system includes:

[0049] The signal acquisition module is used to acquire the original vibration signals of the lifting machinery transmission system under different working conditions and different fault modes;

[0050] The signal processing module is used to preprocess the raw vibration signal, including:

[0051] The original vibration signal is segmented and normalized to obtain the original time-domain signal;

[0052] Perform a Fast Fourier Transform on the original time-domain signal to obtain the corresponding frequency-domain signal;

[0053] The original time-domain signal is denoised by multi-scale wavelet packet transform to obtain the corresponding time-domain signal.

[0054] The dual-domain idempotent loss generator network module is used to construct a dual-domain idempotent loss generator network model based on the generative adversarial network framework. This model includes a generator and a discriminator, and introduces idempotent loss and dual-domain similarity loss into the generator's loss function. The idempotent loss is used to constrain the consistency of the generator's output in adjacent network layers. The dual-domain similarity loss includes time-domain loss and frequency-domain loss. The time-domain loss is the mean square error of the time-domain signal corresponding to the generated data and the real data in the time domain, and the frequency-domain loss is the Euclidean distance between the frequency-domain signal spectra corresponding to the generated data and the real data.

[0055] The sample generation module is used to train a dual-domain idempotent loss generation network model based on a dual-domain idempotent loss mechanism consisting of idempotent loss and dual-domain similarity loss, and to generate fault samples of crane transmission systems using the trained dual-domain idempotent loss generation network model.

[0056] According to a third aspect of the present invention, a computer device is provided, comprising: a processor and a memory, the memory storing a program or instructions executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the method for generating crane machinery fault samples based on a dual-domain idempotent loss generation network as described in any one of the first aspects.

[0057] According to a fourth aspect of the present invention, a computer-readable storage medium is provided having a program or instructions stored thereon, which, when executed by a processor, implement the steps of the method for generating crane machinery fault samples based on a dual-domain idempotent loss generation network as described in any one of the first aspects.

[0058] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects:

[0059] This invention constructs a generative adversarial network model that incorporates a dual-domain idempotent loss mechanism (idempotent loss and dual-domain similarity loss). This not only uses generative adversarial loss to drive the model to learn the distribution of real data, but also uses idempotent loss to constrain the stability of the generator's internal output, preventing mode collapse. At the same time, the introduced dual-domain similarity loss supervises the generation process from both the time and frequency domains, ensuring that the generated fault samples are highly consistent with the real data in terms of waveform morphology and spectral structure. This fundamentally improves the diversity and fidelity of the generated samples, providing high-quality and sufficient training data for subsequent fault diagnosis models.

[0060] Furthermore, this invention employs Daubechies wavelet decomposition for multi-scale signal processing and utilizes a thresholding method to filter out noise-dominated subband coefficients. This adaptively preserves key time-frequency information reflecting fault characteristics while effectively suppressing background noise interference. This processing provides the subsequent generative model with clean, fault-feature-rich, and realistic data targets, ensuring the physical meaning and diagnostic value of the generated samples and enhancing the applicability of the entire method under complex operating conditions. Attached Figure Description

[0061] Figure 1 A schematic diagram of the overall process of a method for generating fault samples of lifting machinery based on a dual-domain idempotent loss generation network provided in an embodiment of the present invention;

[0062] Figure 2 A flowchart for fault monitoring and diagnosis of a lifting machinery transmission system is provided in this embodiment of the invention;

[0063] Figure 3 This is a schematic diagram of the hardware structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments provided by this invention without inventive effort are within the scope of protection of this invention.

[0065] Obviously, the accompanying drawings described below are merely some examples or embodiments of the present invention. Those skilled in the art can apply the present invention to other similar scenarios based on these drawings without any inventive effort. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this invention, modifications to design, manufacturing, or production based on the technical content disclosed in this invention are merely conventional technical means and should not be construed as insufficient disclosure of the present invention.

[0066] In this invention, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this invention may be combined with other embodiments without conflict.

[0067] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "a," "an," "an," "the," and similar words used in this invention do not indicate quantity limitation and may indicate singular or plural. The terms "comprising," "including," "having," and any variations thereof used in this invention are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. The terms "connected," "linked," "coupled," and similar words used in this invention are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "A plurality" used in this invention refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships may exist; for example, "A and / or B" can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following objects have an "or" relationship. The terms "first," "second," and "third" used in this invention are merely to distinguish similar objects and do not represent a specific ordering of the objects.

[0068] This invention provides a method for generating fault samples for crane machinery based on a dual-loss idempotent generator network, aiming to solve the problem of low diagnostic accuracy caused by the scarcity of fault samples in the transmission system of crane machinery under varying operating conditions. The method includes: building an experimental platform to simulate the operating state of the equipment under different operating conditions and fault modes, and collecting raw vibration signals; preprocessing the raw vibration signals, including data segmentation, normalization, fast Fourier transform, and multi-scale wavelet packet transform, to obtain time-domain and frequency-domain signal pairs; constructing a dual-loss idempotent generator network model, introducing idempotent loss to ensure the consistency of the generated data output in adjacent network layers, and combining time-domain loss and frequency-domain loss to jointly optimize the generator and improve the quality of generated samples; finally, using the trained model to generate samples highly similar to the distribution of real fault data. This invention can effectively enhance the diversity and realism of fault data, thereby improving the generalization ability of the fault diagnosis model under varying operating conditions.

[0069] like Figure 1 As shown, the method for generating mechanical fault samples of a crane transmission system based on a dual-domain idempotent loss generator network of the present invention includes the following steps: (1) collecting the original vibration signals of the crane transmission system under different working conditions and different mechanical fault modes; (2) preprocessing the original vibration signals, including signal segmentation, normalization, fast Fourier transform and multi-scale wavelet packet denoising, to obtain the corresponding time domain signal and frequency domain signal; (3) constructing a dual-domain idempotent loss generator network model, which includes a generator and a discriminator, and introducing idempotent loss, time domain loss and frequency domain loss; (4) training the dual-domain idempotent loss generator network model using the preprocessed vibration signals; (5) generating mechanical fault samples of the crane transmission system using the trained generator.

[0070] Figure 2 This is a flowchart illustrating the fault monitoring and diagnosis process for lifting machinery transmission systems. It clearly demonstrates a complete technical loop from data preprocessing to model innovation training and sample generation applications. Its core implementation logic lies in: using multi-scale wavelet packet transform for denoising to provide high-quality time-domain signals for the model, and Fourier transform to provide frequency-domain signals. Furthermore, by introducing a dual-domain idempotent loss mechanism consisting of idempotent loss and dual-domain similarity loss, the training process of the generative adversarial network is strictly constrained, ultimately generating high-quality mechanical fault samples with both time-domain and frequency-domain fidelity stably and efficiently. This process fundamentally solves the problem of poor performance of fault diagnosis models under small sample conditions and achieves effective data augmentation under varying operating conditions.

[0071] Figure 1 This is a flowchart illustrating a method for generating mechanical fault samples in a lifting machinery transmission system. It demonstrates the specific implementation of the core algorithm of this invention—the dual-domain idempotent loss generation network—and details the entire process from raw signal input to the generation of high-quality fault samples. This process can be divided into the following four key stages:

[0072] Phase 1: Real Data Acquisition and Fault Simulation. This phase forms the data foundation for the entire methodology, aiming to obtain real and diverse raw data from physical devices.

[0073] (1) Process start point and equipment: The process starts with the lifting machinery transmission system. In order to simulate its performance under different health conditions, an experimental platform is built to controllably reproduce the equipment state.

[0074] (2) Operating conditions and fault simulation: ① Variable operating conditions: The experimental platform simulates the operating state of the equipment under different working conditions (such as different speeds and loads). ② Multi-mode fault injection: By prefabricating faulty parts (such as pre-damaged bearings and worn gears), various mechanical fault modes are simulated to ensure that the collected data can cover the complete spectrum from normal operation and minor damage to severe damage.

[0075] (3) Signal acquisition: The original vibration signals of the equipment under various conditions are acquired using a vibration acceleration sensor (selected according to the frequency response of the equipment). The output of this step constitutes the real dataset for all subsequent processing.

[0076] In this phase, to address the problem of scarce fault signal samples in the background technology, a high-quality raw data pool was built for subsequent algorithm models through active experiments and sensor technology.

[0077] The second stage: multi-dimensional signal preprocessing and feature enhancement. This stage involves cleaning, standardizing, and extracting features from the raw data, transforming it into a format suitable for input to deep learning models.

[0078] (1) Input: The input for this stage is the original vibration signal output from the first stage.

[0079] (2) Preprocessing pipeline: ① Signal segmentation: Based on the operating frequency and sampling frequency of the equipment, the continuous vibration signal is divided into data segments of fixed length to prepare for subsequent batch processing. ② Data normalization: The signal is normalized using a normalization formula to eliminate dimensions, accelerate model convergence, and obtain preprocessed time-domain information. ③ Frequency domain transformation: The normalized signal is subjected to a fast Fourier transform to transform the signal from the time domain to the frequency domain, obtaining a frequency domain signal to reveal its frequency components and structure. ④ Time-frequency domain analysis and denoising: Multi-scale wavelet packet transform is applied, and Daubechies wavelets are used to decompose the signal into multiple layers. Threshold denoising technology is used to filter out noise interference and simultaneously extract high-frequency and low-frequency components that are sensitive to faults, achieving effective feature enhancement and obtaining a high-quality time-domain signal.

[0080] (3) Output: The output is a high-quality time-domain signal and frequency-domain signal pair after deep processing, which provides a comprehensive and robust learning target for the generative model.

[0081] This stage addresses the problem that the background techniques could not fully capture the complexity and details of the signal. Through joint processing of the time, frequency, and time-frequency domains, the model is provided with training data far exceeding the information from a single domain.

[0082] Phase 3: Construction and Training of the Bi-Domain Idempotent Loss Generative Network Model. This phase is the core of the entire algorithm, demonstrating how to train a stable and efficient generative model using preprocessed data.

[0083] (1) Model framework: The basic framework of generative adversarial network is adopted, which includes generator and discriminator.

[0084] (2) Training process and innovative loss function: ① Adversarial training: The generator receives random noise and attempts to generate fake data, while the discriminator tries to distinguish between real preprocessed data and generated data. The two compete with each other and evolve together. ② Dual-domain idempotent loss mechanism (innovation): a. Idempotent loss: Inside the generator, its output at different levels is calculated. and Consistency between them. This loss ensures the stability of the generator training process, prevents output oscillations or pattern collapse, and guarantees generation quality. b. Two-domain similarity loss: This loss consists of temporal domain loss. and frequency domain loss Weighted composition. It forces the generator to produce data that is not only close to real data in terms of time-domain waveform, but also highly similar in terms of spectral characteristics. This makes the generated samples more realistic in terms of physical properties. c. Parameter update: The total loss of the generator is jointly determined by the generative adversarial loss, idempotent loss, and bi-domain similarity loss. The model parameters are continuously updated iteratively through backpropagation of the optimization algorithm until the model converges.

[0085] This stage introduces a dual-domain idempotent loss mechanism to ensure the high fidelity and physical rationality of the generated samples, thereby effectively solving the problems of low accuracy and poor generalization ability of the diagnostic model caused by the small sample size.

[0086] Phase 4: Fault Sample Generation and Diagnostic Model Enhancement. This phase utilizes the trained model for data generation to improve the performance of the final diagnostic task.

[0087] (1) Sample generation: Switch the trained generator from training mode to generation mode. Input a new random noise vector into it, and a large amount of generated fault data can be output as needed.

[0088] (2) Data augmentation: These generated data are merged with the original real data to form a larger and more diverse augmented fault diagnosis training set.

[0089] (3) Final application: Using this enhanced dataset to train the final fault diagnosis model (such as a classifier) ​​can significantly improve the diagnostic accuracy and generalization ability of the model in real variable working conditions, and complete the closed loop from data generation to intelligent diagnosis.

[0090] This stage achieves the ultimate goal of the methodology—"data augmentation"—which ultimately enables and improves the performance of another key task (fault diagnosis).

[0091] Specifically, such as Figure 1 and Figure 2 As shown, the method for generating mechanical fault data of a lifting machinery transmission system based on a dual-domain idempotent loss generator network model according to an embodiment of the present invention includes the following steps:

[0092] Step S1: Set up an experimental platform to simulate the state of the lifting machinery transmission system under different working conditions, and acquire the vibration signal of the equipment through a vibration acceleration sensor. The vibration acceleration sensor is selected according to the frequency response range of the equipment to ensure that the signal characteristics can be effectively collected.

[0093] Then, prefabricated parts are used to simulate the equipment's performance under different failure modes. Possible failure modes include bearing damage, gear wear, and impeller failure. Each mode has a different impact on the vibration signal, thus obtaining the raw signals corresponding to various modes in this way, which will form the basis of the training dataset. In addition, more diverse data can be obtained by adding different types of sensors (such as temperature sensors and pressure sensors), providing more dimensional input features for model training.

[0094] Step S2: Preprocess the acquired raw signal.

[0095] Step S21: Divide the data length reasonably according to the device's operating frequency and sampling frequency.

[0096] Step S22: Normalize the partitioned data as shown in the following formula:

[0097] ;

[0098] in, This represents the original time-domain signal after normalization. This represents the original vibration signal collected. This represents the mean. It represents the standard deviation.

[0099] Step S23: The normalized signal is time-domain information. To extract frequency-domain features, a Fast Fourier Transform is performed. The specific process is as follows:

[0100] ;

[0101] in, It is the normalized original time-domain signal. It is a frequency domain signal obtained through Fast Fourier Transform. It is the imaginary unit. It refers to frequency.

[0102] Step S24: To obtain a high-quality time-domain signal, multi-scale wavelet packet transform is applied to extract the high-frequency and low-frequency components in the original time-domain signal, remove the noise, retain the effective time-frequency features, and finally obtain the corresponding time-domain signal.

[0103] Step S241: Select one with Daubechies wavelet decomposition of layers.

[0104] Step S242: For the signal Perform wavelet packet transform to obtain coefficients at different scales and frequency bands, using the following formula:

[0105] ;

[0106] in, It is a wavelet packet basis function. It is the first Layer, First The coefficients of the subband.

[0107] Step S243: During the multi-scale decomposition process, denoising can be performed based on the coefficient magnitudes of each frequency band. For sub-bands with high noise, a threshold denoising method is used, as shown in the following formula:

[0108] ;

[0109] in, The noise reduction threshold is determined based on the noise level.

[0110] The denoised signal is:

[0111] ;

[0112] In the formula, This is the denoised time-domain signal.

[0113] Step S3: Construct a loss function to train a bi-domain idempotent loss generative network model.

[0114] Step S31: Construct the generator loss function to train the model. The standard generative adversarial network loss function is:

[0115] ;

[0116] in, The discriminator output is the response to the input data. The probability of real data. This indicates that it follows the true data distribution. Real data, Indicates the generation of data. It is to follow the noise vector distribution A random noise vector, For random noise vectors Expectations To verify the actual data The expectation.

[0117] This invention introduces an idempotent loss to ensure the consistency of generated data at different stages, based on the above. The idempotent loss is as follows:

[0118] ;

[0119] in, and The generator is in the th Layer and first Data output generated by the layer Euclidean distance; It is a random noise vector that follows a distribution. , The expectation of the generator layer output is given by the distribution. All noise vectors sampled in the middle Its corresponding generator is in the first... Layer and first The expected (average) value of the Euclidean distance between layer outputs.

[0120] The loss function after introducing idempotent loss is:

[0121] ;

[0122] in, It is a hyperparameter that controls the weight of idempotent terms.

[0123] Step S32: Construct the time-domain loss and frequency-domain loss;

[0124] The time-domain loss is calculated using the mean squared error (MSE):

[0125] ;

[0126] in, It is the value of the generated data in the time domain. It is the value of the actual data in the time domain. This represents the number of time-domain data points.

[0127] Frequency domain loss is calculated by measuring the difference between generated data and real data in the frequency domain feature space:

[0128] ;

[0129] in, This represents the number of frequency domain data points. Represents the Fast Fourier Transform. To generate the frequency domain signal of the data, The frequency domain signal corresponding to the actual data; It is the square of the Euclidean distance.

[0130] Step S33: Take into account both time-domain loss and frequency-domain loss to ensure that the generated data maintains a high similarity to the input data:

[0131] ;

[0132] in, For dual-domain similarity loss; For time domain loss; For frequency domain loss; and These are the weighting coefficients.

[0133] Step S4: Train the dual-domain idempotent loss generator network model. During training, a random noise vector is input into the generator to obtain the generated data in the time domain, and the time-domain loss is directly calculated. Then, the generated data in the time domain is subjected to a Fast Fourier Transform to obtain the frequency domain signal of the generated data, and the frequency domain loss is calculated again. The dual-domain similarity loss is calculated by combining the time-domain loss and the frequency-domain loss. The dual-domain idempotent loss generator network model is trained by combining idempotent loss and generative adversarial loss. Finally, the trained dual-domain idempotent loss generator network model is used to generate high-quality fault data, i.e., random noise. Input generator to generate samples These generated samples are highly similar to the real data in terms of data distribution.

[0134] In summary, the method of this invention, centered around a dual-domain idempotent loss generation network, systematically realizes the entire process from raw signal acquisition to high-quality fault sample generation. Its specific working process is as follows:

[0135] First, a data acquisition platform for the lifting machinery transmission system was built, simulating the equipment's operating state under varying conditions by adjusting parameters such as load and speed. Pre-fabricated faulty parts (such as pre-cracked bearings and worn gears) were used to simulate various typical mechanical fault modes, including bearing damage and gear wear. During this process, vibration acceleration sensors were selected based on the equipment's frequency response characteristics to accurately collect raw vibration signals under different operating conditions and fault modes, constructing an initial realistic fault dataset. This step provides a realistic and diverse data foundation for subsequent model training.

[0136] Then, in the multi-dimensional signal preprocessing and feature enhancement section, after obtaining the original vibration signal, the preprocessing stage begins. First, the signal is segmented according to the device's operating frequency and sampling frequency to determine an appropriate sampling duration. Next, the segmented signal is normalized to eliminate the influence of dimensions and accelerate model convergence. Subsequently, a Fast Fourier Transform is performed on the normalized time-domain signal to transform it into the frequency domain, revealing the signal's frequency structure characteristics. To extract features more precisely and suppress noise, a multi-scale wavelet packet transform is further applied: the Daubechies wavelet is used to decompose the signal into multiple layers, obtaining sub-band coefficients at different scales; the coefficients are quantized by setting an appropriate threshold to achieve signal denoising; finally, the denoised signal is reconstructed as the real data. This step simultaneously outputs the time-domain and frequency-domain signal data of the real data, providing a comprehensive learning target for the generator network.

[0137] Secondly, the construction and training of the dual-domain idempotent loss generator network model is the core work link. A dual-domain idempotent loss generator network model based on the generative adversarial network framework is constructed. The model includes a generator and a discriminator. The training process is as follows: the generator receives a random noise vector and tries to generate fault data; the discriminator is responsible for judging whether the input data comes from real preprocessed data or the generator. The innovation of the model is reflected in the dual loss function designed for the generator: (1) Idempotent loss: by constraining the consistency between the output data of adjacent network layers of the generator (such as the nth layer and the (n-1th layer), the stability of the generator training process and the reliability of the output results are ensured, and the mode collapse is effectively prevented. (2) Dual-domain similarity loss: this loss is composed of a weighted combination of time domain loss (calculating the mean square error between the generated data and the real data in the time domain) and frequency domain loss (calculating the difference measure between the generated data and the real data in the frequency domain feature space), which forces the samples generated by the generator to be highly similar to the real data in both time domain waveform and frequency domain characteristics. The generator's total loss function consists of standard generative adversarial loss, idempotent loss, and bi-domain similarity loss. The network parameters are iteratively optimized using the backpropagation algorithm, continuously adjusting the generator and discriminator until the model converges. The generator is then able to produce data sufficient to "deceive" the discriminator and possess high time-frequency domain fidelity.

[0138] Finally, after the fault sample generation and diagnostic model enhancement model training is completed, the sample generation and application stage begins. New random noise vectors are input into the trained generator to generate high-quality mechanical fault samples that are highly similar to the distribution of real fault data in batches as needed. These generated samples are then merged with the original real samples to construct a significantly larger and more diverse enhanced fault diagnosis training set. Ultimately, this enhanced dataset is used to train a fault diagnosis model (such as a deep neural network classifier), which can significantly improve the identification accuracy and generalization ability of the diagnostic model under varying operating conditions, thus achieving a complete closed loop from data generation to intelligent diagnosis.

[0139] In summary, through the above-described systematic working process, this invention creatively utilizes a dual-domain idempotent loss generation network to effectively solve the problem of scarce fault samples in the transmission system of lifting machinery under varying working conditions, providing solid data support for improving the accuracy and reliability of fault diagnosis.

[0140] This invention also provides a crane machinery fault sample generation system based on a dual-domain idempotent loss generation network. Applying the crane machinery fault sample generation method based on a dual-domain idempotent loss generation network described in the above-described method embodiments, the system includes:

[0141] The signal acquisition module is used to acquire the raw vibration signals of the transmission system of lifting machinery;

[0142] The signal processing module is used to segment, normalize, perform fast Fourier transform and wavelet packet denoising on the vibration signal to obtain the time domain signal and frequency domain signal of the real data.

[0143] The dual-domain idempotent loss generation network module is used to train and generate fault samples using the dual-domain idempotent loss mechanism.

[0144] The sample generation module is used to output generated fault data.

[0145] The dual-domain idempotent loss generation network module includes: a generator for generating fault samples from noise vectors; a discriminator for distinguishing real samples from generated samples; and a loss calculation unit for calculating idempotent loss, time-domain loss, and frequency-domain loss.

[0146] It should be noted that the steps shown in the above process or in the flowchart of the accompanying figures can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0147] In addition, combined Figure 1The crane machinery fault sample generation method based on a dual-domain idempotent loss generation network described in this embodiment of the invention can be implemented by a computer device. Figure 3 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Figure 3 As shown, the device may include a processor 301 and a memory 302 storing computer program instructions.

[0148] Specifically, the processor 301 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of the present invention.

[0149] Memory 302 may include a large-capacity memory for data or instructions. For example, and not limitingly, memory 302 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 302 may include removable or non-removable (or fixed) media. Where appropriate, memory 302 may be internal or external to a data processing device. In a particular embodiment, memory 302 is non-volatile memory. In a particular embodiment, memory 302 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable read-only memory (PROM), an erasable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), an electrically alterable read-only memory (EAROM), or flash memory, or a combination of two or more of these. Where appropriate, the RAM can be Static Random-Access Memory (SRAM) or Dynamic Random-Access Memory (DRAM). DRAM can be Fast Page Mode Dynamic Random-Access Memory (FPMDRAM), Extended Data Out Dynamic Random-Access Memory (EDODRAM), Synchronous Dynamic Random-Access Memory (SDRAM), etc.

[0150] The memory 302 can be used to store or cache various data files that need to be processed and / or communicated, as well as possible computer program instructions executed by the processor 301.

[0151] The processor 301 reads and executes computer program instructions stored in the memory 302 to implement any of the lifting machinery fault sample generation methods based on a dual-domain idempotent loss generation network in the above embodiments.

[0152] In some embodiments, the computer device may further include a communication interface 303 and a bus 300. For example, Figure 3 As shown, the processor 301, memory 302, and communication interface 303 are connected through bus 300 and complete communication with each other.

[0153] The communication interface 303 is used to enable communication between the various modules, devices, units, and / or equipment in the embodiments of the present invention. The communication interface 303 can also enable data communication with other components such as external devices, image / data acquisition devices, databases, external storage, and image / data processing workstations.

[0154] Bus 300 includes hardware, software, or both, that couples components of a computer device together. Bus 300 includes, but is not limited to, at least one of the following: data bus, address bus, control bus, expansion bus, and local bus. For example, and not as a limitation, bus 300 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a MicroChannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or a combination of two or more of these. Where appropriate, bus 300 may include one or more buses. Although specific buses are described and illustrated in embodiments of the invention, the invention contemplates any suitable bus or interconnect.

[0155] The computer device can execute the crane machinery fault sample generation method based on a dual-domain idempotent loss generation network as described in this embodiment of the invention, thereby achieving a combination of Figure 1 This paper describes a method for generating fault samples of lifting machinery based on a dual-domain idempotent loss generation network.

[0156] Furthermore, in conjunction with the crane machinery fault sample generation method based on a dual-domain idempotent loss generation network in the above embodiments, this invention can be implemented using a computer-readable storage medium. This computer-readable storage medium stores computer program instructions; when executed by a processor, these computer program instructions implement any of the crane machinery fault sample generation methods based on a dual-domain idempotent loss generation network described in the above embodiments.

[0157] It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification. In addition, depending on the implementation needs, the various steps / components described in this invention can be broken down into more steps / components, or two or more steps / components or parts of steps / components can be combined into new steps / components to achieve the purpose of this invention.

[0158] It will be readily understood by those skilled in the art that the above-described embodiments merely illustrate several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A method for generating fault samples of lifting machinery based on a dual-domain idempotent loss generation network, characterized in that, The method includes: S1. Collect the original vibration signals of the lifting machinery transmission system under different working conditions and different fault modes; S2. Preprocess the original vibration signal, including: The original vibration signal is segmented and normalized to obtain the original time-domain signal; Perform a Fast Fourier Transform on the original time-domain signal to obtain the corresponding frequency-domain signal; The original time-domain signal is denoised by multi-scale wavelet packet transform to obtain the corresponding time-domain signal. S3. Construct a dual-domain idempotent loss generative network model based on the generative adversarial network framework. The model includes a generator and a discriminator, and introduces idempotent loss and dual-domain similarity loss into the loss function of the generator. The idempotency loss is used to constrain the consistency of the generator's output across adjacent network layers, and its expression is: ； In the formula, For idempotent loss; To conform to the noise vector distribution A random noise vector; For random noise vectors Expectations; and The generator is in the th Layer and first Data output generated by the layer; Euclidean distance; The dual-domain similarity loss includes time-domain loss and frequency-domain loss. The time-domain loss is the mean square error of the time-domain signal corresponding to the generated data and the real data in the time domain, and the frequency-domain loss is the Euclidean distance between the frequency-domain signal spectra corresponding to the generated data and the real data. S4. Train a dual-domain idempotent loss generation network model based on a dual-domain idempotent loss mechanism consisting of idempotent loss and dual-domain similarity loss, and use the trained dual-domain idempotent loss generation network model to generate fault samples of crane transmission systems.

2. The method for generating fault samples of lifting machinery based on a dual-domain idempotent loss generation network according to claim 1, characterized in that, The total loss of the bi-domain idempotent loss generative network model includes generative adversarial loss, idempotent loss, and bi-domain similarity loss.

3. The method for generating fault samples of lifting machinery based on a dual-domain idempotent loss generation network according to claim 2, characterized in that, Generative adversarial loss includes generator loss and discriminator loss, where: The generator loss is: ； In the formula, For generator loss; To conform to the noise vector distribution A random noise vector; For random noise vectors Expectations; To generate data; This represents the output of the discriminator; The discriminator loss is: ； In the formula, For discriminator loss; To conform to the actual data distribution Real data; To verify the actual data Expectations; This represents the discriminator's output on the real data.

4. The method for generating fault samples of lifting machinery based on a dual-domain idempotent loss generation network according to claim 1 or 2, characterized in that, The two-domain similarity loss is: ； In the formula, For dual-domain similarity loss; For time domain loss; For frequency domain loss; and These are the weighting coefficients; The time domain loss is: ； In the formula, This represents the number of time-domain data points. To generate the time-domain signal corresponding to the data; This is the time-domain signal corresponding to the actual data; The frequency domain loss is: ； In the formula, This represents the number of frequency domain data points. Represents the Fast Fourier Transform. To generate the frequency domain signal corresponding to the data, The frequency domain signal corresponding to the actual data; It is the square of the Euclidean distance.

5. The method for generating fault samples of lifting machinery based on a dual-domain idempotent loss generation network according to claim 1, characterized in that, Multiscale wavelet packet transform is: Select the Daubechies wavelet pair signal Perform wavelet packet transform to obtain coefficients at different scales and frequency bands, using the following formula: ； In the formula, For the first Layer The coefficients of the subband; These are wavelet packet basis functions; For sub-bands with high noise levels, a threshold denoising method is used, as shown in the following formula: ； In the formula, The noise reduction threshold; These are the denoised coefficients; The denoised signal is then: ； In the formula, This is the denoised signal.

6. The method for generating fault samples of lifting machinery based on a dual-domain idempotent loss generation network according to claim 1, characterized in that, An experimental platform was built to simulate the operation of the transmission system of lifting machinery under different working conditions, and various mechanical failure modes were simulated by prefabricating faulty parts, so as to collect the original vibration signals using vibration acceleration sensors. During training, a random noise vector is input into the generator to obtain the generated data in the time domain, and the time domain loss is calculated directly. Then, the generated data in the time domain is subjected to a fast Fourier transform to obtain the frequency domain signal corresponding to the generated data, and the frequency domain loss is calculated.

7. A crane mechanical fault sample generation system based on a dual-domain idempotent loss generation network, characterized in that, The method for generating fault samples of lifting machinery based on a dual-domain idempotent loss generation network, according to any one of claims 1 to 6, comprises: The signal acquisition module is used to acquire the original vibration signals of the lifting machinery transmission system under different working conditions and different fault modes; The signal processing module is used to preprocess the raw vibration signal, including: The original vibration signal is segmented and normalized to obtain the original time-domain signal; Perform a Fast Fourier Transform on the original time-domain signal to obtain the corresponding frequency-domain signal; The original time-domain signal is denoised by multi-scale wavelet packet transform to obtain the corresponding time-domain signal. The dual-domain idempotent loss generator network module is used to construct a dual-domain idempotent loss generator network model based on the generative adversarial network framework. This model includes a generator and a discriminator, and introduces idempotent loss and dual-domain similarity loss into the generator's loss function. The idempotent loss is used to constrain the consistency of the generator's output in adjacent network layers. The dual-domain similarity loss includes time-domain loss and frequency-domain loss. The time-domain loss is the mean square error of the time-domain signal corresponding to the generated data and the real data in the time domain, and the frequency-domain loss is the Euclidean distance between the frequency-domain signal spectra corresponding to the generated data and the real data. The sample generation module is used to train a dual-domain idempotent loss generation network model based on a dual-domain idempotent loss mechanism consisting of idempotent loss and dual-domain similarity loss, and to generate fault samples of crane transmission systems using the trained dual-domain idempotent loss generation network model.

8. A computer device, characterized in that, include: The processor and memory, wherein the memory stores a program or instructions that can run on the processor, and when the program or instructions are executed by the processor, implement the steps of the method for generating crane machinery fault samples based on a dual-domain idempotent loss generation network as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, It stores programs or instructions, which, when executed by a processor, implement the steps of the method for generating crane machinery fault samples based on a dual-domain idempotent loss generation network as described in any one of claims 1 to 6.

Images