A fault diagnosis method based on diffusion model and 2D-CNN for small sample conditions: Application to reciprocating pumps

Data-driven fault diagnosis is a method widely used in the maintenance and support of radio frequency (RF) front-end. Its effectiveness depends on a large volume of high-quality training data. However, due to the low probability and uneven occurrence of faults, quantity of fault data is often less than the normal data. This data imbalanced problem greatly affects the accuracy of fault diagnosis for RF front-end. Data generation is an important way to address this problem, including generative adversarial networks (GAN), variational autoencoders (VAE), and denoising diffusion probabilistic models (DDPM). However, these methods suffer from issues such as unstable training, mode collapse, and uncontrollable generation. To solve these problems, this article proposed an adaptive conditional DDPM (Ada-CDDPM) based on improved U-Net3+, aiming to enable the model to better extract features of RF fault data and then guide it to generate the modal data required for RF system diagnosis. More specifically, this article proposed an adaptive CDDPM, which uses a conditional function with learnable coefficients vector to guide direction of training data generation. A differentiable loss function is proposed by this article to evaluate the quality of the generated training data. The learnable coefficient vector is optimized adaptively based on the evaluation results, enabling the final generated data to have good fidelity, diversity, and modal characteristics. This article constructed a simulation platform and an experimental platform to test the performance of the method. Comprehensive experiments show that data generated by the proposed method has significant advantages in diversity, fidelity, and is effective in solving RF front-end diagnosis problem under imbalanced data.

Accurate multivariate parametrization of subsurface properties is essential for subsurface characterization and inversion tasks. Deep generative models, such as variational autoencoders (VAEs) and generative adversarial networks (GANs), are known to efficiently parametrize complex facies patterns. Nonetheless, the inherent complexity of multivariate modeling poses significant limitations to their applicability when considering multiple subsurface properties simultaneously. Presently, diffusion models (DM) offer state-of-the-art performance and outperform GANs and VAEs in several tasks of image generation. In addition, training is much more stable compared to the training of GANs. In this work, we consider score-based DMs in multivariate geological modeling, specifically for the parametrization of categorical (facies) and continuous (acoustic impedance – I­P) distributions, focusing on a synthetic scenario of sand channel bodies in a shale background. We benchmark modeling performance against results obtained by GAN and VAE networks previously proposed in literature for multivariate modeling. As for the GAN and VAE models, the DM was trained with a training dataset of 3000 samples, consisting of facies realizations and co-located I­P geostatistical realizations. Overall, the trained DM shows significant improvements in modeling accuracy, for all evaluation metrics considered in this study, except for the sand-to-shale ratio, where the values are comparable to those of the GAN and VAE. In particular, the DM is 26% more accurate at reproducing the average (nonstationary) facies distribution and up to 90% more accurate at reproducing the IP marginal distributions for both sand and shale classes. Higher accuracy is also found in the reproduction of the facies-to-I­P joint distribution, whereas the spatial I­P distributions generated by the DM honour the two-point statistics of the training samples. The iterative generative process in DMs generally makes these networks more computationally demanding than VAEs and GANs. However, we demonstrate that with appropriate network design and training parametrization, the DM can generate realizations with significantly fewer sampling iterations while maintaining accuracy comparable to these benchmarking networks. Finally, since the proposed DM parametrizes the joint prior probability density function with a Gaussian latent space, it is straightforward to perform inversion. In addition to improved modeling accuracy, the mapping between the latent and image representations preserves a better topology than that of GANs, overcoming the well-known limitation of the latter for inference tasks, particularly for gradient-based inversion.

Deep learning fault diagnosis method based on global optimization GAN for unbalanced data

A novel deep learning framework for rolling bearing fault diagnosis enhancement using VAE-augmented CNN model

In gas turbine rotor systems, an intelligent data-driven fault diagnosis method is an important means to monitor the health status of the gas turbine, and it is necessary to obtain sufficient fault data to train the intelligent diagnosis model. In the actual operation of a gas turbine, the collected gas turbine fault data are limited, and the small and imbalanced fault samples seriously affect the accuracy of the fault diagnosis method. Focusing on the imbalance of gas turbine fault data, an Improved Deep Convolutional Generative Adversarial Network (Improved DCGAN) suitable for gas turbine signals is proposed here, and a structural optimization of the generator and a gradient penalty improvement in the loss function are introduced to generate effective fault data and improve the classification accuracy. The experimental results of the gas turbine test bench demonstrate that the proposed method can generate effective fault samples as a supplementary set of fault samples to balance the dataset, effectively improve the fault classification and diagnosis performance of gas turbine rotors in the case of small samples, and provide an effective method for gas turbine fault diagnosis.

The performance of intelligent fault diagnosis models is often limited by the scarcity and inaccessibility of fault data. Although existing data augmentation techniques-such as generative adversarial networks and denoising diffusion probabilistic models (DDPMs)-partially reduce this limitation, the synthesized data frequently lack physical consistency. To address this challenge, this study introduces SemPhyGen, a semantic-guided and feature-corrected denoising diffusion probabilistic model designed for high-fidelity fault data generation. First, a two-dimensional convolutional neural network is employed to extract fault semantics from real data, thereby constructing a semantic library representative of actual fault characteristics. Second, a conditional DDPM is developed, guided by the extracted semantics and optimized using a joint loss function that accounts for both noise and semantic errors, enhancing the realism of the generated data. To further ensure adherence to physical principles, a feature correction mechanism based on time-domain characteristics is incorporated. Experimental results confirm that SemPhyGen produces higher-quality augmented fault data compared to existing methods, offering more reliable inputs for intelligent fault diagnosis systems.

In industrial production systems, rapid and accurate fault detection is crucial for enhancing productivity, ensuring safety, and reducing the risk of accidents. However, since fault data is generally imbalanced, existing sampling methods are typically sensitive to noisy data and prone to generating noisy data. Meanwhile, the high dimensionality of fault data further significantly weakens fault detection performance. To solve the above problems, this paper proposes an improved hybrid sampling method based on diffusion model with two-stage noise processing combined with counterfactual feature optimization for fault detection (CITL-PM-CITL-CE): Firstly, we introduced a confidence evaluation mechanism to optimize the Tomek Links method and proposed a data-cleaning approach named CITL. Based on CITL and the diffusion model, a hybrid sampling method (CITL-PM-CITL) with a two-stage noise processing mechanism is designed to create a balanced dataset and reduce noisy data. Secondly, to eliminate redundant features and improve data quality, a feature optimization method based on counterfactual explanations is employed for dimensionality reduction. Finally, fault detection is performed on the optimized dataset. It is proved that the proposed model has better generalization performance in the quality prediction of the Tennessee Eastman (TE) chemical process.

Low fault diagnosis accuracy in case of the insufficient and imbalanced samples is a major problem in the wind turbine fault diagnosis. The imbalance of samples refers to the large difference in the number of samples of different categories, or the lack of a certain fault sample, which requires good learning of the characteristics of a small number of samples. Sample generation in the deep learning generation model can effectively solve this problem. In this study, we proposed a novel multi-class wind turbine bearing fault diagnosis strategy based on the conditional variational generative adversarial network (CVAE-GAN) model combining multi-source signals fusion. This strategy converts multi-source one-dimensional vibration signals into two-dimensional signals, and the multi-source two-dimensional signals were fused by using wavelet transform. The CVAE-GAN model was developed by merging the variational auto-encoder (VAE) with the generative adversarial network (GAN). The VAE encoder was introduced as the front end of the GAN generator. The sample label was introduced as the model input to improve the model’s training efficiency. Finally, the sample set was used to train encoder, generator and discriminator in the CVAE-GAN model to supplement the number of the fault samples. In the classifier, the sample set is used to do experimental analysis under various sample circumstances. The results show that the proposed strategy can increase wind turbine bearing fault diagnostic accuracy in complex scenarios.

Current data-driven methods for diagnosing gas path faults in aero-engines often rely on large, costly fault sample sets and face challenges related to class imbalances. These include disparities in the quantity of normal and fault data, differences among fault types, and variations in fault severity levels. This paper proposes a novel generative model, TL-GMVAE, which integrates Gaussian mixture models (GMM) with variational auto-encoders (VAE) and incorporates a transfer learning (TL) strategy. Using a large dataset of normal operational data, the VAE learns latent feature mappings of engine behavior and establishes a joint probability distribution across sensor measurements. To account for complexity in engine operating conditions, the GMM is used as the sampling distribution of the VAE. This enhances the model’s ability to represent diverse operating scenarios. The pretrained model is then fine-tuned with a small dataset of gas path fault data, transferring knowledge to fault domains. Each fault-specific TL-GMVAE model serves as an independent generator for synthetic fault samples. The proposed approach is validated using several established classifiers. The impact of different Fault-Normal ratios and imbalances across fault categories on classification accuracy is analyzed. Additionally, the robustness of the method to individual engine variability is evaluated. Results demonstrate that the TL-GMVAE generates high-quality fault samples and significantly improves fault diagnosis accuracy. These findings highlight its potential application in aero-engine health monitoring and fault diagnosis systems.

Safe and stable operation of the aircraft hydraulic system is of great significance to the flight safety of an aircraft. Any fault may be a threat to flight safety and may lead to enormous economic losses and even human casualties. Hence, the normal status of the aircraft hydraulic system is large, but very few data samples relate to the fault status. This causes a data imbalance in the fault diagnosis of the aircraft hydraulic system, which directly affects the accuracy of aircraft fault diagnosis. To solve the data imbalance problem in the fault diagnosis of the aircraft hydraulic system, this paper proposes an improved GAN-LSTM algorithm by using the improved GAN method, which can stably and accurately generate high-quality simulated fault samples using a small number of fault data. First, the model of the aircraft hydraulic system was built using AMESim software, and the imbalanced fault data and normal status data were acquired. Then, the imbalanced data were used to train the GAN model until the system reached a Nash equilibrium. By comparing the time domain and frequency signal, it was found that the quality of the generated sample was highly similar to the real sample. Moreover, LSTM (long short-term memory) and some other data-driven intelligent fault diagnosis methods were used as classifiers. The accuracy of these fault diagnosis methods increased steadily when the number of fault samples was gradually increased until it reached a balance with the normal sample. Meanwhile, three different sample generation methods were compared and analyzed to find the method with the best data generation ability. Finally, the anti-noise performance of the LSTM-GAN method was analyzed; this model has superior noise immunity.

Integrated generative networks embedded with ensemble classifiers for fault detection and diagnosis under small and imbalanced data of building air condition system

The difficulty of obtaining sufficient fault samples is one of the vital problems in mechanical fault diagnosis with machine learning methods. Although fault data generated through simulations can supplement the fault samples, there exists a certain difference between the simulated data and the real ones, due to the lack of noise and other actual working environment factors in simulations. In engineering scenarios, since most equipment operates normally, the fault data is typically difficult to acquire, whereas the normal data is relatively easy to obtain and inherently contains valuable condition information absent in the simulated fault data. Therefore, the paper proposes a 1D-DDcGAN (1-Dimensional Dual Discriminator Conditional Generative Adversarial Network) based signal fusion method fusing the easily obtainable normal data and the simulated fault data to generate near-real fault data for the fault diagnosis of steel wire ropes without real fault data. Firstly, normal data of wire ropes are acquired through experiments and simulated data of different fault types are generated. Secondly, the 1D-DDcGAN is applied to fuse the real normal data and the simulated fault data to generate corrected-simulation fault data which is close to the real fault data. Finally, a convolutional neural network model is trained using these corrected-simulation fault data and tested using real fault data to evaluate the effectiveness of the proposed method. Experimental results demonstrate that the proposed method can fuse real normal data and simulated fault data to generate near-real fault data with closest similarity to real ones, and achieves the highest fault recognition accuracy among all comparative methods.

GAN (Generative Adversarial Network) is widely used in image generation, renowned for its ability to produce high-fidelity details and sharp edges through adversarial training. Unlike Variational Autoencoders (VAEs), which often generate blurrier outputs, GANs excel in visual realism by leveraging a dual-network architecture—a generator and a discriminator—engaged in a competitive learning process. Furthermore, GANs synthesize images in a single forward pass, making them significantly faster than iterative approaches like Diffusion Models, which rely on multi-step denoising. This efficiency enables real-time applications, a critical advantage in fields such as AI-assisted art creation. This essay begins by outlining the foundational concepts of GANs, including their adversarial training mechanism. Next, it explores their methodology, emphasizing key architectures and training techniques that enhance stability and output quality. A comparative analysis with VAEs and Diffusion Models follows, highlighting GANs' superior perceptual quality while acknowledging challenges such as mode collapse and training instability. Finally, the discussion shifts to GANs' transformative role in AI painting, where they facilitate style transfer, photorealistic artwork generation, and interactive digital art tools. By examining these aspects, this essay underscores GANs' unique contributions to generative AI while addressing their limitations and future potential.

An interpretable data augmentation scheme for machine fault diagnosis based on a sparsity-constrained generative adversarial network

Motor bearings are the most failure-prone components in industrial motors, and the accuracy of their fault diagnosis is highly dependent on the effective acquisition and analysis of vibration signals. However, traditional Compressed Sensing (CS) methods face an inherent trade-off between compression ratio and reconstruction accuracy, while deep learning-enhanced CS models generally suffer from complex architectures and insufficient real-time performance. These drawbacks severely restrict their practical application in industrial fault diagnosis scenarios. To address the aforementioned challenges, this study proposes a lightweight compressed sensing framework tailored for motor bearing fault diagnosis-Hybrid Autoencoder Generative Adversarial Network (HAGAN). This framework integrates the core advantages of Autoencoders (AE) and Generative Adversarial Networks (GAN). The AE is responsible for extracting key fault features and compressing high-dimensional vibration signals, while the GAN improves the fidelity of reconstructed signals through an adversarial training mechanism. Meanwhile, a streamlined network structure design is adopted to remove redundant nonlinear modules, thereby reducing computational overhead and ensuring real-time deployment capabilities in industrial settings. In the experiments, this method achieves an ultra-high data compression ratio of 100:1 for motor bearing fault data. Three metrics-Root Mean Square Error (RMSE), Percentage Root Mean Square Deviation (PRD), and Signal-to-Noise Ratio (SNR)-are employed to comprehensively verify the compression and reconstruction performance. The results demonstrate that the HAGAN framework can effectively preserve the characteristic information of bearings in both healthy states and various fault states (including inner race faults, outer race faults, rolling element faults, and combination faults) even under high compression ratios. Its compression and reconstruction performance, quantified and validated by the three metrics, is excellent. The framework can be successfully applied to data compression and reconstruction tasks in motor bearing fault diagnosis, providing an efficient and reliable fault diagnosis solution for resource-constrained industrial environments.