Deep Generative Models in Condition and Structural Health Monitoring: Opportunities, Limitations and Future Outlook

TL;DR

This paper surveys how deep generative models (DAR, VAEs, GANs, diffusion, and LLMs) can enhance condition monitoring and structural health monitoring across rotating machinery, aircraft, wind turbines, and civil infrastructure. It highlights four core applications—data generation, domain adaptation/generalization, multimodal data fusion, and fault diagnosis/anomaly detection—while detailing practical challenges such as fine-tuning efficiency, explainability, and edge deployment. The review provides quantitative and qualitative comparisons to traditional methods, discusses current industrial adopters, and outlines future directions including physics-informed DGMs and reinforcement learning with DGMs. Overall, DGMs hold substantial promise for robust, data-efficient, and multimodal CM/SHM solutions, but require advances in interpretability, computational efficiency, and domain-specific tuning for widespread adoption.

Abstract

Condition and structural health monitoring (CM/SHM) is a pivotal component of predictive maintenance (PdM) strategies across diverse industrial sectors, including mechanical rotating machinery, aircraft structures, wind turbines, and civil infrastructures. Conventional deep learning models, while effective for fault diagnosis and anomaly detection through automatic feature learning from sensor data, often struggle with operational variability, imbalanced or scarce fault datasets, and multimodal sensory data from complex systems. Deep generative models (DGMs) including deep autoregressive models, variational autoencoders, generative adversarial networks, diffusion-based models, and emerging large language models, offer transformative capabilities by synthesizing high-fidelity data samples, reconstructing latent system states, and modeling complex multimodal data streams. This review systematically examines state-of-the-art DGM applications in CM/SHM across the four main industrial systems mentioned above, emphasizing their roles in addressing key challenges: data generation, domain adaptation and generalization, multimodal data fusion, and downstream fault diagnosis and anomaly detection tasks, with rigorous comparison among signal processing, conventional machine learning or deep learning models, and DGMs. Lastly, we discuss current limitations of DGMs, including challenges of explainable and trustworthy models, computational inefficiencies for edge deployment, and the need for parameter-efficient fine-tuning strategies. Future research directions can focus on zero-shot and few-shot learning, robust multimodal data generation, hybrid architectures integrating DGMs with physics knowledge, and reinforcement learning with DGMs to enhance robustness and accuracy in industrial scenarios.

Figures (11)

• Figure 1: The cluster to indicate the co-occurance among keywords (condition monitoring, fault diagnosis, deep generative models, and etc.).
• Figure 2: A deep generative model $\mathcal{G}_{\bf{\theta}}$ is trained to map samples from a tractable distribution $\mathcal{Z}$ to the more complicated distribution $\mathcal{G}_{\bf{\theta}}(\cal Z)$, which approximates the true distribution $\cal X$. Finding an objective function that minimizes the discrepancy between the generated samples and the original samples is the key obstacle to train the DGM.
• Figure 3: Evolving applications of deep generative models in CM/SHM: from focused tasks to integrated industrial systems.
• Figure 4: Data transformation-based augmentation approach. In CM/SHM applications, both time series and 2D image formats are utilized. The 2D representations are typically generated from 1D time series through signal transformation techniques such as the short-time Fourier transform and the wavelet transform. The airplane wing and wind turbine blade figures are reused from Refs. Qing03042022 and Yaowen2021, respectively.
• Figure 5: An illustration of domain adaptation and domain generalization WAN2025109614. Domain adaptation focuses on minimizing data distribution discrepancies between source and target domains, whereas domain generalization seeks to transfer knowledge from multiple source domains to an unseen target domain without utilizing any target domain data.