[Paper Review] Time Series Diffusion Method: A Denoising Diffusion Probabilistic Model for Vibration Signal Generation

Haiming Yi, Lei Hou+4 more|arXiv (Cornell University)|Dec 13, 2023

Spectroscopy and Chemometric Analyses43 references4 citations

TL;DR

This paper proposes Time Series Diffusion Method (TSDM), a denoising diffusion probabilistic model tailored for vibration signal generation using an improved U-Net with attention blocks and ResBlocks. It achieves state-of-the-art performance in generating high-fidelity single- and multi-frequency signals and bearing fault signals, improving small-sample fault diagnosis accuracy by up to 32.38% on public datasets.

ABSTRACT

Diffusion models have demonstrated powerful data generation capabilities in various research fields such as image generation. However, in the field of vibration signal generation, the criteria for evaluating the quality of the generated signal are different from that of image generation and there is a fundamental difference between them. At present, there is no research on the ability of diffusion model to generate vibration signal. In this paper, a Time Series Diffusion Method (TSDM) is proposed for vibration signal generation, leveraging the foundational principles of diffusion models. The TSDM uses an improved U-net architecture with attention block, ResBlock and TimeEmbedding to effectively segment and extract features from one-dimensional time series data. It operates based on forward diffusion and reverse denoising processes for time-series generation. Experimental validation is conducted using single-frequency, multi-frequency datasets, and bearing fault datasets. The results show that TSDM can accurately generate the single-frequency and multi-frequency features in the time series and retain the basic frequency features for the diffusion generation results of the bearing fault series. It is also found that the original DDPM could not generate high quality vibration signals, but the improved U-net in TSDM, which applied the combination of attention block and ResBlock, could effectively improve the quality of vibration signal generation. Finally, TSDM is applied to the small sample fault diagnosis of three public bearing fault datasets, and the results show that the accuracy of small sample fault diagnosis of the three datasets is improved by 32.380%, 18.355% and 9.298% at most, respectively.

Motivation & Objective

• To address the lack of diffusion model applications in vibration signal generation, where signal quality evaluation differs fundamentally from image generation.
• To develop a diffusion-based generative model specifically adapted to the temporal and spectral characteristics of one-dimensional vibration signals.
• To improve the quality of generated vibration signals, particularly for fault features in bearing signals, using architectural innovations in U-Net.
• To validate the effectiveness of TSDM in few-shot fault diagnosis scenarios using real-world bearing fault datasets.
• To demonstrate that architectural enhancements like attention blocks and ResBlocks significantly outperform standard DDPM in vibration signal generation.

Proposed method

• TSDM employs a denoising diffusion probabilistic framework, applying forward diffusion to gradually add noise to time-series data and reverse denoising to generate realistic signals.
• The model uses a U-Net backbone enhanced with residual blocks (ResBlocks) and temporal attention mechanisms to better capture long-range dependencies in 1D signals.
• TimeEmbedding is integrated to condition the denoising process on the diffusion step, enabling step-wise refinement of the signal.
• The architecture is trained end-to-end using a denoising score matching objective, minimizing the difference between predicted and ground-truth denoised signals.
• The model is evaluated on single-frequency, multi-frequency, and bearing fault datasets to assess fidelity and feature retention.
• TSDM is fine-tuned and applied to few-shot fault diagnosis by augmenting small training sets with generated signals.

Experimental results

Research questions

• RQ1 Can a diffusion model effectively generate high-fidelity one-dimensional vibration signals with preserved frequency and fault features?
• RQ2 How does the integration of attention blocks and ResBlocks in a U-Net architecture improve vibration signal generation compared to standard DDPM?
• RQ3 To what extent can TSDM enhance small-sample fault diagnosis performance on real-world bearing datasets?
• RQ4 Does the model preserve fundamental frequency components and fault-related transients in generated bearing vibration signals?
• RQ5 How does TSDM compare to baseline DDPM in generating complex, non-stationary vibration signals?

Key findings

• TSDM successfully generates single- and multi-frequency vibration signals with high fidelity, accurately preserving the target frequency components.
• The model retains essential fault features in bearing vibration signals, including transient impulses and periodic patterns, even after diffusion and denoising.
• The improved U-Net with attention and ResBlocks significantly outperforms standard DDPM, which fails to generate high-quality vibration signals.
• On three public bearing fault datasets, TSDM improved small-sample fault diagnosis accuracy by up to 32.38%, 18.355%, and 9.298% respectively.
• The ablation study confirms that the combination of attention and residual blocks is critical for capturing complex temporal dynamics in vibration signals.
• TSDM demonstrates robust generalization across diverse signal types, including clean, multi-frequency, and fault-impulse signals.