• A small sample fault diagnosis method based on diffusion models and 2D-CNN is proposed. • An imbalanced dataset mixed with original samples and generated samples is constructed to simulate real industrial imbalanced scenarios. • The reciprocating pump fault test bench is built and vibration signals are collected, which verifies the effectiveness of the proposed method. For complex and precise mechanical equipment such as reciprocating pumps, which are critical components in energy conversion and fluid transport systems, it is difficult to perform large-scale fault simulation. The scarcity of fault data poses a serious challenge not only to equipment fault diagnosis but also to the overall reliability and operational safety of the power and energy systems they serve. In the existing research, generating fault data is regarded as an effective way to solve such problems. However, traditional generation methods such as variational autoencoders (VAE) and generative adversarial networks (GAN) and their improved versions, often have the limitations of poor generation quality and unstable training in practical applications. To overcome this technical bottleneck, this study proposes a novel scheme for small-sample fault diagnosis based on a diffusion model and a two-dimensional convolutional neural network (2D-CNN). Firstly, the noise of the reciprocating pump fault data is processed by the diffusion model, and the noise is gradually denoised to generate high-quality samples similar to the original data distribution. Then, the quality of the generated fault samples is evaluated, and an extended fault sample database is constructed on this basis. Finally, a 2D-CNN is used to classify and diagnose reciprocating pump faults. Experimental verification on reciprocating pump fault datasets shows that the proposed scheme is effective and superior. This demonstrates its significant potential for ensuring the operational stability of electrical power and energy infrastructures.
Zhang et al. (Sun,) studied this question.