Machine fault diagnosis from vibration signals typically requires large amounts of labelled fault data, but such labels are often scarce in real-world settings. At the same time, many applications demand that inference happen locally on edge devices, which have limited computational resources. A new paper on arXiv introduces DualRes, a compact oscillatory state-space model that aims to tackle both challenges simultaneously.
The authors position DualRes as a solution for learning from scarce labelled fault recordings while keeping the model lightweight enough for edge deployment. By leveraging an oscillatory state-space design, the model is intended to capture the dynamics of vibration signals without relying on large labelled datasets or heavy computation.
The paper is at an early stage, with the abstract outlining the model's design goals rather than reporting detailed experimental results. Still, the work points toward a practical direction for vibration-based monitoring in resource-constrained environments, where both data scarcity and hardware limits are common obstacles.