PhyNet

PhyNet is a fault diagnosis framework for industrial fluid systems guided by physical variables, whose core component is physical-to-response cross-attention.

PhyNet's design does not require physical variables and response signals to be strictly aligned on the time axis; instead, physical variables serve as global operating constraints and response signals as local diagnostic evidence.

PhyNet encodes physical variables as sequence-level tokens and sets a learnable global response token for the response signal; this global token acts as the Query to retrieve the physical variable tokens (Key/Value), thereby dynamically selecting the relevant operating context when forming the local diagnostic representation.

Through attention and mask analysis, the paper shows that physical variables do play a substantive constraining role in explaining local responses, rather than being redundant inputs ignored by the model.

The paper evaluates PhyNet on four benchmarks, including three real-world cavitation datasets and one public two-phase flow dataset, and PhyNet achieves the best performance on all evaluation metrics.

Class-level analysis shows that PhyNet's ability to distinguish ambiguous states improves markedly, indicating that the performance gain does not come merely from overfitting to easy samples but from improving the decision boundary for hard samples.

PhyNet abandons the assumption of strict temporal alignment, bringing a lower deployment threshold and stronger transfer adaptability, because it is normal for industrial systems to upgrade sensors and adjust sampling strategies.

The PhyNet framework still depends on the availability of physical variables, and its advantages may narrow in scenarios where physical measurements are missing or severely distorted.

PhyNet's core paradigm is to let physical knowledge constrain signal interpretation, rather than letting signals accommodate physical variables.