.png){kind=icon}
.png){kind=icon}
Return
Prototype prediction based fault similarity for zero-shot industrial fault diagnosis
.png){kind=icon}
PRE
AI
C
Y
H
Z
Q
DOI：10.1016/j.jprocont.2026.103708.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In industrial processes, zero-shot fault diagnosis (ZSFD) has emerged as a promising paradigm for the diagnosis of unseen faults. While most existing approaches rely on semantic attribute-enriched fault descriptions to bridge the gap between seen and unseen fault categories, such approaches typically require certain specialized domain expertise or an in-depth understanding for a complex industrial production process. Therefore, a prototype prediction based fault similarity (PPFS) method is proposed for zero-shot industrial fault diagnosis. The PPFS framework learns to determine the fault categories using simple fault descriptions rather than collected fault samples. The core lies in predicting the fault prototypes of unseen categories by leveraging seen fault prototypes and a fault similarity matrix between seen and unseen categories. The seen fault prototypes are obtained through a multi-scale adaptive sparse coding feature extraction method, while the fault similarity matrix is constructed by analyzing variable contributions and their inter-correlations. Experimental results demonstrate that the proposed PPFS advances ZSFD performance without neither unseen-fault training samples nor in-depth industrial process knowledge.
Keywords:
Zero-shot fault diagnosis
Fault similarity
Prototype prediction
Industrial process
Journal
IF:
3.9
Papers:
3.4K
Citations:
7.3K
Organization
Cited Papers