Zero-Shot Fault Diagnosis Model for Manufacturing Process with Multi-Dimension Attributes

The problem of unbalanced fault samples or even zero samples at the initial stage of production line deployment for manufacturing processes will affect the accuracy of fault diagnosis models at batch production. In this case, zero-shot learning is well utilized for fault diagnosis of manufacturing processes. However, existing zero-shot learning methods ignore the representation of fault attributes in different dimensions. Based on this, this paper proposes a zero-shot fault diagnosis method for manufacturing processes based on multi-dimension attribute descriptions to fully utilize the public knowledge to identify unknown faults. First, the fault modes in the manufacturing process are analyzed using process failure mode and effects analysis (PFMEA) to form the multi-dimension fault attribute description space in fault dimension, equipment dimension and process dimension. After that, a multilayer network is utilized to extract the features related to the fault modes in the process quality data and establish the mapping relationship between the attribute space and the feature space in order to realize the zero-shot fault diagnosis of the manufacturing process. Finally, the effectiveness of the proposed method is verified using the Tennessee-Eastman Process (TEP) process.