Weighted alignment-like local domain adaptation mechanical fault diagnosis method

• Main Authors: ZHANG XIAO, HAN BAOKUN, ZHANG ZONGZHEN, WANG JINRUI, JI SHANSHAN, BAO HUAIQIAN
• Format: Patent
• Language: Chinese; English
• Published: 19.11.2021
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; HANDLING RECORD CARRIERS; PHYSICS; PRESENTATION OF DATA; RECOGNITION OF DATA; RECORD CARRIERS

The invention discloses a weighted alignment-like local domain adaptation mechanical fault diagnosis method, and relates to a fault diagnosis technology of rotating mechanical vibration signals. According to the method, firstly, representative features of a source domain and a target domain are extracted by adopting a convolution automatic encoder, a relatively large weight is distributed to a shared class crossing the two domains and a relatively small weight is distributed to an irrelevant class by utilizing target softmax output of a pre-training source classifier, cross-domain global features are aligned by adopting joint maximum average difference to realize knowledge migration to the greatest extent, and finally, a domain difference learning residual block is inserted into a shared classifier to correct and relieve domain differences. The residual block is learned through domain difference. According to the method, the model can be forced to explicitly learn the domain difference, reduction of the chara