Multi-modal industrial process fault diagnosis method based on confrontation of local maximum mean value difference

Abstract

The invention discloses a multi-modal industrial process fault diagnosis method based on confrontation of local maximum mean difference, and the method comprises the steps: carrying out the sliding window interception of process sequence data under a small sample condition, and obtaining a two-dimensional fault diagnosis input sample; local dynamic features of the multi-modal process data are extracted through a convolution feature extractor; introducing a gradient inversion layer to enable a convolution feature extractor and a domain discriminator to form an adversarial relationship, extracting inter-modal domain invariant features in an adversarial manner, and realizing global distribution alignment; and embedding local maximum mean difference (LMMD) measurement into a full-connection layer of a class label predictor, and accurately realizing class-level alignment in combination with pseudo label information of a target modal label-free sample. According to the multi-modal fault diagnosis method, LMMD sub-domain alignment is introduced on the basis of adversarial training to realize migration of key process knowledge, and finally, cross-domain fault diagnosis performance is effectively improved.

Description

technical field [0001] The invention relates to the field of process monitoring and fault diagnosis based on data driving, in particular to a fault diagnosis method for multi-modal industrial process based on resisting local maximum mean difference, which is proposed for fault diagnosis under the condition of small sample of multi-modal industrial process . Background technique [0002] Fault monitoring and fault diagnosis technology play a vital role in the safe production of industrial processes. Factors such as changes in production demand, the impact of the external environment, fluctuations in product raw materials, and damage to process unit equipment cause the process to operate in different working conditions, making the industrial process feature multi-modal. In different operating modes, the statistical characteristics of the process are quite different. If the fault diagnosis model of the same operating mode is used, the fault diagnosis performance of other modes...

AI Technical Summary

Problems solved by technology

[0007] In order to overcome the problem of difficult modeling of fault diagnosis in the target modal field under the condition of small samples in the existing methods, the present invention adopts a multi-modal industrial process fault diagnosis method based on the anti-local maximum mean difference (ALMMD)

Images