Method for predicting residual service life of aero-engine based on graph convolutional network and unsupervised domain self-adaption

Abstract

The invention belongs to the field of residual life prediction of aero-engines, and provides an aero-engine residual life prediction method based on a graph convolutional network and unsupervised domain self-adaption. The method comprises the following steps: firstly, preprocessing tagged source domain aero-engine sensor data and tagged target domain aero-engine sensor data; secondly, correlation among different sensor data is calculated through a maximum information coefficient algorithm, and a graph data set of the aero-engine sensor data is obtained; thirdly, extracting local features of sensor data and global features of a common structure by using a gated loop unit and an improved graph convolutional network, and fusing the features; finally, updating model parameters through a designed target function, training to obtain domain invariant features, and performing high-accuracy prediction on the residual service life label of the non-label target domain aero-engine sensor data. According to the method, the unsupervised cross-domain prediction accuracy of the aero-engine is improved.

Description

technical field [0001] The invention belongs to the field of remaining service life prediction of aero-engines, in particular to an aero-engine remaining service life prediction method based on graph convolution network and unsupervised domain self-adaptation. Background technique [0002] In recent years, with the continuous development of industrial technology, the operating status and maintenance methods of industrial equipment have received people's attention. Therefore, failure prediction and health management technology (PHM) is proposed to predict and manage the health status of equipment in industrial systems. This technology uses degradation management to predictively maintain the functional state of equipment to improve the safety and reliability of industrial equipment. Remaining useful life (RUL) prediction, as one of the fault prediction and health management technologies, has attracted widespread attention and has become a research hotspot in this field. [0...

AI Technical Summary

Problems solved by technology

However, in the scheme proposed in this paper, the target domain data and the source domain data use the same feature extractor. This strongly shared feature extraction layer may not be helpful for feature learning in specific domains.

[0006] In summary, the remaining life prediction method of aero-engine based on the unsupervised domain adaptive method mainly has the following defects: (1) only the RUL label and domain label of the input data are considered, and the data structure information of the input data is not introduced into the aeroengine Engine Unsupervised Domain Adaptation Model

(2) The existing models in the current field only consider how to extract domain-invariant information, but ignore the impact of specific features in the target domain on feature learning, and shared feature extraction parameters or shared feature extraction layers may have significant impact on specific domains. Feature learning is counterproductive

(3) At present, the existing methods for modeling the correlation between aero-engine sensor data are not clear enough, and cannot accurately and reasonably express the information correlation between sensor data

Images