A Method for Identifying Gas Path Faults in Aeroengine Envelope Based on Elm Filtering Algorithm

Abstract

The invention discloses a gas path fault identification method for an aeroengine envelope based on an ELM filtering algorithm. The method comprises training an ELM model topology parameter by using aKalman filter algorithm; and using an ELM model trained well by the filter algorithm for gas path fault identification in the engine envelope. According to the method, the problem that the conventional data-driven engine fault diagnosis is not strong in generalization at different working points and the precision is not high in the existing gas path fault diagnosis of the aeroengine engine in theenvelope is solved, and the method is suitable for engine failure mode identification at different working points in the flight envelope and has a positive promotion effect on engine health managementand reduction of maintenance costs.

Description

technical field [0001] The invention belongs to the technical field of air path fault diagnosis of aero-engines, and in particular relates to an air path fault identification method in an aero-engine envelope based on an ELM filter algorithm. Background technique [0002] As the heart of an aircraft, an aeroengine has a complex structure and a harsh working environment. Engine fault diagnosis technology is an important means to ensure engine performance and reliability, and reduce maintenance costs. During the service life of an aero engine, component performance slowly degrades. In addition, mutations in component health parameters may also occur. The failure of gas circuit components affects the performance and reliability of aeroengines, and it is necessary to diagnose them. Extreme learning machine (Extreme Learning Machine, ELM) is a fast learning method developed on the basis of neural network theory, and has been widely used in data mining, pattern recognition and ...

AI Technical Summary

Problems solved by technology

[0003] The ELM algorithm also has some defects. Its random generation of input weights and biases brings a certain degree of randomness to ELM, and different initialization parameters will cause different learning effects.

And because the precision of the least squares algorithm itself is not high, which affects the stability and generalization ability of the ELM model

One of the ways to make up for this defect is to use the combination of neural network ideas, but this network structure will be redundant in the network structure and increase the computational complexity

Using genetic algorithm or cross-validation method to obtain the optimal hidden node parameters is also a measure to improve the stability of ELM, but these algorithms require a large number of iterative calculations, resulting in low learning efficiency

Images