Rocket health degree evaluation method based on long-short-term memory auto-encoder

Abstract

A rocket health degree evaluation method based on a long-short-term memory auto-encoder belongs to the technical field of engine health diagnosis, and comprises the following steps: taking a long-short-term memory unit as a basic unit for constructing a network model, constructing a model by taking the auto-encoder as a framework, performing compressed encoding on original data, performing decoding, and comparing with the original data, and judging the data health degree through the error level between the two. According to the method, the model of the long-short-term memory auto-encoder is adopted to train the ground data, the obtained model can be used for engine data diagnosis in the rocket flight process, measurement faults can be effectively eliminated through verification of historical flight data, the diagnosis accuracy is larger than 95%, the false alarm rate is lower than 5%, and health degree evaluation can be accurately achieved.

Description

technical field [0001] The invention relates to a training method for a launch vehicle engine health diagnosis model based on a long-short-term memory autoencoder, and belongs to the technical field of engine health diagnosis. Background technique [0002] Aerospace has always been a very important sector in my country's strategic deployment, and the rocket engine, which is the power source of aerospace, is even more important. In the rocket launch mission, engine performance and engine thrust play a decisive role in the launch process. How to evaluate the engine health status in an efficient and reliable way during the launch process is the key to the redundant control of rocket power. The current mainstream detection methods mainly include parameter threshold detection, mathematical model prediction, etc. These methods rely on the fixed model coefficients of each engine component, which cannot dynamically reflect the internal relationship between various engine parameters,...

AI Technical Summary

Problems solved by technology

The current mainstream detection methods mainly include parameter threshold detection, mathematical model prediction, etc. These methods rely on the fixed model coefficients of each engine component, which cannot dynamically reflect the internal relationship between various engine parameters, and cannot be very sensitive to the differences in different engine processes. Good matching, the reliability of the diagnosis result of the engine working condition is low, which is not enough to be used as the basis for the rocket to judge whether the power is abnormal or not

Images