Flying and learning about my country’s intelligent launch vehicle on the road

　　Our reporter Fu Yifei

For the development of learning launch vehicles, the research team has formulated a "two-step" plan: to have strong adaptability in 2020, which means that the rocket will have stronger adaptability in some failure situations and can continue to fly; strive to 2025 Having the ability to learn every year means that my country’s main carrier rockets in service will initially have the ability to learn.

　　In the history of human spaceflight, space launch missions have high risks.

Improving the safety and reliability of launch vehicles has always been a goal pursued by aerospace science and technology workers.

In this regard, the rocket flight control system plays a vital role.

　　According to statistics, from 1990 to 2000, about 40% of the failures of launch vehicles in Europe, the United States, Japan, and Russia were likely to be remedied by advanced navigation guidance and control technology, so as to continue to complete the mission or degrade to complete the mission. .

In recent years, both the Falcon 9 and Delta 4 carrier rockets have experienced engine thrust drop failures in flight, but the implementation of different levels of remedial measures did not affect the main mission.

　　A reporter from Science and Technology Daily learned from the Beijing Aerospace Automatic Control Research Institute that my country is developing a launch vehicle that can learn.

Yu Chunmei, the deputy director of the institute, introduced that the technology introduces intelligent technology into various mission links such as navigation, guidance and control, making the launch vehicle smarter and more autonomous, and has a stronger active adaptation to complex environments and emergencies. Ability to ensure the completion of tasks to a greater extent.

Learn while flying and lifelong learning Perceive the surrounding environment, accumulate data and improve yourself

　　For a rocket, the flight control system is equivalent to its brain.

According to the research results of the research team of the Beijing Institute of Aerospace Automatic Control, this "brain" has two major characteristics for learning.

　　According to Yu Chunmei, the flight strategy and trajectory of traditional rockets are designed in advance. If the flight trajectory of the rocket cannot be adjusted online if it encounters an unexpectedly complex space environment during the flight.

In addition, if there is an unexpected situation on the rocket, such as a power system failure, and measures cannot be taken in time, the mission is likely to fail.

　　But learning about launch vehicles has the characteristics of learning while flying.

Its control system can make full use of the multi-source information of the rocket to enable the rocket to realize the functions of online identification of flight status and environment, online evaluation of carrying and control capabilities, online trajectory planning, online control reconstruction, and online target change.

Simply put, it can perceive its surrounding environment in real time and change its trajectory if there is a threat.

At the same time, the rocket will carry out self-diagnosis while flying, if it finds a problem, it will evaluate its own ability, take corresponding measures according to the actual situation, or adjust the mission plan.

"For example, if the rocket detects that the second-stage engine has failed, it can implement the second- and third-stage separation ahead of time and start the third-stage engine to continue flying. Even if the mission cannot be completed as scheduled, it can also plan a new trajectory based on the reconstruction of the control system. , Fly to a safe orbit or a degraded orbit to reduce losses as much as possible.” Yu Chunmei said.

　　Another characteristic of learning launch vehicles is lifelong learning.

Its control system can make full use of the data generated in the whole life cycle, based on big data, intelligent analysis technology, etc., to realize the functions of intelligent correction of models, intelligent establishment of models, intelligent optimization of schemes and parameters, and continuous self-learning and improvement.

　　Most of the current carrier rockets are for one-time use. During the flight of each rocket, it can transmit its own "learning content" back to the ground for later learners to learn from, and achieve transfer learning between rockets. Follow-up rockets are becoming more and more "smart" and "experience" more and more abundant.

　　The reporter learned that the above two learning characteristics complement and promote each other.

The accumulated experience and data of learning while flying supports lifelong learning; the training and optimization of lifelong learning and the self-evolution of algorithms can promote learning while flying to be smarter.

Rocket learning has a threshold control system that needs to be designed once and applied for life

　　The prospect of being able to learn the launch vehicle is bright, but to be able to learn, the rocket must first have sufficient capabilities.

　　"The first is to put forward higher requirements for the computing power of the control system, which requires the ability to have distributed heterogeneous cross-core high-speed information exchange and multiple heterogeneous memory sharing scheduling management." Yu Chunmei said.

In short, the Rockets have to have a smart enough brain.

　　There are unavoidable errors in the structural installation of the launch vehicle. Elastic vibration, liquid sloshing, unknown environmental disturbances and other factors will affect the control system. Therefore, the rocket is required to have real-time perception of its own health and flight status.

For example, typical non-fatal faults of the rocket power system occur from time to time. It is necessary to deal with the problems of the main engine thrust drop in the ascending phase, the normally open, normally closed, and polarity problems of the attitude control nozzle in the high-altitude flight phase, and the ability to quickly and accurately identify.

　　Based on the results of online identification, perception and self-assessment, the rocket must have the ability to change the target online, trajectory online planning, control online reconstruction and other functions, it must have the ability to respond to sudden changes in the environment, the uncertainty of the rocket's own structural parameters, typical power system failures, and task changes. The strong adaptability to other needs ensures stable flight under complex flight conditions and failure conditions, and can enter the largest semi-major axis elliptical orbit or safe parking orbit.

　　So far, the Long March series of carrier rockets have carried out 349 missions and accumulated a lot of data.

After these data are coordinated and regulated, they become the rocket's knowledge base.

Learn to extract information from the launch vehicle, acquire experience and knowledge, and combine it with intelligent control technology to realize self-learning ability, so as to continuously "grow".

　　In addition, the rocket must have lifelong learning, and the control system has also put forward the "one-time design, life-long application" requirement, which requires it to have the ability to cover the entire model life cycle with one design.

　　For the learning carrier rocket, the intelligent control system architecture is the foundation, the intelligent control algorithm based on full life cycle data is the core, and the strong computing power based on the intelligent algorithm is the carrier.

my country has made certain progress and strive to have the ability to learn by 2025

　　my country has made some progress in the exploration and application of relevant technologies for learning launch vehicles.

The reporter learned that the Beijing Aerospace Automatic Control Research Institute team has begun to improve my country's in-service carrier rockets and popularize their learning capabilities.

　　On July 9 and October 12, 2020, my country successively launched the Long March 3B carrier rocket from the Xichang Satellite Launch Center.

In addition to sending the satellite into the scheduled orbit for these two missions, the research team of Beijing Aerospace Automatic Control Research Institute also carried typical power failure identification and guidance control reconstruction technology on the arrow, successfully completing the flight test verification.

　　"It can be said that the Long March IIIB carrier rocket has certain intelligent features." Yu Chunmei said.

At present, the improvement of the Long March IIIB launch vehicle only modifies the algorithm, and all perceptions are still through the original components.

In the future, if we want to improve more thoroughly, we need to add detection methods to the arrows.

　　For the development of learning launch vehicles, the research team has formulated a "two-step" plan: to have strong adaptability in 2020, which means that the rocket will have stronger adaptability in some failure situations and can continue to fly; strive to 2025 Having the ability to learn every year means that my country’s main carrier rockets in service will initially have the ability to learn.

　　In addition, the concept of being able to learn will also be incorporated into the development of my country's research carrier rocket.