Will there be a space version of ChatGPT?

ChatGPT is very popular now. As an artificial intelligence chat robot, it has information integration and natural language processing capabilities. It can chat, write papers, and compose poems according to requirements. It can also generate game scripts and write program codes.

In aerospace engineering missions, it is necessary to process large amounts of data and coordinate a large number of instruments, equipment and sensors. So can artificial intelligence play a big role in the aerospace field?

Some plots in science fiction movies do not seem to be fantasy, but rather ideal imaginations of the role of artificial intelligence technology in the aerospace field. Some things are also quietly changing.

The desire for self-diagnosis and repair capabilities

Spacecraft are expensive and have a complex working environment. If a fault occurs in the space environment, it is difficult to repair, which will cause significant losses. Therefore, it is particularly important for spacecraft fault diagnosis technology to develop in the direction of intelligence and efficiency.

"Spacecraft are becoming more and more complex, and the number of sensors is increasing. We originally used a fault diagnosis method with pre-set rules, which required setting judgment thresholds for various possible abnormalities. This requires a lot of expert experience, but we usually lack the ability to detect faults beyond experience." Gong Jianglei, deputy director of the Information Office of the Communications and Navigation Satellite Department of the Fifth Academy of China Aerospace Science and Technology Corporation, said.

Artificial intelligence technology will be a "good helper" for satellites to diagnose their own faults. Satellite designers are trying to carry artificial intelligence processing components on spacecraft to achieve autonomous navigation, control, data processing, fault diagnosis and reconstruction and maintenance work.

In the past, the satellite's in-orbit operating status, faults, lifespan, etc. were monitored and warned through ground measurement and control systems. "Artificial intelligence algorithms have the characteristics of continuous learning, reinforcement and self-adaptation. They adjust according to how the data changes, and will become more and more accurate," said Gong Jianglei.

"Especially with the rise of communication, navigation, and remote sensing needs and the increase in network complexity, it is more important than ever to manage remote information transmission well and ensure high-quality transmission performance and reliability." Gong Jianglei pointed out that, for example, the increase in space junk has increased the probability of space debris colliding with spacecraft; the development of satellite Internet has led to increasing risks in the field of information security, etc., all of which indicate that artificial intelligence technology has huge room for development.

Traditional fault feature extraction and identification evaluation methods are based on human experience and are difficult to accurately describe complex fault feature information. However, combining a variety of artificial intelligence learning algorithms under different application scenarios can effectively improve the efficiency of fault detection and feature classification caused by complex impacts of spacecraft, and provide risk warning and repair assistance in addition to detection and evaluation.

The idea of ​​combining ChatGPT with aerospace

The First Academy of China Aerospace Science and Technology Corporation is conducting research on "smart rockets", which will enable the rockets to automatically identify the results of failures and proactively change strategies when failures occur, implement autonomous flight, and make the rockets smarter and more intelligent.

Rocket terminal velocity correction flight process simulation diagram Today, with the high-density launch of launch vehicles becoming a normal practice, rockets are also very "eager" for autonomous fault diagnosis and mission reconstruction capabilities. An artificial intelligence expert from the First Academy Research Center said: "We will combine artificial intelligence and information technology to establish an intelligent health monitoring system for the entire life cycle of launch vehicles, which will run through the entire life cycle of rocket design verification, production and manufacturing, test launch and flight testing."

Can we study a customized ChatGPT?

Aerospace designers envision that by building an intelligent auxiliary design system, integrating the existing massive data and resources, simulating the human brain's thinking process, and forming a "smart brain", artificial intelligence may become an auxiliary tool for designing aerospace products in the future.

After being combined with aerospace, ChatGPT may be like an experienced senior aerospace expert, who can systematically analyze, understand and extract information from text data, and assist designers in the design of launch vehicles and spacecraft. Since the "smart brain" can complete a large amount of knowledge learning in a very short time and find the best solution, the standardization and level of design can also be guaranteed.

After the main engine fails, new parking and degradation orbits are planned online to save the rocket mission.

Taking smart rockets as an example, during the production process, the "smart brain" can collect and ensure information on the entire life cycle of the rocket, establish a comprehensive archive database, collect data and knowledge from each process of manufacturing, assembly, and testing, and build a big data analysis center as the data support and health diagnosis basis for smart rockets, thereby reducing design and development costs, improving test and launch efficiency, and improving the reliability of the rocket.

Customized ChatGPT also needs to use deep learning to model human thinking, so that machines can understand human behavior and apply knowledge to interactions with users, to achieve the ultimate goal of "humanization" of machines and realize the application of artificial intelligence technology in the aerospace field.

Experts pointed out that to realize customized ChatGPT, aerospace workers also need to inject a large amount of training samples and knowledge information into the machine learning process, and continuously improve the generalization and adaptability of the algorithm, conduct intensive training at the application layer, and improve the level of the algorithm.

Concerns and early ice-breaking

The development and application of artificial intelligence technology is quietly changing the original model and pattern of aerospace product design. However, Gong Jianglei pointed out the current difficulties: there is little sample data of telemetry faults of spacecraft, and the effect of deep learning algorithm training and application will be limited.

In addition, there are still many problems that need to be solved when applying artificial intelligence methods to on-orbit spacecraft, such as the insufficient computing power and storage resources of current spacecraft.

Gong Jianglei also said that most of the current artificial intelligence technologies are still in the "weak artificial intelligence" stage. Therefore, in the actual application of the aerospace field, the focus is still on developing artificial intelligence technologies at the auxiliary analysis and auxiliary decision-making levels.

Artificial intelligence with neural networks as its core is "unexplainable", which means that humans cannot know why the intelligent algorithm gives this result. "Unexplainable" means risk, especially in aerospace engineering missions that require "no mistakes", which is also what aerospace designers worry about.

"We need to break through 'small sample learning' and try to achieve better results through training with a limited number of data sets. We also urgently need to improve onboard processing capabilities. On the one hand, we use high-performance aerospace-grade chips to improve onboard computing capabilities. On the other hand, we upgrade the onboard information system architecture and improve onboard data transmission capabilities, laying a solid foundation for the application of artificial intelligence algorithms on satellites." said Gong Jianglei.

In the future, with further breakthroughs in algorithms, computing power and cloud technology, aerospace designers are optimistic that "strong artificial intelligence" will eventually move towards the application level of autonomous decision-making in the aerospace field.

Gong Jianglei has always believed that artificial intelligence technology is penetrating more and more widely into the aerospace field, and in the future it will definitely achieve cross-generational technology, thereby promoting the development of the entire aerospace field.

Aerospace designers' expectations for artificial intelligence technology are also their expectations for exploring the unknown and mysteries of the universe. Space and the sky are bound to become an important stage for artificial intelligence.