Artificial Intelligence and Natural Language Processing and Understanding in Space A Methodological Framework and Four ESA Case Studies

2025-04-30 0 0 4.04MB 65 页 10玖币

侵权投诉

Artiﬁcial Intelligence and Natural Language Processing

and Understanding in Space: A Methodological

Framework and Four ESA Case Studies

Jos´e Manuel G´omez-P´ereza, Andr´es Garc´ıa-Silvaa, Rosemarie Leoneb, Mirko

Albanic, Moritz Fontained, Charles Poncetb, Leopold Summererb,

Alessandro Donatie, Ilaria Romab, Stefano Scaglionie

aLanguage Technology Research Lab, Expert.ai, 3 Poeta Joan

Maragall, Madrid, 28020, Spain

bEuropean Space Research and Technology Centre (ESA-ESTEC), Keplerlaan

1, Noordwijk, 2201 AZ, The Netherlands

cEuropean Space Research Institute (ESA-ESRIN), Via Galileo Galilei,

1, Frascati, 00044, Italy

dEuropean Space Agency (ESA), 24 Rue du G´en´eral Bertrand CS

30798, Paris, 75345, France

eEuropean Space Operations Center (ESA-ESOC), Robert-Bosch-Str.

5, Darmstadt, 64293, Germany

Abstract

The European Space Agency is a powerful force for scientiﬁc discovery in

numerous areas of space. The amount and depth of the knowledge produced

throughout the diﬀerent missions carried out by ESA and their contribution

to scientiﬁc progress is enormous and involves large collections of documents

like feasibility studies, technical reports, scientiﬁc publications, and quality

management procedures, among many others. Handling such wealth of in-

formation, of which large part is unstructured text, is a colossal task that

goes beyond human capabilities, hence requiring automation. In this paper,

we present a methodological framework based on artiﬁcial intelligence and

natural language processing to automatically extract information and enable

machine understanding of space documents. We illustrate such framework

through several case studies implemented across diﬀerent functional areas of

ESA, including Mission Design, Quality Assurance, Long-Term Data Preser-

vation and the Open Space Innovation Platform, and demonstrate the value

of our approach by solving complex information extraction and language

understanding challenges that had not been addressed in space until now.

Preprint submitted to Engineering Applications of Artiﬁcial Intelligence October 25, 2022

arXiv:2210.03640v2 [cs.CL] 24 Oct 2022

Keywords: Space Science and Engineering, Artiﬁcial Intelligence, Natural

Language Processing and Understanding, Information Management

1. Introduction

The European Space Agency (ESA) is Europe’s gateway to space, with

the mission to shape the development of Europe’s space capability and ensure

that investment in space continues to deliver beneﬁts to the citizens of Europe

and the world. ESA consistently helps to answer the biggest scientiﬁc ques-

tions of our time, such as the mysteries of the Universe, the understanding of

our Solar System, and the quest for life outside our home planet. Its space

mission programs are a powerful force for scientiﬁc discovery, both looking

outward to the conﬁnes of our galaxy and beyond to understand the origin

of the Universe, as well as inwards, observing Earth through constellations

of satellites orbiting our planet to study Earth’s climate and deﬁne climate

change mitigation, adaptation and development pathways.

The amount, depth and scope of the data, information and knowledge

generated and managed during such missions is enormous and their contri-

bution to scientiﬁc progress is invaluable. From the announcement of op-

portunity and feasibility study to space and ground segments design, de-

velopment, operation, mission decommissioning, and long-term preservation,

large collections of heterogeneous information are produced. Some examples

include: ideas to develop innovative solutions to technical and operational

challenges, space project design and implementation documents, technical

reports, operational procedures, quality management instructions, and space

records about missions spanning over more than 40 years, like climate data

records, exploitation reports, and scientiﬁc publications. Managing, mining,

and exploiting such wealth of information, of which a large part is free text,

is a colossal task that goes beyond human capabilities.

In this paper, we present a methodological framework based on Artiﬁ-

cial Intelligence (AI) and Natural Language Processing and Understanding

(NLP/U1) to automatically extract information from text documents related

to space missions and enable machine understanding during diﬀerent mis-

sion stages, contributing to create a virtuous circle of knowledge acquisition,

management, and transfer at ESA, as well as scientiﬁc discovery and innova-

1Henceforth, we will use NLP indistinctly for both NLP and NLP/U.

tion worldwide. We demonstrate the added value of this approach through

actual NLP solutions implemented at ESA, with potential impact across a

wide range of space areas. The goals of such systems range e.g. from as-

sisting the evaluation of the innovation potential of ideas submitted to ESA

through the Open Space Innovation Platform (OSIP) to facilitating access

to space mission design information, contributing to the adoption of quality

assurance and training procedures, and helping to preserve heritage space

mission and operation records for long-term archival and exploitation.

This work contributes to unroll the vision of ESA’s 2025 Agenda2in

”adopting fast-learning/higher-risk approach for future technology matura-

tion such as AI” through the implementation of intelligent systems able to

support ESA’s workforce in several tasks, like eﬀortlessly searching and rec-

ommending space information within ESA’s repositories, automatically de-

termining how innovative an idea can be, answering questions about space-

craft design or generating training materials to master space operation proce-

dures. The accomplishments described in the paper represent a step forward

in increasingly intelligent AI focused on NLP and its applications for infor-

mation management in space, from assistants able to structure and facilitate

access to information to intelligent systems capable to understand and reason

with it. We envision a future where AI systems augment human capabili-

ties and engage with human peers in solving challenging technical problems,

joining forces in producing major scientiﬁc discoveries that could eventually

be worthy of a Nobel Prize (Kitano, 2016).

The remainder of the paper is as follows. Section 2 provides an account of

the diﬀerent types of applications of AI and NLP in the scientiﬁc enterprise

and their relation to space. Next, in section 3 we propose our methodolog-

ical framework for the application of NLP technologies in space. Section 4

summarizes the guidelines proposed in section 3 and makes special emphasis

on the aspects to consider in order to decide whether to adopt a machine

learning-based approach to NLP, a symbolic approach or a combination of

them depending on the results of the analysis of the use cases and NLP tasks

to be addressed. Sections 5 to 8 illustrate the application of our framework to

speciﬁc NLP projects recently developed at ESA. Based on such experiences,

section 9 provides a series of recommendations for the successful development

of NLP capabilities in space. Finally, section 10 concludes the paper.

2https://www.esa.int/About_Us/ESA_Publications/Agenda_2025

Figure 1: Foreseen progression of the role of AI systems in the scientiﬁc enterprise.

2. Antecedents and related work

In her presidential address at the AAAI Conference on Artiﬁcial Intel-

ligence, Gil (2022) pondered whether AI will write scientiﬁc papers in the

future. Both her and many others including us believe that we can be hope-

ful that the answer will be yes and that it may happen sooner than we might

expect. Our capabilities to do scientiﬁc and technical breakthroughs need

to be augmented as scientiﬁc questions become signiﬁcantly more complex.

Compare for instance the challenges of formulating Kepler’s laws of plane-

tary motion with demonstrating the existence of binary stellar-mass black

hole systems (Abbott et al., 2016). While the former was achieved by one

scientist, the latter required a large and interdisciplinary team involving the

collaboration of hundreds of scientists from diﬀerent ﬁelds to work together

during years to produce results.

Space science and engineering is no exception. The challenges that need

to be addressed are extremely complex and involve increasingly large and in-

terdisciplinary teams. AI and speciﬁcally NLP become imperative to manage

the large volumes of information that need to be processed during the lifecycle

of space missions. Some examples of scenarios in space where such capabil-

ities are required include the analysis of documents involving e.g. mission

objectives deﬁnition, mission feasibility and concept analysis, ground seg-

ment, space segment and launch segment requirements deﬁnition, design and

development, satellite platform operations or payload space records acquisi-

tion and processing, among many others. In those and other related areas,

NLP technologies are starting to prove their value. In some occasions, to ex-

tract information from large collections of scientiﬁc documents (Gomez-Perez

et al., 2017; Murdaca et al., 2018; Garcia-Silva et al., 2019; Berquand et al.,

2020, 2021b), producing semantic metadata (see section 7) that enables the

development of sophisticated information retrieval applications (Rico et al.,

2017), making research more accessible in accordance to the principles of

FAIR research data (Wilkinson et al., 2016), and contributing to long-term

data preservation. Other increasingly representative scenarios of application

of NLP technologies include systems that address needs related to language

understanding of space mission technical documents, like ESA’s Concurrent

Design Facility (CDF) reports or Quality Management procedures, automat-

ically answering and even formulating questions about space (Garcia-Silva

et al., 2022a,b) (sections 5 and 6).

Like Gil, we foresee a future scenario (see ﬁgure 1) where AI systems will

not only assist but also become an eﬀective part of the scientiﬁc and engi-

neering space ecosystem, collaborating, independently pursuing substantial

aspects of space mission development, operation, and space data records anal-

ysis, and contributing their own discoveries to the space community. Today,

we are already witnessing AI systems that address language understanding

challenges involving scientiﬁc and technical documents. As originally put

by Reddy (1988), ”Reading a chapter in a college freshman text and an-

swering the questions at the end of the chapter is a hard (AI) problem that

requires advances in vision, language, problem-solving, and learning theory.”.

Although this is one of the grand challenges in AI yet to be tackled, recent

AI systems like ARISTO (Clark et al., 2019) and ISAAQ (Gomez-Perez and

Ortega, 2020) are already capable to read a scientiﬁc text and answer ques-

tions related to its content at a level similar to humans. However, none

of such systems have focused on space yet, probably because they rely on

key components of modern NLP like transformer language models (Vaswani

et al., 2017) that until very recently have not been trained on space data.

Such limitation has started to be addressed with the advent of new language

models speciﬁc for space (Berquand et al., 2021a), which on the other hand

still need to prove real-life added value over general-purpose counterparts.

In this paper, we focus on the ﬁrst two of the three steps of the timeline

shown in ﬁgure 1, which represents the evolution of the possible roles to be

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

ArticialIntelligenceandNaturalLanguageProcessingandUnderstandinginSpace:AMethodologicalFrameworkandFourESACaseStudiesJoseManuelGomez-Pereza,AndresGarca-Silvaa,RosemarieLeoneb,MirkoAlbanic,MoritzFontained,CharlesPoncetb,LeopoldSummererb,AlessandroDonatie,IlariaRomab,StefanoScaglionieaLanguageT...

展开>> 收起<<

Artificial Intelligence and Natural Language Processing and Understanding in Space A Methodological Framework and Four ESA Case Studies.pdf

共65页,预览5页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Artificial Intelligence and Natural Language Processing and Understanding in Space A Methodological Framework and Four ESA Case Studies

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: