Legacy

https://github.com/AngelUrq/space-legacy

https://youtu.be/5I-vaAmO9ik

Our team has developed a notebook with Python that can read a collection of PDF files of public NASA Technical Report Server documents , summarize them, and generate a list of topic keywords to help users to check the most important terms in the document that they have chosen. Most importantly, generation of summaries and keywords was made with natural language processing. The summaries were used to compute embeddings, a representation of the text used in semantic search to find relevant and similar papers.

We have also developed a prototype of an 3D interactive platform with three.js that allows users to search for a relevant document in 50 NTRS documents by a phrase. Then, when a user enters a phrase, our platform will position the camera in the relevant node that matches the search criteria and show a window with the paper information (title, authors, keywords, a summary, and a button that allows to see the original PDF). Each document is represented by a node which is a sphere. It can be connected with other nodes if they have content similarities. Therefore, if a node does not have a link with another, they have diferent content.

We believe that our tool will be a valuable and useful resource for democratizing the access to space data.

We used 50 papers of the public documents of the NASA Technical Report Server . Those were chosen randomly, and they had to be read and cleaned. The last was required because sometimes strange characters appeared. Subsequently, those texts were the input of two NLP models, and the results were their summaries and keywords. The summaries were used to compute embeddings, a representation of the text used in semantic search.

These are the documents that we use for our prototype:

Blade Tip Rubbing Stress Prediction

Evolution And Use Of Combined Mechanical And Thermal Codes For Cryogenic Turbopump Bearings

Adjustable Impedance, Force Feedback And Command Language Aids For Telerobotics (Parts 1-4 Of An 8-Part Mit Progress Report)

A Hybrid Architecture For The Implementation Of The Athena Neural Net Model

Methods And Strategies Of Object Localization

Planning 3-D Collision-Free Paths Using Spheres

Computational Neural Learning Formalisms For Manipulator Inverse Kinematics

Techniques And Potential Capabilities Of Multi-Resolutional Information (Knowledge) Processing

Satellite-Map Position Estimation For The Mars Rover

Planning Robot Actions Under Position And Shape Uncertainty

Organising Geometric Computations For Space Telerobotics

Model System Studies With A Phase Separated Membrane Bioreactor

Countermeasures To Microgravity

Fermentation And Oxygen Transfer In Microgravity

Film Boiling Transition Temperature For Tissue Cooled With Liquid Nitrogen

Theory Of Cubical Complexes With Applications To Diagnosis And Algorithmic Description Quarterly Report, 26 May - 10 Aug. 1970

Inflatable Honeycomb Patent

The Prediction Of Coronal And Interplanetary Magnetic Fields

Gas Lubricant Compositions Patent

Trajectory-Correction Propulsion System Patent

Null-Type Vacuum Microbalance Patent

Device For Directionally Controlling Electromagnetic Radiation Patent

Electrically-Operated Rotary Shutter Patent

Thermo-Protective Device For Balances Patent

Si/N,P/ And Si/N,Alpha/ Reaction Cross Sections At 14.5 Mev Neutron Energy

High Altitude Radiation Measurements Quarterly Report, 1 Jan. - 31 Mar. 1970

Some Performance Characteristics Of Divergent- Field Bombardment Thrusters

Radar Data Collection Mission, Addendum Final Report

Psychological Effects Of Substantial And Appetizing Menus For Submarine Personnel

Helicopter Testing In A Wind Tunnel

Flexible Foam Erectable Space Structures Patent

Pivotal Shock Absorbing Pad Assembly Patent

Effects Of Roughness On Heat Transfer In Conical Nozzles

Electrical Discharge Apparatus For Forming Patent

Investigation To Identify Paint Coatings Resistive To Microorganism Growth

Novel Shift Register Eliminates Logic Gates And Power Switching Circuits

The Deterioration Of Intermediate Moisture Foods

Probability Learning - The Shortest Path Hypothesis

Improved Smoke Generator For Low-Speed Wind Tunnels

Communications System For Zero-G Simulation Tests In Water

Digital Decoder For Phase-Delay Coded Data

Double Phase-Lock Loop With Rapid Transient Response - A Concept

Optimum Doping Achieves High Quantum Yields In Gaas Photoemitters

Cast Segment Evaluation

Insulation Assembly Uses Cryopumping To Reduce Heat Transfer In Cryogenic Liquid Line

Voltage Regulator Dissipates Minimal Power And Functions As A Voltage Divider

Use Of Cermet Thin Film Resistors With Nitride Passivated Metal Insulator Field Effect Transistor

New Procedure For Determining Minimum Time Orbit Transfers

Study-Simulation Of Space Station Dynamics

Treatment Of Blueberries Prior To Freeze Dehydration

This hackathon was a great experience. We learnt a lot about using 3D objects in web, working with NASA technical report server, extraction of pdf information, and how to solve problems using deep learning for natural language processing (NLP). Our team decided to choose this challenge because we think that NLP is a powerful branch that fits perfectly with this problem, so we applied it in our project. We are very proud to have won the local competition and to represent our country. 

However, one of the hardest problems we faced was to calculate the coordinates for each node in the 3D space. We solved this problem by brainstorming and using a whiteboard. Also, another problem was the the processing of pdfs since these have different formats and sometimes strange characters were observed when reading them. Thus, we had to process those characters to have a clean text.

Overall, this was a great experience and we are very proud of our achievement.

• 3D Platform: https://threejs.org/
• Web: https://vitejs.dev/
• NLP GPT-3 model: https://openai.com/
• NLP models: https://huggingface.co/
• https://www.python.org/
• API: https://flask.palletsprojects.com/en/2.2.x/
• Textures: http://planetpixelemporium.com/
• Textures: https://www.solarsystemscope.com/textures/
• Dataset: https://ntrs.nasa.gov/

#AI,#NLP,#NTRS,#paper,#visualization