ctbto-min-350x

Comprehensive Nuclear-Test-Ban Treaty Organization (CTBTO)

ctbto-min-350x

Comprehensive Nuclear-Test-Ban Treaty Organization (CTBTO)

Description of Activities on AI

The Comprehensive Nuclear-Test-Ban Treaty (CTBT) bans nuclear explosions on the Earth’s surface, in the atmosphere, underwater and underground. The Treaty has a unique and comprehensive verification regime consisting of three pillars:

  • The International Monitoring System (IMS) will, when complete, consist of 337 facilities worldwide to monitor the planet for signs of nuclear explosions. Around 90 percent of the facilities are already up and running.
  • The International Data Centre (IDC) at the CTBTO’s headquarters in Vienna acquires data from the IMS monitoring stations. The data are processed automatically, reviewed by human analysts and distributed to the CTBTO’s Member States in both raw and analyzed form. On-site inspections (OSI) can be dispatched to the area of a suspected nuclear explosion if data from the IMS indicate that a nuclear test has taken place there. Inspectors collect evidence at the suspected site.

Artificial Intelligence (AI) is applied in all three pillars of the verification regime as outlined below.

Project 1: To detect fall army worm damage using a mobile application

Classifying signals from seismic stations to determine their seismic phase based on features measured automatically. The features include amplitude, frequency content, particle motion parameters, etc. Manual data processing of signals from seismic stations is cumbersome thus the need to automate data processing at ICTBTO’s International data center.

  • Project Type (Status): Software project (Proof of concept)
  • Project Domain: Global nuclear explosion monitoring
  • AI Approach: Artificial Neural Networks (ANN) and Bayesian Classifiers
  • Datasets: Automatic signals from the International Monitoring System (IMS), reviewed and corrected by human analysts.
  • Related SDGs: SDG 16 Peace, Justice and Strong Institutions
  • Resources/Skills: Human experts to review and correct the signals from seismic stations of IMS
  • Technology: Deep Learning

Challenges: Improvement of the current system by retraining the existing ANNs on a per station basis and replacing the ensemble of ANN and Bayes Classifiers with a deeper ANN. Methods are being explored for seismic phase identification directly from the waveform signal. Further studies are being undertaken to determine if the use of additional information, such as the raw waveform data, during classification can further improve performance

Project 2: Network Processing of detected signals to determine the events that have triggered them

Detection of events by on-site inspections for every signal detected is time consuming and expensive hence the need for network processing of signals detected at seismic, infrasound and hydro-acoustic stations in determining the events that have caused these signals to be observed.

  • Project Type (Status): Software project (Deployment)
  • Project Domain: Global nuclear explosion monitoring
  • AI Approach: Rule-based
  • Datasets: Signals detected at seismic, infrasound and hydro-acoustic stations of IMS
  • Related SDGs: SDG 16 Peace, Justice and Strong Institutions

Challenges: Further research is being undertaken on the classification of radionuclide spectra by ANNs

Project 3: NET-VISA (NETwork processing Vertically Integrated Seismic Analysis)

Improvement of the current rule-based system.

  • Project Type (Status): Software project (Deployed)
  • Project Domain: Global nuclear explosion monitoring
  • AI Approach: Machine Learning + physics model. The theoretical underpinnings are based on the “Open Universe Probability Model”
  • Datasets: Signals detected at seismic, infrasound and hydro-acoustic stations of IMS
  • Related SDGs: SDG 16 Peace, Justice and Strong Institutions
  • Project Partners: University of California (developing NET-VISA software)
  • Resource: Bayesian approaches. Knowledge of seismic, infrasound, and hydro data

Challenges: Extending to stations without detailed history from which to derive priors.

Project 4: Automatic triage

Distribute certain trouble tickets based on their content.

  • Project Type (Status): Software project (Deployment)
  • Project Domain: Global nuclear explosion monitoring
  • Datasets: Signals detected at seismic, infrasound and hydro-acoustic stations of IMS
  • Related SDGs: SDG 16 Peace, Justice and Strong Institutions

Project 5: Predicting failure at IMS stations

Predicting failure at IMS stations based on extensive State Of Health (SOH) parameters that are continuously collected and store.

  • Project Type (Status): Software project (Deployed)
  • Project Domain: Global nuclear explosion monitoring
  • AI Approach: Statistical methods and rule-based system; Next approach: ANNs and Support Vector Machines (SVM).
  • Datasets: IMS data and noble gas monitoring system SOH data.
  • Related SDGs: SDG 16 Peace, Justice and Strong Institutions
  • Project Partners: Pacific National Northwest Laboratory (PNNL)

Project 6: Seismic aftershock monitoring

Monitoring changes in the geological structures caused by a possible nuclear explosion and classifying “weak” detections produced to enable separation of noise from signals of interest (aftershocks).

  • Project Type (Status): Software project (Testing)
  • Project Domain: Global nuclear explosion monitoring
  • AI Approach: AI-based technique and Self Organizing Map (SOM)
  • Datasets: IMS raw waveform data
  • Related SDGs: SDG Peace, Justice and Strong Institutions
  • Project Partners: University of Stuttgart (developed AI-based technique)

Project 7: Satellite monitoring for On Site Inspection (OSI)

The use of air-and-spaceborne multispectral imagery (MSIR) for classification and change detection in the inspection area, with the ultimate goal of limiting the search area and detecting features of interest.

  • Project Type (Status): Software project (Ideation)
  • Project Domain: Global nuclear explosion monitoring
  • AI Approach: Pixel-based classification (unsupervised and supervised Machine learning), object-based classification (decision rules and fuzzy-logic) and Change detection techniques using Multivariate Alteration Detection (MAD)
  • Datasets: Air and Space-borne multispectral imagery (MSIR)
  • Related SDGs: SDG 16 Peace, Justice and Strong Institutions
  • Resource: GIS (Geographic Information Systems) operations

Challenges: Timeframe during an ‘On Sight Inspection (OSI)’ (e.g. availability of imagery)

Related Sustainable Development Goals (SDGs)

E-WEB-Goal-16-500px