SMO

Software system that enables both data exploration and generation of a selection of Machine Learning models for spacecraft engineers with no need of data scientist involvement.

Objectives

Improving the data exploitation using the knowledge extracted by both AIT/AIV and operations data to support and optimize the specification, evolution and maintenance of the satellite subsystems: ML models are trained within the application by data scientists and distributed to the spacecraft engineers.

SMO enables the spacecraft engineers to design their own ML pipelines, selecting and configuring pre-defined modules in a simple way, without requiring data scientist competencies.

SMO also ensures the possibility of exploiting AIT/AIV and operations data for the development of Machine learning-based models with anomaly detection and behaviour prediction capabilities. The target use case developed by SATE deals with the detection of degradation of reaction wheels bearings during manoeuvres.

Benefits

Optimization through real data

Possibility to compare engineering models (as-designed models) with test and operational ones (as-built and as-flown models)

Enhancement of as-designed models

Comparison of as-designed models results with as-built and as-flown data allows the optimization and the refinement of the former for obtaining more reliable results from on-ground simulations

Diagnostic capabilities of ML models

The ML models developed within SMO have diagnostic capabilities useful for behaviour prediction and for anomaly detection of modelled systems

Features

SMO is developed with a modular architecture in order to extend its capability adding new modules

Modular architecture

The SMO architecture allows to compare different models and simulation data and to generate a report with the summary of the performed comparison

Report generation

The architecture is scalable from small teams to big organization

Scalability

The ML pipeline can be composed of many ML models developed in SMO

Multimodel interaction

System Architecture

The system is composed of the Big Data subsystem, the Machine learning subsystem, the Microservices and REST API layer and the Web Client Application.

  • The Big Data subsystem is in charge of leading and storing data in the SMO
  • The Machine Learning subsystem is in charge of managing the Machine learning lifecycle
  • The microservices and REST API layer provides the interfaces to interact with the data source software and the Machine Learning subsystem

There are two web client applications, one dedicated to users such as spacecraft engineers and data scientists and the other dedicated to the system administrators.

Main Partners

Prime contractor of the project. ALTEC has coordinated the entire project mainly focusing on the infrastructure and on the integration of the different developed modules in the overall system.

Thales Alenia Space Italia (TASI) has been involved in SMO project as data provider for the development of the use cases and as potential end user of the system.