Alstom Case Study: Industrialization of MobAn Projects

How we used Industrialization of MobAn Projects for enhanced customer experience

Project Description

The project involved working with the Prognostic Healthcare Management and Mobility Analytics R&D program teams from Alstom Digital & Integrated Systems. Throughout the collaboration, the team developed, tested, validated, and industrialized machine learning (ML) algorithms for Odometry and Radioscopy.

Business Objective

The industrialization of the mobility analytics project streamlined  Alstom’s analytics portfolio, improving customer performance and experience. The goal was to achieve these outcomes using existing and new data analytics, supported by advanced simulations and input from domain experts.

Industrialization-of-MobAn-Projects

Pain Points

At the onset of the project and during its execution, the team had to tackle pain points, such as:

  • Scalability challenges:
    • Unstructured code— when deployed in the
    • Production environment— would eat up a lot of resources and be difficult to scale up
  • Time consuming deployment process
  • Unmanageable code in Jupyter Notebooks: While the data scientists focus on solving the problems, most often, modularization, encapsulation, inheritance, etc., take a backseat
industrialized-machine-learning

High-level solution framework used to resolve the issue

Industrialization was undertaken for the client by extracting and converting the Jupyter Notebooks, applying OOPs concepts, hyper-parameterizing the code, transforming to server-less functions (OpenFaas) driven by NiFi Workflows, and deploying them over Azure-based MDP Platform using CI/CD.

Further, a Dash-based UI was developed to display the predictions, capture the user feedback, i.e., labels, and retrain the models. Additional Screens were also designed to display the Model Performance metrics.

Subsequently, ML and Data Pipelines were created for various stages. Training and Inference Pipelines were implemented to automate training/retraining or predictions of Anomaly Detection and Classification Models.

Tech Stack

  • For executing the project, the Teliolabs team employed:
  • Open Faas for serverless function
  • NiFi for workflow management
  • MLFlow for model performance tracking and experimentation
  • CI/CD for continuous integration and deployment
  • Postgre DB for storing structured data and integration of backend with UI
  • Minio buckets for storing Object Data (particularly large)
  • Dash for visualization to capture user feedback, display performance matrix, confusion matrix, Geolocation Maps, etc.
  • Azure for pipeline creation and deployment on Kubernetes

Key business benefits delivered by Teliolabs

  • From the project, the client acquired:
  • A UI for capturing user feedback and displaying the resultant performance matrix
  • A solution that could be automatically deployed
  • Industrialization capabilities

Project Niche

The project belonged to the Smart Mobility domain.

End Customer

Alstom is the end customer for the project.

 

Achievements

The Teliolabs team was able to achieve:

  • A successful launch of the beta version in a record time of 8 weeks, despite tight deadlines
  • A scalable and robust solution
  • An interface with a performance matrix that allowed users to label data and retrain the model effortlessly

Registration

Forgotten Password?