Module 3: Model training

This module explores the model training infrastructure on the Single Node OpenShift (SNO) server. You’ll access a Jupyter notebook environment, import training notebooks from Git, and execute the complete model training workflow for battery health prediction models.

Learning objectives

By the end of this module, you will be able to:

  • Access and configure Jupyter workbench environments in Red Hat OpenShift AI

  • Import training notebooks from Git repositories using JupyterLab

  • Execute model training workflows for Stress Detection and Time to Failure prediction

  • Understand the training pipeline from data collection through model export

  • Export trained models to OpenVINO format for edge deployment

Exercise 3.1: Review and access workbench

Red Hat OpenShift AI provides Jupyter workbench environments for data scientists to develop and train machine learning models. The workbench includes pre-configured runtimes with necessary libraries and tools.

Review the model-training workbench

  1. Ensure you’re in the ai-edge-project project in the Red Hat OpenShift AI Dashboard

  2. Navigate to the Workbenches tab

  3. Locate the model-training workbench

  4. Click on the three vertical dots (⋮) on the right and click Edit workbench

Workbenches tab showing model-training workbench

Understand Workbench configuration

The Workbench is configured with:

Table 1. Workbench configuration
Parameter Value Purpose

Name

model-training

Workbench identifier for model training

Image selection

Jupyter | TensorFlow | CUDA | Python 3.12

Development environment with ML frameworks

Version selection

2025.2

Latest version available

Hardware profile

default-profile

2 CPUs and 4 GiB of memory

Attach DataConnection

Connect the Workbench to the existing DataConnection that contains the credentials to access the MinIO instance in the transportation robot.

  1. Select Attach existing connections

  2. Verify that the vehicle-models connection is selected

  3. Click Attach

  4. To apply the configuration, click Update workbench

Access the model training workbench

  1. Click model-training to launch JupyterLab

JupyterLab dashboard

Verify

Confirm the workbench is accessible and operational:

✓ model-training workbench exists in Workbenches tab

✓ DataConnection has been attached to the workbench

✓ JupyterLab Dashboard is accessible

Exercise 3.2: Import training notebooks

The training notebooks are stored in a Git repository. You’ll clone the repository directly into your JupyterLab environment to access the complete training workflow.

Clone the Git repository

  1. In JupyterLab, click the Git icon in the left sidebar (looks like a branching diagram)

  2. Click Clone a Repository

    JupyterLab Git clone option

  3. Enter the repository URL:

    https://github.com/rhpds/ai-lifecycle-edge-automation.git

  4. Click Clone

JupyterLab clones the repository and creates an ai-lifecycle-edge-automation folder in your workspace.

The ai-lifecycle-edge-automation folder is imported

Verify

Confirm the repository was cloned successfully:

✓ Clone operation completes without errors

✓ ai-lifecycle-edge-automation folder appears in file browser

Exercise 3.3: Execute training notebooks

Run the training notebooks to train both battery health prediction models. The notebooks demonstrate the complete ML workflow from data collection to model export.

Understand the training workflow

  1. In the file browser, expand the ai-lifecycle-edge-automation folder

  2. Navigate to notebooks/training/

  3. Verify you see the training notebooks:

    1. retrieve-influx-data.ipynb - Retrieve battery telemetry from InfluxDB on robots

    2. prepare-data.ipynb - Clean, normalize, and prepare training datasets

    3. bms-stress-training.ipynb - Trains the Stress Detection model and saves it locally

    4. predict-stress.ipynb - Validates the Stress Detection model performance with new data

    5. bms-ttf-training.ipynb - Trains the Time to Failure model and saves it locally

    6. predict-ttf.ipynb - Validates the Time to Failure model performance with new data

    7. save-models.ipynb - Store in MinIO instance in MicroShift for distribution to robots

  4. Open in order each file and press the Play button at the top to manually run each cell or the Double Play icon to run the entire file

Play and Double Play buttons to run the Notebook

At this point, you can follow the instructions contained in each Notebook. Those will explain each step in detail. Pay attention to some of the variables used during the execution. You will need to replace some of them based on your environment. For example, in the first notebook you will need to replace the InfluxDB URL with yours:

https://influx-db-microshift-001.apps.cluster.example.com

Once you have run all of them, the new AI models should have been uploaded to the vehicle’s MinIO storage and are now being used by the Battery Monitoring System for real-time stress detection and time-to-failure predictions.

Verify model export

After all notebooks complete, check that models were exported to OpenVINO format and MinIO contains updated models:

  1. Open MinIO dashboard: https://minio-microshift-001.apps.cluster.example.com

    Credential Value

    Username

    minio

    Password

    minio123

  2. Navigate to inference bucket

  3. Verify stress-detection/1/ contains model.xml and model.bin

  4. Verify time-to-failure/1/ contains model.xml and model.bin

  5. Check file timestamps show recent updates

Verify

✓ Notebook executes all cells without errors

✓ Data collection retrieves telemetry from InfluxDB

✓ Training shows loss decreasing over epochs

✓ Upload confirms storage to MinIO

✓ stress-detection/1/model.xml exists with recent timestamp

✓ stress-detection/1/model.bin exists with recent timestamp

✓ time-to-failure/1/model.xml exists with recent timestamp

✓ time-to-failure/1/model.bin exists with recent timestamp

Summary

You have successfully trained battery health prediction models using Red Hat OpenShift AI:

✓ Workbench Access - Launched Jupyter environment with pre-configured Python runtime

✓ Git Integration - Cloned training notebooks from repository

✓ Stress Detection Training - Trained binary classification model for battery stress

✓ Time to Failure Training - Trained regression model for remaining battery life

✓ Model Export - Converted models to OpenVINO IR format for edge deployment

✓ Model Storage - Uploaded trained models to MinIO for distribution

What You’ve Learned:

  • How to access and use RHOAI Jupyter workbench environments

  • Git integration in JupyterLab for notebook management

  • Complete ML training workflow from data to deployment

  • TensorFlow model training and evaluation techniques

  • OpenVINO model conversion for edge optimization

  • S3 storage integration for model distribution

The models are now trained and ready for serving.

Next, you’ll explore the model serving infrastructure that loads these models and exposes inference endpoints for real-time predictions.

Navigate to Module 4: Model serving.