Failure Prediction App

This web app was written to serve the needs of a hands-on workshop. Much of the data preparation work is automated.

The source data was downloaded as a csv and is served via a PostgreSQL database. The prediction feature can use one of two data sources.

• csv files that contain prediction data with 720 data points
• Apache Kafka which simulates data as it would be streamed from sensors

Prerequisites

Last tested with:

• Python 3.8
• OpenShift 4.10
• Shell: bash
• OS: mac,linux (various)
• Podman 4.4.4
• DBeaver 23.*

Workshop

You can find the workshop instructions here

OpenShift Setup

You can view the default parameters for bootstrap.sh here.

For more info about interacting with the database in OpenShift look here

# workshop user parameters
# note: run this ONLY if you are a WORKSHOP USER before bootstrap.sh

export NAMESPACE=$(oc whoami)
export DB_HOSTNAME="predict-db.edge-failure-prediction.svc.cluster.local"
export KAFKA_HOSTNAME="kafka-cluster-kafka-bootstrap.edge-kafka.svc.cluster.local"

scripts/bootstrap.sh

Local Quickstart

Setup local postgres container

. scripts/bootstrap.sh
container_setup_db

Setup local development

Note: Python version 3.9+ may fail. replace python3 with python3.8 if needed.

The following section only needs to be run once

python3 -m venv venv

# activate your virtual env with the following
. venv/bin/activate

pip install -r src/requirements.txt

Run local web app

# custom connection example
export DB_HOSTNAME="waterpump.ci3tyclo8vsc.us-east-1.rds.amazonaws.com"
export DB_DATABASE="postgres"
export DB_USERNAME="postgres"
export DB_PASSWORD="FR2s3rv2ll3y"
export DB_PORT="5432"

# reactivate your virtual env with the following
. venv/bin/activate

cd src

python3 wsgi.py

Run local Jupyter Notebook

pip install -r notebooks/requirements.txt
jupyter-lab

Acknowledgements

Data preparation to train the model is an adaptation the work done by Xiaxiau: here

Source of the data used in this application is found: here

Future Improvements

The ML model used in this application needs improvement, particularly in the prediction. The next version will include those improvements.

Team Members

1. Audrey Reznik
2. Cameron Garrison
3. Cory Latschkowski
4. Eli Guidera