Machine Predictive Maintenance Classification

Overview

• This repository hosts an Artificial Neural Network model, served via FastAPI, that predicts whether a tool is on the verge of failure or not based on it's working environment. It also predicts the most probable type of failue the tool may experience using the same information.
• The app is deployed on streamlit. Try it out here
• Dataset obtained from Kaggle

Getting Started

Prerequisites

• Docker
• Docker Compose

A. Run with Docker

1. Clone the repository

git clone https://github.com/shaloy-lewis/machine-predictive-maintenance-classification.git
cd machine-predictive-maintenance-classification

2. Build and run the Docker container

docker-compose build
docker-compose up

3. Access the application

http://localhost:8080

B. Run Locally Without Docker

1. Clone the repository

git clone https://github.com/shaloy-lewis/machine-predictive-maintenance-classification.git
cd machine-predictive-maintenance-classification

2. Create and activate virtualenv

pip install virtualenv
python3.12 -m venv venv

For windows

venv/Scripts/activate.bat

For linux

source venv/bin/activate

3. Install all the required packages and dependencies

pip install -r requirements.txt

5. Run the server

uvicorn api:app --reload --port 8080 --host 0.0.0.0

6. Access the application

http://localhost:8080

Getting Predictions

curl -X 'POST' \
  'http://localhost:[hostname]/predict' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -d '{
  "air_temperature_k": 298.9,
  "process_temperature_k": 309.1,
  "rotational_speed_rpm": 2861,
  "torque_nm": 4.6,
  "tool_wear_min": 143,
  "Type": "L"
}'

Change the hostname with the hostname given on your environment

Sample Response

{
    "probability_of_failure": 0.999088,
    "failure_type": {
        "Heat Dissipation Failure": {
            "probability": 0.235062
        },
        "Overstrain Failure": {
            "probability": 0.143143
        },
        "Power Failure": {
            "probability": 0.618992
        },
        "Random Failures": {
            "probability": 0.0009
        },
        "Tool Wear Failure": {
            "probability": 0.001512
        }
    }
}