top of page

Bearing Condition Monitoring and Fault Severity Analysis

Project Overview

This project focuses on monitoring the health of machine bearings using vibration data and analytics. Bearings are critical components in rotating machinery, and their failure can lead to unplanned downtime, safety risks, and high maintenance costs.

The objective of this project is to build an end-to-end condition monitoring and predictive maintenance solution that:

  • Assesses bearing health using vibration signals

  • Identifies fault types and severity

  • Supports maintenance decisions through a clear dashboard

  • Combines analytics, machine learning, and visualization


Business Problem

Traditional maintenance strategies are often:

  • Reactive (fix after failure), or

  • Preventive (replace too early)

Both approaches increase operational cost.

This project demonstrates how data-driven condition monitoring can:

  • Detect faults early

  • Prioritize maintenance actions

  • Reduce downtime and unnecessary replacements


System & Data Description

  • Rotating machine with two bearings

  • Each bearing monitored in X and Y directions

  • High-frequency vibration data collected from sensors

  • Fault conditions included:

    • Normal (healthy)

    • Ball defect

    • Inner race defect

    • Outer race defect

Each fault produces distinct vibration behavior that can be analyzed to assess machine condition.


Data Processing & Feature Engineering (Python)

Raw vibration signals are large and noisy, making direct analysis impractical.

Using Python:

  • Vibration signals were segmented into time windows

  • Statistical features were extracted to represent machine behavior

Key features used:

  • RMS (Root Mean Square) – represents overall vibration energy and severity

  • Kurtosis – captures impulsive shocks caused by bearing defects

These features are standard in industrial condition monitoring.


Health Severity Index

To make vibration data interpretable for non-technical users, a Health Severity index was created by combining RMS values across both bearings.

This index:

  • Represents overall bearing condition

  • Enables risk classification

  • Forms the basis for KPIs and dashboard visuals


Machine Learning Model (Fault Classification)

A machine learning model was trained to classify bearing condition based on vibration features.

  • Algorithm: Random Forest Classifier

  • Inputs: RMS and Kurtosis features

  • Target: Fault Type (Normal, Ball, Inner, Outer)


Model Insights:

  • Inner race faults were detected with very high confidence

  • Ball and outer race faults showed partial overlap, reflecting real industrial behavior

  • Feature importance confirmed RMS and kurtosis as key indicators

The model was evaluated using accuracy, confusion matrix, and confidence (probability) analysis.


👉 View the complete Python implementation on GitHub




Power BI Dashboard (Decision Support Layer)

While analytics and ML provide insights, operational decisions require clear visualization.

A Power BI dashboard was built to:

  • Translate vibration analytics into actionable information

  • Support operators, engineers, and managers

  • Enable quick, data-driven maintenance decisions

Dashboard Explanation

KPI Section

  • Average Health Severity – overall machine condition

  • High-Risk Operating Conditions (%) – proportion of time in critical state

  • Maximum Health Severity Observed – worst vibration event detected


Fault Analysis Visuals

  • Overall Vibration Severity by Fault Type – compares RMS severity

  • Impulsive Vibration Severity by Fault Type – highlights shock-based faults

  • Health Severity Distribution by Fault Type – combined severity comparison


Interactivity

  • Fault Type slicer

  • Risk Level slicer

Allows users to focus on specific faults or high-risk conditions.


Digital Twin & Predictive Maintenance Perspective

This project represents a component-level Digital Twin for bearing condition monitoring:

  • Physical system: bearings and vibration sensors

  • Digital representation: features and health indices

  • Analytics: fault classification and severity assessment

  • Decision layer: interactive dashboard

Together, this enables predictive maintenance and risk-based decision making.


Who This Is For

  • Operators: quick visibility into machine condition

  • Maintenance Engineers: fault identification and prioritization

  • Managers: risk monitoring and maintenance planning


Key Takeaways

  • Raw sensor data was transformed into meaningful health indicators

  • Machine learning validated fault patterns and severity

  • Power BI enabled clear, actionable decision support

  • The project demonstrates end-to-end industrial analytics and predictive maintenance thinking


bottom of page