Highly Scalable ML-based Digital Twin Model for Analyzing and Forecasting Equipment Anomalies Results in USD 1.7MM Savings

Business Objective

Our client is an Indian multinational steel-making company that operates across five continents and is among the largest steel manufacturers globally. The existing process for manufacturing equipment anomaly detection was primarily operator driven manual inspection with a few analytics models for a few sensors.

It took two to three months to develop a model for one equipment, and there was no standardized framework for creation and selection, making it a very analyst dependent solution. The objective was to improve equipment maintenance by building a digital twin model that is analyst independent and automatically selects the best model for equipment anomaly detection.[

Challenges

Not enough data to build the model and missing data in some cases resulting in data inadequacies to build a model
Sensor data for certain equipment was not available due to sensors not being installed
Data from the installed sensors were not flowing into the database adequately

Solution Methodology

Utilized sensor-based, near real-time information (Vibration, Pressure, RPM, Current, Temp. etc.) from the equipment to build a robust framework for automatic model selection for anomaly detection
Implemented models for both anomaly detection and forecasting (anomaly forecast for the next 72 hours) using a multi-model technique. The models analyzed huge packets of data and compared with historical downtimes to detect early warning signals
Initiated proactive user alerts to suggest proactive maintenance operations
Developed a real-time dashboard and updated it to reflect system health, anomalous behavior, 72-hr forecast of reliability, and performed a sensor-based root cause analysis

Business Impact

The solution resulted in an immediate saving of USD 1.7 MM in terms of cost of maintenance and additional production time gained
The scalable models are being deployed in blast furnaces, crushers, fans, transformers, and similar types of equipment across the company
Transparent approach – no dependency on external IPs and solutions. On-demand future upgrades are possible due to open design concepts

Privacy Notice Cookie Policy