Manufacturer- Machine Health Monitoring System

Business Need

Build a near real-time health monitoring system for the automated bottling lines that facilitates predictive maintenance.

Challenges

• None of the 54 automated plants had monitoring systems provided by the OEM.
• Limitations of traditional BI tools (PowerBI, Tableau) to render highly granular data.
• The lack of predictive maintenance capabilities resulted in substantial costs as well as a risk of potential losses from production delays.

Services

• Headless BI with Databricks pipelines were used to render KPIs.
• Streaming ETL with kSQL and Cube connectivity was used for near real-time rendering.
• Using COX was of death prediction, a machine learning pipeline and reinforcement learning algorithm were used to predict machine breakdown about 6 hours earlier.

Results

• Automated Data Aggregation.
• Precision & Granularity of KPIs up to a 5-minute level.
• 10x Faster than PowerBI.
• Saved up to $1.7 Million.
• Up to 65k data points rendered in <7 secs.
• <6 mins latency.