Root Cause Analysis for deviation
in fault rate in high
precision Manufacturing
The Challenge
Our client, a global leader in data storage solutions, had a problem in their manufacturing unit. They were experiencing high variance in failure rate of disks produced, from one day to the next. This caused losses due to scrapping as well as loss of face with customers whose delivery was delayed. The challenge they gave us was to develop machine learning models to identify early signs of variation in the manufacturing environment and to raise an alarm proactively, minimizing impact on production.
Hypothesis
- One or more of the numerous environmental conditions that are finely tuned to ensure low failure in disk production had shifted causing other parameters to also deviate causing faults in products produced.
- While their unit was well equipped to collect lots of data about the manufacturing process, the data was very high dimensional and it was difficult for them to diagnose what was the root case of his change in failure rate.
Execution
- Historical data was extracted from the data logs produced by the manufacturing unit of both, days when the failure rate was within acceptable numbers and those days when failure rate were abnormal.
- Ensemble Classification models were built using methods to deal with the imbalance in normal and abnormal manufacturing days and deployed.
Outcomes
- The client was now able to identify when the manufacturing process was at high risk of producing high failure rates.
- Parameters responsible were identified and helped the engineers reduce the down time by making timely interventions and hence reducing losses.
- Reduction in downtime of manufacturing process.
- Reduced scrapping of product through early onset identification.
- Improved SLAs with customers by achieving delivery timelines.
Project Highlights
20%
less downtime and
65%
reduction in waste output