Applied Data Analysis
Engineering Case StudyBattery Diagnostics
Repeatable analysis of an 84-cell Formula SAE battery pack, separating persistent cell anomalies from session-specific noise.
Named cell outliers across sessions
- My Contribution
- Built shared loading and diagnostic functions, session notebooks, and cross-session comparisons.
- Project Status
- Completed analysis · March 2026 logs
The Problem
One unusual reading cannot tell you whether a battery cell is weak. I compared voltage behavior across sessions to identify persistent outliers, with shared loading and diagnostic functions keeping notebooks consistent.
Method
Per-cell z-scores, rank stability, and loaded-voltage residuals gave different views of pack behavior. The March 11 fault session was excluded from trend metrics: a failed cell’s extreme value would otherwise compress differences among remaining cells.
Findings
Cell 43 was a high-side voltage outlier on March 3, then fell to approximately zero volts on March 11. The earlier recording contained a warning signature one session before collapse.
Cells 31 and 67 were persistent outliers associated with the P0A80 weak-cell fault. Recorded discharge-current limit fell from 198 A to 100 A.
These are findings from recorded logs. This case does not claim a deployed predictive model or that analysis prevented the failure.