A production line was struggling with a self-discharge problem. A process engineer discovered the root cause and isolated the equipment it was coming from with this plot:
Figure 1. A parallel coordinates plot that shows the relationship between self-discharge (measured as ΔV) and electrode slitting equipment IDs. This shows that high ΔV values are all coming from Slitter 1.
The parallel coordinates plot is used to visualize multi-dimensional datasets, like in batteries. You can draw a range on one axis, and it will highlight all the cells that fall in that range. For each highlighted cell, it then connects a line to the next variable. This makes a relationship between the variables visually obvious.
Figure 2. Left plot shows high ΔV (>300µV) only comes from Slitter 1. Right plot shows that normal ΔV (150-200µV) come from all slitters.
Why it matters
When it comes to batteries, time is money. Finding problems faster doesn't just mean solving them faster; it also means less defective cells being produced. This plot allows you to easily scan through tens of variables at a time, quickly screening through thousands of potential root causes.
The easily-grasped visual nature makes it a convincing tool to persuade teammates and managers. This intuition makes it easy to collaborate with teammates in root cause analysis.
How this was worth $2M
All the high ΔV cells came from Slitter 1. Drilling down on Slitter 1 showed that the blade distance was causing burrs with a different morphology than expected. Fixing this improved the yield by 2%, which at $3/cell and 100,000 cells/day, is $2M.
The truth is that in the vast majority of pilot and production lines, most data isn't looked at. There is an overwhelming amount of variables being collected at all times -- process variables, electrochemical variables, equipment settings, build metadata, and more. In a production setting, this means that potential root causes for systemic problems are simply never investigated.
Consider a common problem like self-discharge, measured as a ΔV during the aging step. The list of potential causes of self-discharge is nearly every process, material, and setting. To attempt to find a root cause requires investigating:
| Item to investigate | Mechanism |
| Incoming Quality Inspection | Contamination, impurities |
| Mixing | Contamination, rheology, equipment |
| Coating | Contamination, uniformity, loading |
| Drying | Contamination, temperature, environment |
| Calendaring | Contamination, waviness, thickness |
| Slitting | Contamination, burrs |
| Winding | Electrode cutting, tab welding, overhang |
| Assembly | Contamination, cell can, welding |
| Formation | Formation protocol |
Conventionally, you'd examine each process step individually in a sequential manner over a period of several months. The typical workflow is to gather the data and plot a scatterplot, one variable at a time.
With a parallel coordinates plot, you can combine the electrochemical data and process metadata to generate insights significantly faster. You can do an initial pass to determine the process area, and then a second pass to drill down into the exact cause in that process area.