A refrigerator manufacturer was spending $2.4M per year on compressor warranty claims. A structured investigation using MSE, COV, and DOE cut that to $430K. Here is exactly how they did it.
Case Study · 2025-04-10 · 10 min · OpEx Excellence Team
The Problem: $2.4M in Annual Warranty Claims
A mid-size refrigerator manufacturer was facing a growing warranty crisis. Compressor-related claims had climbed to $2.4 million annually, representing 34% of total warranty costs. Field data showed that compressors were failing or losing efficiency between 14 and 22 months after purchase, well within the 24-month warranty period. The engineering team had already tried three fixes over the previous 18 months: upgrading to a premium compressor supplier, tightening the refrigerant charge specification, and adding a secondary cooling fan. Each fix reduced claims temporarily before they climbed back up. Total cost of these attempted fixes: $1.8 million in tooling, materials, and engineering time.
Step 1: Validate the Measurement System (MSE)
Before investigating the compressor performance, the team ran an MSE study on their end-of-line efficiency test station. Ten compressors were tested twice each by three different operators. The results were revealing: 41% of the total observed variation came from the measurement system itself. Nearly half of the variation they were seeing in efficiency readings was noise from the test station, not real differences between compressors. Units that barely passed the end-of-line test might have been perfectly fine, and units that passed easily might have had real problems masked by favorable measurement error. The team recalibrated the test fixtures and standardized the test procedure. After improvements, the measurement system contribution dropped to 12%, well within acceptable limits.
41% of observed variation was measurement noise. The team had been making accept/reject decisions based on a test station they could not trust.
Step 2: Understand the Variation (COV)
With a trustworthy measurement system, the team ran a Components of Variation study. They tested compressors across three production shifts, four assembly stations, and two refrigerant suppliers. The COV decomposition showed: 8% of variation came from the measurement system (acceptable after MSE improvements), 11% came from shift-to-shift differences, 52% came from assembly station differences, and 29% came from refrigerant supplier differences. This was a breakthrough. More than half the variation was coming from differences between assembly stations, something no one had considered. The previous fixes (better compressors, tighter specs, extra fan) had all been addressing the wrong sources.
Step 3: Map the Process
The team created a detailed process map of the compressor assembly and refrigerant charging process. For each step, they classified every factor as Controlled (C), Noise (N), or Experimental (X). This exercise revealed that the four assembly stations had different fixture ages, different torque wrench calibration schedules, and slightly different refrigerant charging pressures. These were factors everyone assumed were identical across stations but had drifted apart over time.
- Fixture age: 2 years to 7 years across stations (Noise factor)
- Torque wrench calibration: weekly vs. monthly schedule (Controlled factor to standardize)
- Charging pressure: 148-156 psi variation across stations (Experimental factor for DOE)
- Ambient temperature during charging: 68-78F depending on station location (Noise factor)
Step 4: Design and Run the Experiment (DOE)
The team designed a 2^4 full factorial DOE with four factors: charging pressure, charging duration, ambient temperature (noise), and fixture torque setting. With 16 runs plus 4 center points, they could estimate all main effects and two-factor interactions. The ANOVA results told a clear story. Charging pressure and charging duration had a strong interaction: the optimal charging time depended heavily on the pressure setting. At low pressure, longer charging times were better. At high pressure, shorter times were optimal. This interaction had been invisible to one-factor-at-a-time adjustments. The team also discovered that ambient temperature interacted with charging pressure, explaining why the same settings produced different results in summer versus winter.
The Results
After implementing the optimized settings and standardizing all four assembly stations, the team ran a three-month confirmation study. Compressor field failure rates dropped by 82%. Annual warranty claims fell from $2.4M to $430K. The total investment in the structured investigation was approximately $85,000 in engineering time and test materials, delivering a 23:1 return in the first year alone.
82% reduction in compressor failures. $1.97M in annual savings. 23:1 ROI in the first year.
Frequently Asked Questions
What is MSE and why does it matter?
Measurement System Evaluation (MSE) decomposes the total observed variation into its sources: the measurement system itself versus the actual parts. A high measurement system contribution means your readings are unreliable, and decisions based on that data may be wrong. Generally, the measurement system should contribute less than 30% of total variation to be considered adequate.
How many test runs does a DOE require?
A full factorial 2-factor DOE requires 4 runs. A 4-factor design requires 16 runs (or 8 runs with a half-fraction). Adding center points and replication typically brings a practical experiment to 20-32 runs. This is far more efficient than changing one factor at a time, which cannot detect interactions.
Can DOE be applied to existing production lines?
Yes. DOE is commonly run on existing production lines by systematically varying controllable factors during planned production windows. Split-plot designs can accommodate factors that are hard to change (like oven temperature) alongside factors that are easy to change (like cycle time).
Run your own investigation
Load the refrigerator compressor example data into the interactive DOE tool and walk through the complete analysis yourself.
Get started →Tags: case study, DOE, refrigerator, compressor, warranty reduction, cost savings