1 of 10

Sandia Data

Challenge Presentation

Team: PrGu (Jada Giddens, Colin Gu, Maclain Prom, John Bezaitis)

2 of 10

Problem Restatement and Data Overview

We are tasked with utilizing experiment manufacturing data to find the optimal combination of inputs to minimize the probability of scrapping the produced part.

Key Manufacturing Inputs

Build plate layout
Powder (virgin/recycled)
Test artifact presence
Part location on build plate

3 of 10

Understand the Problem

Our Plan

Verify using Statistical Tests

Combine and Process Datasets

Determine Scrapped Cups

Generate Visuals

Identify Optimal Conditions

Our first step was to grasp the problem to approach it effectively

Next, we combined the given datasets and dealt with data entry issues

Third, we filtered for the scrapped cups using given specifications

Create visualizations to clearly represent how factors influence scrap rate

Confirm suspicions with statistical testing to ensure accuracy

Use resources to state formal conclusions and recommendations

4 of 10

Data Cleaning and Organization

Initial Steps:

Merge AllData(Recycled/Virgin) using rbind
Rename measurements for ease of use
Create columns to check if each part clears requirements
General column ‘is_scrap’ to determine if the part is ready for use

Quick Notes:

1560 observations (CUPs)
Three 6x6 plates, three 6x6TA plates, and twelve 11X11TA plates
11 days of pure powder, 7 days of recycled

5 of 10

Takeaways

Thickness and Internal Diameter were the ONLY contributors to scrapping

Thickness above spec (every layout)
Internal Diameter below spec (11x11)

Violin Plot Creation

Filtered dataset to scrap-only cups
Plotted distribution of each measurement (LID, LED, FH, LH, LT1-LT4) by plate layout
Added specification limit lines (red dashed) for visual reference against acceptable ranges
Quantified failure directions showing whether scrap failed above or below spec limits

Specification Violating Measurements

6 of 10

Significance of Coefficients

By forming confidence intervals for each specification variable, we can plot which coefficients are significant.

Observing the plot we can see that Floor Height, LED, and LH were not significant.

This makes sense as they did not play any role in causing a CUP to be scrapped

7 of 10

Bar plots to create basic visualizations on the importance of plate layout and powder.

Takeaways:

Virgin powder brings a clear advantage in the reduction of scrap percentage
6x6 without TA also presents a significant advantage over 6x6TA and 11x11TA

The optimal combination of powder and plate to reduce scrap percentage is 6x6 and virgin powder.

Diving Deeper

8 of 10

Location on Plate

Manufacturers may want to understand which location on the plate leads to the lowest chance of scrapping. This knowledge could lead to changes in manufacturing to produce a more consistent product in the future.

Our heat map suggests that locations near the edge of the plate, and locations near TAs (which act as heat sinks) reduce the quality of the product. This is best seen on the 11x11 plate.

Mac -

Next let's consider WHERE on the build plate parts are positioned. Understanding optimal placement can help manufacturers improve consistency and reduce waste.

Our heatmaps reveal a spatial pattern. The green areas show zero scrap rates. Yellow and red indicate higher failure rates.

Notice on the 6X6 layouts, position doesn't matter much—everything stays pretty green. But look at the 11X11TA plate on the right: we see significantly higher scrap rates along the edges and near the test artifacts, shown here in yellow and orange.

Why? TAs act as heat sinks, creating uneven cooling that affects part quality. Edge positions likely experience similar thermal inconsistencies. For large plates, these visualizations show that center positions dramatically outperform edge locations, reducing scrap rate.

9 of 10

Statistical Validation: Confirming Our Findings

We conducted statistical tests to confirm that our observations weren't due to

random chance. These tests prove which manufacturing factors truly impact scrap rates.

Chi-Squared Test (𝝌²)

Measures if two factors are genuinely related
Higher values = stronger relationship

Cramér's V (Effect Size)

Measures HOW MUCH one factor influences another

All major factors show P-Values < 0.001 (highly significant)

Powder type, layout, and position DEFINITELY affect scrap rates

Strongest effects (Cramér's V ~0.18-0.19):

Powder × Position interaction
Layout × Powder × Position interaction

Position matters on 11X11TA plates but NOT on 6X6 plates

This confirms our heatmap findings from earlier slides

10 of 10

Final Conclusion

Overall, the data indicates that the optimal build conditions occur using virgin powder and placing parts near the center of the build plate.

These parts consistently show the lowest scrap rates across all layout types, suggesting that material quality and part positioning have the strongest influence on build success.

Management may consider solving build plate location issues by modifying equipment

The optimal manufacturing conditions are clear: use virgin powder and 6x6 plate size without test artifacts. On smaller 6x6 plates, position doesn't significantly impact

results. But on larger 11x11 plates, center positioning becomes critical for quality. Manufacturing may consider adjusting constructing methods to ensure consistent quality for CUPs produced on the edge versus center.

Looking at the chart, the best configurations all use virgin powder—achieving as low as 3.9%

scrap rate on 6x6 layouts. Compare that to recycled powder on large plates with edge

positioning at 27.5%—a seven-fold difference in quality outcomes.

This all leads to our takeaway: given production options: stick to 6x6 plates without test artifacts, ensure virgin powder is used across the board. This will result in the lowest percentage of scrap parts produced.

Thank you!