Manufacturing Optimization for Perry’s Ice Cream

Manufacturing Analytics Operations Analytics

I analyzed a year of production data to find where a plant was losing capacity, then built a model to simulate how much faster each line could safely run.

Python Excel Regression Time Series Analysis K-Means Clustering XGBoost Simulation

01 The Problem

Perry’s Ice Cream runs seven production lines, each producing dozens of different products, and nobody had a clear picture of where the plant was losing capacity. Years of run logs and checkweight data existed, but only as raw, disconnected records. Leadership wanted one answer: how can we produce more ice cream with the same equipment, people, and hours?

02 The Approach

1

Core Question

Framed the whole project around one question: capacity-constrained, or losing output to fixable problems?

2

Data Cleaning

Merged run logs and checkweight records across all 7 lines.

3

True Speed Model

Built a regression model isolating each line’s real mechanical speed from downtime noise (91% accuracy).

4

Time Series Check

Converted logs into hourly data to catch unexplained speed drops.

5

Product Clustering

Grouped 74 products into 4 clusters by speed, uptime, and weight.

6

Speed Simulation

Simulated throughput at different speeds to find each line’s true optimal point.

03 The Result

Headroom Found

All 7 lines ran below optimal speed, by 16% to 31%

Every line had unused speed capacity waiting to be unlocked.

Output Gain

Simulated speed fixes lifted output from 34.3M to 41.7M units

Almost no added downtime in the simulation.

Real Bottleneck

Downtime, not slow speed, was the actual constraint

The “run faster” story wasn’t the full picture.

Fixable Issues

Line 1200 startup loss & Line 1201 midweek dip

A slow startup on Line 1200 (6.7% loss) and an unexplained Tuesday/Friday slowdown on Line 1201.

Weight, Not Machines

Heavier products explained most speed differences

Not the lines themselves — product weight drove the gaps.

04 Business Impact

Targeted Fixes

Recommended line-specific speed changes instead of a blanket increase.

SOP Audit

Flagged Line 1200’s startup process.

Investigate Pattern

Flagged the recurring Line 1201 slowdown.

Maintenance Leads

Gave exact timestamps of speed drops on Line 900.

Scheduling Tool

Delivered the product-cluster map in Excel for supervisors.

05 What I Learned

I expected “faster machines” to be the answer. The data showed the real opportunity was cutting idle time, not raw speed.

← more projects

Perry’s · Operations Analytics