Cascadia Medical Devices

Predictive maintenance and OEE analytics on a 5.9M-row manufacturing dataset.

Manufacturing Analytics

Cascadia Medical Devices

Proving that a production-grade BI architecture scales from raw sensor data to boardroom OEE dashboards — in a single, reproducible build.

Overview

A full analytics stack for a fictional 3-line medical-device manufacturer, built on ~5.9M synthetic production events across 29 months, paired with real NASA C-MAPSS turbofan run-to-failure sensor data for predictive maintenance. It demonstrates dimensional modeling at scale, a medallion pipeline, and root-cause operational analytics.

Business Problem

Manufacturing leaders couldn’t see why on-time delivery was slipping on certain product lines. The data existed but was fragmented across systems with no single, trusted view.

Architecture

This build is the portfolio’s full enterprise-BI stack — the one Cascadia domain that runs end to end on Microsoft Fabric:

Cascadia end-to-end data platform

Source (SQL Server): CascadiaMES — star schema with 4 fact tables, 8 dimensions, dim_date, ~5.9M rows, plus a separate schema for NASA C-MAPSS sensor data. Loaded reproducibly via PowerShell; DP-600-aligned.

Pipeline (Microsoft Fabric): Bronze → Silver → Gold medallion lakehouse, implemented in PySpark notebooks.

Semantic model + report (Power BI): Direct Lake connection; DAX measures for OEE, MTBF, defect rate, and on-time-in-full delivery.

Data Sources

Source Type Rows Notes
CascadiaMES (simulated) SQL Server ~5.9M Seed-controlled deterministic generator; 4 fact tables, 8 dimensions
NASA C-MAPSS Public (NASA) 265K cycles Turbofan Engine Degradation dataset (FD001–FD004); cited and disclosed

Headline Skill: Root-Cause Analytics & Predictive Maintenance

The dataset carries an engineered, discoverable story: a single station’s cycle time drifts up (≈58→69s), dragging First Pass Yield down on one line (≈95.5%→90.2%) and on-time-in-full delivery down on the affected product (≈92%→69%) — while control lines stay flat. The reporting isolates the root cause rather than just showing the symptom.

The NASA C-MAPSS layer adds a parallel story: remaining useful life (RUL) modeling on turbofan engine sensor data, surfaced as a reliability/maintenance planning view in the same report.

The Report

The manufacturing analytics report — Power BI. OEE, FPY, and OTIF headline cards; availability / performance / OEE by production line; FPY by month and station (the declining station drives the root-cause story); a downtime Pareto by reason code; and OTIF by customer.

The Power BI report surfaces the OEE and root-cause yield story alongside the NASA C-MAPSS reliability / remaining-useful-life view.

Opportunities for Future Enhancements

  • A recorded 3-minute walkthrough of the root-cause story.
  • A Direct Lake performance writeup on the 5.9M-row model.

Tech Stack

SQL Server 2025 T-SQL Python (simdata) Microsoft Fabric PySpark Power BI DAX PowerShell