Manufacturing Quality Anomaly Detection Dashboard

Manufacturing Data Analytics·2026·Synthetic-data dashboard demo·4 min read

Synthetic manufacturing quality dashboard for yield movement, defect-code concentration, equipment-level failure patterns, false-fail signals, anomaly ranking, and corrective-action validation.

GitHub

Overview

Built a manufacturing quality anomaly detection dashboard using synthetic data to support high-volume electronics manufacturing review. The dashboard combines SPC-style yield monitoring, defect-code Pareto analysis, equipment concentration checks, false-fail pattern review, ML-assisted anomaly ranking, and corrective-action validation so engineering teams can prioritize abnormal patterns for review.

Problem

High-volume manufacturing quality teams need to detect abnormal yield movement early, separate localized equipment issues from product-wide quality risks, and validate whether corrective actions restore expected quality behavior. If yield excursions and defect-code shifts are only reviewed in summary reports, production impact can accumulate before the signal is visible.

Data Used

  • Synthetic daily manufacturing quality records across a 90-day production window
  • Synthetic yield, pass, fail, retest, and retest-pass behavior
  • Synthetic product family, line, station, equipment, lot, and shift dimensions
  • Synthetic defect codes and defect categories
  • Injected anomaly scenarios for yield drop, equipment concentration, false-fail pattern, and corrective-action recovery

Review Scope

  • The project uses synthetic data for portfolio demonstration.
  • No confidential production data, company-specific process details, customer data, supplier information, or real product names are included.
  • ML-assisted anomaly ranking prioritizes abnormal manufacturing patterns for engineering review; it does not automatically determine root cause.
  • No deployed live demo link is included yet.

Approach

Generated synthetic manufacturing data and built an interactive ECharts dashboard that applies 3-sigma SPC control limits, rolling average trend detection, defect-code share comparison, equipment-level failure share analysis, Isolation Forest anomaly ranking, and pre/post corrective-action comparison.

Investigation Focus

  • SPC-style yield monitoring flags days below the lower control limit.
  • Rolling yield movement highlights sustained product quality shifts.
  • Defect-code Pareto comparison identifies concentration changes between baseline and anomaly windows.
  • Equipment concentration analysis separates localized tester or fixture issues from broader quality risks.
  • False-fail pattern review uses elevated retest-pass behavior as an investigation signal.
  • Isolation Forest ranking prioritizes unusual days based on yield, fail rate, retest behavior, defect concentration, equipment concentration, and rolling yield delta.
  • Corrective-action validation compares pre/post yield recovery, defect share reduction, and equipment normalization.

Key Investigation Choices

Use synthetic data to model a realistic quality review workflow.

Reasoning:

Synthetic data makes the dashboard safe for public portfolio use while still demonstrating manufacturing quality logic: yield movement, product quality risk, defect concentration, equipment concentration, false-fail behavior, and corrective-action recovery.

Alternatives considered:
  • Use anonymized real production exports
  • Use generic business dashboard sample data
  • Publish only static screenshots without reproducible analysis logic

Combine statistical signal detection with ML-assisted anomaly ranking.

Reasoning:

SPC limits and rolling averages make yield movement explainable, while Isolation Forest ranking helps prioritize days with multiple concurrent abnormal signals. The ML layer supports engineering triage; it does not decide root cause.

Alternatives considered:
  • Use only KPI cards and trend charts
  • Use only a machine-learning score without explainable quality signals
  • Manually label anomaly days without statistical context

Include corrective-action validation as part of the dashboard flow.

Reasoning:

A quality dashboard is more useful when it follows the issue from detection through review and post-action validation instead of stopping at anomaly discovery.

Alternatives considered:
  • Show only anomaly detection
  • Show only defect Pareto analysis
  • Treat corrective action as a text note outside the dashboard

Methods & Tools

  • Anomaly Detection
  • Defect Pattern Mining
  • Factory Dashboard
  • False-Fail Signal Review
  • Corrective Action Tracking
  • Python Data Pipeline

Result & Impact

  • 90 days
    Simulation window
  • 4
    Anomaly scenarios
  • 100% synthetic
    Data boundary

The demo detects the injected yield excursion, surfaces a concentrated defect-code movement, identifies a suspicious equipment-level failure pattern, ranks unusual production days for engineering review, and summarizes whether corrective actions improved post-action quality trends. The workflow supports failure analysis and corrective-action validation without claiming automated root-cause determination.

Notes

  • Quality monitoring is strongest when statistical detection, defect Pareto, equipment concentration, and retest behavior are reviewed together.
  • ML-assisted anomaly ranking is useful for prioritization, but engineering review is still required to determine root cause.
  • Corrective-action validation should compare post-action behavior against the pre-action anomaly window, not just confirm that an action was logged.
  • Synthetic data can communicate manufacturing quality workflows publicly without exposing real products, customers, equipment, or process details.

Dashboard Screenshot

Manufacturing quality anomaly detection dashboard showing yield movement, defect-code concentration, equipment concentration, ML-assisted anomaly ranking, and corrective-action validation.

Synthetic manufacturing quality dashboard for SPC-style yield monitoring, defect-code and equipment concentration review, anomaly prioritization, and corrective-action validation.

All projects