Manufacturing IoT — Predictive Maintenance & OEE Analytics¶

Source: examples/manufacturing-iot/README.md -- this page is rendered live from that file.

Manufacturing IoT — Predictive Maintenance & OEE Analytics¶

Examples > Manufacturing IoT

[!TIP] TL;DR — End-to-end manufacturing analytics platform that ingests sensor telemetry from OPC-UA and MQTT devices, computes Overall Equipment Effectiveness (OEE), and predicts equipment failures before they occur. Raw readings flow through a medallion architecture into a Lakehouse where ML models score failure probability and recommend optimized maintenance windows.

Table of Contents¶

• Overview
• Architecture Overview
• Prerequisites
• Quick Start
• Data Pipeline
• Sample Analytics Scenarios
• Data Products
• Related Resources

Overview¶

This vertical demonstrates a production-grade IoT analytics platform built on Azure Cloud Scale Analytics (CSA). It models a discrete manufacturing environment where CNC machines, presses, conveyors, and packaging lines emit continuous sensor telemetry. The platform computes OEE metrics in near-real-time, detects anomalous vibration and temperature patterns, and generates ML-scored failure predictions that feed optimized maintenance schedules.

All data in this example is synthetic and intended for demonstration purposes only.

Key Features¶

• Sensor Telemetry Ingestion: High-throughput ingestion of temperature, vibration, pressure, and humidity readings via IoT Hub and Event Hubs
• OEE Calculation: Automated computation of Availability, Performance, and Quality factors at the equipment and production-line level
• Predictive Maintenance: ML-based failure probability scoring using rolling sensor statistics and maintenance history
• Anomaly Detection: Vibration and temperature anomaly flagging using statistical thresholds and z-score analysis
• Maintenance Optimization: Schedule recommendations that balance predicted failure risk against production impact
• Equipment Lifecycle Tracking: Age-based risk profiling and replacement planning

Data Sources¶

Architecture Overview¶

flowchart TB
    subgraph Sources["Sensor & Operational Data"]
        S1[OPC-UA Sensors]
        S2[MQTT Sensors]
        S3[Equipment Registry]
        S4[Maintenance Logs]
        S5[Production Schedules]
    end

    subgraph Ingestion["Ingestion Layer"]
        I1[Azure IoT Hub]
        I2[Azure Event Hubs]
    end

    subgraph Bronze["Bronze Layer - Raw Ingestion"]
        B1[Raw Sensor Telemetry]
        B2[Raw Equipment Master]
        B3[Raw Maintenance Logs]
    end

    subgraph Silver["Silver Layer - Cleansed & Enriched"]
        SV1[Fact: Sensor Readings]
        SV2[Dim: Equipment]
        SV3[Fact: Maintenance Events]
    end

    subgraph Gold["Gold Layer - Analytics & Reporting"]
        G1[OEE Metrics Report]
        G2[Failure Predictions]
        G3[Maintenance Schedule]
    end

    subgraph ML["ML Pipeline"]
        ML1[Feature Store]
        ML2[Failure Prediction Model]
    end

    subgraph Consumers["Consumers"]
        C1[Power BI Dashboards]
        C2[Plant Floor Displays]
        C3[CMMS Integration]
        C4[Alert Notifications]
    end

    S1 --> I1
    S2 --> I1
    I1 --> I2
    S3 --> B2
    S4 --> B3
    S5 --> B3
    I2 --> B1

    B1 --> SV1
    B2 --> SV2
    B3 --> SV3

    SV1 --> G1
    SV1 --> ML1
    SV2 --> G1
    SV2 --> G2
    SV3 --> G1
    SV3 --> G3

    ML1 --> ML2
    ML2 --> G2

    G1 --> C1
    G1 --> C2
    G2 --> C3
    G2 --> C4
    G3 --> C3

Business Drivers¶

Supports ISO 55000 asset management, OSHA process safety, FDA 21 CFR Part 11 electronic records, and ISA-95 equipment hierarchy standards through auditable sensor data lineage and anomaly detection.

Prerequisites¶

Azure Resources¶

• Azure Subscription with IoT Hub provisioned
• Azure Event Hubs namespace (Standard tier or higher)
• Azure Data Lake Storage Gen2 (for medallion layers)
• Azure Databricks Workspace
• Azure Machine Learning Workspace (for failure prediction model)

Tools Required¶

• Azure CLI >= 2.50
• dbt-core >= 1.7 with dbt-databricks adapter
• Python >= 3.10
• Databricks CLI

Permissions¶

• IoT Hub Data Contributor on the IoT Hub resource
• Azure Event Hubs Data Receiver on the Event Hubs namespace
• Storage Blob Data Contributor on ADLS Gen2

Quick Start¶

1. Clone and Navigate¶

git clone https://github.com/your-org/csa-inabox.git
cd csa-inabox/examples/manufacturing-iot

2. Upload Sample Data¶

az storage blob upload-batch \
  --destination bronze/sensor-telemetry \
  --source data/ \
  --account-name csadatalakedev

3. Run dbt Models¶

cd domains
dbt seed --profiles-dir .
dbt run --profiles-dir .
dbt test --profiles-dir .

4. Explore Results¶

Open Power BI or Databricks SQL to query the Gold layer tables: rpt_oee_metrics, rpt_failure_predictions, and rpt_maintenance_schedule.

Data Pipeline¶

Bronze Layer - Raw Ingestion¶

Raw data lands in ADLS Gen2 with no transformations. Each source retains its original schema.

Silver Layer - Cleansed & Enriched¶

Validated, deduplicated, and enriched with rolling statistics and equipment context.

Gold Layer - Analytics & Reporting¶

Aggregated, business-ready views for dashboards and operational systems.

Sample Analytics Scenarios¶

1. Vibration Anomaly Detection¶

Identify equipment exhibiting abnormal vibration patterns that may indicate bearing wear or misalignment.

SELECT
    equipment_id,
    sensor_type,
    reading_value,
    rolling_avg_24h,
    z_score
FROM silver.fct_sensor_readings
WHERE sensor_type = 'vibration_mm_s'
  AND ABS(z_score) > 3.0
  AND reading_timestamp >= DATEADD(DAY, -7, CURRENT_DATE())
ORDER BY ABS(z_score) DESC

2. OEE Calculation¶

Calculate Overall Equipment Effectiveness broken down by Availability, Performance, and Quality for each production line.

SELECT
    equipment_id,
    report_date,
    availability_pct,
    performance_pct,
    quality_pct,
    oee_pct,
    CASE
        WHEN oee_pct >= 85 THEN 'World Class'
        WHEN oee_pct >= 60 THEN 'Acceptable'
        ELSE 'Needs Improvement'
    END AS oee_tier
FROM gold.rpt_oee_metrics
WHERE report_date >= DATEADD(DAY, -30, CURRENT_DATE())
ORDER BY oee_pct ASC

3. Maintenance Scheduling¶

Query the optimized maintenance schedule to find upcoming high-priority work.

SELECT
    equipment_id,
    equipment_type,
    failure_probability,
    risk_tier,
    recommended_date,
    estimated_downtime_hours
FROM gold.rpt_maintenance_schedule
WHERE risk_tier IN ('Critical', 'High')
ORDER BY failure_probability DESC

Data Products¶

Data Contract¶

The contracts/equipment-health.yaml data contract defines the schema, quality thresholds, and SLAs for the core equipment health data product. Key guarantees: scores updated hourly, OEE/failure probability within 0-1 range, and 99.9% data product availability.

Directory Structure¶

manufacturing-iot/
├── README.md                          # This file
├── contracts/
│   └── equipment-health.yaml          # CSA data product contract
├── data/
│   └── sample_telemetry.csv           # 20 rows of synthetic sensor data
├── deploy/
│   └── params.dev.json                # Deployment parameters (dev)
└── domains/
    ├── _sources.yml                   # dbt source definitions
    ├── bronze/
    │   ├── stg_sensor_telemetry.sql   # Raw sensor staging
    │   ├── stg_equipment.sql          # Equipment master staging
    │   └── stg_maintenance_logs.sql   # Maintenance history staging
    ├── silver/
    │   ├── fct_sensor_readings.sql    # Enriched sensor facts with anomaly flags
    │   ├── dim_equipment.sql          # Equipment dimension with risk tiers
    │   └── fct_maintenance_events.sql # Maintenance event facts
    └── gold/
        ├── rpt_oee_metrics.sql        # OEE per equipment per day
        ├── rpt_failure_predictions.sql # 7-day failure probability scores
        └── rpt_maintenance_schedule.sql # Recommended maintenance schedule

Related Resources¶

• Azure IoT Hub Documentation -- Device connectivity and telemetry ingestion
• Azure Event Hubs Documentation -- High-throughput event streaming
• OPC-UA Foundation -- Industrial interoperability standard
• Overall Equipment Effectiveness (OEE) -- OEE calculation reference
• Azure Machine Learning -- ML model training and deployment