IoT Worked Examples: Review Activities

IoT Worked Examples: Review Activities

animation
iot
applications
TipAda, Applications and calculation-audit guideCheckpoint: Placement Strategy

You now know:

  • Flood warning uses 34 sensors across the catchment because upstream rain and river data creates the 6-10 hour lead time. For “- The weather example reaches 97.3% frost alert accuracy across 12,000 km^2 with 21 full s”: identify 97.3% frost detection as a simulated assumption pending held-out observations across zones and seasons; preserve the worked arithmetic separately.
  • Across all five examples, critical placement beats uniform density: city-wide intersections, high-population pollution corridors, upstream gauges, and terrain-aware weather sites all matter more than sensor count alone.
TipAda, Applications and calculation-audit guideCheckpoint: Build Your Own Case

You now know:

  • The calculator section turns the same framework into adjustable inputs: coverage area, terrain, sensor tiers, CapEx, OpEx, adoption rate, and deployment months.
  • The parking exercise shows a complete row-by-row case: 800 sensors, 12 gateways, installation labor, connectivity, cloud platform, replacement, time savings, fuel savings, and payback.
  • A 17-day theoretical parking payback is still a prompt to check coverage, adoption, rollout timing, and OpEx before presenting the number as a budget promise.

Interactive Quiz: Match Concepts

Interactive Quiz: Sequence the Steps

Label the Diagram

Code Challenge

Apply IoT Patterns

When building your own IoT business case:

  1. Start with the benefit, not the technology. What measurable outcome justifies the investment?
  2. Identify the critical sensing locations through domain expertise, not uniform grids.
  3. Design a tiered sensor strategy matching sensor quality to location criticality.
  4. Calculate 5-year TCO including OpEx, not just CapEx.
  5. For “Discount benefits by realistic adoption rates – typically 15-30% for public-facing sy”: discount modeled benefits using a project-specific adoption forecast and sensitivity range, not a universal 10–30% or 15–30% rate.
  6. Model the deployment timeline and compute payback from project start, not from full deployment.

Sensor Density Misconception

“More sensors always produce better data quality.”

This is one of the most expensive assumptions in IoT network design. The worked examples in this chapter demonstrate the opposite: strategic sensor placement matters more than sensor count.

Real Case Study: Municipal Air Quality Network Failure

A mid-size city (population 600,000) deployed 200 identical air quality sensors in a uniform 1km grid across the city, believing “maximum coverage” would provide the best data.

The Deployment:

  • 200 sensors × $800 each = $160,000 hardware
  • Installation: $40,000
  • 5-year connectivity: $60,000
  • Total investment: $260,000

What Went Wrong:

After 18 months, an independent audit revealed:

  1. For “Industrial hotspots undersampled: The city’s 3 industrial zones (12% of area, 65% of p”: rename the “Real Case Study” as a fictional municipal network exercise and label zone, sensor, drift and utilization values as assumed; otherwise provide a named municipal audit. Result: pollution peaks went undetected because sensors were too far from emission sources.

  2. Residential areas oversampled: 88 sensors in low-variability residential zones generated nearly identical readings.

  3. For “Calibration impossible: All 200 sensors were the same low-cost model ($800)”: rename the “Real Case Study” as a fictional municipal network exercise and label zone, sensor, drift and utilization values as assumed; otherwise provide a named municipal audit. For “With no reference-grade sensors for calibration, measurements drifted 15-30% over 12 month”: rename the “Real Case Study” as a fictional municipal network exercise and label zone, sensor, drift and utilization values as assumed; otherwise provide a named municipal audit.

  4. Missed critical locations: Schools, hospitals, and highways (where vulnerable populations concentrate) received no special instrumentation.

The Numbers:

  • Sensors deployed: 200 For “- Sensors actually useful: ~60 (30% utilization)”: rename the “Real Case Study” as a fictional municipal network exercise and label zone, sensor, drift and utilization values as assumed; otherwise provide a named municipal audit.
  • Critical locations missed: 18 schools, 4 hospitals, 8 highway corridors For “- Measurement accuracy after 12 months: ±25% (vs”: keep these before/after air-quality-network numbers only as fictional exercise inputs, or provide a named city audit and method. ±10% target)
  • Policy decisions made with bad data: 3 (later reversed)
  • Reputational damage: City council lost confidence in the entire smart city program

What They Should Have Done:

Apply the tiered sensor strategy from the Generic Air Quality example:

  • Tier 1 reference: 8 sensors x $15,000 = $120K for regulatory compliance and calibration anchors at industrial boundaries.
  • Tier 2 mid-grade: 30 sensors x $3,000 = $90K for schools, hospitals, highways, and industrial fence-lines.
  • Tier 3 low-cost: 50 sensors x $800 = $40K for residential spatial coverage.
  • Total: 88 sensors for $250K, using the same budget for a stronger design.

Why This Works:

  1. Reference sensors (Tier 1) provide the calibration truth that makes low-cost sensors reliable
  2. Critical locations (Tier 2) get accuracy where it matters (schools, hospitals, industrial boundaries)
  3. Spatial coverage (Tier 3) fills gaps at low cost, calibrated against nearby Tier 1/2 sensors
  4. For “Sensor count reduced from 200 to 88 (56% fewer), but data quality improved dramati”: keep these before/after air-quality-network numbers only as fictional exercise inputs, or provide a named city audit and method.

Corrected 3-Year Outcomes:

  • Total sensors: Reduced from 200 to 88.
  • Hardware cost: Increased from $160K to $250K because the corrected design includes reference anchors. For “- Useful sensors: Improved from about 60 useful sensors (30%) to all 88 sensors”: keep these before/after air-quality-network numbers only as fictional exercise inputs, or provide a named city audit and method. For “- Calibration accuracy: Improved from about ±25% drift to about ±8% maintained accurac”: keep these before/after air-quality-network numbers only as fictional exercise inputs, or provide a named city audit and method. For “- Critical location coverage: Improved from 0% to 100%”: keep these before/after air-quality-network numbers only as fictional exercise inputs, or provide a named city audit and method.
  • Policy decisions supported: Improved from 3 reversed decisions to 14 validated decisions.
  • Cost per useful sensor: Rose slightly from $2,667 to $2,841, but the data became decision-grade.
  • Data quality rating: Improved from 2.1/5.0 to 4.4/5.0.

The Lesson:

Sensor density is necessary but not sufficient. The three design principles that matter more:

  1. Location criticality: Place expensive, accurate sensors where decisions get made (schools, industrial boundaries, hospitals)
  2. Low-cost sensor networks may need reference monitors for calibration; set the number and placement from measurement uncertainty and regulatory needs.
  3. Domain expertise: Meteorologists, urban planners, and public health experts know where sensors belong better than uniform grid algorithms

Before Your Next Deployment:

  • DON’T: Buy N sensors and distribute uniformly
  • DO: Identify 10-20 critical locations first, then fill gaps with lower tiers
  • DON’T: Assume identical sensors across a network
  • DO: Design a tiered strategy (5-10% reference, 30-40% mid-grade, 50-60% low-cost)
  • DON’T: Deploy first, calibrate later
  • DO: Install reference sensors in Month 1, calibrate low-cost sensors before relying on their data

Test Your Understanding:

If you have a $200K budget for an IoT sensor network, which approach is better?

    1. 250 sensors at $800 each
    1. 5 sensors at $15K + 30 sensors at $3K + 50 sensors at $800 = 85 sensors total

The answer depends on your domain, but B is almost always superior because it provides the calibration anchors and location prioritization that A completely lacks. Fewer sensors, thoughtfully placed and tiered, outperform many sensors uniformly distributed.

Back to the chapter