🎓 Lesson 19 D5

KPI Dashboard Implementation: From Raw Data to Executive Alerts

A KPI dashboard is a live, visual summary of key performance numbers—like haul truck availability or blast fragmentation quality—that helps mine managers spot problems and make fast, data-driven decisions.

🎯 Learning Objectives

  • Design a KPI hierarchy for blast-to-crusher logistics using SMART criteria
  • Calculate real-time alert sensitivity (false positive rate) given historical downtime and sensor accuracy specs
  • Analyze raw SCADA/ERP data streams to identify and correct three common data quality gaps (e.g., timestamp misalignment, unit conversion errors)
  • Apply ISO 22400-2:2021 guidelines to classify KPIs by governance tier (strategic, tactical, operational)
  • Explain how alert fatigue arises from poorly calibrated thresholds—and redesign one alert rule to reduce false positives by ≥40%

📖 Why This Matters

In modern open-pit mines, delays in detecting underperforming haul trucks or suboptimal crusher feed gradation can cost $250K–$1.2M per day in lost throughput. A well-designed KPI dashboard doesn’t just display data—it acts as the central nervous system of the logistics chain, turning raw sensor feeds into executive alerts that trigger precise interventions. Without it, teams react to symptoms; with it, they prevent failures before they cascade.

📘 Core Principles

KPI dashboards rest on three interdependent pillars: (1) Data Provenance—ensuring traceability from source (e.g., GPS loggers, weighbridges, blast vibration monitors) through ETL pipelines; (2) Metric Governance—defining each KPI’s calculation logic, update frequency, ownership, and escalation protocol per ISO 22400-2; and (3) Alert Architecture—distinguishing between diagnostic alerts (e.g., 'Truck #T782 idle >18 min'), predictive alerts (e.g., 'Crusher liner wear trend exceeds 95% confidence bound'), and prescriptive alerts (e.g., 'Reschedule maintenance for Crusher C1 within 4 hrs'). Success requires co-design with operations—not IT alone—and validation against field truth (e.g., timed manual audits).

📐 Alert Sensitivity & False Positive Rate

The false positive rate (FPR) quantifies alert reliability and directly impacts operator trust. It is calculated from historical alert triggers versus verified incidents. Minimizing FPR without sacrificing detection sensitivity requires balancing threshold placement against natural process variability.

False Positive Rate (FPR)

FPR = (N_alerts − N_true_positives) / N_alerts

Measures reliability of alerting system; critical for sustaining operator trust and preventing alert fatigue.

Variables:
SymbolNameUnitDescription
N_alerts Total alert triggers count Number of times the alert condition was met in evaluation period
N_true_positives Verified true events count Number of alerts confirmed via field audit or secondary sensor as valid incidents
Typical Ranges:
Operational alerts (e.g., equipment temp): 5% – 15%
Predictive alerts (e.g., bearing failure forecast): 10% – 25%

💡 Worked Example

Problem: Over 30 days, the 'Blast Fragmentation Index (BFI) < 65' alert triggered 47 times. Field verification confirmed only 12 were true underperformance events (e.g., oversize >15% at primary crusher). The remaining 35 were due to sensor drift or calibration lag.
1. Step 1: Identify true positives (TP) = 12, total alerts = 47
2. Step 2: Compute FPR = (Total Alerts − TP) / Total Alerts = (47 − 12) / 47
3. Step 3: Calculate: 35 / 47 ≈ 0.745 → 74.5% FPR — unacceptably high (target ≤15%)
Answer: The result is 74.5%, which far exceeds the safe limit of ≤15%. To meet target, threshold must be tightened (e.g., BFI < 60) or sensor recalibration scheduled weekly.

🏗️ Real-World Application

At Rio Tinto’s Pilbara Operations (2022–2023), implementation of a unified KPI dashboard reduced blast-to-crusher cycle time variance by 32% by correlating real-time GPS haul truck positions, crusher feed size distribution (via laser scanner), and blast muck pile density (from drone LiDAR). A critical alert—'Conveyor belt speed drop >12% sustained for >90 sec'—triggered automatic dump point reassignment, preventing 11 unplanned stoppages/month. Root cause analysis revealed that prior to dashboard deployment, 68% of such events were logged >22 minutes after onset—too late for mitigation.

✏️ Data Quality Audit Exercise

You receive a CSV extract from the fleet management system showing 100 haul truck cycles over 24 hours. Columns: 'truck_id', 'load_time_sec', 'haul_time_sec', 'dump_time_sec', 'return_time_sec', 'timestamp_utc'. Your task: (1) Identify 3 data anomalies (e.g., negative durations, duplicate timestamps, inconsistent units); (2) Propose corrective ETL steps; (3) Recalculate average cycle time after cleaning and compare to raw value. Submit annotated output with justification.

📚 References