Fleet Sizing Methodology: Queuing Theory & Bottleneck Analysis
Fleet sizing is figuring out how many trucks, shovels, and conveyors you need to move ore and waste without bottlenecks — like choosing the right number of checkout lanes so no one waits too long.
⚠️ Why It Matters
📘 Definition
Fleet sizing methodology is a systems engineering approach that integrates queuing theory (to model stochastic service times and waiting behavior) and bottleneck analysis (to identify and quantify throughput-limiting resources) to determine the optimal number, type, and configuration of mobile and fixed haulage assets in mining operations. It balances capital expenditure, operational flexibility, and system-level productivity under uncertainty in cycle times, equipment availability, and material variability. The output is a robust, capacity-matched fleet design validated against probabilistic performance targets (e.g., 95th percentile wait time < 2.5 min).
🎨 Concept Diagram
AI-generated illustration for visual understanding
💡 Engineering Insight
Queuing theory alone fails when bottlenecks shift dynamically — e.g., a 'balanced' fleet becomes unstable if a single crusher liner change reduces Q_max by 15%. Always size the fleet *relative to the most constraining, least flexible node*, not the average loader rate. Field validation trumps analytical elegance: a 5-truck fleet with 92% availability often outperforms an 8-truck fleet at 78%.
📖 Detailed Explanation
Going deeper, real mines violate M/M/c assumptions: loading times are bounded (never zero, rarely >120 s), haul distances vary, and breakdowns cluster. This demands M/G/c modeling — where 'G' denotes general service time distribution — requiring CV input to compute expected queue length (Lq) and wait time (Wq) using the Pollaczek–Khinchine formula. Critical here is distinguishing *system* CV (aggregate cycle time) from *service* CV (loading time only); conflating them overstates delays.
At the advanced level, fleet sizing integrates with digital twin frameworks: real-time telematics feed live λ and μ into adaptive queuing models that trigger dynamic dispatch rules (e.g., rerouting trucks away from congested dump points) and feed predictive maintenance alerts when CV spikes — indicating incipient mechanical degradation. Bottleneck analysis evolves beyond static throughput limits to include transient constraints: a 10-minute ramp closure during shift change can reduce daily throughput by 8% even if average Q_max is unchallenged.
🔄 Engineering Workflow
📋 Decision Guide
| Rock/Field Condition | Recommended Design Action |
|---|---|
| Surface mine with steep haul roads (>12% grade) and mixed ore/waste haulage | Apply M/G/c queuing with CV = 0.28; cap fleet at 85% of theoretical max to absorb grade-induced speed variance; verify bottleneck at crusher feed conveyor. |
| Underground block cave with high CV (≥0.5) due to drawpoint variability and ventilation restrictions | Use simulation-based fleet sizing (not analytical queuing); allocate 20% buffer trucks for dynamic reassignment; prioritize bottleneck mitigation at muck pile access ramps. |
| Greenfield open-pit with >90% equipment availability and GPS-guided autonomous trucks | Use M/M/c steady-state model with μ = 22 trucks/hour; validate with 72-hr discrete-event simulation; target average queue length ≤ 1.2 trucks. |
📊 Key Properties & Parameters
Cycle Time (CT)
4.5–12.0 minutes (surface), 8.0–22.0 minutes (underground)Total time for a haul truck to complete one full round trip: loading → hauling → dumping → returning → queuing → reloading.
Directly governs maximum theoretical fleet capacity; small errors compound exponentially in queuing models.
Service Rate (μ)
12–24 trucks/hour (hydraulic shovels), 8–16 trucks/hour (front-end loaders)Average number of trucks a loader can serve per hour, calculated as 60 / average loading time (min/truck).
Sets the upper bound on stable queue length; exceeding μ triggers exponential queue growth per Little’s Law.
Coefficient of Variation (CV)
0.15–0.35 (well-maintained surface fleets), 0.40–0.65 (underground with variable grade/ventilation)Standard deviation of cycle time divided by its mean — quantifies stochasticity in haulage operations.
Higher CV increases queue variance disproportionately; M/G/c models require CV input for realistic delay estimation.
Bottleneck Throughput (Q_max)
2,500–15,000 tph (surface), 800–4,200 tph (underground)Maximum sustainable material flow rate (tph) through the slowest subsystem — e.g., crusher feed conveyor, primary dump pocket, or ramp capacity.
Fleet must be sized to *feed*, not exceed, Q_max — oversizing causes congestion upstream and idle time downstream.
Equipment Availability (A)
85–93% (modern surface fleets), 72–84% (deep underground fleets with ventilation constraints)Fraction of scheduled operating time that equipment is operationally ready (excluding planned maintenance).
Reduces effective service rate; ignored availability leads to chronic underperformance even with ‘correct’ nominal fleet size.
📐 Key Formulas
Utilization Factor (ρ)
ρ = λ / (c × μ)Measures traffic intensity — fraction of time servers (loaders) are busy; determines queue stability.
| Symbol | Name | Unit | Description |
|---|---|---|---|
| ρ | Utilization Factor | Measures traffic intensity — fraction of time servers (loaders) are busy; determines queue stability | |
| λ | Arrival Rate | units/time | Average rate at which customers or jobs arrive |
| c | Number of Servers | Number of parallel service channels (e.g., loaders) | |
| μ | Service Rate per Server | units/time | Average rate at which a single server can complete service |
Expected Queue Length (Lq) – M/G/c
Lq = (ρ² / (1 − ρ)) × (1 + CV²) / 2Average number of trucks waiting to be loaded — critical for pit design and safety buffer zones.
| Symbol | Name | Unit | Description |
|---|---|---|---|
| Lq | Expected Queue Length | trucks | Average number of trucks waiting to be loaded — critical for pit design and safety buffer zones |
| ρ | Traffic Intensity | dimensionless | Utilization factor; ratio of total arrival rate to total service capacity (λ / (cμ)) |
| CV | Coefficient of Variation of Service Time | dimensionless | Standard deviation divided by mean of service time distribution |
Effective Throughput Limit
Q_eff = Q_max × A × ηMaximum sustainable tonnes/hour considering bottleneck capacity, equipment availability, and operational efficiency (η).
| Symbol | Name | Unit | Description |
|---|---|---|---|
| Q_eff | Effective Throughput Limit | tonnes/hour | Maximum sustainable throughput considering bottleneck capacity, equipment availability, and operational efficiency |
| Q_max | Maximum Throughput Capacity | tonnes/hour | Theoretical maximum throughput without constraints |
| A | Equipment Availability | dimensionless | Fraction of time equipment is available for operation (0 to 1) |
| η | Operational Efficiency | dimensionless | Fraction representing real-world performance relative to ideal (0 to 1) |
🏭 Engineering Example
Escondida Mine (Chile)
Porphyry copper deposit (altered andesite/diorite)🏗️ Applications
- Autonomous haulage system (AHS) commissioning
- Mine expansion feasibility studies
- Maintenance shutdown impact assessment
- Fuel consumption optimization
📋 Real Project Case
Chilean Copper Mine: Autonomous Haul Fleet Deployment
A Tier-1 copper mine in the Atacama Desert, northern Chile, deployed an autonomous haul fleet across its open-pit operation. The site processes ~450 ktpd of ore and waste, with a 2.8-km average haul distance and 320-m vertical lift. The project involved retrofitting and integrating 42 autonomous 290-tonne CAT 794 AC electric drive haul trucks into existing dispatch and traffic management systems.