The Underground Mine Planning Problem No One Talks About: How Agent-based Simulation Supports Decision-Making.
Ask any Technical Services Manager what keeps them up at night and the answer is rarely the geology. It is the schedule. Specifically, the gap between what the schedule says will happen and what happens when crews go underground.
The plan looked defensible on paper. The fleet numbers added up. The advance rates were reasonable. Then shift three happened, a truck sat idle waiting for a loader, congestion built on the decline, and by the end of the week the target was behind. The General Manager wants to know why. The Board wants to know what the recovery looks like. And the Planning team is back at the whiteboard, with the same tools they used last time.
This is not a failure of planning effort. It is a limitation of the tools most planning teams rely on.
Why Deterministic Mine Planning Falls Short Underground.
Spreadsheets and deterministic planning models work by calculating a single outcome from a set of inputs. Fleet size, shift length, advance rate assumptions, equipment availability. We feed in the numbers, get back a schedule. It is fast, auditable, and familiar.
The problem is that the output is an average. It tells you what happens if everything performs at its assumed rate simultaneously. Underground operations do not work that way. Trucks queue. Loaders wait at faces that are not ready. Two machines arrive at the same intersection from different directions. A re-entry delay on one heading pushes the sequence on three others. These are not edge cases, this is normal operating behaviour of a complex underground mining ecosystem with competing demands on shared infrastructure.
A deterministic model cannot predict any of this. It assumes equipment moves independently, at modelled rates, without interacting. The result is a schedule that looks achievable but often proves difficult to execute consistently underground. Not because the Planners got the numbers wrong, but because the model is structurally blind to the variation that drives real-world outcomes.
For Technical Services Managers, this creates a specific problem: how do you make a defensible recommendation when the model you are using cannot show you the risk profile of the decisions you are making?
What Spreadsheets Cannot Give to a Technical Services Manager.
As a Technical Services Manager, you sit at a specific cross-departmental intersection. On one side: the Planning Engineers and Schedulers need technical rigour, workflow integration, and outputs that reflect how the mine actually behaves. On the other side: the General Manager and Executive Leadership need recommendations they can act on and defend.
The single-point estimate does not serve either audience well. For the Planning team, it produces a schedule they know is fragile but cannot prove is fragile. For the General Manager, it provides false confidence. A number without a range gives no honest picture of what the downside looks like.
Effective planning software needs to serve both. That starts with replacing the single-point estimate with a range of probable outcomes.
For underground operations running tightly constrained weekly schedules, understanding how equipment interactions affect shift-by-shift execution is often more valuable than a single theoretical production target.
How Probabilistic Modelling Works for Underground Mining.
When evaluating planning tools, Technical Services Managers should be looking for tools that provide a range of probable outcomes from P10 to P90. P10 represents a credible downside outcome, while P90 represents a more optimistic outcome. The gap between them is the risk profile of the decision, and that gap is where capital decisions, fleet sizing calls, and development timeline commitments either hold or fail.
How Agent-based Simulation Differs from Discrete Event Simulation.
Agent-based simulation works differently from discrete event simulation. In a discrete event model, equipment moves through a sequence of pre-defined events. Interactions between machines are typically simplified or pre-defined rather than dynamically emerging from the simulation.
idoba.sim runs on an agent-based modelling engine. Every piece of equipment operates as an independent agent with its own rules, constraints, and decision logic. The behaviour that emerges from those interactions, such as congestion and queuing are not inputs to the model. It is a product of it. That distinction matters when you are trying to understand not just what the outcome was, but why.
How idoba.sim Models Equipment Interactions Underground.
Trucks decide when to load, where to queue, and how to navigate the decline. Loaders operate within configured priorities, availability, and operational constraints. Jumbos cycle through drill, blast, re-entry, and ground support sequences at each heading. These agents interact with each other and with the mine environment in real time, producing outcomes that cannot be predicted from the inputs alone.
Congestion on the decline is not modelled as an assumption; it emerges from actual truck and loader movements in the simulated environment. Queuing at a face is not an input parameter, it happens because two trucks arrived at the same time and only one can load. Shift change effects, re-entry delays, and equipment interactions are all products of the simulation, not inputs to it.
For a Technical Services Manager, this matters for one specific reason: the model provides insight into where operational variability is being generated, not just that a schedule carries risk. When the P10 outcome is materially worse than the P50, idoba.sim surfaces the risk profile of the plan through the P10/P50/P90 distribution.
The P50 median run is available for detailed review, giving the planning team a clear view of how equipment interactions and congestion affected the most likely outcome. That makes the recommendation far more credible than a single-point estimate alone.
Strengthening Capital Decision Recommendations with Probabilistic Outputs.
Capital decisions in underground mining are almost always development dependent. A new fleet purchase, a schedule commitment to the board, a feasibility study with an NPV assumption baked in. All of these depend on a development timeline that is, in most cases, supported by a single-point estimate.
Every week a development schedule slips it pushes cash flow to the right and reduces NPV. If schedule tests are consistently showing they are in the optimistic range, this could result in a multi-week delay to a 2-year mining plan once operations start executing. That is not an abstract risk, it is a quantifiable impact on the value of the project. Being able to demonstrate this means you will have a fundamentally different conversation with the General Manager and the Board than presenting your schedule with no range.
idoba.sim lets planning teams run multiple scenarios against actual mine geometry and fleet configuration. Run multiple scenarios with different heading sequences, fleet sizes, and operational constraints to produce P10/P50/P90 outputs for each. The recommendation that comes out of that process is not a best guess. It is a defensible position, grounded in an agent-based model that reflects how your specific mine behaves.
See idoba.sim Underground Mining Simulation Software in Action.
Book a discovery call. We will walk through your mine, your fleet, and how you can increase confidence in your schedule.
Frequently Asked Questions.
-
Agent-based simulation models each piece of equipment, trucks, loaders, jumbos, spraymecs, as an independent agent operating under its own rules, constraints, and decision logic. Rather than calculating a single average outcome from fixed inputs, the simulation runs those agents through your actual mine geometry, letting behaviour emerge from their interactions in real time.
In practice, this means congestion on the decline isn't an assumption entered into the model, it's a result that occurs because two trucks happened to converge on the same intersection at the same time. Queuing at a face happens because a loader was occupied. Re-entry delays ripple through the heading sequence the same way they do on shift. The model doesn't predict these events; it produces them.
For mine planning, the value is twofold. You get a realistic distribution of outcomes rather than a single-point estimate, and you can see where in the operation variation is being generated, not just that a schedule slipped, but why.
-
In a discrete event simulation (DES), equipment moves through a sequence of pre-defined events — load, haul, dump, return. Interactions between machines are assumed or approximated; the model follows a script. It's useful for high-level throughput analysis, but the equipment doesn't actually respond to each other or to the environment.
In an agent-based model, each machine makes its own decisions in real time. A truck decides when to load based on loader availability and queue position. A loader operates to a defined priority sequence, moving between faces based on configured rules and availability. A jumbo cycles through drill, blast, re-entry, and ground support at each heading on its own schedule. None of this is scripted — it's the product of agents interacting with each other and with the mine geometry simultaneously.
The practical difference: DES tells you what happens if the system behaves as designed. Agent-based simulation shows you what actually happens when it doesn't. Interference, congestion, and cascading delays are outputs of the model, not inputs to it. That distinction is what makes the risk profile operationally meaningful rather than theoretical.
-
P10, P50, and P90 are percentile values drawn from a distribution of simulated outcomes. Running multiple simulation iterations, each with slightly different equipment interactions, timing variations, and operating conditions, produces a range of possible results rather than a single number.
P10 represents a credible downside outcome, with only 10% of simulation runs performing worse. P50 represents the median outcome across all simulation runs. P90 represents a more optimistic outcome, achieved or exceeded by 90% of simulation runs.
The gap between P10 and P90 is the risk profile of the decision. A narrow band means the schedule is relatively robust. A wide band, where the P10 outcome is noticeably behind the P50, means the underlying plan carries material schedule risk that a single-point estimate would hide entirely. For capital decisions, feasibility studies, and board-level commitments, that range is the honest answer.
-
idoba.sim is designed to work alongside your existing planning tools, not replace them. The schedules, heading sequences, and fleet configurations your team already produces in mine planning software or spreadsheet-based workflows serve as direct inputs to the simulation. You're not rebuilding your planning process. You're adding a validation and risk-quantification layer on top of it.
The typical workflow for your Planning Engineers is to produce a schedule using your existing planning tools, the weekly schedule (14-shifts) is imported into idoba.sim against your mine geometry and fleet configuration. The simulation runs and produces a P10/P50/P90 distribution for that plan. The output goes back to the planning team as a risk-graded version.
-
A working simulation model for a new mine can typically be configured and producing initial outputs within days, not weeks. The inputs required include; mine geometry, heading sequences, fleet configuration, shift structures, and equipment productivity assumptions.
Once the mine geometry, heading sequences, fleet parameters, and shift structures are entered, idoba.sim can produce P10/P50/P90 outputs within operational planning timeframes, depending on scenario complexity and simulation scope. View multiple scenario comparisons side-by-side including different heading sequences, fleet sizes, and operational constraints.
For Technical Services Managers evaluating the tool, the discovery call and workshop process is structured specifically around your mine and your fleet. The first session produces outputs against a real planning problem, not a demonstration dataset. You leave with a result that is directly applicable to a decision you are currently making.