When Simulation Beats Hands-On, and When It Does Not

Simulation is usually justified on cost — cheaper than vehicles, tooling and workshop time. That argument is weak, because a poor simulation is expensive and produces nothing.

The stronger case is capability: simulation does several things a training vehicle cannot, and it fails at several things a vehicle does trivially.

Where simulation genuinely wins

Faults you cannot reliably produce. The intermittent condition, the seasonal fault, the failure that occurs at one hundred thousand kilometres. A training vehicle sits in known condition; a simulation can present the fault on demand, repeatedly, to every technician.

This is the strongest case and it is the one most often under-used.

Dangerous conditions. High-voltage fault states, thermal events, anything you would not create deliberately. See retraining for electric vehicles.

Repetition. A technician can work through the same diagnosis twenty times. A vehicle presents it once and then it is fixed.

Scale. One simulation reaches a network; one training vehicle reaches one workshop at a time.

Making reasoning visible. A simulation can record which tests the technician chose and in what order, which is diagnostic capability itself rather than the answer. No vehicle-based exercise captures this, and it is what makes simulation valuable for assessment. See assessments that predict performance.

Safe failure. A wrong decision costs nothing.

And rare vehicles, before the model has reached the network.

Where it fails

Physical skill. Feel, torque, access, working blind, knowing when something is seized rather than tight. No simulation delivers this, and claims otherwise should be treated sceptically.

Judgement about condition. Whether a component is worn enough to replace is learned by handling many of them.

The environment. Working under a vehicle, in awkward positions, with limited access, at temperature. A substantial part of the difficulty of the job is physical and simulation removes exactly that.

Anything the simulation authors did not anticipate. A simulation has defined states. Real vehicles present combinations nobody designed for, and the transfer problem is that technicians learn the simulation's decision space rather than the general skill.

Fidelity failures that teach wrong things. A simplified system model that behaves differently from the real one teaches an incorrect mental model, and that is worse than no training.

The middle option that is frequently the answer

Simulation for diagnosis, vehicle for execution.

Diagnostic reasoning — which test, in what order, what the reading means, what to conclude — transfers well from simulation and is the expensive part to teach on vehicles.

Physical execution is taught at the vehicle, where it is cheap in courseware terms and irreplaceable.

This split is usually better than choosing one, and it costs less than a high-fidelity simulation attempting both.

Evaluating a simulation

Does it record the path, not just the answer? If it only marks the final conclusion correct or incorrect, it is a test with graphics.

Is the underlying model correct? A simulation that behaves in a way the real system does not is teaching a false mental model. Have a technical specialist verify the behaviour, not just the interface.

Can faults be varied? A simulation with one fault is a demonstration.

Does it present realistic data? Clean, unambiguous readings train for a world that does not exist. Noise, marginal values and misleading readings are the actual difficulty.

Does it allow wrong paths? A simulation that prevents an incorrect test removes the learning.

What does it cost to update when the product changes. A simulation built as a monolith becomes obsolete and unaffordable to maintain. See updating courseware.

The cost question, honestly

Simulation is expensive to build and cheap to distribute. Training vehicles are the reverse.

The economics favour simulation where the network is large, the fault is hard to reproduce, or the content is stable enough to amortise.

They favour vehicles where the network is small, the skill is physical, or the model changes frequently.

And the maintenance cost is the one that decides it. A simulation that must be rebuilt with each model year rarely pays back. One covering principles that persist across generations can pay back many times.

What it does not replace

A technician who has passed a simulated diagnosis has demonstrated reasoning under simulated conditions. That is genuinely valuable and it is not the same as having performed the repair.

Combine it with workplace assessment for anything where physical execution matters, and record both. See training records and compliance.

The short version

The case for simulation is capability, not cost — faults you cannot produce, repetition, safe failure, and recording the diagnostic path.

It fails at physical skill, condition judgement and the working environment, and claims otherwise deserve scepticism.

Simulation for diagnosis, vehicle for execution is usually the right split.

Check that it records the path, allows wrong turns and presents ambiguous data — otherwise it is a test with graphics.

And check the maintenance cost, because that is what determines whether it pays back.

For independent research and standards work on simulation, consult NIST modeling and simulation resources.