IronPeak Mining Group · Mining

Safety & Fatigue Intelligence: Protect the Workforce, Predictively

Maturity
L4
Domain
Protect & Disclose
Analytics
predictive
High-risk crew-shift index
~24%
Sponsor
Chief HSE Officer
Confidence
High

The situation

Which crews, shifts and sites carry rising safety and fatigue risk — early enough to intervene — and which factors (fatigue, experience, crew mix, supervision, contractor integration) drive the risk for which population?

Protect & DiscloseL4Sponsor · Tom Bradshaw

The recommendation on the table

Stand up leading safety and fatigue intelligence — explainable, crew/site-level, prevention-only

Intervention before incidents, prioritised by risk, and a safer, more available workforce.

Decision ownerChief HSE Officer · Tom Bradshaw
MaturityL4
Priorityhigh

Trade-offDemands strict governance, glass-box models and worker consultation; trust must be earned to keep reporting honest.

The evidence

IronPeak's workforce is the population physically at risk, and its safety performance is driven by workforce factors it could see coming: fatigue, inexperience, poor crew composition, supervision gaps, roster position and contractor integration. Today IronPeak sees safety only after incidents and manages fatigue with blunt rules. Mining's signature project and most distinctive contribution to the portfolio, MIN-03 makes safety and fatigue leading indicators, predicting where incident risk is rising in time to intervene. It found night-shift contractor crews at roughly double the safety-risk score, fatigue exposure above 20% at remote sites, and near-miss reporting lowest exactly where modelled risk is highest.

Safety & Fatigue Intelligence

Predict and explain crew/shift/site safety and fatigue risk — for prevention and protection, never surveillance.

High-risk crew-shift index· crew-shifts at elevated, driver-attributed safety risk
~24%+15vs target
Critical
Fatigue exposure· workers exceeding bio-mathematical fatigue thresholds (remote site)
~23%+10vs target
Critical
Near-miss reporting gap· reporting lowest on highest-modelled-risk crews
inverse
On watch
Contractor safety integration· contractor crew-shifts fully integrated into safety picture
~78%+95vs target
On watch
Illustrative preview
Safety risk: crew-type × driver
FatigueExperienceCrew mixSupervision
Night contractor
3
3
2
3
Night permanent
3
2
2
2
Day contractor
2
2
2
2
Day permanent
1
1
1
1
Underground
2
2
1
2

Key takeawayNight-shift contractor crews carry roughly double the system safety-risk score — acted on at crew level, never individual.

Crew-shifts by safety-risk tier
100%TOTAL
  • Low52%52%
  • Moderate24%24%
  • High24%24%

Key takeaway~24% of crew-shifts sit at elevated risk — a leading indicator, used for prevention only.

Safety-risk driver attribution (share of high-risk)
012.52537.550FatigueInexperienceCrew mixSupervision gap

Key takeawayFatigue and inexperience dominate — schedulable, relievable drivers.

Interactive view is best explored on desktop.

Key findings

Night-shift contractor crews carry roughly double the system safety-risk score, driven by fatigue and integration gaps, and the most remote site shows ~23% fatigue exposure — all visible in advance from workforce factors.

What we can’t claim

The crews with the highest modelled safety risk report the fewest near-misses: a reporting-culture gap, not low risk. The uncomfortable truth is that this data only works if it is never used punitively — the moment a fatigue or safety score is used against an individual, the reporting and honesty the models depend on collapse. Prevention-only use is both the ethical path and the operational one.

Recommendations

Stand up leading safety and fatigue intelligence — explainable, crew/site-level, prevention-only

high priority

Intervention before incidents, prioritised by risk, and a safer, more available workforce.

Trade-off

Demands strict governance, glass-box models and worker consultation; trust must be earned to keep reporting honest.

Make rostering fatigue-aware and integrate contractor crews fully into the safety picture

high priority

Lower fatigue-driven incident risk, recovered availability, and the largest safety blind spot closed.

Trade-off

Buffer and induction cost up front; fatigue-aware rules constrain scheduling flexibility.

Analytical framework

How we reached this

Predictive, glass-box safety and fatigue intelligence — predict and explain crew/shift/site risk early enough to intervene, for prevention and protection, never punishment.

ConfidenceMedium

Methods applied

Safety-risk driver modellingBio-mathematical fatigue modellingIncident-precursor analysisCrew-composition risk analysisContractor-safety analysisResilience scoring

Statistical techniques

Logistic regressionDriver/feature attributionRecurrent-event/exposure analysisSegmentationCorrelation

Algorithms

Interpretable logistic risk modelsExplainable driver-attributed safety scoringBio-mathematical fatigue models (sleep/wake, circadian, cumulative)

Data sources

Rosters/time-and-attendance (fatigue primitives)Experience/competencyCrew compositionSupervisionNear-miss/hazard/stand-down reportsContractor induction/ticket data

Outputs generated

Driver-attributed safety-risk and fatigue-exposure scores (crew/shift/site)Incident-precursor trendCrew-risk concentrationContractor-safety integrationResilience

Why this confidence

Safety and fatigue rest on partly behavioural and reported signals (near-misses, reporting culture) noisier than MIN-02's roster data; outputs are ranges, framed as risk to relieve, with explicit ethical and measurement caveats. Excludes biometric surveillance and individual punitive use.

The reasoning

Business context

Mining's signature theme, sponsored by the Chief HSE Officer because safety is IronPeak's licence to operate and its deepest duty of care. It owns safety risk, fatigue exposure, incident precursors, crew risk, contractor safety and mining wellbeing. Its outputs exist for prevention, support and protection — never surveillance or punitive action — a hard design constraint.

Expected value

Avoided incidents and their human, legal, production and social-licence costs (a single serious event is catastrophic), plus the availability recovered by scheduling fatigue down. Fatigue is the hinge: it raises incident risk and drains availability, so it pays back twice when scheduled down.

Workforce landscape

Night-shift contractor crews carry roughly double the system safety-risk score; the most remote site shows ~23% fatigue exposure from long commute plus night rotation; near-miss reporting is lowest on the highest-modelled-risk crews (a reporting-culture gap); short-mobilisation contractor crews are only ~78% integrated into the safety picture.

The analytics journey

Level 4, predictive, and deliberately glass-box given the stakes. Predicts crew/shift/site safety and fatigue risk with driver attribution and explicit ethical caveats; outputs are framed as risk to relieve at the crew/site level, never verdicts on individuals, and are used only for prevention and protection.

Under the hood

Interpretable logistic risk models and explainable, driver-attributed scoring identify rising crew/shift/site safety risk; bio-mathematical fatigue models (sleep/wake, circadian, cumulative) compute fatigue from roster/work primitives — not biometric surveillance. No opaque ML on safety or fatigue; every score is explainable, crew/site-level, and prevention-bound, under strict access governance.

Confidence & evidence

Why you can rely on this

82%
Analysis confidenceHigh

The inconvenient truth

The crews with the highest modelled safety risk report the fewest near-misses: a reporting-culture gap, not low risk. The uncomfortable truth is that this data only works if it is never used punitively — the moment a fatigue or safety score is used against an individual, the reporting and honesty the models depend on collapse. Prevention-only use is both the ethical path and the operational one.

Method

Confidence is a deterministic read of KPI strength, target and benchmark coverage across this project — shown on an illustrative reference dataset, computed the same way it would be on live data.

Take this further

Where this project connects