Mine fuel reconciliation: every litre from supplier tanker to machine, and the tolerances that decide what a variance means

In the Mirage Metrics method, mine fuel reconciliation is a volume balance run per tank and per period, then per machine. For each bulk tank and fuel truck: opening stock, plus supplier deliveries and transfers in, minus litres dispensed, gives a book closing stock; the gap with the measured closing stock is the variance. Every volume goes on one basis first, 15 °C: dockets usually are, pump meters usually are not. Then compare the variance with what the instruments alone can explain: 0.5% of the litres through each class 0.5 meter, plus the gauge limit times the tank's liquid surface at each reading. Inside that band, the variance goes on the trend; outside it, a person checks timing, measurement and data before anyone talks about loss. Last, compare what each machine received with its engine hours and its own history: that is where swapped tags and missing fills show. Mirage runs this balance in production (fuel management system).

Published
Updated
Reading time
17 min
Written by
Jules Toussaint

Scope: diesel on a mine or quarry site with bulk tanks, fixed pumps, fuel trucks and machines. Standards are cited for the limits they publish; whether an instrument is legally verified depends on national law. The worked month is invented. Not metrology, tax or legal advice.

In production, at a mining freight operator

Transwin, a Moroccan logistics operator that moves mining, fertilizer, steel and agri-food freight with 550 specialized trucks and more than 100 machines, runs Mirage's fuel system across fuel deliveries, on-site tanks, pumps and vehicle transactions. It reconciles delivered against received volumes, tracks stock movements and helps the team investigate anomalies by site and by vehicle, and it is being extended to more sites and mobile tanks. Transwin reports 12% lower fleet diesel consumption; its case study gives that figure for the whole programme, not for reconciliation alone. Transwin is a logistics operator, not a mine: Mirage publishes no deployment at a mine operator's own fuel bay.

Where fuel moves on a mine site

Draw the path before building any report. Every place fuel is stored has its own balance, and every transfer between two places is a measurement that can be wrong. A site-wide total hides which one is off; a balance per tank shows it.

Fuel path on a mine site: supplier tanker, bulk tank, fixed pump, haul trucks, fuel truck, ancillaries and telematics, feeding one balance per tank and per machineSupplier tankermeter, class 0.5docket at 15 °CBulk tankgauge 4 mm, or dipcalibration tableT1Fixed pumpmeter, class 0.5tag reads the IDP1Haul trucksexcavatorsfill per machine IDFuel truckown meter, dipmobile tankFT-2Ancillariesdozers, graders,generatorsTelematicshours, fuel usedBalance per tank, then per machineevery volume at 15 °C, one cut-off, one perioddocketsreadingsfillsInside the band: on the trendOutside: a person investigates
Each box is a measurement. The balance reads dockets, tank readings and fills, all at 15 °C, and telematics hours for the machines. Inside the band goes on the trend; outside goes to a person.

A common layout on an open-pit site: bulk tanks filled by road tanker, fixed pumps that fill haul trucks and loading tools at a fuel bay, and fuel trucks that carry diesel to machines that do not travel to the bay (dozers, graders, drills, lighting towers, generators). The fuel truck is a tank with its own balance: what it loaded, what its meter dispensed, its dips. Telematics sits beside the path: it measures no litres moved, it reports what the engine says it burned and how long it ran.

What each measurement is allowed to be wrong by

Every number in a fuel balance comes from an instrument with a published limit. Added together, the limits give the widest variance measurement alone can explain.

Measurement points on a mine fuel path: what each measures, the limit a standard or regulator publishes, that limit in litres, and what breaks it in practice. Compiled by Mirage Metrics, October 2026.
PointWhat it measuresPublished limitIn litresWhat breaks it in practice
Supplier deliveryLitres off the road tanker, usually on the docket at 15 °CClass 0.5: 0.5% for the measuring system (OIML R 117-1, Tables 2 and 3; same in MI-005, Tables 2 and 5)±150 L on a 30,000 L deliveryA docket keyed to the wrong tank or day; no gauge reading before and after
Bulk tank, automatic gaugeLevel, turned into litres by the tank's calibration table4 mm once installed, 1 mm before (OIML R 85, 6.2.2)4 mm × liquid surface: 144 L on a 3 m × 12 m horizontal tank near half full, 113 L on a 6 m vertical tankThe calibration table itself; readings taken while fuel moves
Bulk tank, manual dipLevel read on a dipstick or tape, then the chartUS underground tanks: equipment able to read to 1/8 inch over the full height (40 CFR 280.43(a)(2))3.2 mm × 36 m² = 114 L on the same horizontal tankSloping ground, the wrong chart, a dip taken during a transfer
Calibration table (chart)Litres for each millimetre of levelBuilt by ISO 7507-1 (vertical) or ISO 12917-1 (horizontal, tilt corrected)No fixed figure: the error grows when the tank settles or is repairedA chart older than the last repair or relocation
Fixed pumpLitres per fill, tagged to a machineClass 0.5: 0.5% system, 0.3% meter (OIML R 117-1); US tanks: 6 in³ per 5 gallons, 0.52% (280.43(a)(5))±606 L on 121,240 L in a monthAn uncompensated meter at 30 °C counts 1.3% more litres than the same fuel at 15 °C
Fuel truck (mobile tank)Litres loaded at the pump, litres dispensed by its own meter, its dipClass 0.5 for measuring systems on road tankers (OIML R 117-1, Table 2)±166 L on 33,200 L dispensedVerified less often than fixed pumps; more manual entries
Machine ID at the fillWhich machine received the litresNo metrology standard: a site controlNot a volume error, an attribution errorTags swapped, misread or overridden by hand
TelematicsEngine hours, fuel used, idle hours, fuel level, payloadISO/TS 15143-3 defines the data elements and how they are exchanged, not their accuracyWhatever the engine controller estimatesUnits offline; a replaced unit restarting its counters
TemperatureDiesel expands as it warmsFactor to 15 °C: 0.9873 at 30 °C, 1.0084 at 5 °C (Measurement Canada, table 54B)30,000 L at 30 °C is 29,619 L at 15 °C: 381 L of no fuelComparing a meter at fuel temperature with a docket at 15 °C
Measurement points on a mine fuel path: what each measures, the limit a standard or regulator publishes, that limit in litres, and what breaks it in practice. Compiled by Mirage Metrics, October 2026.

Source: OIML R 117-1:2019, OIML R 85:2008, Directive 2014/32/EU, MI-005, 40 CFR 280.43, Measurement Canada diesel table, ISO/TS 15143-3. Litres computed by Mirage on the example quantities.

Which of these limits actually binds your instruments

A class is a limit for an instrument verified to it, not a property of every pump. In the European Union, the Measuring Instruments Directive leaves it to each Member State to prescribe measuring instruments for tasks such as fair trading and levying taxes (Article 3). The supplier's road tanker meter sells you fuel, so it is the instrument most likely to be under legal control. A fixed pump that fills your own machines sells nothing; whether national law requires it to be verified is a question for your country's metrology rules, and many site pumps are held only to whatever maintenance the site gives them. Take the class as the best case, and the meter's last proving as the real one.

Temperature: the variance that is only physics

Diesel expands as it warms. Measurement Canada's diesel table, built on the API Chapter 11.1 method for a density of 840 kg/m³ at 15 °C, gives a factor of 0.9958 at 20 °C, 0.9873 at 30 °C and 0.9830 at 35 °C, and says to multiply an uncompensated meter reading by the factor for the liquid's average measured temperature during the delivery (Measurement Canada). In the EU, the Measuring Instruments Directive refers to the Member States' obligation to require 15 °C under the energy taxation directive (MI-005, note under 1.4), one reason dockets so often say 15 °C. On a site where diesel sits at 30 °C, a meter that does not compensate counts 1.3% more litres than the docket would for the same fuel. Over a month of 120,000 litres that is about 1,500 litres, as much as the whole instrument band of the worked month below. Correct first, compare second.

The one regulator figure for a tank balance

No standard sets a tolerance for a mine's tank balance. The closest published figure is the US EPA's rule for underground storage tanks: monthly inventory control must be able to detect a release of at least 1.0% of flow-through plus 130 gallons, with inputs, withdrawals and stock recorded each operating day, levels read to the nearest one-eighth of an inch, deliveries reconciled by gauging before and after, dispensing metered within local meter standards or 6 cubic inches per 5 gallons, and water measured monthly (40 CFR 280.43(a)). An above-ground tank is outside that rule. Its value here is the method: a daily record, a gauge before and after every delivery, and a monthly threshold that grows with throughput.

The balance, tank by tank

The arithmetic is one line per tank and per period, the same for a 100,000-litre bulk tank and a fuel truck:

  • Book closing = opening measured stock + supplier deliveries + transfers in − litres dispensed, every term at 15 °C.
  • Variance = measured closing stock − book closing. Negative means less fuel than the records account for.
  • Instrument band = deliveries × the delivery meter's limit + transfers in × the pump's limit + litres dispensed × the dispensing meter's limit + 2 × (gauge limit × liquid surface).

The band is the widest gap measurement alone could produce if every instrument sat at its limit in the same direction. Inside it, a variance cannot be called a loss on that period alone; outside it, it cannot be put down to instrument accuracy. Two rules: every source closes at the same cut-off time, or a delivery lands on one side and its fills on the other; and the period is long enough for throughput to outweigh reading error, which is why the EPA's threshold is monthly. Daily runs spot events (a delivery with no rise, a fall overnight); the monthly figure is the one to trend.

One month on one tank, one fuel truck and six machines

Illustrative example, invented figures: bulk tank T1, fuel truck FT-2 and six machines, one month, litres

Assumptions, all invented. T1 is a horizontal tank, 3 m across and 12 m long, so its liquid surface near half full is 36 m² and a 4 mm gauge limit is 144 litres a reading. Its gauge reports volume at 15 °C; dockets are at 15 °C. Pump P1 and FT-2's meter do not compensate, and the fuel averaged 30 °C all month, factor 0.9873. P1 filled three haul trucks and loaded FT-2 with 34,400 litres; FT-2 served an excavator, a dozer and a grader.

Monthly balance for bulk tank T1 and fuel truck FT-2, in litres at 15 °C (invented). Compiled by Mirage Metrics, October 2026.
LineBulk tank T1Fuel truck FT-2
Opening stock41,200 (gauge, at 15 °C)1,481 (dip 1,500 at 30 °C)
Received120,000 (4 dockets at 15 °C)33,963 (34,400 from P1 at 30 °C)
Dispensed121,240 (P1 meter 122,800 at 30 °C)32,778 (own meter 33,200 at 30 °C)
Book closing stock39,9602,666
Measured closing stock38,900 (gauge)2,271 (dip 2,300 at 30 °C)
Variance−1,060 (0.87% of dispensed)−395 (1.2% of dispensed)
Band the instruments explain1,494 (600 + 606 + 2 × 144)334 (170 + 164)
US EPA monthly reference (underground tanks)1,705 (1% of 121,240 + 130 gallons)not applicable
ResultInside the band: on the trendOutside: a person checks
Monthly balance for bulk tank T1 and fuel truck FT-2, in litres at 15 °C (invented). Compiled by Mirage Metrics, October 2026.

Run without temperature, T1's balance reads 41,200 + 120,000 − 122,800 = 38,400 litres against 38,900 measured: a gain of 500 litres, which would close the subject. Corrected, P1 dispensed 121,240 litres at 15 °C and the tank is 1,060 litres short. That is inside the 1,494 litres the instruments can explain and inside the EPA's reference, so the tank is not investigated this month; the figure goes on the trend, and a second month short by the same share would be read differently. FT-2 is 395 litres short against a band of 334: outside, so its meter's last proving and its dips come first.

Fuel by machine for the same month: litres at the meter against engine hours and each machine's own 12-week median (invented). Compiled by Mirage Metrics, October 2026.
MachineReceived (L)Engine hoursOwn median (L/h)Expected (L)DifferenceSite band 5%
HT-07 haul truck31,90022613229,832+2,068 (+6.9%)Outside 5%
HT-11 haul truck30,40023113130,261+139 (+0.5%)Inside
HT-14 haul truck26,10022813330,324−4,224 (−13.9%)Outside 5%
EX-02 excavator18,6002059018,450+150 (+0.8%)Inside
D-03 dozer9,700not available41not computablenot computableData gap
GR-01 grader4,900188264,888+12 (+0.2%)Inside
Fuel by machine for the same month: litres at the meter against engine hours and each machine's own 12-week median (invented). Compiled by Mirage Metrics, October 2026.

Read the machine table as a set. The tank balance hid what it shows: HT-07 received 2,068 litres more than its hours explain while HT-14, at the same pump, received 4,224 fewer. The first hypothesis is attribution, not theft: HT-14 fuelled under HT-07's tag, or fuelled by FT-2 without a transaction, which would also fit FT-2's shortfall. P1's transaction times against both trucks' positions settle the first; FT-2's log the second. D-03's telematics was offline, so it cannot be judged at all this month, and the table says so.

The balance above as a spreadsheet: deliveries, dips, dispensing by machine, tank balance and fuel by machine per period, with temperature correction from Measurement Canada's diesel table. Only the standards' limits are filled in; every volume, hour and band is yours.

Litres per hour, litres per tonne, and what each hides

Once the tanks balance, two ratios carry the machine side, each against the machine's own past.

Litres per engine hour

Litres received in the period divided by engine hours in the same period, against the machine's own median over the previous weeks. A catalogue figure is no baseline: Caterpillar's handbook gives a Cat 793F haul truck from 32.08 to 51.8 US gallons an hour, about 121 to 196 litres, between its low and high estimates, with values drawn from its VisionLink telematics and assuming 50% and 10% idle time at the two ends (Caterpillar Performance Handbook, fuel consumption, page 15). Two identical trucks on different haul roads sit far apart within that range. Telematics supplies the hours: ISO/TS 15143-3, the mixed-fleet telematics standard, defines cumulative operating hours, cumulative idle hours and cumulative fuel used among its data elements (ISO/TS 15143-3), so the share of hours spent idling can sit next to the ratio. It defines how the values are exchanged, not how accurate the fuel figure is: an engine's own fuel estimate is a second opinion on the meter, not a replacement for it.

Litres per tonne hauled

Litres received by the haul fleet divided by the tonnes it moved, from the weighbridge or the trucks' payload records (the same standard carries cumulative load count and payload totals). It ties fuel to production, which litres per hour cannot. It moves with haul distance, grade and road condition, so a rising figure after the pit deepens is not a fuel problem. The tonnes need their own reconciliation: the quarry weighbridge ticket reconciliation guide applies the same balance to tonnes.

The two indicators of the measurement itself

  • Share of litres with a valid machine ID. Below full, part of the fuel cannot be attributed, and machine-level figures are lower bounds.
  • Share of alerts closed as measurement or data. A high share is useful: it says which meter, gauge or tag reader to fix, so the same alert does not return.

Measurement error, leak or theft: read the signature

Most variances are measurement and data. They have shapes, and so do leaks and theft. Name the shape before naming a cause.

Causes of a fuel variance on a mine site, the signature each leaves in the data, and the first check (recommended practice; regulatory references where stated). Compiled by Mirage Metrics, October 2026.
CauseSignatureFirst check
Temperature basisFollows the fuel temperature: a gain in one season, a loss in the other, proportional to throughputPut every volume at 15 °C before comparing
Cut-off timingA variance that reverses the next periodOne cut-off time for every source
Fills not keyed inTank short, and machines short by about the same litresThe paper log against the system
Meter driftThe same sign and roughly the same percentage of throughput, every periodProve the meter with a test measure, then adjust or replace
Gauge or chart errorA variance that depends on how full the tank isGauge against manual dips at several levels, as OIML R 85 checks an installed gauge
Short deliveryTank rise below the docket on particular deliveriesGauge before and after every delivery (40 CFR 280.43(a)(3))
LeakLevel falls while nothing is dispensedA still period: readings at the start and end with nothing added or removed (280.43(b)(1))
Tag swap or misreadOne machine over, another under, same pump, same hoursTransaction times against each machine's position
TheftDiscrete events: a fill larger than the machine's tank, a fill while the machine was elsewhere or off shift, a delivery with no tank riseThe evidence first, then the site's own investigation procedure
Causes of a fuel variance on a mine site, the signature each leaves in the data, and the first check (recommended practice; regulatory references where stated). Compiled by Mirage Metrics, October 2026.

Source: Gauging before and after a delivery and the still-period test: 40 CFR 280.43(a)(3) and (b)(1); checking an installed gauge against a manual reference level: OIML R 85, 6.2.1.

Keep four words apart in reports and meetings. A variance is the gap before any explanation. An explained variance is the part accounted for by temperature, timing, instrument limits and data errors found and corrected. An unexplained loss is what remains, traced to a tank, a fuel truck, a machine or a shift. Suspected misuse is a hypothesis that an investigation tests with evidence: transaction times, machine positions, tank capacity, camera records. A false theft alert costs the credibility of the next, real one.

The records serve twice. In Australia, the tax office asks for records of the type, date and quantity of fuel acquired and how it was used, lists fuel issue records and GPS or telematics data among the examples, and wants them kept five years (ATO, fuel tax credit business records). A reconciled record per tank and per machine is that evidence; check the current rules for your own claims.

What software prepares, and what a person decides

Collecting dockets, readings and fills, putting them at 15 °C on one cut-off, computing the book stock and the band, attributing fills to machines and flagging what falls outside can all run without a person. Four things should not:

  • Setting the bands: the instrument band from the meters' provings, the machine band from judgement about each machine's duty.
  • Explaining an alert, with a cause from a fixed list (temperature, timing, meter or gauge, tag or data, leak, unexplained); a measurement cause opens a maintenance task.
  • Opening an investigation: only for an unexplained loss traced to a place and a time, by a named manager.
  • Any conclusion about a person, which belongs to the site's own procedure and its evidence.

When not to automate this, and what to do first

Automation adds nothing, or adds false confidence, in four cases:

  • One tank, a few machines, a monthly dip: the workbook above is enough, and a person reads it in an hour.
  • No machine ID at the pump: fills cannot be attributed, so the machine side is blind. Fit tags or a mandatory ID entry first.
  • Meters never proved and a calibration table older than the last repair: the band is unknown, and every alert is noise. Prove the meters and check the gauge against dips first.
  • The fleet fills at public stations on fuel cards: there is no site tank to balance, and the job is card reconciliation, covered in the fuel card fraud post linked below.

Automation earns its place with several tanks or sites, fuel trucks, thousands of fills a month and telematics on most machines: more records than a person can line up by hand.

Where Mirage fits

Mirage builds the system that runs this balance on the site's own pumps, tank probes, dockets, transactions, telematics and ERP, per tank per day and per machine, and routes what falls outside to the people who decide: the fuel management system page describes it, the Transwin case study documents it in production. A proof of concept usually runs on one month of transaction logs, dips and dockets from one fuel bay.

Questions readers ask

  • A volume balance for every tank and every period: opening stock plus deliveries and transfers in, minus fuel dispensed, gives the book closing stock, which is compared with the measured closing stock. The difference is the variance. It is then attributed to machines by comparing what each one received with what its engine hours and its own history say it should have used.

  • There is no single figure. The widest variance the instruments alone can explain is the sum of their limits: 0.5% of the litres through each class 0.5 meter under OIML R 117-1, plus the gauge limit times the tank's liquid surface for the opening and closing readings (4 mm each for an installed automatic gauge under OIML R 85). The one regulator figure for a tank balance is the US EPA's monthly inventory control for underground tanks: a method able to detect 1.0% of flow-through plus 130 gallons.

  • Usually temperature or timing. Dockets are often at 15 °C while a pump meter counts litres at the fuel's temperature: at 30 °C, Measurement Canada's diesel table gives a factor of 0.9873, so 30,000 litres at 30 °C are 29,619 litres at 15 °C. Put every volume on the same basis, then gauge the tank before and after each delivery to see a short delivery the day it happens.

  • By the signature. Measurement errors are proportional to throughput and keep the same sign period after period; a leak shows as a falling level when nothing is dispensed; theft shows as discrete events tied to a time, a place and a fill, such as a fill larger than the machine's tank or a fill when the machine was elsewhere. A conclusion about a person belongs to the site's investigation procedure, not to a report.

  • Its own. Caterpillar's handbook gives a Cat 793F from 32.08 to 51.8 US gallons an hour, about 121 to 196 litres, between its low and high estimates, which assume 50% and 10% idle time. Compare each machine with its own median litres per engine hour over the previous weeks, and read litres per tonne hauled next to it, since haul distance and grade move both.

Sources and scope

Checked on . Measurement standards are cited for the limits and methods they publish; whether an instrument must be legally verified depends on national law. The EPA rule applies to underground storage tanks in the United States and is cited as a reference point. Litres in the tables are computed by Mirage on example quantities; the worked month is invented.

This guide describes an operational method. It is not metrology, tax or legal advice, and any investigation involving people should follow your own procedures.

Related on this site

Rebuild a past month on your own records

Send us a month of delivery dockets, tank readings, pump transactions and engine hours for one fuel bay. We return the balance per tank and per machine, at 15 °C, with every variance traced to its records, so your team can judge it against what it already knows.

We use your email to reply to you and, now and then, to tell you about Mirage. You can opt out at any time. Privacy policy