Flow Computer

What a Flow Computer Is and What Is Being Decided

The definition has teeth. A flow computer is not a box that displays a rate; it is a device whose arithmetic is traceable to published measurement standards, whose configuration is sealed against silent change, and whose records are organized so that a third party can audit what happened months later. It totalizes pulses against a K factor, applies correction equations that turn line conditions into base conditions, applies the meter factor from the last proving, and stores the result in tamper evident logs with a real time clock stamp. That last clause is the commercial heart of the product: custody transfer totals move money, and the device exists so that both parties to the money can trust the number without trusting each other.

What is actually being decided in this purchase is where computation is allowed to live. A modern PLC can sample pulses and run arithmetic fast enough; a DCS can host a function block that approximates the corrections. The flow computer is bought when the calculation must carry certified provenance and an audit seal rather than a programmer's initials: the algorithm implementations are tested and versioned against API and AGA documents, the housing carries tamper switches, the configuration changes are logged events, and the whole package holds a measurement certificate, API 21.1 compliance for electronic gas measurement in North American gas practice, or MID and OIML R117 context for liquid metering in Europe. The RFQ decision is therefore not performance, it is evidentiary standing: whether the totals from this point will be defended in an accounting reconciliation, and to which standard.

The naming deserves one paragraph because it confuses procurement. EFM, electronic flow measurement, is the gas field term for the same device class, and "gas flow computer" and "liquid flow computer" usually differ only in firmware pack and I/O population. The batch controller at a loading rack is a different duty with a family resemblance: it stops a preset quantity into a truck compartment, where the flow computer accounts for what left the rack. The two meet in loading systems, where the batch logic fires valves and the fiscal totals land in the flow computer. A gas chromatograph is not a flow computer either, but it feeds one: heating value from the GC arrives over Modbus and the flow computer multiplies corrected volume by energy content to report energy, the unit many gas contracts actually settle.

The honest accuracy position frames everything that follows. The computation error of a certified flow computer is negligible, 0.01 percent class on totalization, while the primary flow meter contributes 0.2 to 1 percent or more depending on technology and proving discipline. The device is bought to protect and correct a measurement, not to create one; money spent here without money on the meter run and its proving is misspent. What the buyer is really purchasing is the certainty that a 0.5 percent meter stays 0.5 percent after corrections, configuration events, clock stamps, and an audit are applied to it.

The station around the device decides half the order. Inputs arrive as pulses over shielded cable, sometimes through a signal conditioner that squares up turbine pickup coils; analog 4 to 20 mA loops from a pressure transmitter and a temperature transmitter; direct RTD inputs for critical temperature; and serial Modbus RTU or TCP streams from multivariable transmitters and analyzers. Outputs go to SCADA, to a pipeline host, to valve control in skid duty. In hazardous areas the field wiring lands through approved barriers or the whole device sits in an Ex d enclosure. Chapter 4 prices these rows; chapter 6 wires them.

How the Calculation Engine Works

Everything starts at the inputs. Pulse inputs are counted against a K factor, pulses per unit volume, entered per meter run and sealed as a configuration event; typical cards accept 0 to 10 kHz with scalable amplitude thresholds, and serious streams are wired dual channel, two independent counters on the same pickup, so that a lost or extra pulse is detected as a mismatch instead of entering the totals silently. A turbine flowmeter and an oval gear flow meter both arrive this way, as pulses whose unit value is known from proving. A coriolis flowmeter may deliver frequency pulses for mass or volume, or higher resolution values over Modbus. The orifice path is different: a differential pressure transmitter, a static pressure transmitter, and a temperature input feed the AGA 3 equation set directly, and the device integrates flow from those live values rather than from counted pulses.

The liquid path is gross to net. Volume counted at line conditions is corrected to base conditions using API MPMS Chapter 11.1: the temperature correction factor CTL and the pressure correction factor CPL, computed from measured temperature and pressure with the fluid properties in the standard. Multiply gross volume by CTL and CPL and the result is net standard volume, the number a crude oil contract settles. Where a density meter or a coriolis density output exists, mass follows directly; LPG and light hydrocarbon service often settles in mass for exactly this reason. The meter factor from the last proving multiplies in as a sealed correction, and every application of a new factor is an audit event with a timestamp and an author.

The gas path is a stack of AGA documents. AGA 3 computes flow from an orifice run: differential, static pressure, temperature, bore, and plate geometry into mass flow. AGA 7 covers linear meters, turbine and rotary positive displacement, driven by pulse inputs with compressibility applied downstream. AGA 8, the detail characterization method, computes the compressibility factor from gas composition or agreed properties; AGA 10 provides complementary methods for other gas applications; a full composition from a chromatograph lets the device use AGA 8 detail properly instead of a generalized correlation. AGA 9 governs multipath ultrasonic custody metering where an ultrasonic flowmeter is the primary. Energy reporting multiplies corrected volume by heating value, per the contract unit, and AGA 5 frames that energy computation context. The firmware names these algorithm versions in the configuration record; the RFQ should name them too.

Recording discipline is what API 21.1 actually regulates for electronic gas measurement. The device accumulates flow on fixed intervals, hourly and daily records being the settlement norm, logs unaccounted time when inputs are invalid under defined rules, and stores configuration changes, alarms, and events in a sealed audit log. Retention is concrete: API 21.1 expects at least 35 days of onboard configuration and event records, and serious devices hold far more, with battery backed real time clocks, 5 to 10 year battery life typical, keeping timestamps honest across power loss. Totals live in nonvolatile memory so a supply failure costs history of milliseconds, not months. Communications use store and forward buffering: when the SCADA link drops, the device keeps recording and backfills the host on reconnect.

The arithmetic closes with checks the buyer rarely sees until they matter. Square root extraction on differential, low flow cutoffs, pulse validity windows on dual channels, input range and rate alarms, and watchdog supervised execution all sit under the certified firmware. The result is a computation error around 0.01 percent of total, an order of magnitude below any field meter, which is why the metering skid budget belongs on the primary and its proving, and the flow computer budget on features, certifications, and audit depth rather than on chasing decimal places.

Types, Construction, and Duty Boundaries

The catalog divides by stream count, firmware pack, and area rating. Single stream units meter one run and are the tank truck and small skid standard. Multi stream units host 2, 4, 8, or more runs in one chassis, sharing the display, communications, and audit subsystem, which is how a pipeline station meters six meter tubes with one certified device and one clock. Allocation class units trade custody certificates for price and live inside plants on steam, water, and fuel gas balancing. The EFM flavor packages the same gas math in a field housing with solar and battery options for remote wells. Combined liquid and gas firmware exists and suits terminals that load both, at the cost of a longer validation conversation with the metering authority.

Hardware follows metering room practice rather than panel fashion. Power is redundant, dual 24 V DC feeds or AC with DC backup, diode auctioneered, with a watchdog that alarms on either supply failing. Totalizers and configuration live in nonvolatile memory; the real time clock rides a lithium battery specified for 5 to 10 years of retention. Enclosures are DIN rail or rack format in control rooms, or stainless and composite field housings with convection cooling, ambient ranges of roughly minus 40 to 60 deg C for outdoor work. Displays are segment LCD or graphic, and every serious unit keeps a local display because commissioning a meter run without one is misery.

Communication is layered. Modbus RTU over RS-485 remains the field lingua franca to transmitters and chromatographs; Modbus TCP and Ethernet carry the same registers to SCADA faster; digital I/O runs valve commands, proving signals, presets, and alarms; and printer or serial ports survive at loading racks. Where a remote PLC hosts skid logic, the flow computer remains the totalizing authority: the PLC sequences valves, the certified device counts and corrects. On the input side, noisy turbine pickups over long runs go through a signal conditioner before the pulse card, and intrinsically safe field loops land through a safety barrier appropriate to the gas group, or the whole instrument wears Ex d explosion proof certification with sealed cable entries.

The duty boundaries are worth writing down. This is not a batch controller: it accounts, it does not sequence valves to a preset fill, though the two cooperate at loading racks. It is not a gas chromatograph: composition arrives from the analyzer, the flow computer consumes it. It is not a general purpose controller and should not be bought as one; a site that needs PID loops should buy a PLC for the loops and let this device do measurement. And it is not where money should be saved on a custody point: the price difference between an allocation unit and a certified custody unit is small against a single month of disputed volume on a large line.

Key Specification Parameters

The datasheet rows below decide whether a quoted unit can actually serve the metering point. The paragraphs after the tables cover the two rows auditors read first, algorithm identification and audit retention.

Algorithm identification is the row that makes totals defensible. A compliant device does not merely implement AGA 8; it implements a stated implementation of it, and the configuration record identifies the algorithm and version used for every interval. The RFQ should require the vendor to state, per calculation pack, the standard, the implementation version, and the validation evidence, because "AGA 8 compliant" without a version is a phrase, not a specification. The same logic applies to liquid packs: API MPMS Chapter 11.1 has implementation editions, and the two parties to a custody transfer should be running the same one or should agree in the contract which applies.

Audit retention decides what an auditor can reconstruct. The requirement to write is straightforward: sealed configuration and event logging with tamper detection, at least 35 days of onboard records per API 21.1 with a preference for devices holding months, battery backed clock through outages, store and forward buffering on communication loss, and an export format the host or an audit tool can read without the vendor present. The table below compresses the calculation standards into the form a specification section can carry.

Close the specification with the document list: algorithm versions per pack, type certificates for the area rating, the factory calibration and firmware checksum procedure, and the retention depth actually fitted. A quotation that cannot name the algorithm implementation versions is answering a different question than the one the RFQ asked.

Selection Decision Factors and RFQ Language

The decision sequence runs as a checklist, and each step eliminates a class of quotations before price comparison begins.

1. State the duty honestly: custody, allocation, or check metering, and name the governing contract standard the totals must satisfy.
2. Count the streams and the meter technologies on them, then size the chassis and the pulse, analog, and serial cards from that list.
3. Pick the calculation packs: API MPMS 11.1 and 21.1 for liquid and gas settlement, AGA 3, 7, 8, 10 as the gas metering requires.
4. Define the audit scope: sealed events, retention depth, export format, and who is allowed to change configuration under which procedure.
5. Fix the proving interface: prover signals, meter factor entry as a sealed event, and the proving cadence the contract expects.
6. Resolve the area classification: Ex d enclosure or Ex ia loops with a safety barrier per gas group, and name the certificate.
7. Plan integration: Modbus registers for the host, digital I/O to the skid PLC, chromatograph and analyzer links, and store and forward depth.
8. Require the documents: algorithm versions, type certificates, firmware checksum practice, and battery replacement schedule.

Against neighboring computation platforms the honest map runs as follows, and it belongs here rather than in the page title. A PLC can count pulses and apply corrections with good code, and sites do it for internal flows; what the general purpose PLC does not carry is certified algorithm provenance, a sealed audit trail, and measurement certification, which is exactly why fiscal points do not use one. A DCS hosts the supervisory layer beautifully and remains the wrong home for the certified totals. The batch controller is the mirror image at a rack: it lands a preset quantity into a compartment and hands the fiscal accounting to the flow computer that shares the same pulses. A gas chromatograph feeds the energy computation rather than performing it. The flow computer occupies the narrow ground where arithmetic becomes legally defensible measurement.

Meter pairing drives the I/O list, and the pairings repeat across the industry. A turbine flowmeter delivers clean pulses at 0 to 10 kHz class and wants dual channel counting plus periodic proving. An oval gear flow meter does the same for LPG and viscous products at lower rates. A coriolis flowmeter hands over mass directly, by pulse or Modbus, and makes the density question disappear. An orifice plate flowmeter run with a differential pressure transmitter, a pressure transmitter, and a temperature transmitter is the classic AGA 3 station, fully computed in the device. An electromagnetic flowmeter feeds pulse or frequency for water and conductive product loading. An ultrasonic flowmeter, in multipath custody form under AGA 9, reports over serial with its own diagnostics. A vortex flowmeter covers steam and gas allocation duty from pulse inputs. None of these pairings constrains the vendor list; all of them constrain the card count.

The small end of the flow world deserves one honest sentence: a gas mass flow controller meters and controls small gas doses in laboratories and pilot plants, but it is a dosing instrument and never a custody device; a gas mass flow controller belongs at the bench, while totals that move money come from the meter and computer pairing above. Where the flow meter population is mixed and growing, buy the chassis with spare card slots; retrofitting streams into a full unit costs a second device.

The application table turns the pairing logic into the form a project file carries.

The RFQ block should be written once and reused: number of streams and duty class; meter technologies with pulse, analog, and serial counts; calculation packs with implementation versions; audit scope with retention depth and export format; proving interface and meter factor procedure; area classification with certificate basis; power redundancy; communications to host and skid including a safety barrier schedule where applicable; environment; and the document list. A quotation returned without algorithm versions or retention numbers is incomplete at any price.

Installation, Standards, Failure Modes, and Maintenance

Signal wiring decides the first year of reliability. Pulse and analog cables are shielded twisted pair, grounded at one end, routed away from power circuits, and kept within the vendor length tables; turbine pickups over long runs get a signal conditioner near the meter to square the coil output before transmission. Surge protection goes on every field cable that leaves the skid, because the most common hardware failure in the field is not the computer, it is an input card killed by a nearby lightning induced surge on a long line. Multi point grounding, shared neutrals, and convenient but wrong cable shortcuts account for most commissioning noise tickets; the fix is discipline, not filtering.

Hazardous area installation follows two accepted routes. Explosion proof, Ex d, puts the certified enclosure and sealed cable glands at the hazardous location and keeps the wiring method strict. Intrinsically safe, Ex ia, keeps energy out of the field loop entirely: each loop passes through a safety barrier mounted on the safe side, the barrier entity parameters match the field device, and the loop documentation lives in the inspection file. Mixing the two philosophies on one skid is common and fine; mixing them on one loop is not. The RFQ should state the gas group, temperature class, and certificate basis, and the verifier will find the barriers on the drawing or they do not exist.

Commissioning and proving give the device its numbers. Configuration is entered under procedure and sealed: K factors, fluid properties, algorithm selections, contract units. Proving runs the meter against a prover, ball or small volume, and the resulting meter factor is entered as a sealed event with an author and timestamp; between proofs, the factor stands. The real time clock is checked against a reference at every proof, because interval totals inherit clock error. Analog inputs are ranged and checked against calibrated sources, orifice impulse legs are proven filled, and dual channel pulse counters are verified against each other with the meter flowing. The commissioning record, configuration listing, and first audit log export are filed together; that bundle is what an auditor asks for first.

Failure modes are few, known, and mostly detected by the device itself. Pulse input failure announces as a dual channel mismatch or a dead counter, and the totals flag unaccounted time per API 21.1 rules instead of guessing. Communication loss to the host triggers store and forward; the buffer depth against the longest credible outage is a selection number, not a comfort. Power failure exercises the redundancy and, on total loss, the nonvolatile totals and battery clock; restoration is an event in the log. Real time clock drift accumulates quietly between proofs and is corrected on the proof schedule, which is one reason the schedule exists. Surge damage presents as one dead input channel after a storm, which is why chapter wiring budgeted the protectors. Firmware faults are rare and are managed by version control: updates on custody devices follow procedure, checksum verification, and a log entry, never a casual download.

Maintenance is calendar driven and cheap next to what it protects. Review the audit and alarm logs monthly for unaccounted time and input faults. Prove meters on the contract cadence, common practice monthly to quarterly for liquid custody, and enter factors under seal. Replace RTC batteries on schedule, typically every 5 years well inside the 10 year retention claim, and verify the clock at every proof. Exercise and inspect surge protectors after electrical storms, check barrier continuity annually where Ex ia is used, and keep one cold spare of the chassis or at least the I/O cards for single stream sites. Export and archive the configuration and logs off device on a fixed cycle; retention onboard is a floor, not an archive strategy. The file that leaves this maintenance loop, factors, proofs, clock checks, and log exports, is the evidence that keeps the next contract negotiation short.