OEE was designed for discrete manufacturing, but it is not forbidden territory for continuous flow plants. The owner of oee.com answers the question directly: "OEE can also be applied to continuous processes (e.g., refineries)" [S1]. The metric is just the ratio of Fully Productive Time to Planned Production Time, so any line with a defined theoretical throughput has an OEE to calculate.
What changes is the construction of each factor. Continuous plants measure output in volume, weight, or length rather than in countable units, and they typically run 24/7 with minimal changeovers, which reshapes how Availability, Performance, and Quality are defined in practice [S3].
Why the Discrete Origin Matters for Continuous Plants
The 85% "world-class" benchmark that OEE practitioners chase was set in Japanese automotive plants in the 1970s, not in refineries or paper mills, and most discrete manufacturers still only average 60% against that line [S3]. Bringing the same number into a continuous plant without re-baselining is the most common misapplication: a paper machine or hydrocracker running 92% schedule utilisation is not necessarily world-class for its own process class.
The second legacy issue is unit count. OEE's Performance factor uses Ideal Cycle Time multiplied by Total Count, which assumes you can count what you made [S3]. A catalytic cracker that pushes 14,000 barrels per stream day (BPSD) does not have a discrete unit to count, so Performance has to be re-expressed against a design mass or volumetric rate, not a cycle time.
Where Standard OEE Breaks Down on a Continuous Line
Three structural mismatches surface on continuous flow assets. First, changeovers are rare, so the classic "Setup and Adjustments" loss from the Six Big Losses almost vanishes; the real losses shift to grade transitions, fouling, and ramp/derate windows, which the discrete Six Big Losses taxonomy does not enumerate [S1][S3]. Second, scheduled versus unscheduled downtime collapses differently: a hydrocracker may run for 36 months between turnarounds, so Planned Production Time must be defined against the campaign, not the shift, otherwise Availability turns meaningless. Third, Quality in a continuous plant is a property of the stream (sulphur content, melt index, basis weight), not a pass/fail on a finished part, so Good Count over Total Count does not translate without a lab or inline analyser closing the loop.
For a worked example, a process control loop on a distillation column can be at its setpoint and still be producing off-spec product because the feed composition drifted; the OEE numerator has to count only the time the stream was inside spec, which means OEE in this setting is really OEE on spec, not OEE on flow.
The OPE Alternative and Throughput-Based Variants
Process industry practitioners commonly substitute OEE with OPE (Overall Process Effectiveness) or with throughput-weighted variants. Continuous processes often use OPE that accounts for throughput rate rather than unit count, which is the same Availability multiplied by Performance multiplied by Quality multiplication but anchored to a design rate in t/h or m³/h instead of a cycle time [S3]. The calculation engine is identical, only the Performance denominator changes.
In practice that means the flow meter on the line becomes the primary OEE input, not a part counter. Coriolis, magnetic, and ultrasonic meters on a calibrated flow measurement skid feed the actual rate into the MES, and the OEE Performance factor becomes (actual mass flow / design mass flow) over the run window. This is also where a multifunction process calibrator earns its keep during commissioning: it verifies the flow and temperature transmitters against the design rate so the OEE baseline is grounded in traceable measurement, not in a nameplate guess.
Who OEE-on-Continuous Fits, and Who Should Pick a Different Metric
OEE on continuous flow makes sense for plants with a stable design rate, a clear in-spec window, and a constraint step that gates throughput: refineries, paper machines, steel rolling mills, glass furnaces, and cement kilns [S1][S3]. A process calibration programme around the constraint instrument is non-negotiable here, because a 1% drift on the constraint flow transmitter can swing the OEE score by a full point.
OEE is a poor fit for job shops, R&D pilot plants, and lines that run campaigns with very different design rates from batch to batch, since the Performance denominator has nowhere stable to sit [S4]. For these, OLE (Overall Labour Effectiveness), takt-based scheduling, or a pure throughput-yield metric is usually a cleaner answer than forcing OEE onto a process that does not have a constant theoretical output.
Implementing OEE on a Continuous Line: Practical Adjustments
Three adjustments make the metric defensible. Define Planned Production Time against the campaign (start of feed-in to planned shutdown or turnaround), not against the shift or the day, so that long steady runs are not penalised. Define Ideal Rate from the design nameplate of the constraint, or, better, from a V-process line acceptance test or performance test run, with a documented engineering tolerance of typically 2-5% for flow transmitters. Define Quality window from the product specification (lab or inline), with explicit time-bounding so the numerator reflects only in-spec time. [S2]
On instrumentation, anchor each leg to a traceable source. Pressure and differential-pressure transmitters on the constraint should be process calibration verified at 0%, 25%, 50%, 75%, and 100% of span, and flow meters should be proved against a reference standard at least once per campaign. Without this, the OEE number you report to corporate is a function of your instrument drift, not of your operations.
Comparison: OEE vs OPE vs Throughput-Yield on Continuous Lines
Three options line up against the decision criteria a process engineer actually faces. (1) OEE, Availability × Performance × Quality, with Performance anchored to a design rate: best for plants with one stable product and a clear in-spec window, easiest to roll up to corporate dashboards, weakest when design rate changes between campaigns. (2) OPE, same multiplication but Performance is throughput-weighted against a rate baseline: best for multi-grade continuous plants where a single design rate does not exist, requires more configuration in the MES. (3) Pure throughput-yield (good mass / total mass over a defined window): best for simple lines where Quality is the only thing the team can act on, but it loses the Availability and Performance decomposition that drives root-cause work. [S1]
For a brownfield continuous plant with an existing DCS, the layered protocol choice for pushing OEE data to the cloud matters as much as the formula. Practical guidance on what survives OPC UA versus what gets pushed over MQTT is covered in MQTT cannot replace OPC UA in brownfield plants, and where OPE/throughput dashboards pair with a digital-twin layer for scenario work, the trade-offs are laid out in digital twins in process development.
Trackable Signals for the Next Six Months
Two signals to watch. First, vendor and standards-body publications (e.g. ISA, API) on OPE and throughput-weighted effectiveness for continuous process industries, since the formal naming and any new benchmark values for 24/7 plants will land there first. Second, the published 2026 cohort numbers from software vendors that monitor OEE across 3,500+ machines in 50+ countries, which give a free cross-industry baseline to sanity-check any single-plant OEE result before it goes into a board pack. [S3]