If you are planning to install furnace atmosphere monitoring in a Thai plant, a quotation for an oxygen probe is only the beginning. In a gas carburizing furnace, the useful system connects probe voltage, furnace temperature, the assumed or measured CO level, reference air, carbon potential calculation, gas control, alarms and the heat-treatment lot. Without that chain, a stable value on a screen cannot reliably explain what happened to a batch.
This guide is for plant engineering, quality and IT teams preparing an RFP for an existing or new gas carburizing furnace. It focuses on in-furnace atmosphere measurement and control. The broader design of heat-treatment process records and AI predictions of finished hardness are separate subjects.
In this article, FAT means Factory Acceptance Test (pre-shipment factory acceptance) and SAT means Site Acceptance Test (on-site acceptance); these are the project acceptance stages.
Define the scope of furnace atmosphere monitoring first
“Heat-treatment furnace” is too broad for a procurement specification. Gas carburizing, low-pressure carburizing, nitriding and tempering have different atmospheres and control methods. This article addresses gas carburizing with carrier and enrichment gases. Do not assume that the same zirconia probe arrangement transfers to a vacuum or low-pressure process. Record the furnace type, gas train, current controller, instrumentation and customer requirements during the site survey.
Monitoring means more than seeing a trend. Operators must know whether the measurement is valid, how the carbon potential (CP) was calculated, which recipe segment and heat lot it belongs to, what the gas valves did and who responded to an abnormal condition. A controller, historian and MES can be separate systems, but a reviewer must be able to reconstruct the same batch across them.
An acceptable CP trend does not by itself certify hardness or effective case depth. Material grade, loading, temperature, time, quenching and product tests remain part of quality control. ISO 20431:2023 treats heat-treatment quality as a broader system, while SAE AMS2759/7F covers carburizing of carburizing-grade steel parts across several processes. Check the applicable standard edition against the customer contract and drawing; a standard name alone is not an acceptance criterion.
What carbon potential represents
Carbon potential is a process value describing the atmosphere’s ability to supply carbon at the steel surface. A zirconia oxygen probe does not measure carbon directly. It produces a millivolt signal related to the oxygen partial-pressure difference between furnace gas and reference air. The controller combines that signal with temperature and a CO input or assumption, plus its calculation parameters, to estimate CP. Eurotherm’s technical material explicitly distinguishes that calculation from a direct carbon measurement and discusses the effects of the CO factor and a non-equilibrium atmosphere.
Two screens reading the same CP need not represent identical evidence. Check whether CO is a fixed setting or an analyzer input, which probe and algorithm are configured, where temperature is measured, whether probe cleaning occurred and whether gas supply was stable. More decimal places do not improve measurement integrity. CP is also not measured surface carbon or finished hardness. Diffusion takes time and depends on steel and geometry. The process engineer defines the recipe, permissible band and treatment of transition periods; quality approves the relationship to product inspection.
Specify the complete probe measurement chain
Probe position, signal and temperature
An RFP should name the probe model or required characteristics, insertion position and length, connection, service access, signal range and diagnostics. A probe in a stagnant or unrepresentative location cannot describe the atmosphere the controller is meant to regulate. The furnace maker and instrumentation team should review heat exposure, seals and safe replacement access.
Archive raw probe mV, the associated temperature, calculated CP and a validity flag as distinct tags. A temperature-input failure must not be hidden by silently retaining the last good value. Recording only the displayed CP makes it harder to distinguish a real atmosphere change from sensor drift or a calculation error.
Reference air is a measurement condition
The zirconia method depends on the oxygen difference between the furnace and reference sides. Reference-air supply, blocked or leaking tubing, pump failure and the model-specific recommended flow therefore belong in the measurement chain. Do not copy a generic flow setting into the RFP; use the selected probe manual. If reference-air status is unknown, define how the CP value is marked invalid and what operators must do with the affected lot.
Health checks and independent comparison
Soot, ageing and high probe impedance can undermine a credible-looking control trend. Eurotherm describes probe cleaning, soot alarms, impedance diagnostics and CO input from a gas analyzer as available implementation methods. They are product examples, not universal maintenance intervals or alarm thresholds.
Choose an independent comparison that suits the process and customer requirement. A gas analyzer may measure CO, CO2 and CH4; dew-point checks and shim-stock or foil checks answer different questions and have different response times. State the method, sampling conditions, allowed discrepancy and disposition when methods disagree. Keep calibration evidence for the reference instrument.
Fixed CO assumption or gas-analyzer integration?
One design calculates CP from probe voltage and temperature using an engineering CO setting. Another takes measured CO, and possibly other gases, from an analyzer. Neither architecture is automatically correct for every furnace. Consider changes in the gas source, recipe transitions, evidence required by the customer and maintenance capability.
| Decision point | Configured CO value | Analyzer integration |
|---|---|---|
| Calculation premise | CO treated as a selected setting | Measured gas value can update the calculation |
| Failure modes to test | Wrong setting or changing gas composition | Sample train, calibration, invalid value, communications loss |
| Acceptance evidence | Periodic independent comparison of the premise | Verified updates and explicit fallback behavior |
| Ongoing work | Maintain the premise and comparison schedule | Maintain analyzer and sampling system |
An analyzer is only as trustworthy as its sample path. Specify the sampling location, residence time, leakage, filtration, cooling, calibration gas and exhaust arrangements. Align timestamps and furnace state when comparing analyzer readings with the probe. A three-gas scheme may use CO, CO2 and CH4, but its calculation and connections are equipment-specific. If the analyzer is only a future option, state what this project actually uses to compute CP.
Link atmosphere history to recipe and heat lot
Minute-by-minute data are not useful if nobody can tell which parts were exposed. Record furnace ID, furnace batch ID, production lot, part and material, load location where needed, recipe ID and revision, segment targets, actual temperature and CP, alarms and operator decisions. Where the furnace batch and MES production lot differ, preserve the mapping as controlled master data.
A recipe name is insufficient. Save the approved revision, who changed it, when it became effective and which revision actually ran. Detect a failed download that leaves an older version in the controller. Distinguish the planned recipe from the executed recipe. Choose sampling frequency and retention based on transients, hold periods, gas-valve behavior, customer requirements and investigation needs rather than maximum tag throughput. Preserve timestamps, timezone treatment, missing-data flags and recovery after a power or network interruption.
For the wider record architecture, see our heat-treatment process-record guide. For modelling finished-product outcomes, see our hardness-prediction AI guide. This article addresses the reliability of the furnace atmosphere value that may feed those records and models.
Design alarms through lot disposition
CP high and low limits are only part of the alarm plan. Evaluate probe signal faults, reference-air failure, invalid temperature, analyzer fault or calibration state, gas-pressure or flow fault, CP deviation, soot and cleaning alerts, logging failure and recipe mismatch. Engineering and quality must approve which conditions warn, stop the process or place a lot on hold. Do not transfer an example threshold from another furnace.
Store alarm onset and recovery times, affected lot and segment, measured values, acknowledgment, action, quality disposition and release approval. Silencing a sound is not evidence that the cause was removed. A night-shift event should be reviewable by morning quality staff without reconstructing the story from screenshots. Automatic recovery of a controller should not automatically release the lot.
Keep local furnace safety control independent of a dashboard or cloud connection. If remote changes to CP targets or valves are allowed, define authority, access logs, approval, rollback and behavior during connection loss. NIST SP 800-82 Rev.3 provides guidance for OT security in systems where performance, reliability and safety matter.
Survey the existing furnace before requesting quotations
Collect the P&ID, wiring diagrams, controller and PLC details, probe information, gas-generator and enrichment lines, recorder, network diagram and maintenance history. Verify drawings against the physical plant during an agreed outage. Plant safety procedures and the furnace maker’s requirements govern modifications to a hot or pressurized installation.
Observe who creates and approves recipes, who can alter CP, how shifts hand over alarms, and how cleaning and probe replacement are documented. Handwritten correction factors, separate calibration spreadsheets or screenshot-based audit evidence reveal gaps a feature list will miss. Map the system of record for controller values, MES heat lots, ERP part data and laboratory hardness or case-depth results. Assign a data owner and reconciliation method to each interface.
What a comparable RFP should request
An RFP saying only “visualise CP” produces incomparable bids. State the baseline for each furnace and separate mandatory from optional scope: probe replacement, controller update, analyzer and sample train, reference-air system, historian, MES link, alarms and support. Include outage time, commissioning, training, spares, language needs and support in Thailand.
| RFP area | Response required from the vendor |
|---|---|
| Calculation | Probe mV, temperature, CO premise, algorithm and valid range |
| Instrumentation | Position, reference air, plumbing, cleaning, replacement and comparison |
| Control | CP targets, gas-valve behavior and failure-state behavior |
| Records | Raw values, calculated values, validity, recipe revision, lot and timestamps |
| Alarms | Priority, delay, recipient, lot hold and release authority |
| Integration | PLC, historian, MES and ERP interface ownership and recovery |
| Acceptance | FAT/SAT cases, expected results, evidence and corrective action |
| Service | Calibration, consumables, spares and controlled remote access |
Normalize exclusions before comparing price. One supplier may reuse a probe, another may include a new one, while a third omits analyzer sampling hardware. Ask for compliant, conditional, non-compliant and extra-cost responses to the same checklist. Compare recurring calibration and maintenance as well as the first installation. Do not justify a purchase with an invented payback figure.
FAT: prove logic and records before shipment
FAT can test controller and software behavior with simulated inputs; it cannot reproduce the exact gas conditions of the installed furnace. Inject probe mV, temperature and CO inputs and verify calculation, display, history, deviations, gas-valve commands and invalid-data handling. Label every simulated input in the evidence.
Test broken probes, missing temperature, reference-air faults, analyzer communication loss, implausible inputs, recipe-revision mismatch, clock drift, network outage, power restart and duplicate lot IDs. Verify that missing history is flagged rather than silently interpolated. Exercise recipe creation, review, approval, download, revision rollback and rejection of unauthorized changes. Retrieve one example lot with raw mV, CP and temperature curves, alarms and operation history.
SAT: prove installation and a real heat lot
At SAT, inspect installed wiring, probe position, reference air, gas connections, temperature synchronization, analyzer sampling point and valve response. A correct simulated calculation cannot compensate for a leaking sample path or an unsuitable probe location. Agree an outage window, safety lead and stop conditions before tests.
Use real lots across representative recipes, loads and segment changes. The allowed CP deviation comes from the product and customer specification, not a general web article. Align independent comparisons with their sample time and furnace condition. Retain the relationship between atmosphere history and final product inspection, while keeping product release under the approved quality plan. Log SAT defects, owners, deadlines and retest evidence. Hand over usable local instructions, appropriate Thai/English/Japanese documentation and spares.
Maintain the system and control changes
After go-live, review probe cleaning and diagnostics, reference air, temperature input, analyzer calibration and independent comparison. Set intervals using the chosen equipment, furnace contamination and customer requirements. A stable trend can also mean a frozen signal or last-value substitution, so validate data freshness.
Treat changes to the CO factor, calculation logic or recipe as quality-relevant. Record rationale, before and after settings, approvals, affected furnaces and lots, tests and rollback. Review repeated alarm causes, held-lot disposition time, missing data and discrepancies in cross-checks. If several suppliers share responsibility, specify who diagnoses probe, gas, controller and network failures first. Require a data dictionary, units, timestamps, validity flags and export of joined lot history so the plant retains usable evidence during future equipment changes.
Put data handover and responsibilities in the contract
Installation is not the end of the project. If ownership and export format remain ambiguous, the plant will face the same problem at the next furnace replacement. Require a data dictionary covering historical field definitions, units, time, validity flags, recipe revision and heat-lot join keys, and alarm-state transitions. “CSV export” or “API available” is not an acceptance result. Select a sample heat lot and require an export that a third party can follow from raw measurements through the recorded quality decision.
When the controller maker, furnace maker, integrator and plant quality team are separate parties, specify who performs first-line diagnosis for each failure. If replacing a probe does not restore credible values, who checks reference air, temperature, gas composition, calculation settings and the network, and in what order? A single service desk does not replace an internal responsibility matrix or the logs needed to resolve an incident. Agree night-shift support, escalation and the location and custody of spare parts in Thailand.
For expansion, validate measurement integrity and lot linkage on one well-understood furnace before defining a rollout standard. Standardization means common data fields, alarm ownership and test procedure, while documenting each furnace’s installation, calculation and gas-condition differences. It does not mean copying thresholds to every furnace. Perform a separate SAT for each furnace added to the system.
Make the data usable as quality evidence
An operations dashboard serves immediate decisions; a quality record must reconstruct an earlier lot. Define which values are primary evidence and retain a data dictionary with tag names, units, range, sample interval, instrument ID, conversion and invalid-value codes. Preserve an instrument replacement boundary so a later reviewer knows which probe generated each period of history. A graph screenshot without raw values and calculation premises is insufficient for investigating a later probe fault.
Treat a communication gap as missing data, not zero CP or a continuation of the last valid sample. Record its start, end, recovery and any interval that could not be backfilled. Synchronize controller, collector and MES clocks and store an explicit timezone. A furnace batch can contain several production lots, and one production lot may be split across batches. Model those as relationships in load and unload records rather than forcing a one-to-one ID field.
For a first deployment, narrow scope to one well-understood furnace and representative recipes while retaining the essential probe, temperature, reference-air, CP calculation, lot ID, raw history and critical alarm evidence. Confirm independent comparison and the product-inspection link before expanding dashboards across the plant. Reuse the data format and test method at the next furnace; revalidate its own probe position, gas conditions and acceptance settings.
At the furnace door, verify that the recorded load actually matches the physical work basket or tray before the cycle starts. A barcode on a tag is insufficient if no one confirms which parts went into the closed furnace. Include an approved correction process for a wrong scan or a split lot, with the original and corrected records preserved.
Priorities when narrowing the first installation
A limited budget or outage window is a reason to select one furnace and representative recipes, not a reason to omit the measurement foundation. Confirm probe, temperature and reference-air health, the CP calculation premise, heat-lot ID, raw history and critical alarms. Then verify independent comparison, linkage to product inspection and the shop floor’s ability to place a lot on hold after an abnormal event. More elaborate dashboards and plant-wide analytics can come later; multiplying displays will not multiply trustworthy evidence if the first furnace value cannot be explained.
Do not reuse the first furnace’s acceptance result as proof for another furnace. Furnace design, probe position, gas supply, instruments and loading can change the measurement chain even with identical software. Reuse common data fields, test records and approval procedures; recheck alarm thresholds and comparison conditions on each furnace. Feed exceptions and rework found in the first trial back into the next furnace’s RFP rather than treating rollout as a copy operation.
Frequently asked questions
Does an oxygen probe directly measure carbon potential?
No. It produces an oxygen-related voltage. The controller estimates CP using temperature and the configured or measured gas assumptions. Evaluate the full measurement chain and comparison method.
Is a gas analyzer mandatory?
That depends on gas variability, customer requirements and the required confidence in the calculation. Compare both architectures and their maintenance commitments before deciding.
What should be checked first on an existing furnace?
Check furnace type, mounting point, reference air, temperature source, controller, gas plumbing, access for maintenance, available outage and the existing heat-lot identity.
Are FAT and SAT interchangeable?
No. FAT tests simulated signals, control logic and data linkage. SAT tests the installed measurement chain, response and actual lots. The test plan should distinguish them.
Can product inspection stop when CP stays in band?
Not on that fact alone. CP is an atmosphere-management value, not the measured finished-property result. Follow the drawing, customer and approved process requirements for inspection.
What changes when several furnaces share one monitoring screen?
Show each furnace’s probe type, calculation settings, CO handling, recipe revision, timestamp and alarm thresholds distinctly. Do not assume identical-looking CP values were calculated under identical conditions. Keep the furnace and heat lot visible whenever comparing trends.
Conclusion: specify the evidence before buying the sensor
A useful furnace atmosphere monitoring project defines the furnace process, probe and reference-air chain, CP calculation premise, independent comparison, recipe and lot linkage, alarm disposition and acceptance tests. Make bids comparable with a concrete RFP, test abnormal inputs at FAT and prove the installation against real lots at SAT. The deliverable is an explainable quality history, not merely a trend display.
If your Thai plant is reviewing an existing carburizing furnace, probe replacement, analyzer connection or heat-lot history, contact TOMAS TECH while the RFP is still being shaped. We can help organize measurement points and acceptance evidence.
Sources
- ISO 20431:2023, Heat treatment — Control of quality
- SAE AMS2759/7F, Carburizing and Heat Treatment of Carburizing Grade Steel Parts
- AIAG CQI-9, Heat Treat System Assessment
- Eurotherm, Carburising Gas Reactions
- Eurotherm, Carburising and Quench Control
- Eurotherm, Atmosphere Carburizing Control Solutions
- NIST SP 800-82 Rev.3, Guide to Operational Technology Security
