Every plant owns a heat treatment furnace that has run the same cycle for fifteen years, never tripped an alarm, and quietly become the most trusted equipment on the floor. Nobody has mapped its temperature uniformity since commissioning. Nobody has verified its load thermocouples against a calibrated reference. The process sheet says 845 °C for 90 minutes, the recorder draws a flat line, and parts pass hardness checks, so the furnace must be fine.

That reasoning is where the trouble starts. A flat line proves only that the control thermocouple saw what it saw. It says nothing about the corner of the load basket, nothing about the mass of the workload acting as a heat sink, and nothing about whether the control probe has drifted over a decade in carburising atmosphere. Across two decades of auditing heat treatment cells in automotive and aerospace, the equipment that worries me most is never the one with documented problems. It is the one with no documented history at all.

What a CQI-9 Style Assessment Actually Asks

Heat treatment assessment in the automotive sense, with AIAG's CQI-9 as the reference most suppliers work to, is built on one premise: you cannot inspect quality into a heat-treated part. Once the load comes out, the microstructure is fixed. If the process drifted, no amount of downstream measurement recovers it. The assessment therefore concentrates on process understanding rather than part inspection.

The core questions are deceptively simple. When was the last temperature uniformity survey performed, at how many points, and at what temperatures? Was it done empty or with a representative load? What was the acceptance band, plus or minus eight degrees, or something tighter because the process demands it? A supplier who answers that the OEM approved the plant years ago has said everything an auditor needs to know. Approval documents age, and furnace elements age faster.

Special process status is the other pillar. Personnel who set cycles, load furnaces and interpret charts need demonstrated competence, not a signature on a training matrix. If the night-shift operator cannot explain what a sagging load thermocouple trace means, or why a heavily loaded basket needs a longer recovery allowance, the process is not under control regardless of what the paperwork claims. Competence is observed at the furnace, not filed in a binder.

Both pillars converge on the same evidence set: temperature uniformity survey (TUS) results, system accuracy test (SAT) records, instrumentation classification of recorders and controllers, and operator qualification records. Where any one of these is missing, the assessment finding is not the gap itself but the absence of any mechanism that would have detected the gap.

Quality in heat treatment is decided inside the chamber, before the first hardness check is ever performed.
Quality in heat treatment is decided inside the chamber, before the first hardness check is ever performed.

Load Thermocouples: The Only Truth You Get

Control thermocouples live in protected zones, often near the elements, and they keep the furnace happy. Load thermocouples, sometimes called workload or part thermocouples, tell you what the steel actually experienced. The distinction dominates carburising and through-hardening. A control probe reading 850 °C while the core of a dense load sits at 790 °C for the first twenty minutes produces exactly the sort of shallow case depth that surfaces months later as a field failure on a gear tooth.

Practical verification means attaching thermocouples to actual parts or dummy parts of similar mass, positioned in the coldest and hottest locations established by the survey, typically the centre of the load and near the elements. The soak clock should start when the part reaches temperature, not when the furnace does. If the cycle timer starts on furnace recovery, every heavy load is systematically under-soaked, and the error grows with load size.

Thermocouple condition deserves equal attention. Type K probes contaminated by carburising atmospheres drift, and a drifted load probe delivers false confidence with a perfectly flat trace. Check for embrittlement, verify against a calibrated reference on a defined frequency, and replace on condition rather than calendar convenience. In my experience, the thermocouple that has always worked is the first one to pull.

Load thermocouple verification cycle

  1. 01Survey the zoneNine-point map establishes hottest and coldest locations in the working volume.
  2. 02Place probes on massAttach to real or dummy parts at the extremes, not to furnace furniture.
  3. 03Start soak at part temperatureTime the cycle from part recovery, not furnace recovery.
  4. 04Verify against referenceCompare probe readings to calibrated instrumentation on a set frequency.
  5. 05Replace on conditionSwap drifted or embrittled probes and restart the baseline.
The loop closes only when replacement triggers a new baseline, because a fresh probe changes the evidence, not the process.

Case Depth: Verifying What the Cycle Claimed to Do

Effective case depth is the property the customer bought, and surface hardness cannot confirm it. A surface reading tells you the carbon potential and the quench behaved; it says nothing about the gradient beneath. The accepted verification is a microhardness traverse, a Vickers or Knoop line scan from the surface inward, plotted against depth, with effective case depth defined at the hardness limit agreed on the drawing, commonly 550 HV for carburised cases.

Sample preparation is where labs cheat themselves. A tapered or poorly polished mount skews the entire traverse. Cutting must avoid tempering the sample, which means abrasive cutting with adequate coolant, or the depth you measure reflects the metallurgy of the metallographic lab rather than the furnace. Section location matters as well: a gear tooth flank and root will show different depths, and the test plan should specify where the traverse runs rather than leave it to whoever is holding the part.

Frequency is a judgement call I see botched in both directions. Testing one part per month from a furnace running eight loads a day is decoration. Testing every load when the process has years of stable traverses and no variable changes is waste. Tie sampling to what varies: new part numbers, load density changes, alloy changes, atmosphere recalibration, or any event that touched the process. When something changes, verify; when nothing changes, trust the stability evidence you have built.

Distortion: Reading the Part After the Quench

Distortion is heat treatment's tax on geometry, and it arrives in two forms: thermal distortion from uneven heating and cooling, and transformation distortion from the volume change as austenite converts to martensite. Quench severity dominates the second. A part quenched harder than necessary, oil too cold or agitation too aggressive for the section thickness, will warp, grow or crack in ways no straightening press fully recovers, and straightened parts carry residual stresses that release later during machining or in service.

Control starts with fixturing and load arrangement. Long slender parts hung vertically behave differently from the same parts lying flat in a basket. Parts stacked so quenchant cannot flow between them cool unevenly side to side. Racking density, part orientation and agitation pattern are process parameters in their own right, yet I rarely see them documented anywhere except in the loader's head. Write them down, control them, and audit them like any other parameter.

Post-treatment measurement should be risk-based. Check the geometric features the drawing flags as distortion-sensitive, flatness on plates, runout on shafts, bore roundness, on a defined sample with fixtures that support the part consistently. If distortion trends upward, look first at quench oil temperature and condition. Oil oxidises, water contamination creeps in, and cooling curves shift long before anyone notices. A periodic cooling curve test on the quenchant is cheap insurance against a very expensive sorting exercise.

Mapping the Furnace You Inherited

Temperature uniformity surveying is the core discipline. Map the working zone with a minimum nine-point array, eight corners plus centre, at each operating temperature you actually use, and record both the uniformity band and the time to reach it. Repeat periodically with a loaded furnace, because an empty survey certifies the furnace, not the process. If the TUS rig cannot survive a loaded run, dummy parts with embedded thermocouples give a workable approximation.

System accuracy tests close the loop between the survey and daily operation. Compare control and recording instrumentation against a calibrated reference at temperature, and correct drift before it compounds. Calibration intervals should reflect furnace duty, not a generic annual default. A carburiser running continuously at 900 °C in endothermic atmosphere punishes thermocouples far harder than a low-use temper oven, and the intervals should say so.

Any intervention on the thermal system, elements, refractory, probes or controller, resets the clock on uniformity evidence.

Keep the evidence alive. A survey from four years ago, taken before the elements were replaced and the retort rebuilt, describes a furnace that no longer exists. Suppliers who resent the reset rule usually have something in the survey history they would rather not discuss. Fresh evidence after intervention is not bureaucracy; it is the only honest statement that the furnace you are running is the furnace you mapped.

Habit versus evidence in furnace control

Trusting the flat chart

  • Uniformity survey dated to commissioning
  • Soak timed from furnace recovery
  • Racking knowledge held by one loader
  • Quenchant condition unknown between oil changes
  • Case depth checked on calendar, not on change

Running on evidence

  • Current TUS with loaded-run verification
  • Soak timed from part temperature
  • Racking density and orientation documented
  • Periodic quenchant cooling curve tests
  • Traverses tied to process change events
The left column describes equipment nobody can defend; the right column describes a process an auditor or customer can walk through without a single open question.

Pulling It Together on the Floor

The through-line is simple: trust evidence, not habit. A furnace with current uniformity data, verified instrumentation, competent operators, load thermocouples on critical cycles, hardness traverses tied to change events, and quenchant condition monitoring is a furnace you can defend. A furnace with a flat chart and institutional confidence is a liability wearing a reliability costume. The difference between the two is rarely capital investment; it is discipline applied to records that already should exist.

If you run heat treatment in-house, start this week with one question: when did a calibrated probe last touch a part inside that chamber? If the answer involves shuffling feet, you have found your first improvement project. If you subcontract, ask your heat treater for their last survey report and load thermocouple procedure before the next PPAP-style review. Their furnace becomes your process the moment your parts enter it, and their evidence gaps become your audit findings.

I have watched unmapped furnaces produce good parts for years and then lose an entire week's output to a drifted probe nobody was checking. The failure was never sudden. It was always invisible, and it was always checkable. That combination, invisible but checkable, is the whole argument for treating heat treatment as a controlled special process rather than a trusted black box.