Jim Roberts of U.S. Ignition engages readers in a Combustion Corner column about the modes of heat transfer at play in the heat treating world — breaking down advection and thermal conduction, from how forced air movement drives heating and cooling in furnaces to how Fourier’s law governs heat flow through direct contact, setting the stage for next month’s look at convection and radiation.
This editorial was first released in Heat Treat Today’sAugust 2026 Annual Automotive Heat Treating print edition.
A furnace guy walks into a heat treat facility and asks, “tropos metadosis thermotitas?” The furnace operators stare and say, “What are you saying? It’s all Greek to me!” But many of us know what it means: What mode of transfer? And that was the question of the last article where we looked at an intro to thermodynamics. This set us up for discussions on the four modes of heat transfer, a non-stop set of interactive processes that we then refine for our use in the heat treating world.
In that earlier discussion, we explained that there are several modes involved in heat transfer. Some are so simple to describe, but there may be several modes working at the same time to target the heat transference we seek. Modes of heat transfer and the Laws of Thermodynamics differ. Think of modes as the actual physical properties of heat transfer and thermodynamics that we can see and feel. The laws are the calculations and theory that allow us to check and produce the heat that is needed or to prove how much we need.
Our first mode of heat transfer is advection. Advection is the mechanism of thermal transfer that occurs when the thermal energy is transferred or transported from one space or object via the movement and the motion of a fluid. Pure convection, in the scientific sense, is strictly the heat transfer between heated bodies by gravitational lift or natural fluid movements.
In other words, heat rises… right? We’ve all heard it, witnessed it in a campfire or elsewhere. Advection is when the transfer media, mostly hot gases, is forced mechanically to swirl around and the item we desire to heat or cool. Speaking of heat transfer, we cannot forget that the door swings both ways. We can affect heat transfer from the standpoint of cooling things down too, and at that point, we discover that the hot gases or heat exchangers with air of liquid are using advection to pull the heat out. The plunge cool function of many furnaces relies on advection to pull the heat out of the parts at a controlled rate to cause a metallurgical change. So, the same laws apply, just in reverse!
One of the easiest ways to remember this difference is that advection actually adds energy to the process. It may be in the form of a blower providing combustion air for a burner, which creates the swirl of gases that are providing the heat transfer to the parts. Or, think of a fan pulling heat from a circuit board of electrical components. Advection is actually forced convection. If you have a modern convection oven or an air fryer, it is not really a convection oven. It’s an advection oven in the scientific world, since the fans involved are adding energy to move the fluid (air/gas) around those chicken wings and fries. The concern over this kind of added energy is in part what makes Data Centers such a volatile topic.
This leads us to our second heat transfer mode, thermal conduction. Thermal conduction is the transfer of heat by direct contact of two objects where there is a differential in temperature and energy. In other words, the temperature difference between two objects is the mechanism for heat transfer. Energy wants equilibrium, so it will always try to bring an item to its temperature by maintaining contact. It can be identified as thermal diffusion in Fourier’s law for heat conduction.
Fourier’s law states that eventually, through contact, the handle on the cooking pot will match the temperature of the pot surface because it is conducting the heat into that handle via thermal conduction. These types of considerations take place during calculations for furnace design, engine components, or household items; anything with a heat differential must be accounted for in our everyday lives.
As they say in the old country: choris thermotita, den yparchei ergo. Without heat, there is no work. That’s the loose translation, but it points out that we need to have all the components for heat transfer to be working for us to do this wonderful work called heat treating. And that shouldn’t be Greek to us.
Next month we will tackle convection and radiation.
About The Author:
Jim Roberts President US Ignition
Jim Roberts president at U.S. Ignition, began his 45-year career in the burner and heat recovery industry focused on heat treating specifically in 1979. He worked for and helped start up WB Combustion in Hales Corners, Wisconsin. In 1985 he joined Eclipse Engineering in Rockford, IL, specializing in heat treating-related combustion equipment/burners. Inducted into the American Gas Association’s Hall of Flame for service in training gas company field managers, Jim is a former president of MTI and has contributed to countless seminars on fuel reduction and combustion-related practices.
Heat Treat Today launches a new contributing experts series on emerging technologies reshaping heat treat operations. In this first installment, Zach Menard, CEO and co-founder of HRC Labs, demystifies deep neural networks (DNNs)— a technology quietly used in industrial settings for years — and explains how they turn accumulated furnace cycles, recipes, and quality records into predictive models that help heat treaters develop new recipes, minimize distortion, and cut cycle times.
This informative piece was first released in Heat Treat Today’sAugust 2026 Annual Automotive Heat Treating print edition.
Large language models (LLMs) have dominated recent conversations around artificial intelligence, but they are only one application of deep neural networks (DNNs). Years before DNNs were applied to human language, they had been deployed in industrial settings to optimize processes, improve yield, and support quality control.
These same, well-established approaches can be applied to heat treatment by learning relationships between process inputs — such as temperature, time, atmosphere, and alloy composition — and resultant metallurgical properties like hardness, distortion, or case depth.
It’s the first word in the acronym, DNN, that makes this learning possible. The neural network is said to be “deep” because of its many successive layers of neurons. Each neuron on its own is very simple. All it can do effectively is multiply and add together the numbers it has received from earlier neurons. However, when this process is repeated across many layers, the overall network is able to learn complex relationships that influence the heat treatment process.
By training on historical production data, DNNs can transform years of accumulated process knowledge into predictive models. Patterns that may otherwise remain buried across thousands of furnace cycles, recipes, and quality records can be captured quantitatively by the model itself.
In this way, operational experience and historical records become structured, reusable intelligence that can aid decision-making. Examples include when heat treaters are developing a new recipe, considering how to minimize distortion, or lowering cycle times.
As heat treatment operations continue accumulating larger volumes of process data, applying DNNs to this data offers a clear path toward AI-assisted process optimization, operational intelligence, and more adaptive heat treatment systems.
About the Author
Zachary Menard CEO and Co-Founder HRC Labs
Zachary (Zach) Menard is the CEO and co-founder of HRC Labs, an industrial AI company focused on process optimization and decision support for heat treatment. He holds a B.S. in physics from Vanderbilt University with a background in computational physics and machine learning research.
Ask The Heat Treat Doctor® has returned to bring sage advice to Heat Treat Today readers and to answer your questions about heat treating, brazing, sintering, and other types of thermal treatments as well as questions on metallurgy, equipment, and process-related issues. In this Technical Tuesday installment, Dan Herring explores the fundamentals of vacuum oil quenching — the three stages of oil quenching, the properties that make oil an effective quenching medium, and the ways pressure control, oil circulation, and temperature can be used to minimize part distortion — and offers step-by-step guidance on properly conditioning and degassing vacuum quench oil to ensure consistent, repeatable results.
This informative piece was first released in Heat Treat Today’sAugust 2026 Annual Automotive Heat Treating print edition.
Oil quenching in vacuum furnaces is a technology that has been around for over 65 years. The popularity of oil as a quench medium is due to its excellent performance and stability over a broad range of operating conditions. Proper application, testing, and conditioning of the oil are necessary to ensure its effectiveness. Let’s learn more.
For many, the choice of oil is not only due to metallurgical transformation and resultant properties it can achieve, but also to a number of other factors, including:
The first criterion that any quenchant must meet is its ability to approach an ideal quenching medium.
Figure 1a-b. Three stages of oil quenching
Regardless of the type of liquid (brine, water, polymer, oil, molten salt, or air mist), the ideal quenching medium (Figure 1) is one that would exhibit high initial quenching speed in the critical hardening range (Stages 1 and 2) and a slow final quenching speed through the lower temperature range (Stage 3). Thus, the ideal quenchant is one that exhibits little or no vapor blanket stage, a rapid but controlled nucleated boiling stage, and a slow cooling rate during convective cooling.
The high initial cooling rates allow for the development of full hardness by ensuring the steel misses the “nose” of the time-temperature-transformation diagram (i.e., quenching faster than the so-called critical transformation rate). This is followed by cooling at a slower rate beginning at the time the steel is forming martensite, which allows for better stress equalization so that the potential for distortion and cracking is reduced.
When conventional quenching oils are used, the duration of Stage 1 is longer, the cooling rate in Stage 2 is considerably slower, and the duration of Stage 3 is shorter. As such, the “quenching power” of oil is far less drastic than other quenchants. Water and water-based quenchants exhibit high initial cooling rates. Unfortunately, because of water’s low boiling point, this fast cooling persists until the steel is cooled to below 300°F (150°C). As most steels have formed or are forming martensite by this point, stresses are given little time to equalize. Thus, water is typically limited to low hardenability materials and relatively simple geometries.
Oil has a major advantage over water or polymer due to its higher boiling range. A typical oil has a boiling range between 450°F (230°C) and 900°F (480°C). This causes the slower convective cooling stage to start sooner, enabling the release of transformation stresses. Oil, therefore, is able to quench intricate shapes and high hardenability alloys successfully.
As it is heated, oil has a proportional drop in viscosity. This allows the quenchant to move more freely, increasing, in general, the tendency to break the vapor blanket layer. The nucleate boiling stage is not drastically altered by changes in bath temperature. The cooling rate in the convection stage, however, will slow as the bath temperature increases. This is advantageous for obtaining a slower rate of cooling through the austenite-to-martensite transformation range.
In general, as the temperature of a quenching oil increases, the overall quenching rate increases. Practical heat transfer coefficient (α) values are in the 1,000 to 2,500 W/m²K range depending on oil characteristics and degree of agitation. Peak α values are in the order of 4000 to 6000 W/m²K, or a cooling rate greater than 100°C/sec (180°F/sec).
Figure 2. Typical commercial heat treat load | Image Credit: The HERRING GROUP, Inc.Figure 3. Die cutting punches benefiting from controlling the pressure over the oil | Image Credit: The HERRING GROUP, Inc.
The use of vacuum oil quenching has been found to reduce distortion in many components including gears, pinions, and shafts (Figure 2). Oil quench vacuum systems also offer an attractive alternative to conventional atmosphere oil quenching, given their ability to vary a quench related variable not otherwise possible, namely controlling the pressure over the oil (Figure 3). This technique can be used to extend the range of part cross sections and materials that can be successfully hardened (Herring 1987, Sugiyama and Uchigaito 1987).
Pressure control along with oil temperature and oil circulation improves the predictability of distortion. The lower pressure allows for longer “vapor blanket” stages and a somewhat long “vapor transfer” stage, due to the reduced boiling point of the oil. This may reduce distortion and provide the desired hardness if the material’s transformation ranges are accommodating.
Distortion minimization methods have been used in combination with changes to flow characteristics. Some manufacturers pull oil down through the workload as opposed to pushing it upward. Also, oils formulated for vacuum service typically have a low vapor pressure, allowing them to be easily degassed.
Finally, vacuum systems do not permit the buildup of water in the quench tanks. In a vacuum furnace system, where vacuum is used to process the work or purge the quench environment, moisture will be removed as the system is evacuated and the oil circulated. Circulating the oil carries any moisture to the oil surface, where it vaporizes and is removed by the vacuum pumping system.
Maintaining Your Vacuum Quench Oil
One of the aspects of vacuum oil quenching that is seldom documented is how to condition the quench oil. See “Step-By-Step Instructions for Conditioning and Degassing Vacuum Quench Oil” below.
Final Thoughts
As with all quenching, the key is to understand and control the key process variables. Proper selection of the type and use of oil under ideal conditions in a well-designed and well-maintained quench tank will ensure consistent and repeatable results.
Oil quenching should be applied in applications where its advantages outweigh its disadvantages and, as with all technologies, should be as completely understood as possible with respect to the performance requirements of the product so as to meet the application end use.
References
Brian Barlow. 2025. Private correspondence. Gasbarre Thermal Processing Solutions.
Herring, Daniel. H. 2002. “A Review of Factors Affecting Distortion in Quenching.” Heat Treating Progress Magazine. December. 2012. Vacuum Heat Treatment, Volumes I. BNP Media Group.
Herring, Daniel H. 2016. Vacuum Heat Treatment, Volume II. BNP Media Group.
Herring, Daniel H., Steven D. Balme. 2007. “Oil Quenching Technologies for Gears.” Gear Solutions. July.
Herring, D. H., Sugiyama, M., Uchigaito, M. 1986. “Vacuum Furnace Oil Quenching – Influence of Oil Surface Pressure on Steel Hardness and Distortion.” Industrial Heating Magazine. June.
Sugiyama, M., Uchigaito, M. 1987. “Controlling Oil Surface Pressure in Vacuum Oil Quenching.” Heat Treating Magazine. July.
About the Author
Dan Herring “The Heat Treat Doctor®” The HERRING GROUP, Inc.
Dan Herring has been in the industry for over 50 years and has gained vast experience in fields that include materials science, engineering, metallurgy, new product research, and many other areas. He is the author of six books and over 700 technical articles.
Ask The Heat Treat Doctor® has returned to bring sage advice to Heat Treat Today readers and to answer your questions about heat treating, brazing, sintering, and other types of thermal treatments as well as questions on metallurgy, equipment, and process-related issues. In this installment, Dan Herring explores the optimal dew point for running an endothermic gas generator — comparing the low dew point practices of decades past with today’s modern operating range of +40°F to +45°F — and explains how dew point, temperature, and air/gas ratio affect catalyst life, generator maintenance, and the stability of the process gas.
This informative piece was first released in Heat Treat Today’sJuly 2026 Annual Super Brands Issue print edition.
A critical consideration in heat treatment is the type, consistency, and control of the furnace atmosphere. The purpose of a furnace atmosphere varies with the desired end result, so it begs the question if there is an optimal dew point to operate an endothermic gas generator to balance gas quality, performance, and maintenance life. Let’s learn more.
Purpose of an Endothermic Gas Atmosphere
In general, furnace atmospheres are used:
To protect the components being processed from chemical reactions that could occur on their surfaces (e.g., oxidation or carburization), that is, to be passive (chemically inert) to the metal surface.
To interact with the surface of the component (e.g., adding carbon, or nitrogen, or both), that is, to be reactive (chemically active) to the metal surface.
Table A. Common Types of Furnace Atmospheres
There are many types of furnace atmospheres available for use in heat treating (Table A). By far, one of the most common is endothermic gas.
Endothermic Gas Atmospheres
Figure 1. Endothermic gas generator schematic piping arrangement | Image Credit: The HERRING Group, Inc.
Endothermic gas generators are common equipment in the heat treat shop, with one of the most well-known generators being the RX®. The main components of an endothermic gas generator (Figure 1) are relatively simple, consisting of:
Heated reaction retort with catalyst
Air-gas proportioning control components
Pump to pass the air-gas mixture through the retort
Cooler to “freeze” the reaction and prevent soot formation
Table B. Compositional Ranges for Endothermic Gas
Endothermic gas (aka endo) is produced when a mixture of air and fuel is introduced into an externally heated retort at such a sufficiently low air-to-gas ratio that it will normally not burn. The retort contains an active catalyst, which aids in cracking the mixture. Leaving the retort, the gas is cooled rapidly to avoid carbon reformation (in the form of soot) before it is sent to the furnace. The endothermic gas composition (Table B), by volume, varies depending on the type of hydrocarbon gas feed stock.
Endothermic gas is typically used for applications such as neutral hardening gas carburizing and carbonitriding (as a carrier gas), and for certain types of brazing to name a few. It is generally produced so that its composition is chemically inert to the surface of the steel and can be made chemically active by the addition of enrichment (hydrocarbon) gas that is usually added at the furnace.
What is the Optimal Generator Dew Point?
A reader asked this important question: “I’m doing some searching on endothermic gas generation and the advantages/disadvantages of low/high dew point set points. We normally run a dew point of +50°F in our generator and have had success historically. We installed a new generator and were told to drop the dew point to +40°F as this would extend the life. Do you have any experience, or can you point me to any references listing some of the pros and cons of running lower and higher dew points?”
Past Thinking
The operating philosophy in yesterday’s heat treat shop was to run the generators at a low dew point, typically in the range of +30°F to +35°F, so that gas (or air) additions at the furnace could be minimized. The belief was that the furnace atmosphere was much more responsive at this range and controlling certain processes (e.g., carburizing) was easier.
The nickel content of the catalyst years ago varied by manufacturer but was typically in the 5–7.5% range. Given the operating parameters used, it was mandatory to run weekly air burnout cycles to minimize soot formation in the catalyst bed, which took place quickly at these low dew points. The catalyst also needed to be changed on average about every 12 months. This dew point range was considered a good compromise for heat treat shops that ran both carburizing and neutral hardening.
As a parenthetical note, the very first endothermic gas generators (which were charcoal fired) used 100% nickel balls as a catalyst.
Current Thinking
Today, and for roughly the last 20+ years, the “modern” thinking is to run endothermic gas generators at temperatures in the range of 1900°F–2000°F, depending on the manufacturer’s design and the materials and size (diameter and length) of the retort. Air/gas ratios (for natural gas) should be between 2.5:1 to 3.5:1 (Figure 2). Ratios as low as 2.0:1 can be run at higher generator operating temperatures. Gas and/or air additions are then done at the furnace for better process control.
Figure 2. Typical endothermic gas generator control panel | Image Credit: The HERRING Group, Inc.
Experience has shown that a generator running at 1900°F with an air gas ratio between 3.0:1 and 3.5:1 and an output dewpoint of +42°F provides the best combination of a stable process gas and maximum generator life.
A generator dew point range held between +40°F and +45°F reduces maintenance on the generator and decreases the frequency of performing air burnouts of the catalyst bed, which at these dew points only needs to be done approximately once a month. It also extends the life of the catalyst (upwards of 2+ years) and makes it easier for the modern controls to regulate the system. Remember, however, that gas transmission from the generator to the furnace can often raise the incoming furnace dew point by +5°F to +10°F.
The advantage of performing air burnouts less often is that the nickel coated catalyst lasts longer. A major reason for this is that today’s catalyst averages around 3% nickel. The catalyst is insulating firebrick dipped in a nickel sulfate bath to allow nickel absorption on the outer surfaces of the cubes or spheres.
Miscellaneous Remarks
One must be very careful when running dew points at +50°F or above. In the writer’s experience, the gas is much more unstable, the water (moisture) content of the gas can rise quicker than one might anticipate, and you can literally start “raining water” inside the furnace or in the transmission lines. In literally minutes, the endo dew point can rise from +50°F to +70°F if one is not careful.
In Summary
The subject of endothermic gas generators never grows old. The key to their success is finding an operating dew point that provides both a stable process gas and little maintenance downtime. However, planned preventative maintenance is still required and should be performed on a schedule determined by the number and type of problems that arise.
A future column will discuss specifics of generator control, and we will broaden the subject by providing a guideline to the selection of furnace atmospheres in heat treating, brazing, and sintering. Look for it in next month’s publication.
References
Herring, Daniel H. 2015. Atmosphere Heat Treatment, Volume II, BNP Media Group.
Herring, Daniel H. 2009. “Furnace Atmosphere Considerations During Heat Treating”, Furnaces International, March/April.
Herring, Daniel H. “Understanding Furnace Atmospheres, Atmosphere Operation and Atmosphere Safety,” Heat Treating Hints, Vol. 1 No. 7. Manuscript forthcoming.
About the Author
Dan Herring “The Heat Treat Doctor®” The HERRING GROUP, Inc.
Dan Herring has been in the industry for over 50 years and has gained vast experience in fields that include materials science, engineering, metallurgy, new product research, and many other areas. He is the author of six books and over 700 technical articles.
Machine learning is poised to become a powerful tool for modern heat treatment, transforming decades of production data into actionable process intelligence. In this Technical Tuesday installment, Zachary C. Menard, CEO and co-founder of HRC Labs, draws on real-world case studies in carburizing and tempering to explore how predictive models can accelerate recipe development, optimize operations, and support more consistent decision-making. Rather than replacing metallurgical expertise, data-driven tools offer a pathway to continuous improvement, greater efficiency, and stronger competitive advantage.
This informative piece was first released inHeat Treat Today’sJuly 2026 Annual Super Brands Issue print edition.
Introduction
Heat treatment is one of the most operationally demanding and variable processes in modern manufacturing. It operates at the critical intersection of process design, material variability, and the strict performance requirements of industries as diverse as automotive, aerospace, and construction.
As a result, even highly experienced heat treaters must routinely make high-stakes decisions: how to converge on a new recipe without expensive trial-and-error; how to increase throughput without sacrificing quality; and whether a deviation warrants rework or scrap.
Heat treaters are uniquely positioned to benefit from advances in machine learning when confronting these decisions. Years of archived process data, hardness measurements, and quality records represent a largely untapped resource.
When systematically modeled, this data can power predictive models that map recipe parameters to metallurgical outcomes.
This approach is not without precedent. Similarly complex industrial processes — aggregates mining, cement manufacturing, and even steel casting — have increasingly adopted machine learning to reduce variability, optimize cycles, and proactively detect process drift. These gains have not replaced domain expertise; they have amplified it.
The following sections outline how machine learning, when thoughtfully integrated with metallurgical understanding and domain expertise, can serve as an enabling infrastructure for modern heat treatment.
What is Machine Learning?
Machine learning (ML) refers to computer algorithms designed to learn how a system behaves by leveraging that system’s data. Among the most relevant ML models for heat treatment are multilayer perceptrons (MLPs), which are designed to learn complex, nonlinear relationships between process inputs and observed outcomes.
Image Credit: HRC Labs
In heat treatment, outcomes such as surface hardness, effective case depth, distortion, or residual stress do not depend on any single variable. They emerge from the simultaneous interactions between variables like temperature, time, atmosphere composition, material composition, part geometry, and quench severity and uniformity, to name a few.
MLPs are well suited to modeling precisely this type of behavior. Rather than assuming a predefined relationship or physical law, they learn the mapping from process parameters to measured outcomes directly from historical production data. Layer by layer, the model captures increasingly subtle interactions among variables, enabling it to represent effects that are difficult to express with predetermined equations.
Importantly, once trained on sufficiently representative data, MLPs can:
Predict part properties for proposed recipes in milliseconds.
Evaluate hundreds of candidate parameter combinations rapidly.
Provide explicit uncertainty estimates alongside predictions.
Be tuned to individual furnaces, load configurations, or heat treatment techniques.
While MLPs are well suited to modeling the complexity of the heat treatment process, another class of machine learning models called decision trees are particularly effective for structured operational decisions. Heat treatment routinely requires categorical judgments, from rework decisions to detecting process drift to root cause analysis.
Decision trees can categorize heat treatment runs based on how they differ from historical runs. For example, the tree may organize a run based on its temperature deviations, fluctuations in carbon potential, or variation in effective case depth from the average. Each terminal node or “leaf” of the tree represents a cluster of historically similar runs.
Because these clusters are associated with prior corrective actions and results, the tree can provide probabilistic guidance grounded in historical precedent. For example, when analyzing a recent batch of parts that missed their target case depth, the tree might find that among comparable historical runs, 80% were successfully re-processed under a modified cycle, while 20% were scrapped. This enables faster root cause analysis and supports decision making.
Heat Treatment Applications
ML is particularly well suited to modeling and managing the many interacting variables that determine heat treatment outcomes. These capabilities translate into three practical applications: recipe development, process optimization, and auxiliary decision support.
Recipe Development
Heat treaters rely on operational experience, test coupons, and simulation tools to inform recipe design. However, new alloys, part geometries, or throughput demands often require iterative trial runs. Each additional load dedicated to experimentation represents furnace time removed from production.
Finite element simulations can narrow candidate recipes, but their accuracy depends on material properties, boundary conditions, and calibration to production conditions.
ML models such as MLPs offer a complementary approach to physical simulations. Trained on historical production data, they learn how a specific furnace, material, and process behave under real production conditions. Once trained, these models can rapidly evaluate candidate recipes against hardness, case depth, and distortion targets.
Importantly, predictions can be accompanied by uncertainty estimates, allowing heat treaters to assess risk alongside expected performance. By reducing the number of physical trial runs required to converge on a robust recipe, ML can accelerate the path from testing to production while preserving metallurgical oversight.
Process Optimization
Once a recipe has been demonstrated to meet specification, the focus shifts from feasibility to efficiency and robustness. The objective becomes reducing cycle time and energy consumption while maintaining process stability and required part properties.
In practice, optimization often proceeds through incremental parameter adjustments informed by experience. Soak times may be shortened or quench conditions modified in pursuit of efficiency gains. Yet the interdependence of heat treatment parameters makes it difficult to predict how such changes will affect process robustness.
ML models provide a structured framework for navigating this landscape. Once trained, they can:
Evaluate a multitude of potential changes to a recipe.
Estimate the impact of efficiency-driven adjustments on final properties.
Quantify uncertainty, identifying parameter sets that are not only efficient but stable.
Rather than relying solely on sequential physical trials, heat treaters can assess proposed modifications computationally before allocating furnace capacity.
Each additional production run further refines the model. As data accumulate across varying materials, geometries, and operating conditions, predictive fidelity improves. Over time, optimization transitions from periodic study to continuous, data-driven improvement.
Auxiliary Decision Support
Recipe design and optimization are only part of the operational challenge in heat treatment. A substantial portion of heat treaters’ time is devoted to diagnosing deviations, determining appropriate corrective action, and monitoring for process drift.
When a batch fails to meet specification, multiple correlated factors may plausibly explain the deviation. As a result, root cause analysis often requires time-consuming review of process logs and historical records.
Image Credit: HRC Labs
Decision-tree-based models provide a structured method for accelerating this analysis. By learning from historical runs and their documented outcomes, the model can classify a new deviation according to its similarity to past cases. For example, the model may determine that comparable historical runs were most frequently associated with temperature non-uniformity rather than alternative causes.
This approach offers three practical advantages:
Image Credit: HRC Labs
Faster root cause analysis and reduced diagnostic downtime.
More consistent, data-backed rework decisions.
Earlier detection of process drift and emerging equipment issues.
Importantly, these models do not override heat treaters’ judgment. Rather, they surface structured historical context that may otherwise remain fragmented across logs and institutional memory. Over time, this improves consistency, auditability, and knowledge transfer.
Machine Learning in Heat Treatment
The preceding sections have outlined several proposed applications of ML in heat treatment. The following sections examine documented cases of successful ML application in heat treatment.
Case Study 1: Carburizing
Jia et al. (2023) demonstrated the use of multilayer perceptrons to model the vacuum carburizing process for a low-alloy carbon steel cylinder. Using primarily simulation-derived data supplemented by limited experimental runs, the authors trained an MLP to predict case depth, surface carbon concentration, and surface hardness from process parameters including carburizing and diffusion temperatures, cycle durations, and carbon concentration levels.
Image Credit: HRC Labs
Once trained, the model rapidly identified parameter combinations that satisfied specified hardness and case depth targets while reducing computational cost relative to traditional simulation tools.
Zhang et al. (2025) took a similar approach but focused on atmospheric carburizing. The study’s goal was to train an MLP to predict post-carburizing carbon concentration across multiple part geometries for SAE AISI 3316 gear steel. Representative test coupons were used to validate the simulation-derived training data against production runs.
The model predicted post-carburizing carbon concentration with average errors near 2%. While limited in scope, both studies demonstrate that ML can replicate complex diffusion-driven behavior and accelerate recipe evaluation. These investigations focused primarily on carbon concentration and hardness prediction; the same modeling frameworks could be extended to distortion and residual stress prediction.
Case Study 2: Tempering
To further explore the performance of ML in heat treatment, publicly available tempering data from the classic study of Hollomon and Jaffe (1945) were analyzed. Their model combines temperature and time into a single tempering severity parameter, M. While effective within specific steels, prediction errors increased substantially as carbon contents varied. This model failed in particular to extrapolate to the behavior of lower carbon steels, with errors surpassing 10 HRC in some cases.
Image Credit: HRC Labs
To address this limitation, ML was used as a corrective layer on top of the traditional Hollomon-Jaffe model. The Hollomon-Jaffe equation provided baseline hardness predictions, while the ML model learned systematic corrections based on alloy composition and processing conditions. This approach reduced prediction error dramatically — from over 10 HRC in some cases to 2 HRC — while preserving the underlying metallurgical framework.
A common concern with ML models is that they behave like “black boxes.” However, modern interpretability tools allow engineers to examine how models arrive at their predictions. In this study, SHAP (SHapley Additive exPlanations) analysis was used to examine how the model corrected the Hollomon-Jaffe prediction. This revealed clear trends in how quantities like carbon concentration informed the model’s corrections. Rather than behaving like a black box, the model learned consistent metallurgical patterns in the data.
This case study illustrates how ML models can improve traditional approaches when they are limited by assumptions about physical parameters or cannot accommodate certain inputs. Unique furnace behavior, fixturing configurations, and part processing history are difficult to capture and calibrate using traditional simulation tools. Additional work should explore the integration of ML as a corrective layer for simulation software.
Strategic Implications and Competitive Advantage
The case studies presented here illustrate a broader principle: ML is most powerful in heat treatment not as a replacement for metallurgical understanding, but as a mechanism for formalizing and extending it.
When applied thoughtfully, ML can accelerate recipe development, reduce trial-and-error, and support structured diagnostics grounded in historical precedent. Unlike static optimization studies, trained models improve continuously as new production data become available. Over time, they become increasingly plant-specific, capturing nuances in furnace behavior, fixturing, material sourcing, and operating conditions that are difficult to encode with predefined equations.
Successful implementation will require high-quality process data and close collaboration between heat treaters and data scientists. ML is not a substitute for metallurgical expertise, but a tool for extending it.
The opportunity is substantial. Heat treatment has generated decades of process data that remain underutilized. Shops that convert this historical record into predictive intelligence will gain measurable advantages in efficiency, robustness, and process consistency. As competitive pressures increase and tolerances tighten, the ability to continuously learn from operations may become a defining differentiator.
Heat treatment has a rich history of innovation and experimentation. The integration of ML represents the next step in that progression.
Acknowledgements: The author would like to thank Dan Herring, The Heat Treat Doctor®, whose insight and guidance were instrumental in refining the direction of this work.
References
Hollomon, J. H., and L. D. Jaffe. 1945. “Time-Temperature Relations in Tempering Steel.” Transactions of the American Institute of Mining and Metallurgical Engineers 162: 223–49.
Jia, Hongliang, Dong Ju, and Jian Cao. 2023. “Machine Learning Based Optimization Method for Vacuum Carburizing Process and Its Application.” Journal of Materials Informatics 3: 9. https://doi.org/10.20517/jmi.2022.43.
Zhang, Y., Z. Tang, Q. Yin, et al. 2025. “Machine Learning Based Prediction of Carbon Concentration in Carburized Steel.” Scientific Reports 15: 33678. https://doi.org/10.1038/s41598-025-18531-8.
About The Author:
Zachary Menard CEO and Co-Founder HRC Labs
Zachary (Zach) Menard is the CEO and co-founder of HRC Labs, an industrial AI company focused on process optimization and decision support for heat treatment. He holds a B.S. in physics from Vanderbilt University with a background in computational physics and machine learning research.
Hardness results that fall below specification are not always caused by heat treatment errors. In this Technical Tuesday installment, Ana Laura Hernández, founder of Consultoría Carnegie, examines how decarburization develops, how it affects hardness and surface properties, and the practical methods heat treaters can use to evaluate, control, and prevent it through metallurgical analysis and process discipline.
This informative piece was first released in Heat Treat Today’sJuly 2026 Annual Super Brands Issue print edition.
In heat treatment operations, hardness results outside of specification are often attributed to deviations in process parameters, operational errors, or variations in the chemical composition of the material. However, there are cases where the root cause is related to metallurgical phenomena that directly affect the surface of the material.
A representative case involved a batch of ten forged components weighing more than 5 tons each, manufactured from AISI 4340 steel. The applied heat treatment process was known and previously validated: normalizing, quenching, followed by tempering in a range of 600–610°C (1112–1130°F), with the objective of meeting the required mechanical properties. After completing the heat treatment, the inspection team reported hardness values approximately 20–30 HB below specification. This result was unexpected, considering that AISI 4340 steel, due to its alloying elements, maintains relatively high hardness values even at relatively high tempering temperatures. A hardness reevaluation was requested, increasing the surface preparation depth up to 5 mm. However, the hardness values only increased slightly (~5 HB), without reaching the required levels.
Under pressure due to delivery timelines, it was decided to reprocess the components. The complete heat treatment cycle was applied again, adjusting the tempering temperature to 580°C (1076°F), which, from a metallurgical standpoint, should increase hardness. Nevertheless, the results remained practically unchanged. The chemical composition was verified using PMI, confirming that the material was indeed AISI 4340 steel. Additionally, the evaluation piece “Qualification Test Coupon” (QTC) indicated high hardness values, consistent with the expected behavior of the material.
Given the inconsistency between experimental results, metallurgical fundamentals, and prior experience, a deeper analysis of the process history was conducted, correlating variables from each stage of processing. This is where the key to a proper failure analysis lies: thoroughly understanding the history of the component.
It was identified that, prior to heat treatment, the components had remained inside a furnace within a temperature range of 800–900°C (1472–1652°F) for approximately three days, due to delays in the rolling process caused by maintenance issues. As a result of prolonged exposure and poor communication between shift operators, the components remained inside the furnace at “lower temperatures” while waiting to complete the final rolling step. Although not at rolling temperatures, this prolonged exposure in a non-controlled atmosphere promoted decarburization of the material.
To validate this hypothesis, a longitudinal cut was performed on one of the components, and hardness measurements were carried out in the core. The results confirmed that the material exhibited significantly higher values in the interior, revealing a decarburized surface layer. This case demonstrates the importance of: Analyzing the full material and process history before concluding a non-conformance.
What Is Decarburization?
Figure 1. Representative schematic of the decarburization and oxidation process. | Image credit: Consultoría Carnegie
Steel is a solid solution based on iron and carbon, where mechanical properties depend directly on the carbon content and the alloying elements present. Decarburization is a metallurgical phenomenon consisting of the loss of carbon from the surface of steel when it is exposed to high temperatures in atmospheres with oxidizing potential or low carbon potential (Figure 1). This phenomenon commonly occurs during forging, hot rolling, normalizing, annealing, and heat treatment in conventional furnaces without atmosphere control.
From a thermodynamic perspective, decarburization occurs when the chemical potential of carbon in the steel is higher than that of the surrounding environment, which drives carbon diffusion toward the surface. In the presence of oxidizing gases, carbon reacts to form gaseous species:
It can also react with water vapor or carbon dioxide, accelerating the process. As a result, a carbon gradient is generated (surface vs. core), the microstructure is altered, and mechanical properties are degraded, especially at the surface.
Oxidation and Its Relationship with Decarburization
Decarburization is closely related to oxidation processes. Oxidation occurs when oxygen reacts with the metal, forming oxides on the surface. This process depends on the thermodynamic equilibrium between the partial pressures of the gases present, which can be analyzed using diagrams such as the Ellingham-Richardson.
An increase in temperature accelerates both oxidation and decarburization, which is why the use of controlled atmosphere furnaces is essential to prevent these phenomena.
Process Kinetics: Carbon Diffusion
Decarburization is a diffusion-controlled process. When steel is at high temperatures, atoms gain higher mobility and carbon diffuses from regions of high concentration (core) to regions of lower concentration (surface). This process intensifies when the material is fully in the austenitic phase (above Ac₃), where carbon diffusivity is significantly higher.
The depth of the decarburized layer depends on temperature, exposure time, steel composition and furnace atmosphere conditions. This phenomenon is critical because the reduction of carbon content at the surface directly affects wear resistance and fatigue performance of the material.
Methods for Determining Decarburized Layer and Oxides
Prediction of Oxides Using Computational Tools
The use of tools like Thermo-Calc software allows the prediction of stable phases as a function of temperature and oxygen partial pressure. These diagrams enable identification of which oxides will form at the surface, evaluation of whether these oxides act as a barrier or facilitate carbon loss, and optimization of heat treatment conditions.
Figure 2. Thermodynamically stable phases formed during high-temperature oxidation as a function of oxygen activity for H11 steel at 600°C (1112°F) (Balaško, et al. 2021)
For example, in steels such as H11, phase stability diagrams allow visualization of how oxide formation varies as a function of temperature and oxygen activity (Figure 2). Understanding these relationships is key to preventing decarburization, designing controlled atmospheres, and avoiding undesirable mechanical property outcomes.
Determination of Decarburized Layer Depth
The ASTM E1077 standard establishes different methods to evaluate the depth of the decarburized layer, including microscopic methods, microhardness, and chemical analysis. These methods determine whether a component meets specifications and define the proper preparation required for hardness measurement. The following sections present an example of the use of these methods. The objective of this analysis was to determine the depth of this layer to generate internal instructions for proper sample preparation through grinding, ensuring accurate hardness measurements.
A QTC sample of AISI 4340 steel in forged condition was selected and subjected to normalizing, quenching, and tempering. Samples were obtained through cross-sectional cutting to perform hardness testing, optical microscopy, and carbon chemical analysis (by LECO, Figure 3).
Results
Hardness Profile
Measurements were taken at 1 mm intervals from the surface toward the interior. The results showed 0–3 mm low hardness values (~30 HRC) and an increase up to 35–36 HRC at 5 mm, remaining constant (Figure 4). This indicates that the surface region does not represent the actual properties of the material.
Figure 4. a) Hardness profile results, b) sample used for hardness profile analysis. | Image credit: Consultoría CarnegieFigure 5. a) Hardness evaluation zones every 1 mm, b) hardness results shown as a color map (blue represents the softer surface region and red represents the harder region at 5 mm depth). | Image credit: Consultoría Carnegie and Mikra QATM
One of the challenges in this method is the precision of hardness indentations. It can be observed in Figure 4b that a fine marker was used to define the measurement zones; however, nowadays the use of advanced equipment allows automated programming of measurement locations (Figure 5).
Chemical Analysis (LECO)
Figure 6. a) Carbon analysis results using a LECO system, b) sample used for carbon analysis. | Image credit: Consultoría Carnegie
The LECO analysis showed an increase in carbon percentage starting at approximately 5 mm depth, confirming the presence of a decarburized layer (Figure 6).
Optical Microscopy
Figure 7. Results of decarburized layer analysis using optical microscopy. | Image credit: Consultoría Carnegie
The optical microscopy analysis allowed observing microstructural changes between the surface and the core associated with carbon loss (Figure 7).
Practical Application: Shop Floor Control
Table A. Work Instruction in the Hardness Testing Area
Based on the results, work instructions were established to ensure reliable hardness measurements (Table A).
Operational Implementation
It is essential to involve operational personnel in understanding these phenomena, explaining why proper surface preparation is critical, how it impacts hardness measurement, and how this helps prevent rework. Additionally, it must be considered whether the dimensions of the components allow such preparation. Otherwise, it is necessary to communicate possible deviations or limitations to the client.
This analysis should be conducted for different steel grades, as well as at different temperatures and processing times, in order to estimate the effect on decarburization depth. This will facilitate the implementation of preventive actions for the operational team, instructing them to avoid overheating the parts or exposing them in the furnace for prolonged periods.
Conclusion
It is highly recommended that heat treatment companies conduct this type of metallurgical study, since it allows a deeper understanding of the real behavior of their processes. As previously mentioned, the depth of the decarburized layer is influenced by multiple variables, which may impact each company differently depending on factors such as steel type, furnace type, and processing time, among others.
Based on the results, it is possible to establish clear guidelines for operators, specifically regarding the depth at which samples must be prepared or ground to ensure that hardness values reported to the client are accurate and representative. Likewise, these analyses can be extended to different steel grades, allowing the development of specific work instructions indicating the required grinding depth depending on the material and processing conditions.
Decarburization is a critical phenomenon in heat treatment processes that can lead to incorrect interpretation of hardness results. A comprehensive analysis combining metallurgical fundamentals, standard methodologies, computational tools, and process history allows avoiding unnecessary rejection, improving process control, and ensuring final product quality.
References
Balaško, T., Vončina, M., Burja, J., Šetina Batič, B., & Medved, J. 2021. High-Temperature Oxidation Behaviour of AISI H11 Tool Steel. Metals, 11(5), 758. https://doi.org/10.3390/met11050758.
Herring, Daniel H. 2014. Atmosphere Heat Treatment: Atmospheres, Quenching, Testing. Vol. 2. Troy, MI: BNP Media.
Juan Hou, Fen-Fen Han, Xiang-Xi Ye, Bin Leng, Min Liu, Yan-Ling Lu, Xing-Tai Zhou. 2019. Effect of Surface Decarburization on Corrosion Behavior of GH3535 Alloy in Molten Fluoride Salts[J]. Acta Metallurgica Sinica (English Letters). 32(3): 401-412. https://doi.org/10.1007/s40195-018-0814-5.
Krauss, George. 2015. Steels: Processing, Structure, and Performance. 2nd ed. Materials Park, OH: ASM International.
About The Author:
Ana Laura Hernández Sustaita Founder Consultoría Carnegie
Ana Laura Hernández Sustaita holds a Master’s degree in Materials Science and engineering. She is the founder of Consultoría Carnegie, a technical consulting and training firm specializing in steel heat treatment in Mexico. Additionally, she works as a technical support engineer at Thermo-Calc Software, providing assistance to clients across México, Canada, and United States of America. Ana actively promotes metallurgical education throughout Latin America and advocates for the integration of computational tools into industrial heat treatment practice.
Los resultados de dureza inferiores a las especificaciones no siempre se deben a errores en el tratamiento térmico. En esta entrega de Technical Tuesday, Ana Laura Hernández Sustaita, fundadora de Consultoría Carnegie, analiza cómo se desarrolla la descarburización, cómo afecta a la dureza y a las propiedades superficiales, y los métodos prácticos que los técnicos de tratamiento térmico pueden utilizar para evaluarla, controlarla y prevenirla mediante análisis metalúrgicos y una estricta disciplina en el proceso.
Este artículo informativo se publicó por primera vez enHeat Treat Today’sJuly 2026 Annual Super Brands Issue print edition.
Introducción: cuando los resultados no coinciden con la teoría
En operaciones de tratamiento térmico, los resultados de dureza fuera de especificación suelen atribuirse a desviaciones en los parámetros del proceso, errores operativos o variaciones en la composición química del material. Sin embargo, existen casos donde la causa raíz está relacionada con fenómenos metalúrgicos que afectan directamente la superficie del material.
Un caso representativo involucró una carga de diez piezas forjadas de más de 5 toneladas cada una, fabricadas en acero AISI 4340. El proceso térmico aplicado era conocido y previamente validado: normalizado, temple seguido de revenido en un rango de 600-610°C (1112–1130°F), con el objetivo de cumplir con las propiedades mecánicas requeridas por el cliente. Tras finalizar el tratamiento térmico, el equipo de inspección reportó valores de dureza de las piezas aproximadamente 20-30 HB por debajo de la especificación. Este resultado fue inesperado, considerando que el acero 4340, debido a su contenido de elementos aleantes, mantiene valores elevados de dureza incluso a temperaturas relativamente altas de revenido. Se solicitó una reevaluación de dureza, incrementando la profundidad de preparación de la superficie hasta 5 mm. Sin embargo, los valores de dureza únicamente aumentaron ligeramente (~5 HB), sin alcanzar los niveles requeridos.
Ante la presión por los tiempos de entrega, se decidió reprocesar las piezas. Se aplicó nuevamente el ciclo completo de tratamiento térmico, ajustando la temperatura de revenido a 580°C (1076°F), lo cual, desde el punto de vista metalúrgico, debería incrementar la dureza. No obstante, los resultados permanecieron prácticamente sin cambios. Se procedió a verificar la composición química mediante PMI, confirmando que el material correspondía efectivamente a un acero 4340. Adicionalmente, la pieza de evaluación “Qualification Test Cupon” (QTC) indicaba valores de dureza elevados, consistentes con el comportamiento esperado del material.
Ante la inconsistencia entre resultados experimentales, fundamentos metalúrgicos y la previa experiencia personal, se decidió profundizar en el análisis del historial del proceso y correlacionar las variables de cada uno de los procesos. Y aquí es donde se encuentra la clave de un buen análisis de fallas, el detallar el historial de la pieza con la que se está trabajando.
Se identificó que, previo al tratamiento térmico, las piezas habían permanecido dentro de un horno en un rango de 800-900°C (1472–1652°F) durante aproximadamente tres días, debido al retraso que el equipo de mantenimiento tenía en la máquina de rolado, debido a esto y a una falta de comunicación entre los operadores de turno, la pieza continuaba en espera en el horno de calentamiento a “baja temperaturas” para concluir su último paso. Aunque no fue a temperaturas de rolado, esta exposición prolongada en una atmósfera no controlada promovió la descarburización del material.
Para validar esta hipótesis, se realizó un corte longitudinal en una de las piezas y se llevaron a cabo mediciones de dureza en el núcleo. Los resultados confirmaron que el material presentaba valores significativamente mayores en el interior, evidenciando una capa superficial descarburada. Este caso demuestra la importancia de: Analizar integralmente la historia del material y del proceso antes de concluir con una no conformidad.
¿Qué es la descarburización?
Figura 1. Esquema representativo del proceso de descarburización y oxidación. | Referencia: Consultoría Carnegie
El acero es una solución sólida basada en hierro y carbono, donde las propiedades mecánicas dependen directamente del contenido de carbono y los elementos aleantes presentes. La descarburización es un fenómeno metalúrgico que consiste en la pérdida de carbono desde la superficie del acero cuando este es expuesto a altas temperaturas en atmósferas con potencial oxidante o bajo potencial de carbono (Figura 1). Este fenómeno ocurre comúnmente durante: forja, laminación en caliente, normalizado, recocido y tratamientos térmicos en hornos convencionales sin control de atmósfera.
Desde el punto de vista termodinámico, la descarburización ocurre cuando el potencial químico del carbono en el acero es mayor que el del ambiente circundante, lo que impulsa la difusión del carbono hacia la superficie. En presencia de gases oxidantes, el carbono reacciona formando especies gaseosas:
También puede reaccionar con vapor de agua o dióxido de carbono, lo que acelera el proceso. Como consecuencia: se genera un gradiente de carbono (superficie vs núcleo), se altera la microestructura, se degradan propiedades mecánicas, especialmente en la superficie.
Oxidación y su relación con la descarburización
La descarburización está estrechamente relacionada con los procesos de oxidación. La oxidación ocurre cuando el oxígeno reacciona con el metal formando óxidos en la superficie. Este proceso depende del equilibrio termodinámico entre las presiones parciales de los gases presentes, lo cual puede analizarse mediante diagramas como el de Ellingham-Richardson.
El incremento de la temperatura acelera tanto la oxidación como la descarburización, por lo que el uso de hornos con atmósfera controlada es fundamental para prevenir estos fenómenos.
Cinética del proceso: difusión del carbono
La descarburización es un proceso gobernado por difusión. Cuando el acero se encuentra a altas temperaturas los átomos adquieren mayor movilidad y el carbono difunde desde regiones de alta concentración (interior) hacia regiones de menor concentración (superficie). Este proceso se intensifica cuando el material se encuentra completamente en fase austenítica (por encima de Ac3), donde la difusividad del carbono es significativamente mayor.
La profundidad de la capa descarburada depende de: temperatura, tiempo de exposición, composición del acero y condiciones de la atmósfera del horno. Este fenómeno es crítico, ya que la reducción del contenido de carbono en la superficie afecta directamente la resistencia al desgaste y fatiga del material.
Métodos para la determinación de la capa descarburada y óxidos
Predicción de óxidos mediante herramientas computacionales
El uso de herramientas como Thermo-Calc software, permite predecir las fases estables en función de la temperatura y la presión parcial del oxígeno. Estos diagramas permiten identificar que óxidos se formarán en la superficie, evaluar si los óxidos actúan como barrera o facilitan la salida del carbono y optimizar las condiciones del proceso térmico.
Figura 2. Fases termodinámicamente estables formadas durante la oxidación a alta temperatura (600°C [1112°F]) en función de la actividad del oxígeno para el acero H11 (Balaško, et al. 2021)
Por ejemplo, en aceros como H11, los diagramas de estabilidad de fases permiten visualizar cómo varían los óxidos formados en función de la temperatura y la actividad de oxígeno (Figura 2). Comprender estas relaciones es clave para prevenir la descarburización, diseñar atmósferas controladas y evitar resultados no deseados en propiedades mecánicas.
Determinación de la profundidad de capa descarburada
La norma ASTM E1077 establece diferentes métodos para evaluar la profundidad de la capa descarburada, incluyendo: métodos microscópicos, microdureza y análisis químico. Estos métodos permiten determinar si una pieza cumple con las especificaciones y definir la preparación adecuada para medición de dureza. A continuación, se muestra un ejemplo del uso de los métodos. El objetivo de este análisis fue determinar la profundidad de esta capa para generar instrucciones internas para la preparación adecuada mediante del pulido de las piezas para la medición correcta de dureza.
Metodología experimental
Figura 3. Preparación de QTC para extracción de muestras para análisis. | Referencia: Consultoría Carnegie
Se seleccionó un QTC de acero AISI 4340 en condición forja, sometido a un ciclo de normalizado, temple y revenido. Las muestras se obtuvieron de un corte transversal para realizar el análisis de dureza, microscopía óptica y análisis químico de carbono (LECO, Figura 3).
Resultados
Perfil de dureza
Se realizaron mediciones a intervalos de 1 mm desde la superficie hacia el interior. Los resultados mostraron: 0–3mm valores bajos de dureza (~30 HRC), a 5 mm: incremento hasta 35–36 HRC, manteniéndose constante (Figura 4). Esto indica que la zona superficial no representa las propiedades reales del material.
Figura 4. a) Resultados de análisis de perfil de dureza, b) muestra utilizada para el análisis de perfil de dureza. | Referencia: Consultoría CarnegieFigura 5. a) Zonas para evaluación de dureza a cada 1 mm b) resultados de dureza en mapa de color (en azul la zona más suave es la superficie y en rojo la zona más dura a 5 mm de profundidad). | Referencia: Consultoría Carnegie and Mikra QATM
Uno de los retos en este método es la precisión de las huellas de dureza, se puede observar como la muestra Figura 4b, se usó un marcador fino para determinar las zonas a medir, sin embargo, hoy en día el uso de equipos sofisticados nos permite realizar una programación automática de las zonas a evaluar (Figura 5).
Análisis químico (LECO)
Figura 6. a) Resultados de análisis de %C en un equipo LECO, b) muestra utilizada para análisis de %C. | Referencia: Consultoría Carnegie
El análisis mostró un incremento en el porcentaje de carbono a partir de aproximadamente 5 mm de profundidad, confirmando la existencia de una capa descarburada (Figura 6).
Microscopía óptica
Figura 7. Resultados de análisis de capa descarburada mediante microscopía óptica. | Referencia: Consultoría Carnegie
Este análisis permitió observar cambios microestructurales entre la superficie y el núcleo, asociados a la pérdida de carbono (Figura 7).
Aplicación práctica: control en planta
Tabla A. Instrucción de trabajo en área de pruebas de Dureza
Con base en los resultados, se establecieron instrucciones de trabajo para asegurar mediciones confiables de dureza (Tabla A).
Implementación operativa
Es fundamental involucrar al personal operativo en la comprensión de estos fenómenos, explicando por qué es crítico el pulido adecuado, cómo impacta en la medición de dureza y cómo esto evitará retrabajos. Así mismo, se debe considerar si las dimensiones de las piezas permiten dicha preparación. En caso contrario es necesario comunicar al cliente posibles desviaciones o limitaciones en la medición.
Se recomienda llevar a cabo este análisis para los diferentes grados de aceros, temperaturas y tiempos de procesamiento, que ayude a estimar el efecto sobre la profundidad de capa descarburizada y con esto implementar acciones preventivas al equipo operativo sobre evitar un sobrecalentamiento o someter las piezas a periodos prolongados de tiempo en un horno.
Conclusión
Es altamente recomendable que las empresas dedicadas al tratamiento térmico lleven a cabo este tipo de estudios metalúrgicos, ya que estos permiten comprender con mayor profundidad el comportamiento real de sus procesos. Como se mencionó previamente, la profundidad de la capa descarburada está influenciada por múltiples variables, las cuales pueden impactar de manera distinta en cada empresa, dependiendo de factores como el tipo de acero, el tipo de horno utilizado, los tiempos de tratamiento entre otros.
A partir de los resultados obtenidos, es posible establecer lineamientos claros para los operadores, específicamente en relación con la profundidad a la que deben preparar o pulir las muestras, con el objetivo de asegurar que los valores de dureza reportados al cliente sean representativos y correctos. Así mismo estos análisis pueden desarrollarse para distintos grados de acero, permitiendo la generación de fichas o instrucciones de trabajo específicas para cada caso donde se indique de manera precisa la profundidad de pulido requerida en función del material y las condiciones de proceso.
La descarburización es un fenómeno crítico en procesos de tratamiento térmico que puede llevar a interpretaciones incorrectas de los resultados de dureza. El análisis integral que combine: Fundamentos metalúrgicos, normativas estándar, herramientas computacionales, historial del proceso, permite: evitar rechazos innecesarios, mejorar el control del proceso y asegurar la calidad del producto final.
Referencias
Balaško, T., Vončina, M., Burja, J., Šetina Batič, B., & Medved, J. 2021. High-Temperature Oxidation Behaviour of AISI H11 Tool Steel. Metals, 11(5), 758. https://doi.org/10.3390/met11050758.
Herring, Daniel H. 2014. Atmosphere Heat Treatment: Atmospheres, Quenching, Testing. Vol. 2. Troy, MI: BNP Media.
Juan Hou, Fen-Fen Han, Xiang-Xi Ye, Bin Leng, Min Liu, Yan-Ling Lu, Xing-Tai Zhou. 2019. Effect of Surface Decarburization on Corrosion Behavior of GH3535 Alloy in Molten Fluoride Salts[J]. Acta Metallurgica Sinica (English Letters). 32(3): 401-412. https://doi.org/10.1007/s40195-018-0814-5.
Krauss, George. 2015. Steels: Processing, Structure, and Performance. 2nd ed. Materials Park, OH: ASM International.
Acerca de la autora:
Ana Laura Hernández Sustaita Fundadora Consultoría Carnegie
Ana Laura Hernández Sustaita cuenta con Maestría en Ciencia e Ingeniería de los Materiales, Es fundadora de Consultoría Carnegie, una firma de consultoría y capacitación técnica especializada en el tratamiento térmico de aceros en México. Asimismo, se desempeña como Ingeniera de Soporte Técnico en Thermo-Calc Software, brindando asistencia a clientes en México, Canada y Estados Unidos de América. Ana promueve activamente la educación metalúrgica en Latinoamérica y fomenta la integración de herramientas computacionales en la práctica industrial del tratamiento térmico.
In this installment of Answers in the Atmosphere, David (Dave) Wolff, an independent expert focusing on industrial atmospheres for heat treat applications, examines what furnace owners should expect when their gas supplier conducts a technical assessment. Drawing on insights from Messer LLC, Wolff walks through the step-by-step process suppliers use to diagnose furnace atmosphere issues, from verifying process parameters to running detailed diagnostic surveys, and outlines the standards of technical competence, communication, and accountability that furnace owners should hold their industrial gas partners to.
This informative piecewas first released in Heat Treat Today’sAugust 2026 Annual Automotive Heat Treating print edition.
This column picks up the conversation leveraging the technological know-how of your gas supplier. If you missed it, read Part 1 in the Heat Treat Super Brands (July 2026) print edition. What follows are the insights and best practices that Messer LLC shared with me on what you need to prepare for a technical assessment with your supplier.
Technical Assessment
First, since business models vary throughout the industry, be sure you understand whether or not technical support will entail a specific fee. When establishing your gas supply contract, furnace owners should seek out suppliers who provide a flexible approach that will be aligned with their needs.
A supplier will often follow these steps when providing technical support to identify, prevent, or correct issues in your heat treat furnace:
Define and verify the problem to determine whether it originates from heat treatment, material handling, or operator error.
Verify incoming material to ensure the issue is not caused by raw material condition or pre-processing steps.
Check heat treatment process control parameters.
Verify furnace atmosphere integrity, inspecting gas supply lines, furnace connections, piping leaks, deliveries, and any repairs.
Evaluate furnace component, identify any changes in furnace structure or maintenance activities.
Review process parameters to confirm whether temperature, time, gas composition, or gas purity have changed.
Inspect exhaust systems, doors, and airflow, as disturbances can impact furnace performance.
After reviewing initial process data, a furnace atmosphere survey or targeted spot check may be conducted. This step may include:
Establishing furnace baseline conditions
Running furnace temperature profile tests
Measuring oxygen, CO, CO2, hydrogen content, and dew point
Conducting detailed modeling based on gas flows, heat transfer, or mass balances
Your provider may use advanced tools, such as Computational Fluid Dynamics (CFD) modeling, to analyze and resolve complex furnace atmosphere issues.
What to Expect: Evaluating Supplier Capabilities
Not all technical support delivers the same value. While many suppliers have strong technical capabilities, access can vary based on internal structures or business models. Some suppliers provide expanded support.
Furnace owners should expect their industrial gas partner to:
Demonstrate strong technical competence and ability to identify root causes
Communicate clearly in practical terms familiar to operations teams
Deliver recommendations with a complete implementation plan including safety, training, costs, timelines, risks, and benefits
Clearly define responsibilities between the furnace owner and gas provider
Final Thoughts
For furnace owners, the industrial gas supply space can seem like a lot of the same offerings, but a closer look at how technical support varies from one supplier to the next can significantly fine tune your operation’s ability to adapt and respond to your furnace needs. The key is to be honest about what your own operations can accomplish on its own before reviewing industrial gas contracts (more on contracts in the June 2026 “Answers in the Atmosphere” installment) and committing to a too low or too flexible plan for technical support.
About The Author:
David (Dave) Wolff Industrial Gas Professional Wolff Engineering
Dave Wolff has over 40 years of project engineering, industrial gas generation and application engineering, marketing, and sales experience. Dave holds a degree in engineering science from Dartmouth College. Currently, he consults in the areas of industrial gas and chemical new product development and commercial introduction, as well as market development and selling practices.
A small amount of hidden water at the bottom of a quench tank can rapidly escalate into a violent fire hazard. In this Technical Tuesday installment, Bruno Scomazzon, general manager of Precision Heat Treat Ltd., discusses how free water develops, why conventional testing can miss it, and the practical steps heat treaters can take to detect and eliminate the risk.
This informative piece was first released inHeat Treat Today’sJuly 2026 Annual Super Brands Issue print edition.
Most heat treat shops never see the danger building beneath the surface of their quench oil. Drop by drop, water stratifies, tipping the balance and awakening the dragon.
In an integral quench furnace, quenching is a controlled process. When a hot load is submerged, vapor is generated below the surface, rising and mixing with the furnace atmosphere. Oil temperature, atmosphere, and vapor generation are managed so that any combustion remains contained, with gases and vapor generated and relieved in a controlled manner. Under normal conditions, the process is stable and predictable.
AI-generated illustration based on a potential quench fire scenario. No actual integral quench furnaces were harmed in the making of this article. | Image Credit: Precision Heat Treat Ltd.
Add water, and you introduce a completely different hazard. It can turn violent before you understand or can react to what you are seeing. In this scenario, when the hot load is submerged, water at the bottom of the tank flashes instantly to steam, expanding roughly 1,600 times in volume. That expansion happens almost instantly, and the resulting increase in volume overwhelms the system.
As the steam rises, each bubble becomes coated with oil. Rapid expansion displaces oil and generates a large volume of vapor in a very short period of time. In a confined quench chamber, that surge carries oil and vapor together toward the burn-off vent and the doors.
There are typically two doors in the system, and they behave very differently. The inner door separates the hot zone from the quench tank. During quenching, the hot zone is typically operating around 1550°F. If oil is forced into the hot zone, it will vaporize and burn, generating products of combustion that can lead to a more severe internal event. However, this is not where the external fire develops. The outer door separates the quench chamber from the outside. During a surge, oil and vapor push upon this door and can be forced out. When the flammable oil vapor and furnace atmosphere reach an ignition source such as the flame curtain pilot, it will ignite violently.
At that point, the fire is no longer confined or controlled. Oil that reaches the exterior can spread along the floor and around the base of the furnace. Once outside the chamber, any oil present becomes fuel, and the fire can spread quickly.
How Water and Oil Interact
Two different conditions exist: dissolved water and free water. Dissolved water is moisture within the oil, typically when the oil is hot. It is part of normal operation and must be monitored. Elevated dissolved water levels are mainly a performance and control problem, not the immediate hazard. But if it continues to rise, it leads to free water, which is the hazard. Free water is water that has separated from the oil and settled to the bottom of the tank.
As a general guideline for dissolved water:
Below 100 ppm (0.01%): very good
100–200 ppm (0.01–0.02%): acceptable
200–500 ppm (0.02–0.05%): caution range
Above 500 ppm (0.05%): corrective action required
These values apply to dissolved water in the oil — not the free water condition at the bottom of the tank, which is the condition of greatest concern. As the oil cools, its ability to hold moisture decreases. Excess water comes out of solution, forms small droplets, and over time settles to the bottom. This creates stratification — the formation of distinct layers. Oil floats above, leaving a layer of free water at the bottom. That bottom layer is the dangerous condition, the same condition that drives the surge event described earlier.
Water Does Not Just Appear — It Gets Introduced
Common water sources include condensation during shutdowns and startups, along with operator or maintenance oversights that allow water or contaminated oil to enter the system. On older units, water-cooled components can develop leaks over time, allowing water to enter the oil slowly and go unnoticed. Leaking roof components or failed hood caps can also allow water to reach the furnace and make its way into the tank.
Water does not always find its way into a quench tank through a leaking cooler, heat exchanger, or outside source. During extended shutdowns, particularly during cooler months or periods of high humidity, moisture can condense on the ceiling and sidewalls of the quench vestibule and tank area. As the furnace heats up and the colder oil tank lags behind, water droplets can form and eventually fall into the oil. | Image Credit: Precision Heat Treat Ltd.
Because this develops gradually and often out of sight, everyone in the shop needs to stay alert. If something does not look right, or behave abnormally, report it immediately.
Detecting Water: What the Operator Sees First
Detecting free water in a quench tank is not as straightforward as it sounds. In many cases, the first indication is not a test result but a change in furnace behavior.
Changes in burn-off flame height, flame recovery time, unusual oil discharge (“burping”), or abnormal sounds during quenching are often early warning signs that something in the system has changed. All furnace operators should be trained to recognize these changes and treat them as indicators that the condition of the quench may no longer be normal.
Confirming with Sampling and Testing
Most monitoring systems measure dissolved water in circulating oil — not free water at the bottom. That distinction matters. Water that is mixed in the oil can be measured and trended. Water that has separated and settled to the bottom may not be detected by standard sampling methods. A sample taken from a circulating line or mid-depth in the tank can show acceptable results while free water remains undetected. These values are typically determined through lab analysis or in-shop test kits that measure dissolved water in quench oil. Standard tank sampling does not distinguish between dissolved and free water. That is why where you take the sample matters as much as how you test it.
Bottom sampling is critical. Pulling oil from the lowest point in the tank after the furnace has been idle, such as over a weekend, is often the best way to identify free water. If you do not already have a way to sample from the bottom of your tank, you should plan to install a dedicated drain or sample port. A bottom sample should be part of your weekly oil monitoring.
In the absence of a dedicated bottom drain, a simple method can still be used. Tubing can be inserted down through the fill or access point until it reaches the bottom. By sealing the top of the tube, the oil column inside is held in place as it is withdrawn, allowing a sample from the lowest point in the tank. The sample must be taken before any agitation or circulation begins. If the oil has been disturbed, the water can be temporarily mixed and the true condition at the bottom may not be seen.
The sample is placed in a clear glass beaker and allowed to stand. If water is present in any significant amount, it will usually be obvious, especially when compared to a sample taken from mid-depth or circulating oil. If it is not obvious, a crackle test can be used as a quick field check. A small sample of oil is placed on a hot surface, typically around 300–350°F. If free water is present, it flashes to steam and produces visible bubbling or crackling.
In practice, the severity of the reaction gives a clear visual indication of the condition:
No reaction: acceptable, continue to run
Light fizz or fine bubbles: trace water present, monitor
Moderate crackle or popping: plan corrective action
Strong crackle or aggressive bubbling: correct the condition before continuing
A consistent reaction across samples is the key indicator. This is not a precise measurement, but it is a reliable field guide. If it is reacting hard, you are already past where you want to be.
Sending samples for quarterly lab analysis provides a more complete picture of oil condition, including water content, oxidation, viscosity, contamination, and quench performance. This helps track dissolved water levels and overall oil health. In-shop test kits and commercial monitoring systems are also available, but none replace the need to understand what is happening at the bottom of the tank.
Removing Free Water from the Quench Tank
When free water is present at the bottom of the tank, it must be removed. This is an immediate hazard. In practice, removing oil from the top-down is often the most controlled approach. Oil is siphoned from the surface, working downward and stopping about a foot above the bottom to avoid disturbing the settled water layer.
This allows clean oil to be removed first while leaving the water undisturbed. Attempting to remove water directly through a bottom drain is not always effective, particularly in larger tanks, as it can pull both water and usable oil and disturb the interface between the two.
Left: Accumulated oil sludge, soot, and scale deposits inside a quench tank. Right: The same area after cleaning. These deposits can build up over time and provide fuel for a fire. Regular tank cleaning is an often-overlooked part of quench oil stewardship and fire prevention. | Image Credit: Precision Heat Treat Ltd.
The removed oil can be placed into totes and allowed to sit undisturbed for several days so any remaining water can separate and settle. Clean oil can then be recovered from the top. The remaining oil and water mixture in both the tank and tote should be recycled. Other methods, such as vacuum dehydration or oil reclamation systems, can be used where available and are often more effective at removing both free and dissolved water, but are not always practical in every shop.
With the oil removed, this is an ideal time to carry out thorough tank cleaning. Over time, quench oils form sludge, a combination of oxidation byproducts, degraded oil, carbon, scale, and fines from processed parts. This material settles to the bottom, can trap and hold water, and hide it from normal sampling. As it builds up, it interferes with oil flow and agitation, affecting quench performance.
Tanks require periodic cleaning, typically every 12 to 18 months, depending on usage and condition. This is the time to remove sludge and clean deposits from the walls and ceiling.
It is also an opportunity to inspect agitation systems, elevators, rollers, and other components, and carry out preventative maintenance. Proper lockout procedures and confined space protocols must be followed.
The Bottom Line
Free water is an immediate hazard. If allowed to accumulate, it can trigger a rapid pressure event and an uncontrolled fire. Once it starts, it escalates quickly and is difficult to contain, putting personnel and the entire operation at risk.
Regular bottom sampling for free water must be part of your quality control.
Acknowledgements
The author would like to thank Daniel H. Herring, “The Heat Treat Doctor®” at The HERRING GROUP, Inc.
About The Author:
Bruno Scomazzon General Manager Precision Heat Treat Ltd.
Bruno Scomazzon is the general manager of Precision Heat Treat Ltd. in Surrey, British Columbia, Canada, with over 40 years of experience in metallurgical processes and heat treating operations.
Jim Roberts of U.S. Ignition engages readers in a Combustion Corner column about the basics of heat transfer — breaking down the First Law of Thermodynamics into practical terms for heat treaters, then using a real-world example to show how ambient load temperature can meaningfully shift BTU energy requirements and furnace performance.
This column was first released in Heat Treat Today’sJuly 2026 Annual Super Brands Issue print edition.
A furnace guy walks into a heat treat plant and says to the group of operators, “I just transferred here.” One of the furnace operators says, “Perfect, it’s what we do best.” Huh…? Well, of course they transfer heat. That’s what heat treaters do better than anyone else — we transfer heat. And the science of this transfer is called thermodynamics.
In the world of physics, there are four laws of thermodynamics, which are centered on the movement and flow of heat between objects. We’ll start with the concept of heat transfer, based on the First Law of Thermodynamics.
The concept of thermal conservation states that energy cannot be created or destroyed; it can only be transferred or transformed. In other words, whatever we put into the furnace in the form of heat will be the same amount that comes out or is absorbed. Additionally, the part we want to heat treat has thermal mass and therefore has heat as well. We say that the load is “cold,” but really, it’s normally coming into the process at room temperature, which means there is energy there that gives us a head start in heating it up.
Then, we know that the heat we feel from the shell of the furnace was not absorbed by the load but is part of the energy that we put into the furnace, it just wasn’t absorbed by the load. So, the heat exits the furnace into the room to be absorbed by other items that are not in equilibrium. The energy is always there, continuing on. It’s a wild concept, isn’t it?
Figure 1. The First Law of Thermodynamics | Image Credit: Jim Roberts
As shown in Figure 1, heat enters the furnace (designation “Q”). Heat then enters the work, which is designation “W” (e.g., load, furnace). Work absorbs most of the heat, but it also releases energy since it is starting to go towards a state of equilibrium, meaning heat in and heat released are equal. Then, the work releases energy into the area where it is not as hot and tries to heat it up and gain equilibrium. That’s the furnace guy standing there, the room, the building, etc. All of these things become the next stage of “work.”
So, when we get to the point of calculating the input (energy usage), we generally use BTU or KW ratings. We also must consider ambient temperature of the work because that Delta T, or temperature variance, is what we are having to account for. If that load is sitting at 70°F, it has value as a heat source, so we need to account for that. You will recall that the formula that is commonly used for calculating heat load is:
This will give you BTU requirements after you then apply an efficiency. Sometimes that’s an estimated efficiency. Let’s show the difference in that energy requirement that needs to be provided when the latent heat in the load is different.
Let’s suppose we are a heat treater in central Michigan. It’s December. We have been accustomed to staging our bulk parts for heat treating out on our open loading dock. The furnace is suddenly not performing like it did earlier in the year. It’s the same 1,000 lb load. Earlier in the year, our formula accounted for the 70°F load temperature coming in. Our equation would be:
In this example, if we bring the work in from the frozen loading dock at 20°F, the heat required jumps to 401,231 BTU energy required per hour. It’s not a lot, but the furnace will notice and not perform as well since the burners tend to run at a fixed setting.
Even slight variations can make a big difference in cost and performance. Simple and yet slightly confusing science is behind it all.
About The Author:
Jim Roberts President US Ignition
Jim Roberts president at U.S. Ignition, began his 45-year career in the burner and heat recovery industry focused on heat treating specifically in 1979. He worked for and helped start up WB Combustion in Hales Corners, Wisconsin. In 1985 he joined Eclipse Engineering in Rockford, IL, specializing in heat treating-related combustion equipment/burners. Inducted into the American Gas Association’s Hall of Flame for service in training gas company field managers, Jim is a former president of MTI and has contributed to countless seminars on fuel reduction and combustion-related practices.
