Welcome to Heat Treat Today’sThis Week in Heat Treat Social Media. We’re looking at a new manufacturing center, an AFC-Holcroft farewell, fun heat-treatment quizzes, and more.
As you know, there is so much content available on the web that it’s next to impossible to sift through all of the articles and posts that flood our inboxes and notifications on a daily basis. So, Heat Treat Today is here to bring you the latest in compelling, inspiring, and entertaining heat treat news from the different social media venues that you’ve just got to see and read!If you have content that everyone has to see, please send the link to editor@heattreattoday.com.
1. New Advanced Manufacturing & Aerospace Center at UTEP
Check out the exciting inauguration of UTEP’s Advanced Manufacturing and Aerospace Center (AMAC). On the website, heat treatment is listed as one of the future undertakings of the AMAC. Some new heat treat trainees may be coming your way out of UTEP!
2. Goodbye is The Hardest Word
You know someone is special if they are with you for almost 50 years! Congratulations to Jerry Waineo from AFC-Holcroft on his retirement. We hope you ate an extra slice of cake for us here at Heat Treat Today!
3. Try Your Hand at Heat Treat Quizzes
If you’re ever looking for something light during your work week, Paulo Heat Treating, Brazing and Metal Finishing posts fun heat treat related quizzes on their LinkedIn page on a regular basis. Each quiz shows how the respondents on LinkedIn have answered.
4. We Love Seeing Friendly Faces
Oh, how we do enjoy seeing all of our friends at conferences and tradeshows. We have always said, “I get by with a little help from my friends.” (Well, it may have been the Beatles, but we agree!)
5. Scary: PFAS. NOT Scary: Heat Treat Radio
Tune in to Listen to Heat Treat Radio #119: Solvent vs. Aqueous Cleaning: Choosing the Best Method for Your Process. This helpful information shared by Fernando Carminholi on Heat Treat Radio, will keep you well informed!
Three endeavors are underway to deliver advanced roller hearth furnace technology, featuring large-scale atmosphere quench and temper, annealing, and solution treatment systems to support the automotive and energy industries.
CAN-ENG Furnaces International, Ltd. is engaged in providing three separate systems amounting to over 450 feet of roller hearth furnace capacity.
Tim Donofrio Vice President of Sales Can-Eng Furnaces International, Ltd. Source: Can-Eng Furnaces International, Ltd.
The first of the three undertakings is a high capacity, roller hearth lamination annealing system. Produced for the transformer manufacturing industry, it will be supplied to one of North America’s largest transformer core manufacturers with operations in the USA, Canada, Mexico and China. The lamination annealing process is applied to transformer cores to enhance their magnetic properties and reduce core loss, ultimately improving efficiency and performance. The system integrates a pre-heat system, high temperature annealing furnace with integrated protected atmosphere-controlled cooling and accelerated cooling chamber. It is more than 300ft long and is capable of producing over 14,000 lbs/hr of atmosphere protected electrical steel core laminations used to manufacture high voltage transformer stations needed in the development of North America’s electrical distribution infrastructure.
“This project is being carried out largely to satisfy the demand for improving America’s aging and overburden power distribution network. As part of improving the nation’s power distribution network, consideration is being made to increase access to electric vehicle charging stations of which transformer cores are a critical part,” shared Tim Donofrio, Vice President of Sales, Can-Eng Furnaces International, Ltd.
The second endeavor is a roller hearth solution treatment furnace system for a tier one supplier of cast aluminum automotive components. The furnace will be used as part of a continuous T-6 heat treatment system for the manufacturing of cast aluminum, safety critical suspension components, and will serve to supply for one of America’s largest Japanese-based automotive manufacturers. These suspension components are to be integrated into future hybrid and electric automobiles produced in America by the manufacturer.
The third project is a roller hearth atmosphere mesh belt furnace system for a tier one supplier of safety critical, high value fasteners. This system is used to atmosphere, quench and temper high value, safety critical fasteners used in automotive manufacturing. The system is rated for processing over 6000 lbs/hr, and is being supplied to an existing customer with four similar roller hearth furnaces. The system supports the ongoing demands for domestic-supplied, high volume, heat treatment capacity.
What is missing from induction heat-treating maintenance? Learn seven methods for improving your induction tooling component performance in today’s article by David Lynch, Vice President of Engineering at Induction Tooling, Inc.
This informative piece was first released inHeat Treat Today’sApril 2025 Annual Induction Heating & Melting print edition.
The heat-treating industry is constantly evolving, whether it is due to the influx of AI or the introduction of new materials. The field of induction is not exempt from this constant change. Yet there remains a constant — induction tooling components need to be tough to resist harsh environments comprised of high frequencies, high power, heat, smoke, steam, dirt, oil, quench fluid and additives, and contaminants. It’s been almost four years since we visited the topic of induction tooling equipment longevity and maintenance (see the May 2021 print edition). Amidst the constant change, how do we protect against the same old toxic environment?
Let’s explore some new methods of improving the performance and longevity of induction tooling components:
Figure 2. Break-Away bolts designed to fail beneath the washer if over tightened
More than coils — When working to optimize the life of induction equipment, don’t focus solely on the coils. Bus bars, inductors, and quenching equipment are also key to success.
Austenitic stainless steel — Use austenitic stainless steel for fasteners, fittings, and hose clamps, and remember, non-ferrous is the way to go.
CNC machining — Manufacturing with a 5-axis CNC machine ensures quality and consistency.
“Break-Away” bolts — For fasteners, use “Break-Away” bolts on contact surfaces. These bolts are designed to fail beneath the washer if they are overtightened, a design that prevents damage to the threaded insert inside the copper contact.
Cooling water — For cooling the inductor coil, bus bars, and adapters, reverse osmosis and distilled and deionized water are overkill. Stick with keeping the water below 70°F. This may require a separate cooling supply. Through laboratory experimentation and real-world production trials, it has been proven that lower cooling water temperatures can drastically increase the life of these components, especially in high-volume, high-power, and short cycle applications. In some hard water areas, this may not be possible. Typical cooling-conductivity for the inductor and bus bar is 200–800 microsiemens per centimeter (μS/cm).
Non-ferrous fittings — Use non-ferrous fittings on cooling and quenching water connections, as well as color-coded hoses.
Cleaning — Design with cleaning in mind. Designing a quench with bolted removable quench plates ensures easy clean out. As the heat-treating industry continues to evolve, our practices and technologies for optimizing the performance and longevity of induction tooling equipment evolve with it. Whether it’s using a new method or revamping a tried-and-true practice, we can continue to produce strong induction tooling components to sustain these harsh environments.
Figure 3. 5-axis CNC machining of water-cooling passages to maximize water flow and minimize sharp transitionsFigure 4. CVJ inductor with non-ferrous fittings and color-coded hoses on cooling and quench water connectionsFigure 5. Barrel style OD quench ring with bolted removable quench plates to allow easy clean outFigure 6. Multi-turn OD scanning inductor with integrated quench featuring removable quench plate for easy clean outFigure 7. Heavily used inductor
About The Author
David Lynch Vice President of
Engineering Induction
Tooling, Inc.
David Lynch is Vice President of Engineering at Induction Tooling, Inc. He has over 36 years of experience and is the deputy of the ISO quality system. He has created and developed the system and templates being used today for creating and tracking engineering drawings, job history, rate tracking, and job performance. David holds several design patents, has authored several published articles, and has often presented at technical sessions. He enjoys working closely with customers to develop valued solutions across a wide range of induction heating applications from initial design concepts to implementation, customer support, and troubleshooting.
This article showcases the utilization of data-driven modeling assisted by machine learning (ML) to simulate the melting process of different steel grades in a medium-frequency induction furnace concerning the target variable of energy consumption. The results of the predictive models developed are presented in this article, along with the possibility of producing optimized charge combinations through the use of predictive outcomes and backward analysis.
This informative piece was first released inHeat Treat Today’sApril 2025 Annual Induction Heating & Melting print edition.
Introduction
The steel and foundry industry faces several ongoing challenges, such as escalating costs of raw materials and energy, CO2 emission regulations, and fierce global competition. To tackle these challenges, there is a continual demand for improving current production processes. Despite the strides made by smelters and equipment manufacturers in enhancing plant technology, there is still room for improvement in production procedures and processes. One possible approach to enhancing these processes is through modeling.
In recent times, digitalization and machine learning (ML) have emerged as a promising modeling method, as demonstrated in material development or process optimization in the steel industry (Lee et al., “A Machine-Learning-Based Alloy,” 11012; Klanke et al., “Advanced Data-Driven Prediction,” 1307-1313; Yingjun et al., “A Machine Learning and Genetic,” 360-375.) The subsequent example showcases how the melting process in a medium-frequency induction furnace can be modeled with respect to energy consumption, using specific process data obtained from a steel foundry, and subsequently optimized through synthetic data generation and backward analysis.
Modeling
Various ML algorithms [including Random Forest, Extra Trees, LightGBM, XGBoost, MLP (Neural Network), K-Nearest Neighbors] were trained and their hyperparameters optimized for modeling Erikson et al., “Autogluon-tabular”. The hyperparameters are set before the learning process begins and influence how well a model can represent a process. Some examples of hyperparameters in ML are the learning rates, the number of hidden layers in a deep learning neural network, or the number of branches in a decision tree.
For the modeling, data from the process, induction furnace, and charge database of a medium-sized steel foundry were consolidated and pre-processed. The melting unit is a medium-frequency induction furnace with a capacity of approximately 7 tons. Figure 1 depicts an overview of the modeling workflow. The process and furnace databases contain data, such as the chemical analysis of the input raw materials, the measured melting time, and the measured melting energy requirement of the respective melts. The required melting energy (kWh), the required melting time (min) to reach the required tapping temperature, and the alloy quantities (Co, FeCr, FeV, FeSi, etc. kg) to be added after the melting process to correct the chemical analysis were selected as target values. The charge database contains data on the charge scrap for each melt, with details of its quantity and chemical analysis.
At the foundry, the scrap is roughly pre-sorted in separate bins so that the scrap used in a batch can be distinguished. As an intermediate step, the expected melting enthalpy of the scrap was calculated from the chemical analyses using simulation software (CALPHAD method). Based on the thermodynamic data, an ML model was trained to also consider the theoretically required melting enthalpy of the raw materials based on the chemical analysis. The measured raw data consist of approximately 10,000 individual melts of different steel grades. After pre-processing (removing outliers, formatting the data, etc.) and applying domain knowledge, about 70% of the data was used for process modeling. Domain knowledge in this context means reviewing and filtering the data with an understanding of the process. For example, some obvious outliers or erroneous data, such as negative charge quantities or charges of allegedly more than 7 tons of material, were not detected by the pre-processing algorithms and were manually filtered out.
Figure 1. Process modeling of the melting process
*The ML Models are a function of the change of energy, melting time, and chemical composition of the ferro alloys
The remaining dataset contains information on approximately 7,000 individual steel melts with approximately 300 influencing variables (columns in the dataset influencing variables, also called “features”) of which approximately 200 are charge-relevant influencing variables. In addition, the steel scrap was divided into groups, such as recycled or foreign scrap and alloys, characterized based on its empirically assessed geometry (cut-offs, “bones,” plates, chipped scrap, etc.), and added to the dataset as information.
Table 1. Overview of the prediction metrics of the energy models for 1.2379 steel
The data were subsequently split as 70% training, 20% validation, and 10% test data. The algorithms were trained on the training data and then tested for prediction quality on the test data. This step is necessary because overfitting or underfitting can occur when training ML algorithms. This means that the models perform well on the training data but poorly on unknown future process data. The test data, representing completely unknown future process data or states, are separated beforehand. In each case between the prediction of the model and the “real” measured value of the respective melt in the test dataset, the prediction of the models was evaluated with the metrics MAE (mean absolute error), MSE (mean square error), RMSE (root mean square error), and R2 (coefficient of determination).
The energy consumption during melting in an induction furnace depends on many factors, such as the raw materials being fed (fine or bulky material, impurities, etc.), the sequence in which they are charged, the actual process control (e.g., is the furnace lid open for an unnecessarily long time due to bulky scrap in the charge), and many other direct and indirect influencing factors. Due to this complexity, modeling
Figure 2. Result of the process model in terms of predicting the energy required to melt 1.2379 steelFigure 3. Residuals of the energy model test data for 1.2379 steel
Figure 2 depicts the comparison of the values (1.2379 steel) of energy consumption (x-axis) predicted by the best ML model based on the input features and the actual measured values (y-axis) in the test and training datasets, respectively. For the training dataset the R² is 0.92 with an RMSE of 58 kWh, and for the test dataset, the R2 is 0.69 with an RMSE of 119 kWh (Table 1). For the test data, this corresponds to a relative prediction error of about 5%–10%.
Figure 3 illustrates the residual distribution (difference between the actual measured value and the model prediction) of the 1.2379 energy model in the test dataset. The residual distribution gives an indication of the prediction quality of a model. If the expected value of the residuals is not close to zero and they are not approximately normally distributed, this means that the model has a systematic tendency to either over- or under-predict. Furthermore, if there is a pattern in the residuals, the model does not appear to be able to explain some relationships within the data and is therefore qualitatively inconsistent. In the generated model, the residuals are almost normally distributed, and the prediction error does not appear to follow any pattern, suggesting good prediction quality.
Charge Optimization
After a trained ML model has been prepared for use, backward analysis can be applied to the training dataset to determine the range of values of the independent variable that corresponds to a given target variable. Consequently, backward analysis offers an inverse function of the prediction function, which can be leveraged to determine optimized process values. In this scenario, the optimized process value is the charge composition, with the target variable being energy consumption.
There are multiple mathematical optimization methods that can be employed for this purpose based on a well-trained ML model. A straightforward and easy-to-understand approach is to create a dense set of independent process variables using linear interpolation within a given range, such as the minimum and maximum values of a variable. The target variable is subsequently predicted based on this set of generated variables. This method can be computationally intensive and time consuming, and it does not consider the hidden patterns within the dataset, resulting in some useful information being disregarded.
In order to capture the hidden information and accurately reflect the true value of the original data set, a deep learning-based method called SDV (Synthetic Data Vault) is used in this work Patki et al., “The Synthetic Data Vault,” 1-10. Various synthetic data generation algorithms, such as Gaussian copula, are used in the SDV library. Mathematically, a Gaussian copula is a distribution over the unit cube between 0 and 1 in dimensions generated by applying the probability integral transformation to a multivariate normal distribution over all real numbers (R). Intuitively, the Gaussian copula is a mathematical function that can describe the joint distribution of several random variables by analyzing the dependencies between their marginal distributions. It can learn the intrinsic information of the original dataset to generate new synthetic data that have the same format and statistical properties as the original dataset.
Figure 4. The KDE comparison of real and synthetic variable values of the target variable energy consumption
Since the SDV library learns probabilistic rules, most of the synthesized data is general. To improve the quality of the synthesized data, some technical constraints can be defined when generating the data. For example, constraints can be set so that the values of a column in the generated data set are always larger or smaller than another column. Figure 4 shows the comparison of the Kernel Density Estimation (KDE) of the target variable energy consumption in the real and synthetic data. The distributions are very similar. In the current production process, the SDV dataset can be used to quickly determine the values that best approximate the required quality according to the prediction for the independent variables. This selection can then be further optimized, for example, in terms of cost and energy efficiency.
Figure 5 illustrates the process of backward analysis and resulting optimization based on the target, the required chemical elements, and the amount of scrap that must be included in a melt to ensure the target composition. The aim of data-driven optimization is to determine the most cost-effective scrap mix or “recipe,” taking into account the predicted energy consumption and metal yield. The database, which contains information from scrap suppliers, is constantly updated and fed new data. Because of this, the results of the optimization are automatically adjusted on an ongoing basis.
Figure 5. Sequence of backward analysis and large-scale charge optimization
The solution to the programming problem should indicate from which scrap supplier and which type of scrap combinations should be purchased to maintain the desired stock levels of the steel producer that will meet the above conditions (desired chemical analysis, minimized energy consumption, post-gating with ferro-alloys) or minimize costs Goutam and Fourer, “A Survey of Mathematical Programming,” 387-400.
The availability of individual scrap suppliers, the market price, and the levels of elements, such as chromium, vanadium, sulfur, and phosphorus, all affect how economically recovered steel scrap can be used. To ensure that steel grades are consistent throughout the cast or semi-finished product and to meet a client’s criteria, such as weldability and hardness, it is critical to control the amount of these elements in the final melt.
A model-based linear first-order (LP) optimization problem has been developed as a tool for scrap purchasing decision makers Applegate et al., “Practical Large-Scale,” 20243-20257 and Miletic et al., “Model-Based Optimization,” 263-266. The computations are performed by the model using the results of the ML model shown earlier in terms of energy consumption and scrap quality data, market prices, and supplier availability information. Prices, quality, and supplier information are included in the model along with quality and density constraints and a production schedule. The LP considers the following convex quadratic programming problem:
where A is an m × n Matrix and Q is a symmetric and n × n positive semidefinite matrix. The vectors of the input features have the upper bounds uc and uv and the lower bounds lc and lv , which have values in R U+∞ and in R U–∞ respectively. This equation assumes that lc ≤ uc and lv ≤ uv. For example, it can take into account that a particular scrap is only used between 500 kg and 1,500 kg, which may be due to process-related circumstances, and is therefore added as a constraint to the optimization objective.
An approximate linear cost equation is used in the model. Scrap costs are determined by market prices and availability, as well as internal storage costs; energy costs are calculated by estimating the energy consumption predicted by the ML model for each type of scrap and the amount of electricity required. The work described above ends with a web-based user interface (Figure 6) that displays the concrete purchase plan. Transportation of the purchased scrap could be considered through route planning. Finally, an estimate of the quantitative carbon footprint of the melting process and the logistics of scrap delivery can be calculated and tracked.
Figure 6. Overview of the entire software: In the background, the software accesses the generated ML model and the database of available scrap and generates an optimized shopping list for the scrap. Factors such as energy consumption predicted by the ML model, scrap prices, calculated CO2 emissions, the distance, and desired chemical composition are considered.
Conclusion
This article has shown how the melting and purchasing process in the steel industry can be modeled and optimized using modern methods from the field of artificial intelligence and mathematical optimization methods. Based on the input features (the scrap composition and the process parameters), the generated models can predict the expected energy consumption with a relative error of about 5%–10%. The optimization software can then be used to generate a scrap composition. Here, the composition of the purchase list from a purely monetary point of view (scrap prices) is supplemented by the consideration of resource and energy efficiency.
The foundry and steel industry naturally must go through a large number of (partial) processes on the way from raw material to finished casting or semi-finished product, where a large amount of production data is generated. For a future data-driven optimization of foundry processes, it is therefore necessary to consolidate this data to make the available knowledge usable for process optimization with the help of tools such as ML. This can provide foundries and their staff with another useful tool, like casting simulation, to further improve existing processes and procedures.
References
Applegate, David, Mateo Díaz, Oliver Hinder, Haihao Lu, Miles Lubin, Brendan O’Donoghue, Warren Schudy. “Practical Large-Scale Linear Programming Using Primal-Dual Hybrid Gradient.” Advances in Neural Information Processing Systems 34 (2021): 20243-20257.
Dutta, Goutam, and Robert Fourer. “A Survey of Mathematical Programming Applications in Integrated Steel Plants.” Manufacturing & Service Operations Management 3, no. 4 (2001): 387-400.
Erickson, Nick, et al. “Autogluon-tabular: Robust and accurate automl for structured data.” arXiv preprint arXiv:2003.06505 (2020).
Klanke, Stefan, Mike Löpke, Norbert Uebber, and Hans-Jürgen Odentha. “Advanced Data-Driven Prediction Models for BOF Endpoint Detection.” Association for Iron & Steel Technology Proceedings (2017), 1307-1313.
Lee, Jin-Woong, Chaewon Park, Byung Do Lee, Joonseo Park, Nam Hoon Goo, and Kee-Sun Sohn. “A Machine-Learning-Based Alloy Design Platform Th at Enables Both Forward and Inverse Predictions for Thermo-Mechanically Controlled Processed (TMCP) Steel Alloys.” Scientific Reports 11, no. 1 (2021): 11012.
Miletic, I., R. Garbaty, S. Waterfall, M. Mathewson. “Model-Based Optimization of Scrap Steel Purchasing.” IFAC Proceedings 40, no. 11 (2007): 263-266.
Patki, Neha, Roy Wedge, and Kalyan Veeramachaneni. “Th e Synthetic Data Vault.” IEEE International Conference on Data Science and Advanced Analytics (DSAA) (2016): 1-10.
Yingjun Ji, Shixin Liu, Mengchu Zhou, Ziyan Zhao, Xiwang Guo, and Liang Qi. “A Machine Learning and Genetic Algorithm-Based Method for Predicting Width Deviation of Hot-Rolled Strip in Steel Production Systems.” Information Sciences 589 (2022): 360-375.
This article content is used with permission by Heat Treat Today’s media partner heat processing, which published this article in February 2023.
About The Authors:
For more information: Contact Tim Kaufmann at tim.kaufmann@hs-kempten.de Dierk Hartmann at dierk.hartmann@hs-kempten.de Shikun Chen at shikun.chen@uni-due.de or Johannes Gottschling at Johannes.gottschling@uni-due.de
Heat Treat Today publishes twelve print magazines a year and included in each is a letter from the publisher, Doug Glenn. This letter is from the April 2025 Annual Induction Heating& Meltingprint edition.
This magazine has been a blessing to publish. The industry is niche, yes, but it is far-reaching and impactful on nearly every aspect of life. There isn’t a place where I can go where heat treating and thermal processing have not made life better and/or possible. The people are by and large good people, fun to work with, and interesting to talk to. The content written in these pages is a good mix of challenging technical content, as well as general interest information.
Our target audience, the 15,000–20,000+ engineers and managers that make purchasing decisions for the vast number of manufacturers who have their own in-house thermal processing operations, is highly engaged with Heat Treat Today. While it is impossible to gauge the engagement of our monthly print editions, when readers respond, it is usually because of something they read in one of our print editions.
Granted, it’s easier to see if/when someone opens or clicks on any of our e-newsletters, but it is surprising the number of people who email us about something they’ve seen in the print editions.
The bottom line is: All of our audience (whether print or digital) is an engaged bunch.
Suppliers who want the attention of these manufacturers with in-house heat treat departments would do well to remember the reach of Heat Treat Today. Whether you use Heat Treat Today or some other format to tell your story, you can’t sell if you don’t tell.
This may seem to be an abundantly obvious statement to many, but you would be shocked at the number of engineering-based companies who believe, “If we build it, they will come.” Thank you, Kevin Costner — that may be true in a field of your dreams, but it is untrue in the real world. Your selling story needs to be told.
Just having a great product — even the best product of its kind — is not a guarantee of success. Sooner or later, you must let people know, somehow, you exist and your product is unparalleled.
Here are a few of the ways that companies typically spread the news:
E-blasts: While e-blasts are low cost and convenient, there are a few challenges: 1. Reaching new people, and 2. People will look at your in-house e-blast as a purely promotional effort, and because of that, will not give it the full attention it might deserve.
Advertising in magazines or on a website: A decent way to tell since they boast a targeted audience, however, magazines do not offer metrics, and websites, while able to provide numbers, typically reach far fewer people than a print version of a magazine and are not consumed for as long as print.
Word of mouth: This method is typically slower and dependent on others talking about you. If your product is that good and it causes a buzz in the industry, then word of mouth may be all you need. It is, however, a passive form of telling, which you do not control and depends on others.
Representatives: Assuming your reps are giving you a large enough portion of their time and are knowledgeable in your capabilities, this is a good way to tell your story. The downside is the rep’s reach. Even if a rep were making four calls a day every weekday of the year, that totals up to just over 1,000 visits a year. And let’s be honest, most companies would be thrilled if a rep made 500 calls a year.
Internal sales teams: Assuming they’re on the phone consistently and not fulfilling orders or being distracted by other internal demands, an in-house sales team, although potentially expensive to maintain, is one of the better options a company has for telling their story. Nearly everyone I know has an internal sales staff. It is pretty much a must.
Website: Websites are not as good at getting the message out as many think, but they are still absolutely necessary. Websites are the most misunderstood marketing tool in the marketing toolbox. Most people think if they have a website, they’re good. Please remember, website “advertising” is passive marketing. Once a website is built, it just sits there until someone decides to come look at it.
Marketing materials: Similar to a website, literature sits there until someone decides to look at it. It’s a passive form of marketing. The bottom line here is this — as you plan for the success of your business, don’t forget it is not just about product quality. You must also remember that “telling” is just as important, if not more so, than building the best product out there.
Our audience of in-house heat treaters is interested in hearing your story. Remember, you can’t sell if you don’t tell.
What’s the relationship between excess air and your bottom-line? In this article, Jim Roberts, President, U.S. Ignition, shares how to increase efficiency and reduce waste in your heat-treating operations.
This informative piece was first released inHeat Treat Today’sApril 2025 Annual Induction Heating & Melting print edition.
A furnace guy walks into a heat treat shop . . . and notices there is a little bit of a yelp to the burners, or the furnace operator mentions the furnace is slowing down on heat up recovery times from a cold load. Or, if you are responsible for fuel costs and monitoring the gas meters, you might notice that situation is slipping in the wrong direction. Or, the burners seem to be dumping soot on your floor. We discussed that in past columns — remember?
Well, it’s all got to do with air. It may seem odd to talk about air when the objective is to utilize fuel at an optimum efficiency, but that’s how we intend to get combustion under control. Let’s go after the air. You remember that we talked about making sure that combustion air sources (blowers, eductors, etc.) were all operating at optimum performance, so the air remains supplied as engineered when the equipment was new. So, now we have our air being delivered at the peak levels we want, but it looks like one of the air valves has shifted, which we covered in the last column on keeping the air sources clean.
This next little tidbit of information is intended to show us all how much this little-considered entity we call AIR can affect the bottom line. Here’s some info you might find interesting.
Eliminate Excess Air
If controls have moved or another phenomenon has caused the burners to lean out, it could cost you a fortune. Most burners are designed to burn with a small percentage of excess air (less than 15%).
Exceptions would include air heating equipment and low temperature drying operations where the excess air is used to control the temperature of the flame. If you operate a burner that has been designed to run at 10–15% excess air and the burner controls or settings drift into the range of 50% excess air (that is a difference of 2–3% O2 or 7.5% O2 in the products of combustion), the difference in an 1800°F oven operation is a calculated 9% loss of fuel efficiency. If you operate a 1 million BTU/hr burner, firing at 75% of the time six days a week for 50 weeks a year, your gas usage would be approx. 5400 therms a year. If we calculate that your gas costs (delivered) are in the range of $4 per 1,000 cu/ft, keeping one burner in tune would save approximately $1,950 per year.
What!!! If you are running a good-sized batch furnace with four burners, that’s a cool $7,800 dollars per year. A ten burner continuous line is going to save almost $20,000 dollars per year. All that just because you cared enough to check excess air levels regularly.
Of course, wasting fuel because you are heating air instead of product is a terrible thing. But don’t forget you can go the other way, too, and go fuel rich with the settings. Then, you take the chance of actually damaging equipment with the carbon you could be producing in a reducing (excess fuel) situation. Carbon can affect all sorts of equipment life, including shortening burner component life and reducing radiant tube and fixture life. It’s not good. Don’t do it. No excess air and no excess fuel will lead you to a happier and more profitable life.
As always, I recommend that you associate your business with the furnace and combustion technicians in your area who can help you make sure everything stays in tune. We’ll chat in the next edition of Heat Treat Today about how to keep a handle on this in-house, so you can tell your experts what you are seeing and start saving yourself gobs of fuel!
For more information: Contact Jim Roberts at jim@usignition.com
A top-tier supplier in the aerospace industry has placed an order for a 6-bar vacuum furnace designed to meet the rigorous requirements of aerospace component manufacturing.
Mark Hemsath President Nitrex/UPC-Marathon
Nitrex’s G-M Enterprises division in Corona, California, has seen new vacuum furnace orders as aerospace demands accelerate.
Mark Hemsath, President of Nitrex/UPC-Marathon, commented on the current trend: “The surge in the aerospace sector is a very welcomed occurrence.”
Press release is available in its original form here.
A modular NANO vacuum furnace system was commissioned for increasing in-house heat treat demands in drive technology. The furnace has reached its final acceptance.
SEW-EURODRIVE completed their fully automated in-house vacuum furnace system integrated with their patented MOVI-TRANS® inductive energy power transfer system (pictured parallel with ECM’s transfer system rails). SEW EURODRIVE partnered with ECM USA to commission the furnace which is completely integrated with advanced automation for their Lyman, South Carolina facility.
Source: ECMSource: ECMSource: ECMSource: ECM
This 6 chamber, 20 bar quench NANO vacuum furnace system provides flexibility and integration utilizing the addition of 16 tempering positions, advanced solvent based washer (both oil and water based contaminants), and robotic workload assembly/disassembly. . . Specifically designed to run multiple materials (including carburized grades and tool steels) this system has modular flexibility to adapt to increased production demands for various load scenarios and processes.
Press release is available in its original form here.
This instructive piece was first released inHeat TreatToday’sApril 2025 Induction Heating & Melting print edition.
In heat treating, clean parts are essential for repeatable, high-quality results. Yet cleaning is often seen as a necessary evil rather than a strategic process. That mindset must change.
For decades, gas carburizing with oil quenching has dominated the industry despite challenges, such as contamination, environmental concerns, and part distortion. These issues have driven growth in alternative processes, such as vacuum carburizing with gas quenching and nitriding, which eliminate post-quench oil contamination. However, not all metallurgical requirements can be met with these newer processes, and gas carburizing with oil quenching remains necessary for many part sizes, geometries, and material types. Furthermore, some alternative processes require more advanced pre-cleaning technology, adding complexity.
I briefly left heat treating in 1998 — I call it my sabbatical from heat treating — to work in advanced industrial cleaning and automation. When I returned about five years later, I was struck by how far behind the industry was in cleaning technology. While other manufacturing sectors had embraced modern solvent and hybrid cleaning systems, heat treating continued to rely on outdated aqueous washers that struggle to clean oil-contaminated parts effectively. It goes back to the old axiom: oil and water don’t mix. Spraying harder only emulsifies the oil further, making separation and disposal even more difficult, increasing costs, and creating sustainability concerns.
Paths Forward
Process shift — Where practical, companies have transitioned from oil quenching to vacuum carburizing with gas quenching, nitriding, and other alternative processes that reduce contamination issues. Of course, such changes are not driven solely by cleanliness — metallurgical requirements and process economics are complex topics. Gas carburizing with salt quenching is an often overlooked alternative, which offers superior heat transfer over gas quenching, reduces distortion, and is environmentally sustainable. Unlike oil quenching, cleaning aft er salt quenching is far simpler, as hot-water washers reclaim over 99% of the salt in a closed-loop system. The old negative mindset about salt, which questions the safety and toxicity of high temperature salt, has restrained process growth in this area. New equipment designs could create interesting, alternative paths with multiple benefits.
Mindset shift — If oil quenching remains necessary, cleaning processes must improve. Conventional aqueous washers are inefficient, and while modern cleaning systems are effective, they are costly. However, when considering part quality, sustainability, efficiency, and long-term cost savings, these systems provide a strong ROI and should not be dismissed.
Technology shift — Sustainability in cleaning cannot be ignored. Water-based systems with distillation attempt to recycle but have high energy costs, making solvent-based systems with integrated distillation more practical for higher efficiency and lower hazardous waste output.
Rugged environments (left) require cleaning systems that modern washers are not often built for. Many new washers are more suited to clean controlled environments like vacuum heat treating (right). (Images from “All About IQ Furnace Systems,” 16)
Challenges with Modern Washer Designs — Thoughts for Manufacturers
One major barrier to adopting advanced cleaning systems is cost, driven by their design. Many new washers are built for clean, controlled environments like vacuum heat treating but are poorly suited for traditional heat treat shops using oil quenching. Th ese shops have different requirements — floor space constraints, varied load configurations, and harsher conditions — meaning rugged, adaptable, and cost-effective solutions are needed. Function must take priority over aesthetics.
Washer manufacturers should rethink their designs to better fit conventional operations by focusing on durability, modularity, and cost-conscious engineering. Doing so could lower costs while improving adoption rates and accelerating industry-wide improvements in part cleanliness, quality, and sustainability.
Conclusion
Heat treating is changing, and cleaning technology must evolve with it. Whether by adopting better process alternatives, improving cleaning methods, or rethinking equipment design, companies that embrace innovation will reduce waste, improve efficiency, and ensure long-term success with a stronger commitment to sustainability and environmental responsibility.
The industry is evolving. It’s time to evolve with it.
References
“All About IQ Furnace Systems.” Heat Treat Today, 4, no. 1 (February 2021): 16-28. https://www.heattreattoday.com/ february-2021/
About The Author:
William (Bill) Disler President WDD Consulting LLC
William (Bill) Disler entered the heat treat industry as a young engineer, quickly establishing himself as a hands-on expert and eventually leading an international heat treat supplier company as CEO/president. He now serves the industry as a strategic advisor and partner to the C-Suite, as an engaged board member, through his consultancy, WDD Consulting, and in roles where he can make a positive impact.
For more information: Contact Bill Disler at wdisler@wddconsulting.com
In this Technical Tuesday installment, Thomas Wingens, Founder & President, WINGENS CONSULTANTS; Dr. Dermot Monaghan, Managing Director, and Dr. Erik Cox, Manager of New Business Development, Gencoa, train readers for finding both real and evasive virtual vacuum leaks.
Leak detection is difficult enough with a “real” leak, but “virtual” leaks present their own challenges. To enhance cost savings and further process efficiencies, it’s essential to have leak sensor technology that can effectively monitor the vacuum chamber and pinpoint these problematic leaks.
This informative piece was first released inHeat Treat Today’sMarch 2025 Annual Aerospace Heat Treating print edition.
Uncontrolled impurities in a vacuum furnace can significantly affect the quality of vacuum heat treating and brazing processes. They can compromise the integrity of the processed material, leading to defects, reduced performance, and increased costs.
Real vs. Virtual Leaks
Real leaks are physical openings in the vacuum system that allow external gases to enter the chamber. These can be cracks, weld failures, improperly installed fittings, faulty seals from damaged or worn O-rings on doors, rotating assemblies, or other components of the vacuum furnace.
The impact on quality includes:
Oxidation and contamination: Real leaks introduce atmospheric gases (like oxygen, nitrogen, and moisture) into the vacuum chamber, which can lead to oxidation of the materials being treated or brazed, as well as other forms of contamination.
Inconsistent results: The presence of unwanted gases can interfere with the chemical processes required for proper heat treatment or brazing, leading to inconsistent metallurgical results.
Reduced mechanical properties: Contamination and oxidation can weaken the materials being processed, leading to defects and reduced mechanical properties of the final product.
Difficulties in achieving desired vacuum: Real leaks can prevent the system from reaching or maintaining the necessary vacuum levels, leading to longer cycle times or failed processes.
Figure 1. Pumping times based on residual water vapor
Real leaks are often easier to detect, especially larger leaks, which can be identified by hissing sounds or the inability of the furnace to pump down. They can be located using methods such as pressure rise tests, solvent detection, or helium leak detectors.
Virtual leaks, however, are much harder to detect as they are not physical openings but rather trapped volumes of gas within the vacuum system that slowly release over time. These trapped volumes are typically found in blind holes, porous materials, or unvented components. Even more problematic are leaks from internally sealed systems, such as water cooling or hydraulics. Leaks from these areas cannot be detected via a leak detector, as the water or oil media can “mask” the leak site and prevent the tracer gas from penetrating.
Aside from increasing the pump time it takes to reach the required vacuum levels, leaks can be a continuous source of contamination within the vacuum chamber. Outgassing can be especially problematic during the heating cycle as it can lead to large vacuum “spikes” or a rise in pressure, affecting the stability of the process environment. Gases released from virtual leaks can contaminate the materials being treated. For example, residual solvents or water vapor from cleaning or incomplete drying can lead to contamination and outgassing. It can be small volumes of air or gas trapped at the bottom of threaded holes or trapped volumes between two O-rings that are not properly vented. Also, outgassing from various hydrocarbons in porous materials such as low-density graphite or powder metallurgy components can release unwanted gases when heated up.
They usually become apparent during the pump-down cycle when the ultimate pressures are not reached or when it takes a long time to reach blank-off pressure. Traditional leak detectors will not pick up virtual leaks.
Detecting Virtual Leaks Accurately
However, residual gas analysis (RGA) and remote plasma emission monitoring (RPEM) can identify virtual leaks by monitoring the composition of gases in the chamber. RPEM offers advantages over traditional quadrupole mass spectrometry (QMS) RGA, particularly in large vacuum systems. Unlike RGAs, RPEM technology operates over a much wider pressure range (50 mbar to 10-7mbar) without requiring additional pumps. The RPEM detector is located outside the vacuum chamber, making it more robust against contamination and high pressures, which commonly damage RGA detectors. This external setup also reduces maintenance needs, as RPEM avoids frequent rebuilds required for traditional RGAs in volatile environments.
Figure 2. Functionality and pressure range of the OPTIX sensor
An example of this newer sensor is the OPTIX, which enables real-time monitoring and process control by providing immediate feedback to maintain chemical balance and ensure product quality. By identifying specific gas species, the sensor allows versatile leak detection with faster problem-solving and continuous system monitoring. Determining the nature of the gas leak will be a clear indication of where the problem originates. Also, whether the gas levels are stable or decreasing will point towards either a real leak or outgassing problem. Unlike RGAs, this sensor does not require highly skilled staff for operation, further lowering the technical burden. Its effectiveness in harsh environments with volatile species makes it a robust and versatile tool for industrial vacuum processes.
Conclusion
By understanding the differences between real and virtual leaks, and their specific impacts on vacuum heat treating and brazing, operators can implement more effective detection and prevention strategies, ultimately leading to improved product quality and process efficiency.
Attention to design, manufacturing, and assembly processes is critical to minimize the occurrence of leaks. This includes proper venting of components, use of appropriate sealing methods, and high-quality welding. Ensuring that components and materials are properly cleaned and dried before being introduced into the vacuum system can reduce outgassing.
Regular leak checks, including leak-up-rate tests, are essential for identifying both real and virtual leaks. Advanced gas analysis techniques are very useful for identifying the type of leak and its source through analysis of the gases in the vacuum chamber. Th e method provides continuous on-line monitoring, rather than periodic leak testing when there is a “suspicion” of a problem.
In the demanding environment of vacuum heat treating and brazing, the OPTIX sensor’s advanced technology not only simplifies leak detection and process control, but also delivers significant cost savings through reduced maintenance and operational expenses. Adopting this type of technology gives operators the ability to enhance vacuum system performance, improve product quality, and achieve greater process efficiency.
About The Authors:
Thomas Wingens Founder & President Wingens Consultants Industrial Advisor Center for Heat Treating Excellence (CHTE)
Thomas Wingens is the Founder and President of Wingens Consultants, and has been an independent consultant to the heat treat industry for nearly 15 years and has been involved in the heat treat industry for over 35 years. Throughout his career, he has held various positions, including business developer, management, and executive roles for companies in Europe and the United States, including Bodycote, Ipsen, SECO/WARWICK, Tenova, and IHI-Group.
Dr. Dermot Monaghan founded Gencoa Ltd. in 1994. Following completion of a BSc in Engineering Metallurgy, Dermot completed a PhD focused on magnetron sputtering in 1992 and went on to be awarded with the C.R. Burch Prize from the British Vacuum Council for “outstanding research in the field of Vacuum Science and Technology by a young scientist.” He has published over 30 scientific papers, delivered an excess of 100 presentations at international scientific conferences, and holds a number of international patents regarding plasma control in magnetron sputter processes.
Eric Cox Manager, New Business Development Gencoa
Dr. Erik Cox is a former research scientist with experience working in the U.S., Singapore, and Europe. Erik has a master’s degree in physics and a PhD from the University of Liverpool. As the manager of New Business Development at Gencoa, Erik plays a key role in identifying industry sectors outside of Gencoa’s traditional markets that can benefit from the company’s comprehensive portfolio of products and know-how.
