A System for the Thermal Analysis of Steels

Het Kapadia and Robert Tuttle

There has been growing interest in steel grain refinement research. This interest stems from the relentless drive for improved mechanical properties, specifically strength and ductility. Grain refinement research has been focusing on determining which compounds can heterogeneously nucleate steel. Depending on the alloy, the candidate phase must match with δ-ferrite or austenite. Rare earth phases have been found particularly suitable for austenitic steels. There is also some evidence that they are effective in assisting ferritic steels. Titanium nitrides and carbides have been effective for ferritic steels, and methods for stabilizing them in liquid steel are being examined. Thus, there appears to be several different routes for the grain refinement of steels.

Industrial adoption will eventually require a method for assessing grain refinement additions as part of process control. Production foundries must be able to verify a certain level of refinement will occur. Errors in the production process should be detected and corrected prior to pouring. A method for this verification will require rapid testing time and ruggedness for melt deck use in addition to the normal requirements for industrial process control. To date, researchers have primarily examined the final microstructure, which is time consuming. It also only provides information on refinement after castings are poured. Additionally, this method has limitations on interpretation of the grain refinement process. The slow speed and after casting basis of this technique is unacceptable in industry. Therefore, a method must be developed to support research and industrial adoption.

Looking at other metallurgical areas using additions for promoting heterogeneous nucleation provides a possible suitable method. Thermal analysis has been successfully employed in aluminum and iron foundries. Aluminum foundries utilize the technique for assessing the effectiveness of grain refining and eutectic modification additions. Cast iron foundries use the same technique to determine the success of inoculation additions for promoting graphite. These methods employ a single thermocouple in a shell core cup. A sample is poured into the cup and the cooling curve recorded. This curve is then analyzed to determine the effect additions have had on the melt. This technique has proven very effective for measuring heterogenous nucleation on the foundry floor.

Research examined incorporating the single thermocouple technique on steels, with a system used in aluminum and cast irons, to develop a more detailed understanding of solidification. 4130 steel was selected because of its broad application. This system utilizes an existing commercial DAQ system specializing in the thermal analysis of metals and includes advanced analysis techniques and algorithms. The data obtained then was compared to thermodynamic predictions and the Jenkortoret reference data for this alloy.


A 50-lb. (23 kg.) heat of 4130 steel was melted in a 3kHz induction furnace with an Al2O3 crucible in air. An initial charge of 1010, FeCr, and FeMo was added. At 1700C (3100F), additions of FeSi, FeMn, graphite, and aluminum shot were added. Heating continued to 3,150F (1,730C) where it was then tapped into a preheated
2.3kg capacity hand ladle. To maintain deoxidization, a 1g piece of aluminum shot was put in the hand ladle prior to tapping.

Liquid steel then was poured into the thermal analysis (TA) cup which is mounted on a stand. Once the alloy was poured, the data acquisition system (DAQ) recorded and analyzed the cooling curve of the solidifying steel. To determine the correct cooling rate for the single thermocouple method being used, the TA cups were uncovered, covered, and wrapped and covered. A ceramic fiber blanket with a density of 128kg/m3 was used as the covering material. The top cover had a thickness of 50 mm and a length and width of 150 mm. The run order was randomized. Two TA cups were analyzed for each condition, and a total of six  cups were poured per heat. After filling the first cup, the remaining metal was poured into a bar cavity from which a chemical sample was then extracted for optical emission spectroscopy.

Results and Discussion

Figure 1 depicts the cooling curves of the normal cups. The curves are similar, which indicates the data obtained had acceptable reproducibility. The only difference noticed in the curves are in the form of a slight time difference. This variance is due to the difference in superheat; the black line had a superheat of 64F (17.8C) while the red line had a super heat of 50.4F (10.2C).

In the original experiment, six cups were poured. While the cooling curves appeared similar to each other, examination of the cooling rate found a significant issue (Figure 2). The cooling rate signatures were very different for the first and second wrapped and covered samples, which were the second and sixth cups poured during the experiment. It appears that holding steel in the furnace for more than four TA cups resulted in significant alloying loss. This was confirmed by conducting spectroscopy on the fifth and sixth cups. These cups had much lower contents of the oxidizable elements of C, Si, Mn, Cr, and Al compared to the heat chemistry. Thus, the last two cups poured had a drastically different composition than the first four. This was likely because the initial heat had been held for over an hour due to the freezing times of these cups. A second heat was poured to ensure the covered and wrapped and covered conditions were done with steel of a similar composition. The uncovered cups were close enough to each other to be considered the composition, so they did not require repeating.

Figure 3 illustrates the cooling curves for the second run of covered cups. Figure 4 displays the results obtained from wrapped and covered TA cups.

The DAQ system currently does not have an algorithm for automatically detecting the peritectic reaction temperature. For the peritectic temperature, the authors examined the cooling curves for a peak just prior to the solidus. Then, the corresponding time stamp for that peak were taken. The time was then used to read the
temperature from the cooling curve. Figures 5, 6, and 7 illustrate the cooling rates for the normal, covered, and wrapped and covered cups. The curves plotted display the cooling rate of each cup over a period of five minutes.

Several thermal arrests are identifiable on the cool rate curves. The first is the initial heating of the thermocouple followed by the liquidus arrest. Afterwards, there is a small peak that corresponds to the peritectic transformation (Figure 5). The largest peaks are associated with reaching the solidus. Overall, very little noise is present in the normal cup curves.

Figures 6 and 7 depict cooling rate curves that are far noisier than the normal cup. Since these cups were covered to reduce their cooling rate, the strength of the solidification reactions were less, which created a lower signal-to-noise ratio. This lower signal-to-noise ratio would make analysis, especially automated, more
difficult. Thus, it would appear they are not as appropriate for process control or even research. In both the covered and wrapped and covered conditions, the second cup had a much smaller peak. For the wrapped and covered, it appears that the difference in cooling curve shape is related to the large variation in superheat between the samples. Therefore, normal cups were considered for further analyses due to their higher signal-to-noise ratio.

Table 1 provides the liquidus temperatures from the thermodynamic predictions, Jenkortoret data, and these experiments. The Jenkortoret data includes phase reaction temperatures at 0.1, 0.5, and 2C/s. The 2C/s data has been selected for comparison since it is close to the 1.8C/s average cooling rate of the normal cup samples.

Table 2 lists the solidus temperatures from the thermodynamic predictions, Jenkortoret data, and this work. There is significantly less agreement in the temperature of the solidus. The thermodynamic predictions indicate a very low temperature while the Jenkortoret data and this work find the solidus at a higher temperature. Considering the 68F (20C) difference between the thermodynamic conditions and the Jenkortoret data, it appears the disparity is related to technique. Thermodynamic software are based on the CALPHAD method and frequently use data from DSC experiments for their dataset. Sample sizes for DSC methods are often in the milligram range. The samples for the reference data were approximately 35 g, while these experiments use a sample that is approximately 280 g. As solidification occurs, the latent heat of fusion is released. Assuming the only heat generated during solidification occurs from this, then Equation 1 holds true: Q=mHf

Where, Q is the energy released during a phase change (J), m is the mass of the substance (kg), and Hf is the latent heat of fusion (J/kg). Computing the energy evolved in each sample indicates a large variation in the total energy emitted during solidification. The energy released increased with sample size and matches the increase of the solidus temperature with sample size. It appears that the measured solidus may be affected by the sample size in addition to the documented effect of cooling rate. This would make sense since the smaller samples might evolve less heat during solidification to cause a detectable difference in temperature. Thus in smaller samples, the peak might appear lower since sufficient heat must be released to cause the formation of a detectable peak. Smaller peaks are harder to detect and could be the source of the differences observed. This would also explain the high solidus measured by the system despite having a similar cooling rate to the Jenkortoret data.

Another factor that might contribute to the differences in the measured solidus temperatures stems from the location of the thermocouple in each technique. In DSC, the thermocouple is located within the holder and the pan and sample rest on top. The Jenkortoret experiments had a thermocouple within an alumina protection tube inserted into the sample. The current work also had a thermocouple directly inserted into the sample, but in a quartz tube. The insertions of
a thermocouple within the sample should result in less lag between the thermocouple reading and the sample.


The results demonstrate the applicability of single thermocouple thermal analysis. The cooling rate curves provided distinct peaks for the liquidus, peritectic, and solidus reactions. The normal cup condition produced curves with the least amount of noise when compared to the covered and wrapped and covered cups. Thus, using cups without any covering appears to be the most appropriate for steels. Excessive holding times were found to cause the heat composition
to change significantly. Future work should attempt to keep the number of thermal analysis cup measurements to a minimum to prevent this effect.

The measured reaction temperatures were compared to reference data and thermodynamic predictions. The liquidus temperatures of all three had good agreement. There was more variation with the solidus temperatures among the data sources.

The difference in the solidus may be attributed to the size of the samples, which are much bigger in this work than either the reference data or the DSC data that forms the basis of the thermodynamic predictions. It was also determined that the values for the peritectic temperature were different from each other. These deviations may be attributed to a number of factors such the low height of the peritectic peak, differences in sample size, cooling rate differences, and scarce DSC data, which makes it harder for the thermodynamic prediction software to predict an accurate value for the peritectic. The lack of overall data on the peritectic likely reflects that it is a more difficult reaction to characterize. Further work is required to understand the reason for the difference. It was concluded the disagreement between techniques may reflect experimental difficulties with characterizing the peritectic and solidus and not the experimental system.

The single thermocouple thermal analysis system employed in this work provided consistent data that can be used for the analysis of steels. The robustness of the system indicates it could be incorporated in assessing individual steel heats for process control in an industrial setting. It also provides an additional technique for
characterizing the solidification characteristics of steels. 

Click here to see this story as it appears in the March 2020 issue of Modern Casting.