How Do Isothermal Calorimeters Age? Repeatability and Reproducibility Across Calorimeter Generations

Reliable experimental data is essential for reproducible science. As our lab has access to five instruments of the isothermal (heat conduction) type of different age and product generations, we became interested in the quality of experimental data obtained from the different instruments. Each instrument consists of 8 individual calorimeters. An intralaboratory study of the first 72 hours of cement hydration was conducted with all five instruments. The objective was to determine the differences between the available instruments with respect to repeatability (precision of the 8 calorimeters within one instrument) and reproducibility (precision between the different instruments). The experimental program comprises 120 calorimetry runs in total (5 instruments each with 8 calorimeters, triple measurement performed). Results show that the oldest instrument version with its 8 calorimeters has a stronger higher standard deviation of the heat released after 72 hours compared to the latest state‐of‐the‐art instrument model (246.7 ± 2.7 vs. 232.3 ± 0.5 Jg–1). This is attributed to the technical design differences of the instruments and continuous use over time. In terms of reproducibility, an increasing standard deviation for the heat released from 1 day to 3 days (136.2 ± 2.6 and 236.4 ± 5.9 J g–1) is observed. From our point of view, this test study indicates that isothermal heat conduction calorimetry is a robust and precise measurement method to determine the heat of cement hydration for 72 hours.


Introduction
In nature, almost every physical, chemical, or biological event either emits or consumes heat.In the scientific literature this is referred to as exothermic or endothermic processes.[1] To monitor these phenomena, different types of calorimeters are used, the naming and description of which would go beyond the scope of this section.[1][2][3] A list and explanation of various calorimeter types and their measuring principle can be found elsewhere.[3][4][5] The basic principle of an isothermal (heat conduction) calorimeter is to record the heat flow from a sample through a heat flow sensor to an isothermally held heat sink, or vice versa (from the heat sink to the sample).[1 -5] Isothermal (heat conduction) calorimetry has found its way into various scientific research fields, such as pharmaceutics and food or cement science.[2,5] As cement emits heat from its first contact with water and throughout the hydration period, different types of calorimeters, other than isothermal, have been used to measure the heat of hydration in cement and their methods are standardized in various European and American standards.[4] Isothermal calorimetry presents a more recent method compared to the other calorimetry methods, such as solution calorimetry.[6,7] If cement calorimetry is combined with other techniques such as quantitative phase analysis via X-ray diffraction (XRD) or rheology, a deeper understanding of ongoing processes can be acquired.[8,9] Another example is the linear correlation between emitted heat and compressive strength, which is not only found for ordinary Portland cement, but also blended cement types.[10] As isothermal calorimetry is easier to perform than for example solution calorimetry and less dangerous (solution calorimetry uses nitric and hydrofluoric acids), many scientific researchers use isothermal calorimetry to characterize cement, study the reaction of cement and investigate the influence of different chemical admixtures onto the hydration process.[10][11][12] A study published by Wadsö and Arndt presents the result of an interlaboratory round robin test and shows that isothermal calorimetry is a reliable method to determine the heat of hydration for cement pastes when considering certain aspects regarding the evaluation of obtained data.[4] Measurements were performed to determine the precision within one laboratory (repeatability) and among different laboratories (reproducibility).The participants used different types of isothermal heat conduction calorimeters and repeated at least 6 measurements for the studied ordinary Portland cement.Regarding the different evaluation procedures described, one procedure calculates the emitted heat by starting the integration after 1 hour to avoid artefacts due to the initial disturbance, when a sample is placed into the isothermal calorimeter.Consequently, it is important to note that the obtained heat values are lower compared to values obtained from other calorimetry methods due to missing heat information of the first hour of cement hydration.[4] In our laboratory, we have access to five TAM Air calorimeters (eight-channel isothermal calorimeter) from four different product generations, covering a period of over 20 years.Furthermore, each instrument consists of 8 calorimeters (channels).Therefore, we became interested in determining the variance between the different multichannel instruments.In this contribution we examined the repeatability (precision of the eight calorimeters within the multi-channel instrument) and reproducibility (precision between the different instruments).

2
Instruments, Methods, and Materials

Instruments
To measure the heat of hydration in cement pastes five instruments of the isothermal (heat conduction) type are used (four different models of TAM Air calorimeters; back then: Thermometric, Sweden; now: TA Instruments, USA).[4,5] In the context of the following sections, the term calorimeter means "isothermal heat conduction calorimeter" and refers to the five different instruments used, labelled as calorimeter 1 to 5. [5] Every instrument/calorimeter has 8 calorimeters of the heat conduction type in it, labelled as channel 1 to 8. The individual channels contain a sample and a reference side (twin type calorimeter).[5] The main differences between the different calorimeters are listed below.1).[13] No noisy signal was observed for calorimeter 3 to 5. Environment: The calorimeters are loacated in different laboratories: calorimeter 1 to 3 are located in one room and calorimeter 4 and 5 in a second one.Both rooms present regular laboratory rooms without air conditioning.

Calibration:
The calorimeters are calibrated with empty channels (empty sample and reference sides) before starting the experiment.Calorimeter 1 and 2 were calibrated manually according to the manufactures recommendation in a stationary state (calorimeter 1 with an external power supply: 20 V DC, current of 1 A; calorimeter 2 contains an internal power supply).The calibration of calorimeter 3 to 5 was executed within the TAM Assistant software by an impulse (gain calibration).
As the computer records the electric signal of the calorimeters in volt, the calibration and software provides the user with the thermal power in milliwatt (mW).The heat generated during the measurement is the time integral of the thermal power and is given in joule (J).In this context, the recorded thermal power and calculated heat are normed to the amount of cement binder in the paste, indicated by mWg -1 and Jg -1 , respectively.
Baseline: Before each measurement a steady baseline is recorded with no ampoule in the sample side but a quartz sand filled ampoule in the referene side.For calorimeters 3 to 5 the TAM Assistant software was used with moderate baseline conditions.This means that the absolute value of the slope is < 2 µW/h and the standard deviation is < 4 µW.The values are calculated based on a linear fit within a window length of the last 20 minutes.When the conditions are met the software starts to record a baseline for 30 minutes.For calorimeter 1 and 2 the baseline was considered stable if the recorded signal of the individual channels did not deviate more than ± 0.04 mW from zero for at least 1 hour.DIN EN 196-11 lists technical parameters for the baseline condition, however, also states no experimental evidence for connecting the baseline requirements with a precise measurement.[14] The average value of the recorded baseline (30 minutes duration for calorimeter 3 to 5, 20 minutes duration for calorimeter 1 to 2) is determined as offset and substracted from the subsequent measured values.No final baseline was considered after the 72 h measurement has finished.
Temperature control: All calorimeters were set to a temperature of 20.0 °C with the control pad of the calorimeter.The temperature fluctuations of the air thermostats in calorimeter 4 and 5 are in the range of ± 0.001 °C (monitored by software).The temperature fluctuations of the air thermostats in calorimeter 1 to 3 were not controlled and assumed to be ± 0.02 °C over 24 h.[5] According to , the temperature stability of the thermostat in the calorimeter should not exceed 0.2 °C.The temperature during the measurement should be in the range of 20.0 ± 0.2 °C.[14] Heat flow sensors: Another aspect which is attributed to the age and frequent use of the calorimeters in the past is the quality of the individual calorimeter channels.This means, that the heat flow sensors in some channels of calorimeter 1 to 3 showed no smooth surface and an unknown amount of residue, which we were not able to fully remove.The heat flow sensors in calorimeter 4 and 5 showed a shiny and clear surface.

Methods
To study the repeatability (precision of the 8 channels in one calorimeter) and reproducibility (precision between the different calorimeters) of isothermal calorimetry for the ex-situ measurement of cement hydration over three days, the following experiment setup was used.After calibration and before starting the experiment, the reference side of all calorimeter channels was loaded with a 20 mL glas ampoule (27.5 mm in diameter) containing quartz sand (12.43 ± 0.01 g).The reference ampoules are used to decrease the baseline drift as well as the baseline noise and were not removed until all 120 measurements were finished.The 120 measurements can be derived from the following experiment plan: each calorimeter has 8 channels and by studying 5 calorimeters, run 1 comprises of 40 measurements.Thereby, is one cement paste prepared for one channel of calorimeter 1 to 5, resulting in 8 independently mixed cement pastes.The cement pastes were also transferred into 20 mL glas ampoules (27.5 mm in diameter) and then loaded into the respective channels.To obtain the (arithmetic) mean ( ) and (sample) standard deviation (σ) for the performed measurements, run 1 is repeated 2 times, herein refered to as run 2 and run 3.In between runs, the calorimeters were left to thermally equlibrate over night, after removing the sample ampoule with cement paste.The 120 measurements were completed within 12 days.The folloing experimental procedure was conducted to prepare the cement paste and subsequently load the calorimeters with the sample ampoule (ampoule containing the cement paste): 1. Cement, water, and other items were stored in the laboratory without air conditioning prior to the measurement.2. 100 ± 0.01 g cement is transferred into a 250 mL polyethylene cup. 3. 40 ± 0.01 g water is transferred into a 250 mL stainless steel cup. 4. Cement is added to the water within 10 seconds.
The first contact of cement with water describes the hydration start.5.The paste is then stirred with an overhead-stirrer equipped with a 4-blade propeller (5.50 cm in diameter) for 2 minutes at 600 rpm (revolutions per minute).6.A 20 mL glass ampoule (27.5 mm in diameter) is charged with 6 ± 0.01 g of the freshly prepared paste and then sealed.This step is repeated four additional times (one ampoule with cement paste for one channel of the respective calorimeter, resulting in 5 sample ampoules in total).7. The individual sample ampoules are loaded into channel 1 of the respective calorimeters within 10 minutes after hydration start.8.A new paste is prepared for the second channel of the respective calorimeters (return to point 2) and repeated until channel 8 of the respective calorimeters is filled.Note that all calorimeters were loaded within a time frame of 2 h (starting count when opening channel 1 and finishing count when closing channel 8).
As no air-conditioned laboratory was available, the temperature of the cement at the beginning of each run was measured with a comercially available thermometer (22.4,22.2, and 22.4 °C for run 1, run 2, and run 3, respectively).The room temperature in the laboratory, the temperature of the water before preparing the cement paste and the temperature of the cement paste itself (before loading the sample ampoules into the selected channel of respective calorimeters) was also monitored.The mean value ( ) and standard deviation (σ) within one run is given in Table 1.The room temperature for the two laboratories in which the calorimeters were located is also displayed in Table 1.
The performance of the calorimeters and their channels was addressed by calculating the heat from the recorded heat flow values.By performing the ex-situ method for the calorimetry measurement (mixing the cement and water outside the calorimeter) the first hour after hydration start was not considered for the evaluation.[4] Thus, the integration starts 1 h after hydration start and proceeds until 72 h.

Materials
All experiments were conducted with the same batch of a commercially available ordinary Portland cement (OPC), denoted as CEM I 42.5R according to EN 197-1.For this, 5.20 kg of cement was transferred from a plastic Hobbock drum into a vacuum food bag.Subsequently, a vacuum was applied to the bag and then sealed.The bag is opened when performing one experiment run and then vacuumized and sealed after the run has finished.

Results and Discussion
We were interested in the variation of the heat flow data of 24 independently mixed cement pastes measured by different TAM Air isothermal calorimeter models, labelled as calorimeter 1 to 5, and their respective built-in 8 isothermal heat conduction calorimeters, labelled as channel 1 to 8.For this we studied the reproducibility (precision between the different instruments) and repeatability (precision of the 8 calorimeters within one instrument) of the obtained results.

Reproducibility
The first part focuses on the differences between the available calorimeters (reproducibility).For this, two groups were formed.Group A comprises all 5 available calorimeters and provides 120 measurements in total.Group B is a selection of calorimeter 3 to 5, presenting 72 of the 120 measurements (the 48 measurements obtained from calorimeter 1 and 2 are not considered in the evaluation of this group).The (mean) heat flow curve for group A (calorimeter 1 to 5) and B (calorimeter 3 to 5) with their respective (sample) standard deviation is displayed in Figure 2. The data for the heat flow of group A shows a small standard deviation during the complete measurement time of 72 hours but group B presents a smaller one (see Figure 2).Throughout the first 72 hours of cement hydration, the heat flow curve obtained from group B also displays slightly lower values compared to group A. This becomes more evident when the (mean) heat is calculated and compared for both groups (see Figure 3).At 72 hours the heat obtained from group A is larger than for group B. The same applies to the standard deviation.This indicates that calorimeter 1 and 2 (oldest calorimetry models within this study) slightly shift the data set to higher values and are responsible for the increase in the standard deviation.Statistic values for the heat at 24, 48, and 72 h of group A and B are listed in Table 4.By comparing the heat values at 24, 48, and 72 h for the individual calorimeters (24 measurements), it becomes clear that calorimeter 1 always exhibits the largest cumulative heat, followed by calorimeter 2 and 3 (see Figure 4).The calorimeters 4 and 5 (calorimeters with state-of-theart electronics and better stability compared to the previous TAM Air models, e.g., calorimeter 1 to 3) show the lowest heat values of all tested calorimeters.The mentioned increase of the standard deviation assigned to the calorimeters 1 and 2 in group A becomes also evident when the standard deviation between the calorimeters at different times is compared (Figure 4).As calorimeter 4 and 5 show a low and uniform standard deviation (< 1.0 J g -1 ) at 24, 48, and 72 h, calorimeter 1 and 2 display higher deviations as well as a steady increase in standard deviation with ongoing hydration time (see Table 5).Calorimeter 3 shows heat values comparable to calorimeter 4 and 5, however, also an increasing standard deviation with time.Apart from the total heat emitted by the cement hydration after a certain time, a second important information obtained from heat flow curves are time-related events such as the silicate peak and sulfate depletion point.By using different chemical admixtures, one can see a retardation or acceleration of the cement hydration by shifting the silicate peak to different times.[12] Therefore, it is important that isothermal calorimetry also monitors these heat events as accurately as possible.We determined the maxima from each of the 120 heat flow curves and obtained a total of 240 datapoints (one maximum each for silicate reaction and sulfate depletion point).The calculated mean values and standard deviation for the times of respective maximum for each calorimeter are listed in Table 6.For the maximum of the silicate reaction, the (mean) heat flow curve recorded by calorimeter 1 shows the earliest maximum at approximately 11 h, followed by calorimeter 2 (11 h and 10 min).Heat flow curves for calorimeter 4 and 5, show the maximum of the silicate reaction at around 11 h and 15 min, followed by calorimeter 3 (11 h and 20 min).The same order of observed maximum for the individual calorimeters is also true for the sulfate depletion point.The standard deviation in time for the recorded maxima is less than 5 minutes for each calorimeter.The time window for the observed maxima recorded by the 5 calorimeters increases in case of the sulfate depletion point (all 5 maxima within ~ 25 min) compared to the silicate reaction (all 5 maxima within ~ 15 min).This becomes more apparent when collecting the respective 120 maxima for the silicate reaction and sulfate depletion point and forming the mean values, presented by group A (see Table 7).Here, the standard deviation increased in case of the sulfate depletion point compared to the silicate reaction.Group B with 72 collected maxima for the silicate reaction and sulfate depletion point, respectively, shows its maximum at earlier times and with a lower standard deviation compared to group A (see Table 7).The deviation in time and recorded heat flow value for the respective groups is displayed in Figure 5.As cement hydration is very temperature dependent and higher temperatures accelerate the cement hydration this could be an indication for a less precise temperature control in the calorimeters 1 and 2, where a precise monitoring of the thermostat temperature was not available.However, it is important to note that this shift in time is only marginal.

Repeatability
Until now we discussed the reproducibility by comparing the five calorimeters with each other and dividing them into group A and B (all calorimeters and only a selection of models).Group B (calorimeter 3 to 5) indicated a slightly higher precision than group A (see Table 4).Group A, which comprises all available calorimeters (old and new models) showed a good reproducibility after 72 h (standard deviation = 5.9 Jg -1 ).Group B, which does not consider calorimeter 1 and 2 (two oldest calorimeters available within this study) performed even better (standard deviation = 1.1 Jg -1 ).As all studied calorimeters consist of 8 channels each, we now consider the performance of a single calorimeter by comparing the individual channels in the calorimeter (repeatability).The mean heat values and the standard deviations of channel 1 to 8 in the respective calorimeters (1 to 5) after 72 h are listed in Table 8.In the previous section, it became clear that calorimeter 1 recorded the highest heat values of all studied calorimeters after certain times (see Table 5).This trend continues as all channels of calorimeter 1 also yield the highest heat values after 72 h (> 240 Jg -1 ) compared to the channels of the other calorimeters (see Table 8).The channels of calorimeter 4 and 5 show the lowest heat values (< 233 Jg -1 ) with the smallest standard deviation compared to calorimeter 1 to 3. To address the performance of the individual channels in one calorimeter the following method was applied.The mean heat value of one calorimeter after a certain time was determined by taking the single heat values of all 24 measurements (8 channels, 3 runs) performed in one calorimeter.The mean heat value of a single channel is therefore calculated based on the triple measurement (3 runs).The deviation in percent of the mean heat value of a single channel (3 measurements) from the mean heat value of the respective calorimeter (24 measurements) can thus be displayed schematically as depicted in Figure 6.It shows that the individual channels of calorimeter 4 and 5 show almost no deviation from the respective calorimeters mean values (apart from channel 1 in calorimeter 4).The channels of calorimeter 3 show slightly stronger deviations, apart from channel 2 which shows a strong deviation from the calorimeters mean value at 24, 48, and 72 h.Calorimeter 2 presents a similar picture compared to calorimeter 3, however, here channel 7 shows the strongest deviation of all built-in channels from the calorimeters mean value.In calorimeter 1 all channels apart from channel 3 and 4 show a more pronounced deviation compared to the other calorimeters.The observed deviation from the calorimeter's mean value in case of calorimeter 1 also shows an increase with time.Note, that this illustration only shows the performance of the individual channels within one calorimeter (repeatability, precision of the calorimeter) but does not permit any conclusions about the accuracy (reproducibility) of the respective calorimeter.For the performance of the overall calorimeter see Figure 4.The deviation of the individual channels from the calorimeters mean values can be mainly ascribed to the quality of the heat sensors in the individual channels, with respect to the unknown residue that could not be removed prior to the actual measurement nor the calibration procedure.The obtained low standard variation for calorimeter 4 and 5 as well as the similar recorded heat values indicate a high precision and accuracy for the measurement of cement hydration for 72 hours (see Figure 4).The outcome is also ascribed to the overall performance of the individual channels, showing a high level of agreement, with almost no deviation from the respective calorimeters mean value (see Figure 6).Data recorded by calorimeter 3 showed a slight deviation from calorimeter 4 and 5, affecting the reproducibility of group B (calorimeter 3 to 5) to a minor extent (see Table 4).However, note that standard deviation of calorimeter 3 increased steadily with time (see Table 5) which was not observed for calorimeter 4 and 5 within the three-day measurement.This indicates a poorer stability of calorimeter 3 compared to calorimeter 4 and 5 in terms of longer measurements.A stronger overall deviation was observed when also considering the oldest calorimetry models available within this study (calorimeter 1 and 2).Both models (calorimeter 1 and 2) showed a larger recorded heat value as well as an increasing standard deviation with time (see Table 5).These heat values slightly distorted the precise measurement series of group B (calorimeter 3 to 5) and resulted in group A (calorimeter 1 to 5) with higher mean heat values with a higher standard deviation.
There are different explanations for the observed stronger fluctuations in heat flow values recorded for the older calorimeters (1 and 2).First, these instruments are calibrated manually to determine the calibration coefficient and baseline value.Second, the temperature in the calorimeter is probably less accurately controlled compared to the newer models (calorimeter 4 and 5).Furthermore, does the quality of the heat flow sensor influence the measurement, which can be derived from calorimeter 1 and 2 but also calorimeter 3 (see Figure 6).
It is very important to note, however, that the overall performance of all five calorimeters examined is very much acceptable.According to EN 196-11 the deviation of two measurements within one laboratory (repeatability) should not exceed 10 Jg -1 after 7 days and not 37 Jg -1 for two different laboratories (reproducibility).Apart from calorimeter 1 and 2 one can safely assume that the other 3 calorimeters match these criteria when measuring the cement hydration for 7 days (apart from channel 2 in calorimeter 3 due to a bad quality of the heat flow sensor).
All experiments were performed by the ex-situ method, resulting in missing heat information of the first hour of cement hydration.[4] To collect data within the first hour the in-situ method (mixing cement and water in the calorimeter) would be a good choice.Here, it would be interesting to see how the age of the individual calorimeters affects the results.

Conclusions
The main outcome of this study confirms that state-of-theart isothermal calorimeters represent a very precise method to measure cement hydration.Older calorimeters perform slightly worse than newer ones but still with acceptable results.It is important to keep the heat flow sensors of an isothermal calorimetry free from any residues and mechanical damage, as this is the main communication interface between the sample and the device.Furthermore, it is recommended to control the temperature in the calorimeter as precisely as possible to avoid measuring artefacts regarding the cement hydration rate.This study confirms an earlier study and shows that isothermal heat conduction calorimetry is a reliable and precise measurement tool for recording the heat of cement hydration.

Figure 1
Figure 1 Exemplary comparison of raw data obtained from calorimeter 2 (blue) and calorimeter 3 (green).To smoothen the noisy signal recorded by calorimeter 2 a Savitzky-Golay filter (orange) is applied.

Figure 2
Figure 2 Mean heat flow curve for group A (calorimeter 1 to 5, blue) and group B (calorimeter 3 to 5, orange).The respective colored shaded area shows the standard deviation for each group.

Figure 3
Figure 3 Mean heat curve for group A (calorimeter 1 to 5) and group B (calorimeter 3 to 5).The colored shaded area shows the standard deviation for each group.

Figure 4
Figure 4 Mean heat values for each calorimeter (24 measurements) at different times.Black error bars show the respective standard deviation for the individual calorimeters at certain times.Gray solid lines in the back present the mean heat value of group A (calorimeter 1 to 5, 120 measurements) at respective times.

Figure 5
Figure 5 Mean heat flow curves (colored lines) obtained from group A (calorimeter 1 to 5, blue) and group B (calorimeter 3 to 5, orange) zoomed in to highlight the relevant time events.The colored shaded area shows the standard deviation for each group.The first maximum refers to the silicate reaction, the second maximum to the sulfate depletion peak.Maxima are marked with colored dots.The black crosses show the standard deviation for the respective determined maxima along the heat flow (y-axes) and time (x-axes).

Figure 6
Figure 6 Graphical illustration for the deviation (in percent) of the mean heat value of an individual channel (colored dot) from the respective mean heat value of the respective calorimeter (gray line) at a certain time.Deflections of the dots to the right of the gray line indicate software record data approximately every second during the measurement.Subsequently the data set is reduced to one data point every 10 seconds in the TAM Assistant software.As the recorded signal from calorimeter 1 and 2 showed a noisy signal throughout the complete measurement time, a third order Savitzky-Golay-Filter (filter of width: 100) was applied to smoothen the data (seeFigure th TAM Air) with state-of-the-art electronics and better stability compared to the previous models.Here, a newer TAM Assistant software is used for data recording (V3.0.4).As the PicoLog (data loggers) software collects data every 10 seconds, both versions of the TAM Assistant

Table 1
Monitored temperature throughout the experiment series.

Table 2
Phase and chemical composition of the cement.

Table 2 .
Physicalproperties are listed in Table3.The specific surface area and compressive strength is determined according to method respectively.

Table 3
Physical properties of the cement, such as specific surface area, density, distribution values, and compressive strength.

Table 4
Mean heat values and standard deviations for group A (calorimeter 1 to 5) and group B (calorimeter 3 to 5) at different times.

Table 5
Mean heat values and standard deviation after 24, 48, and 72 h for calorimeter (C) 1 to 5.

Table 6
Mean time values and standard deviation for determined maxima of the mean heat flow curve (24 measurements) for the individual calorimeters (C).The first maximum refers to the silicate reaction, the second maximum to the sulfate depletion point.

Table 7
Mean time values and standard deviation for determined maxima of the heat flow curve for group A (calorimeter 1 to 5) and group (G) B (calorimeter 3 to 5).The first maximum refers to the silicate reaction, the second maximum to the sulfate depletion point.

Table 8
Mean heat values and the standard deviations of channel 1 to 8 in the respective calorimeters (C) 1 to 5 after 72 h.