MOLECULAR CHARACTERIZATION OF ENERGETIC MATERIALS A Dissertation by SANJEEV R. SARAF Submitted to the Office of Graduate Studies of Texas A&M University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY December 2003 Major Subject: Chemical Engineering
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
MOLECULAR CHARACTERIZATION
OF
ENERGETIC MATERIALS
A Dissertation
by
SANJEEV R. SARAF
Submitted to the Office of Graduate Studies of Texas A&M University
in partial fulfillment of the requirements for the degree of
DOCTOR OF PHILOSOPHY
December 2003
Major Subject: Chemical Engineering
MOLECULAR CHARACTERIZATION
OF
ENERGETIC MATERIALS
A Dissertation
by
SANJEEV R. SARAF
Submitted to Texas A&M University in partial fulfillment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY
Approved as to style and content by: _____________________________ _____________________________ _____________________________ _____________________________ _____________________________
December 2003
Major Subject: Chemical Engineering
M. Sam Mannan (Chair of committee)
David M. Ford (Member)
Michael B. Hall (Member)
Dan F. Shantz (Member)
Kenneth R. Hall (Head of Department)
iii
ABSTRACT
Molecular Characterization of Energetic Materials. (December 2003)
Sanjeev R. Saraf, B. Chem. Engg., U.D.C.T, Mumbai, India
Chair of Advisory Committee: Dr. M. Sam Mannan
Assessing hazards due to energetic or reactive chemicals is a challenging and
complicated task and has received considerable attention from industry and regulatory
bodies. Thermal analysis techniques, such as Differential Scanning Calorimeter (DSC),
are commonly employed to evaluate reactivity hazards. A simple classification based on
energy of reaction (-∆H), a thermodynamic parameter, and onset temperature (To), a
kinetic parameter, is proposed with the aim of recognizing more hazardous
compositions. The utility of other DSC parameters in predicting explosive properties is
discussed.
Calorimetric measurements to determine reactivity can be resource consuming,
so computational methods to predict reactivity hazards present an attractive option.
Molecular modeling techniques were employed to gain information at the molecular
scale to predict calorimetric data. Molecular descriptors, calculated at density functional
level of theory, were correlated with DSC data for mono nitro compounds applying
Quantitative Structure Property Relationships (QSPR) and yielded reasonable
predictions. Such correlations can be incorporated into a software program for apriori
prediction of potential reactivity hazards. Estimations of potential hazards can greatly
help to focus attention on more hazardous substances, such as hydroxylamine (HA),
which was involved in two major industrial incidents in the past four years. A detailed
discussion of HA investigation is presented.
iv
To mom, dad, aka, aaji
and
all my family members in Nagpur
v
ACKNOWLEDGEMENTS
I would like to thank Dr. Sam Mannan for all his guidance and encouragement
during this study and for his invaluable mentorship. I’m grateful to Dr. Rogers for his
contributions to all the projects and for being extremely patient. I would like to thank Dr.
Ford for serving on my committee and allowing me access to his SGI. I would like to
express my appreciation to my committee members: Dr. Dan Shantz and Dr. Michael
Hall.
I learnt a lot from my collaborators and would like to express my gratitude to
individuals I worked with: David Frurip, Sima Chervin, Seshu Dharmavarm, and
Abdulrehman Aldeeb. I would like to thank Dr. Lisa Pérez for her help with theoretical
calculations; she has always been a source of inspiration. I thank the Texas A&M
Supercomputing Facility for computer time and the Laboratory for Molecular Simulation
(LMS) for software and support.
I would like to acknowledge my family members, friends, roommates, former
teachers, and students and staff of the Mary Kay O’ Connor Process Safety Center for
being supportive of my work.
Everyday I promise myself not to commit any faux pas and tell God that I’ve
been good so far; but soon I wake up. Therefore I would like to take this opportunity to
2. Classifying Reactive Hazards………………………………………. 16 2.1 Classification Based on Sensitivity Tests…………………... 17 2.2 Classification Based on Calorimetric Data…………………. 17 2.3 Proposed Classification……………………………………. 21
3. Further Investigation of DSC Parameters to Quantify Reactivity…. 27 3.1 Experimental Data…………………………………………... 29 3.2 Results and Discussion……………………………………… 31
4. Conclusions………………………………………………………… 35
III STRUCTURE BASED PREDICTION OF REACTIVITY
HAZARDS ……………………………………………………………
38 1. Review of the Available Methods…………………………………. 39
1.4 CART……………………………………………………… 46 2. Advanced Prediction Techniques…………………………………… 48 3. Application of Transition State Theory for Thermal Stability
Prediction…………………………………………………………… 49
3.1 Model Development……………………………………….. 50 3.2 Experimental Details……………………………………….. 58 3.3 Results and Discussion…………………………………….. 59
4. Correlating Calorimetric Data with Molecular Descriptors……….. 64 4.1 Data Set Selection………………………………………….. 65 4.2 Discussion of a Few Descriptors…………………………… 66 4.3 Correlations………………………………………………… 69 4.4 Correlations Using the Semi-empirical Method, AM1……. 73
5. Conclusions and Future Work……………………………………….. 76
IV DETAILED INVESTIGATION OF A REACTIVE SYSTEM……… 81
1. Background………………………………………………………… 81 2. Ab initio Heat of Formation for HA………………………………. 82
2.1 Computational Methods…………………………………… 84 2.2 Results and Discussion…………………………………….. 87 2.3 Choice of Best Values……………………………………… 91
The reactive chemical section of the TCPA rule utilizes lists of chemicals and certain
functional groups, along with threshold values, as a trigger to perform reactive hazard
assessment. But neither the New Jersey regulation nor the CSB proposal consider
reaction kinetics.
It is important to point out the extreme difficulty to list properties characterizing
reactivity hazards. For example, commercial explosives are typically characterized by
2000 cal/g or more of energy; however, most of the chemicals leading to incidents in the
chemical industries have energies between 500 and 1500 cal/g. Thus, there are a variety
of reasons, besides the energy content, that can pose chemical reactivity hazards, and
recognizing chemicals and hazardous conditions is an area of considerable research. The
3
objectives of this work are to study the characteristics of reactivity hazards, expedite
hazard assessment, develop guidance for recognizing, and evaluating more hazardous
compositions and thus enable better utilization of resources.
Chapter II discusses experimental characterization of reactive hazards and
proposes a classification to rank reactive hazards. Chapter III highlights the role of
molecular modeling to predict reactive hazards and presents results of molecular
modeling. Chapter IV discusses the hydroxylamine system, as an example of a highly
reactive system.
4
CHAPTER II
EXPERIMENTAL CHARACTERIZATION OF REACTIVE HAZARDS
1. Experimental Techniques
There are a variety of experimental techniques to characterize and quantify
hazards due to chemical reactions. Experimentation provides a better understanding of
the energy content of a substance and its behavior under various conditions. Such
information is extremely useful for assessing reactive hazards and managing risks.
Several popular experimental techniques are discussed in Sections 1.1 and 1.2.
1.1 Thermal Analysis
A reactivity hazard involves conversion of stored chemical energy of the
components into mechanical or heat energy, and it is the uncontrolled release of this
stored energy that causes the damage in a reactive chemical incident. The reactivity of a
substance is normally assessed by performing calorimetric measurements.6 Information
about the amount of energy released and the rate of energy released for a process
chemical can be obtained by performing calorimetric tests. A small amount of the
sample is heated over a range of temperature (usually within 30 oC – 400 oC), and
temperature, pressure, and time data are recorded. This information is then used for
alarm settings, relief sizing, and process modeling. Overall thermodynamics and kinetics
of a reaction can be estimated from temperature-time data obtained from a calorimeter,
and this information is used to identify the material hazards posed by a composition and
risk of potential runaway reactions.
5
1.11Types of Calorimeters
There are various calorimeters available for performing reactive hazard
assessments. Prior to detailed testing, screening tests are performed7 using calorimeters
such as a Differential Scanning Calorimeter (DSC) or the Reactive System Screening
Tool (RSST) from Fauske and Associates (http://www.fauske.com). Such screening tests
are relatively inexpensive and can be performed quickly. Detailed testing can be
performed using other calorimeters such as the Automated Pressure Tracking Adiabatic
Calorimeter (APTAC) from TIAX (http://www.tiax.biz) or the Vent Sizing Package
(VSP) from Fauske and Associates. A comparison of three available calorimeters is
presented in Table 2.1 and a brief discussion of the various calorimeters is provided in
the following paragraphs.
Table 2.1. Comparison of Available Calorimeters
Calori- meter
Capital Cost
Time for a run
Sample size
Scanning Rate
(oC/min)
Data obtained
Comments
DSC 1 $ 1 hr 1-10 mg 10 T vs. time
Popular method to screen
reactive hazards
RSST 1.5/2 $ 6 hrs Up to 10 ml 1-5 T,P vs.
time
Open cell; data can be used for
relief sizing
APTAC / VSP 5 $
12-16 hrs
Up to 130 ml 1-2
T,P vs. time
Maintains adiabatic
conditions; maintenance
intensive
6
DSC
A DSC run can provide an overall indication of exothermic activity of the
composition being tested and can help assess potential reactive hazards. In a DSC, a
sample and a reference are subjected to a continuously increasing temperature and heat
is added to the reference to maintain it at the same temperature as the sample. This added
heat compensates for the heat lost or gained as a consequence of an overall endothermic
or exothermic reaction. When the rate of heat generation (Watts) in the sample exceeds a
particular value, the heat supply to the sample is cut off and this additional heat gain is
attributed to exothermic activity within the sample. This cut-off value depends on the
sensitivity of the particular instrument. For an exothermic reaction, a heat vs. time curve
exhibits a peak as shown in Figure 2.1. A base line is constructed from the initial heating
mode, and another line is drawn to coincide with the initial rise due to the exotherm. The
temperature, at the intersection of the two lines is called the onset temperature and
corresponds to a detectable level of heat due to a chemical reaction. The detected onset
temperature is thus a measure of the reaction kinetics and serves as a guideline for
selecting process or storage temperature. The energy released (-∆H) during the process is
calculated as the area under the heat-supplied (Watts) and time curve. DSC is a popular
screening tool because it is safer, since it involves a small amount of sample (1-10 mg),
and is faster, since with 10 oC/min scanning rate, a DSC run can be completed in an
hour. Normally, during a DSC experiment, pressure data are not recorded.
7
Figure 2.1. A typical DSC run.
RSST
The RSST uses a 10-ml sample cell contained in an open, well-insulated glass
test cell. A cross-section of the RSST is shown in Figure 2.2. The RSST is a ramping
calorimeter and ramps the temperature of the sample at a fixed rate using an electric
heater. It allows scanning rates up to 2 oC/min and can generate temperature, pressure,
and time profiles. Output from the RSST for 50 wt% hydrogen peroxide-water system is
illustrated in Figure 2.3. The RSST is used for screening reactive hazards, since it
provides temperature, pressure vs. time data at a moderate cost compared to the APTAC
or VSP.
Baseline Onset temperature (To)
Temperature
Hea
t sup
plie
d (W
)
8
Figure 2.2. RSST.
0
50
100
150
200
250
300
Time (min)
0
50
100
150
200
250
300
350
400
450
500
Figure 2.3. RSST run on 50 wt% hydrogen peroxide.
Sample cell
Container vessel
Temperature
Pressure
9
APTAC
The DSC and the RSST, discussed earlier, are quasi-adiabatic calorimeters, since
the sample cell losses heat to the surroundings. Adiabatic calorimeter minimizes the heat
loss to the surrounding by maintaining the surrounding temperature as close to the
sample temperature, and has proven to be an extremely useful tool to assess thermal
hazards. Following the screening tests, detailed measurements are generally performed
for more hazardous compositions using an adiabatic calorimeter such as the Automated
Pressure Tracking Adiabatic Calorimeter (APTAC). A cross-section of the APTAC is
shown in Figure 2.4 and a typical output is illustrated in Figure 2.5.
Figure 2.4. APTAC.
Reaction vessel Side
Stirrer Pressure
vessel
Top
Bottom
heater
Tube heater
10
The APTAC can be operated in a variety of test modes, such as heat-wait-search,
heat-ramping, and isothermal. If the self-heat rate of the sample is greater than a preset
threshold (0.1 °C/min), the apparatus tracks the reaction adiabatically until the reaction
is over or if one of the shutdown criteria is met. If no exotherm is detected, the sample is
heated to the next search temperature and the steps are repeated until one of the shut-
down criteria is met. Besides the temperature, the pressure outside the sample cell is
controlled to match the pressure inside the sample cell.
Figure 2.5. APTAC run on 50 wt.% hydroxylamine-water system
Hydroxylamine decomposition test
0
50
100
150
200
250
300
400 600 800 1000 1200 1400 1600
Time (min)
Onset TemperatureTo
Maximum TemperatureTmaxStep Heating
Wait
Search
Adiabatic Mode Begins
Adiabatic Mode Ends
11
1.2 Sensitivity Tests
Calorimetric tests capture temperature-time response of a substance and are
performed to detect thermal instability. However, the energy stored within a substance
can be released by a variety of stimuli. Sensitivity is defined as the ease with which a
substance subjected to external stimuli, such as shock, impact or heat, can undergo
detonation.8 A few of the techniques used to determine sensitivity9 of a material are
discussed below.
1.21 Shock Sensitivity10
The Gap test determines initiation of an explosion of a substance due to
detonation in the vicinity. Two cartridges of the smallest commercially manufactured
diameter are coaxially attached on a rod made of soft iron, wood, or plastic, as illustrated
in Figure 2.6. The gap value is the distance between the two cartridges. The gap medium
is such that it stops flying particles and direct heat transmission completely, thus serving
as a heat filter. Consequently, the shock wave is the only energy transmitted to the
substance being tested. The donor charge is a well-characterized explosive, for example
50 g RDX, that generates a known pressure wave (shock wave), and is set off during the
test. The resulting shock wave, generated during this explosion, is transmitted to the
testing material and may trigger a detonation.
12
Figure 2.6. Gap test.10
Detonation in the test sample is verified on the basis of observed mechanical effects. The
test results, based on degree of fragmentation observed, are typically reported as follows:
‘Yes’ or ‘+’ Tube completely fragmented or fragmented at both ends
‘Partial’ Tube only fragmented at booster end but fragmented length is more
than 1.5 times the average fragmented length found with an inert
material.
‘No’ or ‘−’ Fragmented length is less than 1.5 times the fragmented length with
an inert substance.
1.22 Impact Sensitivity11
During impact tests, the impact of a drop-weight on a substance is assessed. The
sample, placed between two flat, parallel, hardened steel surfaces, is subjected to an
impact by dropping a weight. The impact may result in initiation depending on
sensitivity of the material, weight mass, and its drop height (impact energy). Initiation is
13
observed by sound, light effects, or smoke, or by inspection. The BAM impact
apparatus, known to give fairly reproducible results, is shown in Figure 2.7. Typically
drop weights having a mass of 1, 2, 5, or 10 kg are used and the lowest impact energy
required to create a detonation is recorded. Thus drop-weight and drop-height at which
the initiation of the sample occurs are the main parameters determined from impact
testing. The drop height at which detonation is observed is thus a measure of impact
sensitivity of an explosive.
Figure 2.7. BAM impact sensitivity apparatus.11
1. Guiding rods 2. Locking and unlocking devices 3. Drop weight 4. Calibrated scale 5. Indented rod 6. Piston device 7. Anvil 8. Steel block 9. Steel Base
14
1.23 Heat Sensitivity10
Heat sensitivity tests, such as Koenen and time/pressure, are performed to assess
the role of heat in initiation of explosives.
Koenen test
The Koenen test measures the effect of strong heating under confinement. The
test sample is contained in a drawn steel tube (27 ml) equipped with a closure, which
allows orifice plates with various apertures of diameter 1.0, 1.5, 2.0, 2.5, 3.0, 5.0, 8.0,
12.0, or 20.0 mm. The tube is heated with four calibrated propane burners. The result
reported from such a test is the largest size orifice at which the tube is fragmented, and
the following guidelines are used for reporting:
‘Violent’ Limiting diameter greater than or equal to 2.0 mm.
‘Medium’ Limiting diameter is 1.5 mm.
‘Low’ Limiting diameter is equal to or less than 1 mm but an effect is observed on
the tube.
‘No’ Limiting diameter is less than 1 mm and in all tests the tube is unchanged.
A schematic of the apparatus used for performing Koenen tests is shown in Figure 2.8.
15
Figure 2.8. Koenen Test.
Time/Pressure test
This test measures the ability of a material to deflagrate under confinement. A 5
g sample is subjected to a flame in a pressure vessel (20 ml) fitted with a pressure
recording device and a bursting disc (2200 kPa). Based on the shortest time in three runs
for the pressure to rise from 690 to 2070 kPa, test results are classified as follows:
‘Rapid’ Time is less than 30 ms.
‘Slow’ Time is 30 ms or more.
‘No’ A gauge pressure of 2070 kPa is not achieved.
A schematic of assembly used for the time/pressure test is shown in Figure 2.9.
Nozzle
Burner
Sample test tube
16
Figure 2.9. Time/Pressure test assembly.
Sensitivity tests typically require more sample, elaborate testing facilities, and
are more expensive than calorimetric tests. The next section discusses hazard
classification of substances based on sensitivity tests, calorimetric tests, and
interrelationship between the two.
2. Classifying Reactive Hazards
Based on tests discussed in the earlier section, researchers have attempted to
develop a classification to enable ranking of chemicals based on material hazards. Such
a ranking can help can help to develop guidelines for handling, storage, and
transportation of materials.
Burst Disc
Sample (5 g.)
Pressure Transducer
Ignition system
17
2.1 Classification Based on Sensitivity Tests
Table 2.2 provides an explosivity-ranking scheme based on the three
recommended UN sensitivity tests, namely Gap, Koenen, and time/pressure. It is
identical to that used by Whitmore12, except this scheme gives precedence to the UN
Gap result over the BAM 50/60. Ranks A and B identify potential Class 1 substances:
“A” indicates substances that detonated, and “B” indicates those substances that did not
detonate but were strongly positive in the Koenen and/or Time/Pressure tests. Rank C
substances, which had milder results in the Koenen and Time/Pressure tests, are not
Class 1 but are candidates for classification as Self-Reactives or Organic Peroxides.
Rank D substances exhibited no positive results but may still be Self-Reactives or
Organic Peroxides, based on the results of the other recommended tests. This
classification is popular for categorizing substances for transportation.
2.2 Classification Based on Calorimetric Data
Unlike sensitivity tests, hazard classification based on calorimetry is not
well-established. There is disagreement among researchers regarding parameters
characterizing reactive chemicals. A part of the problem is that results from calorimetric
studies are highly dependent on the calorimeter and other conditions during the
experiment. The remainder of this chapter discusses approaches for developing a hazard
classification and associated issues.
18
Table 2.2. Explosivity Rank
Explosivity Rank
Severest Class 1 Property Correspondence to UN Classification
A Detonates (positive result in UN Gap, or BAM 50/60 or TNO 50/70 if UN Gap unavailable)
Potentially Class 1
B Heating under confinement: Violent (Koenen limiting diameter >2 mm), and/or Deflagration: Rapidly (pressure in Time/Pressure >2070 kPa in <30 ms)
Potentially Class 1, but does not detonate
C Heating under confinement: Medium or Low (Koenen limiting diameter <1.5 mm), and/or Deflagration: Slowly (pressure in Time/Pressure >2070 kPa in >30 ms)
Not Class 1
D No effect of heating under confinement, and does not deflagrate (pressure rise in Time/Pressure <2070 kPa)
No explosive properties with respect to transport classification
As discussed in Section 1, calorimetry or thermal analysis techniques represent
temperature, pressure vs. time behavior of a substance. From the temperature-time data,
the energy released (-∆H) during the process is calculated using the following formula:
)( max op TTmCH −Φ=∆−
where
p
ppss
mC
mCCm +=Φ – Phi factor
ms – Mass of the sample cell
19
Cps – Heat capacity of the sample cell
m – Mass of the sample
Cp – Heat capacity of the sample
Tmax – Maximum temperature attained by the sample during the reaction
From the temperature-time data, the rate constant for the reaction is obtained using the
following formula13:
( )o
n
o
TTTTTT
dtdT
k
−
−−
=
maxmax
max
where
n – order of the reaction
Thus, the overall thermodynamics and kinetics of a reaction can be estimated from
temperature-time calorimetric data.
The National Fire Protection Association (NFPA) recommends a classification
for intrinsic thermal instability14 of a substance based on Instantaneous Power Density
(IPD), which is defined as
RateHIPD *∆−=
RTEordero AeACHIPD /*** −∆−=
where
-∆H – enthalpy of reaction (cal/g)
Rate – Rate of reaction = A*exp(-EA/RT)*Coorder (g/ml s)
A – Arrhenius pre-exponential factor
20
EA – Activation energy
Co – Initial concentration of the material
R – Gas constant
Therefore IPD has units of W/ml, and the rate of reaction can be obtained from
calorimetric data. Based on the IPD, the classification illustrated in Table 2.3 is applied
for rating thermally unstable compounds.14
Table 2.3. Instability Rating Based on IPD
Instability Rating IPD at 250 oC
W/ml
Decomposition
initiation temperature (oC)
4 IPD ≥1000
3 100 ≤ IPD < 1000
2 10 ≤ IPD < 100 < 200
1 0.01 ≤ IPD < 10 200 ≤ IPD < 500
0 < 0.01 < 500
As an example, the IPD is calculated for the following system14:
Enthalpy of decomposition (-∆H) : -80.5 cal/gm
Arrhenius activation energy (EA) : 36.4 kcal/mol
Arrhenius pre-exponential (A) : 1.6 * 1015 /s
Reaction order : 1
Initial concentration or density of pure material: 0.8 g/ml
19 Malononitrile 276 1848 4.43 6.7 111.2 No No No 20 Organic perchlorate # 2 229 1465 0.07 6.4 96.8 No No No 21 Dilauroyl peroxide 88 721 0.13 8.2 76.9 No+ No No 22 3-Thiosemicarbazide 176 908 0.24 5.2 68.4 No No No
+BAM 50/60, ++TNO 50/70 instead of UN Gap
31
3.2 Results and Discussion
Earlier correlations failed to recognize a few of the Class A compounds. As
suggested above, additional information from calorimetric data, besides the onset
temperature and decomposition energy, should correlate more effectively with the UN-
recommended tests. The following parameters were obtained for each compound from
DSC experiments:
1. Decomposition energy (∆H or E, J/g)
2. Extrapolated onset temperature, (To, oC)
3. Peak height (W/g)
4. Peak width, at half the peak height (oC)
5. Aspect ratio (peak height/peak width at half peak height)
6. Initial slope (W/g/oC)
7. Maximum slope (W/g/oC)
Various combinations of the above set of variables were used to investigate the
correlation among DSC data and UN tests (Gap, Koenen, and Time/Pressure)
individually. A possible relationship was investigated by employing one of the above
parameters in combination with decomposition energy, which was employed for all the
trials. No more than two parameters were used during a trial, except for the proposed
correlation for UN Gap test. A visually satisfactory separation of data points on the
graphs was considered an acceptable solution. An energy threshold of 500 J/g is
employed for all the correlations, and any composition with energy less than 500 J/g is
not considered to be in Class 1.
Correlations among UN tests and DSC data, based on this analysis, are discussed
below:
32
3.21 Gap Test
For the Gap test, it was found that ∆H/To0.5 and the aspect ratio, obtained from
the DSC data, separated the detonating compounds (“Yes” on the Gap test), as shown in
Figure 2.13. The decomposition energy is a measure of the heat content of a system and,
therefore, the energy available for detonation. A lesser onset temperature is indicative of
faster kinetics and, consequently, a more rapid energy release for high energetic
materials.
The aspect ratio (peak height/peak width) represents the decomposition behavior
of a substance. For an ideal explosive, the peak height would be infinite and the peak
width would be zero, and therefore the higher the aspect ratio, the higher the tendency
for a violent decomposition. Compounds having ∆H/To0.5 > 88 J/g oC0.5 and aspect ratio
> 0.2 W/g oC are expected to exhibit a positive Gap test.
10
100
1000
0 0 1 10 100
Aspect ratio (W/g oC)
∆
Yes
No
y = 88
x = 0.2
Figure 2.13. Correlation between UN Gap test and DSC data
33
3.22 Koenen test
The onset temperature and heat of reaction, as shown in Figure 2.14, yielded
optimum separation for compounds exhibiting “Violent” behavior on the Koenen test.
Based on limited data set, it is proposed that compounds above the line, ∆H = 12.4 To –
796, are expected to exhibit a violent behavior on the Koenen test. Thus a Koenen
potential (KP) can be defined,
KP = ∆H – (12.4 To – 796)
such that if KP > 0, then the compound will display a violent behavior on the Koenen
test.
0
500
1000
1500
2000
2500
3000
3500
4000
0 50 100 150 200 250 300 350 400 450
To(oC)
∆
Violent
Medium
Low
No
y = 12.4 x - 796
Figure 2.14. Relationship between the Koenen test and DSC data.
34
3.23 Time-pressure
As seen from Figure 2.15, a ∆H/To ratio of 8 (J/g oC) separates the compounds
Figure 2.15. Relationship between the Time/Pressure test and DSC data.
Based on the above recommendations, the predicted class are listed in Table 2.6. It is
apparent that by applying the proposed correlations, unlike previous correlations, all of
the Class A and B compounds can be identified. The correlations proposed in this work
screen out 5 of the 11 Class C and D compounds and all of the Class A and B
35
compounds. A summary of the predictions using earlier correlations and this work is
summarized in Table 2.7.
UN-recommended tests for transportation are time consuming and expensive.
Earlier researchers have proposed correlations to predict UN test outcomes using
parameters from DSC data, such as onset temperature and heat of reaction. Based on the
above correlations, DSC data can be used effectively to recognize Class 1 compounds.
Therefore, these correlations can be employed as a screening tool to reduce testing and
focus resources on more hazardous chemicals.
4. Conclusions
Various experimental techniques are available for characterizing reactive
hazards, such as sensitivity tests and thermal analysis. A classification, based on To and
-∆H obtained from calorimetric data is proposed to facilitate ranking of reactive
chemicals. Although it is true that criteria utilizing multiple parameters, such as
pressure change and rate of heat generation, obtained from calorimetric data would lead
to a better classification scheme, but the main aim was to establish a basis for
classification. By neglecting pressure effects, the classification scheme could miss
mildly exothermic but pressure generating reactions. The pressure criterion was not
included because DSC is often employed to gather data and it does not gather pressure
vs. time data. The correlations discussed in Section 3 of this Chapter guide the selection
of parameters for classification. For example, the aspect ratio correlates well with the
Gap test, a measure of explosive tendency of a material and is therefore a potential
candidate for refining the proposed classification. Since the sensitivity tests are
typically more resource consuming, correlations among DSC and sensitivity tests, can
serve as an effective tool for screening compounds for sensitivity testing.
36
Table 2.6. Explosion Propagation vs. Actual Rank
Sr. no.
Material Predicted Class
Explosion Propagation
Actual Class
UN Gap
Koenen Time/ Pressure
EP19 EP’22
1 2,4-Dinitrophenyhydrazine A 0.33 0.36 A Yes+ Violent NA 2 2,4-Dinitrotoluene A 0.30 0.33 A Yes+ Medium Slow 3 4-Nitrophenylhydrazine 100% A 0.20 0.24 A Yes Violent Rapid 4 3,5-Dinitrobenzoic acid A 0.20 0.23 A Yes Low NA
5 2-Bromo-2-nitropropane-1,3-diol A
0.18 0.22 A Yes+ Low No
6 2,2’-Dithiobis(4-methyl-5-nitrothiazole) A
0.18 0.22 A Yes Medium Slow
7 Benzoyl peroxide 100% A 0.06 0.13 A Yes+ Violent Rapid
8 Benzoyl peroxide 70% with H2O A
-0.05 0.01 A Yes (No+) Violent Rapid
9 2-Chloro-5-nitrobenzoic acid A -0.02 0.01 A Yes Medium No 10 t-Butyl peroxybenzoate B 0.07 0.13 B No+ Violent Slow
11 2-Diazo-1-napthol-5-sulphochloride B
-0.13 -0.07B No+ Violent NA
12 4-Nitrophenylhydrazine 76% with H2O B 0.18 0.23 C No Medium No
13 2-Amino-4-chloro-5-nitrophenol A 0.06
0.10 C No Low Slow
14 Di-t-butyl peroxide A -0.01 0.05 C No+ No Slow 15 1-Phenyl-5-mercapto tetrazole A -0.01 0.04 C No++ Low Slow 16 Organic perchlorate # 1 C/D -0.07 -0.03 C No Low No
18 3-Nitrobenzenesulfonic acid sodium salt C/D -0.21
-0.19 D No No Slow
19 Malononitrile A 0.06 0.10 D No No No 20 Organic perchlorate # 2 C/D 0.00 0.04 D No No No
21 Dilauroyl peroxide B -0.12 -0.05 D No+ No No
22 3-Thiosemicarbazide C/D -0.16 -0.11 D No No No
+BAM 50/60, ++TNO 50/70 instead of UN Gap, Bold EP and EP’ values indicate incorrectly categorized Class A and B compounds
37
Table 2.7. Summary of Sensitivity Tests Predictions
A further advancement to reduce experimentation and expedite hazard evaluation
is prediction of experimental data using computational techniques, which is discussed
in the next chapter.
Yoshida19 Bodman22 This work
Number of Class A and B compounds successfully categorized
8/11 10/11 11/11
Number of Class C and D compounds successfully screened 7/11 5/11 5/11
38
CHAPTER III
STRCUTURE BASED PREDICITION OF REACTIVITY HAZARDS
As discussed earlier, a reliable experimental technique for assessing reactivity is
calorimetric analysis, which can be resource consuming and thus possible only for a
limited set of compounds. A computational screening tool can reduce experimentation
and screen out benign compounds. This chapter provides a brief review of computational
methods that can be employed quickly to estimate reactive hazards and a description of
efforts to develop methods for predicting calorimetric data. A goal of this research is to
develop a computerized program for screening reactive hazards.
Generally, rules of thumb based on prior experience and chemical knowledge are
used for screening and estimating reactive hazards of compounds. For example, the
presence of a ‘nitro’ group is regarded as an indicator of potential energy, as in
trinitrotoluene (TNT). Attempts have been made to develop a generalized framework for
estimating reactive hazards based on molecular structure, such as the oxygen balance
method25, Chemical Thermodynamic and Energy Release Evaluation (CHETAH)18, and
Calculated Adiabatic Reaction Temperature (CART).26 A brief description of these
methods is provided in Section 1 of this chapter. The above methodologies have
limitations, and considerable chemical intuition and experience are required for their
effective use. Also, the reliability of estimations for a range of compounds and process
conditions varies significantly.27 Sections 2 through 5 of this chapter discuss
Part of this chapter is reprinted with permission from, “Prediction of reactive hazards based on molecular structure”, by S.R. Saraf, W.J. Rogers, and M. Sam Mannan, 2003, J. Hazardous Materials, A98, 15-29. Copyright 2003 by the name of Elsevier. Part of this chapter is reprinted with permission from “Application of transition state theory for thermal stability prediction”, by S.R. Saraf, W.J. Rogers, and M. Sam Mannan, 2003, I & EC Res., 42, 1341. Copyright 2003 by the name of American Chemical Society (ACS).
39
improvements and advancements in predictive techniques for reactive hazard
assessment.
1. Review of Available Methods
This section reviews some popular methods for reactivity hazard evaluation,
including their strengths and limitations, and it attempts to provide the reader with
enough information to choose a method for a particular application. Some of these
methods have been reviewed previously.26
1.1 Rules of Thumb
The presence of certain functional groups is considered an indicator of reactivity.
This is the simplest possible reactivity screening method and serves as a guideline for
further analysis. For example, chemicals containing the following functional groups can
be considered potentially reactive:
-NO2 : organic nitro compounds
-O-O-, -O-OH : organic/inorganic peroxide and hydroperoxide compounds
-C C- : triple bonded carbon atoms as in acetylene and acetylenic compounds
A comprehensive summary of reactive groups can be found in Bretherick’s handbook28,
and a few of the reactive groups are summarized in Table 3.1.
40
Table 3.1. Functional Groups Indicative of Reactive Hazards 4
Groups containing Carbon -C≡C- Acetylenic compounds -C≡C-M Metal acetylides -C≡C-X N=N C
Abbreviations: Ar = aromatic (benzene); M = metal; R = organic chain; X = halogen; E = nonmetal; Z = anion; n = integer variable; all other abbreviations are for the element symbols from the periodic table of elements Note: Not all chemical bond symbols are shown .
Table 3.1. (Contd.)
42
1.2 Oxygen Balance Method
A quantitative correlation has been demonstrated between oxygen balance and
various measures of explosive effectiveness for several classes of more than 300
compounds organic explosives25, and the following formula was recommended for
calculating oxygen balance for a compound:
where
X – Number of atoms of carbon
Y – Number of atoms of hydrogen
Z – Number of atoms of oxygen
MW – Molecular weight
The above formula yields a value of zero for oxygen-balanced compounds, negative for
oxygen-poor, and positive for oxygen-rich compositions. This method is a criterion for
evaluating self-reactivity in the CHETAH program, and the classification indicated in
Table 3.2 is recommended for estimating hazard potential based on oxygen balance.29
Table 3.2. Oxygen Balance and Hazard Rank
Oxygen Balance Hazard Rank
More positive than +160 Low
+160 to +80 Medium
+80 to –120 High
–120 to –240 Medium
More negative than –240 Low
MWZYX
BalanceOxygen)2/2(1600 −+−
=
43
1.21 Strengths
The oxygen balance method is useful for estimating hazards of organic nitro
compounds and is universally employed in the explosive industry. In general, this
method is applicable to compounds containing C, H, N, and O.29
1.22 Weaknesses
However, it has been shown that there is no necessary connection between
oxygen balance and self-reactivity29. For example, water (H2O) has an oxygen-balance
value of 0 and is given a ‘high’ hazard ranking by this criterion. Also, the method cannot
be applied to oxygen free but hazardous compounds such as acetylene. Application of
the above oxygen balance equation to low-oxygen content or oxygen-free compounds
produces a highly negative, non-hazardous ranking regardless of the actual hazard
potential.
1.3 CHETAH18
CHETAH is a popular program available from the American Society for Testing
and Materials (ASTM) for prediction of reactivity hazards. The software uses the
‘Benson group contribution method’ 30 to estimate heat capacity, heat of formation, and
heat of combustion for a multitude of compounds. Also, the program includes a database
of thermo-chemical properties for selected organic and inorganic compounds. CHETAH
classifies chemicals based on their potential for violent explosion and includes the
following hazard evaluation criteria:
§ Maximum heat of decomposition
§ Oxygen balance
44
According to the first criterion, compositions with heats of reaction more
negative and therefore more exothermic than – 2.929 kJ/g are placed in a ‘high’ hazard
category. A detailed explanation of the above evaluation criteria and three additional
ones can be found in the CHETAH reference manual, and a critical review of CHETAH
for predicting reactivity hazards is available.31 The first criterion has proved to be a
reliable indicator of potential reactive hazards. The other criteria, however, are not
effective for all chemicals and compositions.31
1.31 Input to the Program
The molecular formula of a compound is the only input to the program. From the
included database, the thermodynamic properties are estimated and the hazard criteria
are determined from these values. An energy release evaluation sheet from
CHETAH 7.2 is shown in Figure 3.1. Based on its maximum heat of decomposition,
H2O2 is given a “high” hazard classification. However, it should be noted that the
estimated product spectrum might be incorrect. Thus, one problem is the thermodynamic
feasibility of a proposed stoichiometry under process conditions. Further, the program
gives no indication about the sensitivity to reaction initiation or process conditions to be
avoided. It is difficult to determine conditions under which H2O2 may pose reactive
hazards based only on such an analysis. Thus, the problem of reactivity is not just a
combinatorial problem (stoichiometric analysis) as implicitly suggested by this method.
It is important to realize that CHETAH provides an estimate of a material hazard but not
an estimate of the process risk in using the material.
1.32 Strengths
The software is user friendly and offers the flexibility to include user-defined
group values. It is computationally inexpensive and can be installed on a standard PC.
45
ENERGY RELEASE EVALUATION
Compound Name: Hydrogen Peroxide Formula: H2O2 Molecular weight: 34.015 Amount: 1 Mole(s) Heat of formation at 25 oC: -32.530 kcal/mol
PLOSIVE HAZARD CLASSIFICATION: Over-all Energy Release Potential is HIGH Value: -1.375
Contributing Details: Criterion Value Units Hazard Classification Maximum Heat of Decomposition (#2) -0.743 kcal/g HIGH Fuel Value – Heat of Decomposition 0.000 kcal/g HIGH Oxygen Balance (#3) 47.037 percent HIGH CHETAH ERE Criterion 4 46.934 kcal2/mol MEDIUM Total Number of Peroxide Bonds 1.000 Net Plosive Density (#4) 0.435 PLOSIVE Warning: These ratings only apply to hazards associated with strong mechanical shock. This does not
imply the absence of other hazards. Notes #1 This evaluation was developed to classify a composition as able or not able to decompose with
violence, if subjected to the proper conditions. Information on the interpretation of hazard classification criteria used by CHETAH may be found in the CHETAH documentation. (ASTM publication DS-51A) and in J. Chem. Ed., v66, A137 (1989).
#2 For decomposition products shown. #3 Experience has shown that the oxygen balance criterion is useful only for compounds composed
of the elements C, H, N, and O. #4 Sum of auxoplosive and plosphoric weights per gram of mixture.
Decomposition Products (chosen to maximize heat of decomposition) Moles State Species 0.500 ref-gas O2 Oxygen 1.000 gas H2O Water
HEAT OF COMBUSTION SECTION Fuel Value (Net Heat of Combustion)
* Standard heat of formation values at 1 atm and 298.15 K are from the NIST Chemistry WebBook. † Calculated by CHETAH. ‡ Because the heat of formation value for 2 nitrotoluene was not available, an average of decomposition
values for 3 nitrotoluene and 4 nitrotoluene was used. § Because the heat of formation value for 3,4 dinitrotoluene was not available, an average of
decomposition values for 2,4 and 2,6 dinitrotoluene was used.
54
Figure 3.2. Hypothesized potential energy surface (PES) for a runaway reaction.
Reaction co-ordinate
Ene
rgy
EA
1st step
55
The rate constant for an unimolecular reactions under high pressure conditions is given
by the equation40:
RTEB A
A
TS eqq
hTk
k /*
−= (3.3)
where
kB - Boltzmann constant (J/K) = 1.38 x 10-23
h - Planck constant (Js) = 6.63 x 10-34
R - Gas constant (cal/mol K) = 1.987
T - Temperature (K)
qTS* - Partition function for the transition state (with 1 degree of freedom, along the
reaction coordinate, removed)
qA - Partition function for the reactant
EA - Activation energy (cal/mol)
In all further equations, the activation enthalpy is represented by EA and its
approximated value is based on bond dissociation energy. Each partition function is a
product of translational, vibrational, rotational, and electronic partition functions. For
unimolecular decomposition, the ratio of qTS and qA differs by one degree of vibrational
freedom - the one along which reaction occurs, and this ratio can vary between ~ 0.1 and
~1. For the simulations discussed here, this ratio is approximated by 1 to obtain a
conservative estimate for k. Substituting the values for kB and h, the rate constant can be
written as:
min)(/10*25.1)(/10*08.2 1210 RTE
RTE
RTE
BAAA
eTseTehTk
k−−−
=== (3.4)
Thus, EA is the only missing parameter and is estimated as discussed below.
Following the Polanyi type equation41,
56
EA = EA0 + γP?Hrxn
where
EA - Activation enthalpy for an elementary step
EA0 - Intrinsic activation enthalpy for a reaction class
γP - Proportionality constant, called the transfer coefficient, for a reaction class
?Hrxn - Heat of reaction for the elementary step
For a bond scission reaction the above equation reduces to42:
EA = γP * BDE (3.5)
where EA0 ~ 0
We further assume that the overall kinetics can be approximated by applying the
TST to evaluate the kinetic parameters for the rate-limiting first step. The activation
energy is therefore a fraction of the bond dissociation energy (BDE) of the weakest
bond, C−NO243. Computational chemistry calculations were performed to calculate the
BDE at the B3P8633 level of theory with the cc-pVDZ34 basis set, and the quantum
mechanical calculations were performed using the Gaussian 9844 suite of programs on
the Texas A&M supercomputer. The Gaussian software calculates energies, optimized
molecular structures, and vibrational frequencies, together with molecular properties that
are derived from these three basic computation types for a chemical formula. Optimized
geometries were obtained for the reactant (R−NO2) and the two fragments
(R• and NO2• ). The BDE is then calculated as
BDE = ENO2• + ER• - ER-NO2 (3.6)
BDE values can be calculated if the experimental heats of reaction are available.
Equation (3.4) reduces to
57
min)(/10*25.1 12 RTBDEP
eTkγ
−= (3.7)
Although the above TST equations apply for reactions in the gas phase, it is assumed
that the rate in the gas phase approximates the rate in a condensed phase. With the
current reaction-solvation theories, this is the best working assumption at this level of
analysis.
3.12 Physico-chemical Properties
The concerned physical properties, density (ρ) and heat capacity (Cv) are
available in the open literature37,38 or can be estimated with reasonable accuracy. For the
compounds considered in this research the Cv varied between 0.25 – 0.35 cal/gm K, and
the density varied between 800 – 1200 kg/m3.
Equations (3.1) and (3.2) can be further simplified to calculate onset temperature.
We assume that the concentration of the reactant (CA) is equal to the initial concentration
(CA0) until the temperature equals the onset temperature. This assumption decouples
Equation (3.1) and (3.2), and Equation (3.2) reduces to
MWCkH
CkCH
CkCH
dtdT
V
rxn
V
Arxn
V
Arxn3
0 10∆−=
∆−≈
∆−=
ρφρ (3.8)
MWC A
3
010*ρ
=Q
Substituting the expression for k from Equation (3.7) in Equation (3.8), we obtain
MWCeTH
dtdT
V
RTBDErxn
3/12 10***10*25.1* γ−∆−= (3.9)
58
Thus, γP is the only undetermined variable, and it serves as an adjustable parameter for
the simulations. When the rate of temperature increase exceeds a particular amount (ε),
the calorimeter detects the exothermic reaction. Thus, when dT/dt ≥ ε, T = To, and the
value of ε depends on the sensitivity of the calorimeter. For a given compound the above
equation has the form
TBeTAdt
dT /** −= (3.10)
where A and B are constants. Therefore at higher temperatures the exponential term
dominates, and values of dT / dt from Equation (3.9) are sensitive to the BDE values.
3.2 Experimental Details
For this work, we used available DSC data on aromatic nitro compounds. Pure
organic nitro compounds, aliphatic and aromatic, decompose at high temperatures and
exhibit large exotherms45. These compounds are identified as energetic materials, and
trinitrotoluene (TNT) is used as an explosive. Since a graphical detection procedure is
employed to obtain To, a variation of 5-30 oC is possible in the reported To values for the
same compound. The energy of reaction (-∆Hrxn) is the net heat released during the
reaction and is not the thermodynamic heat of reaction but includes other effects such as
sublimation, evaporation, adsorption, and enthalpy of mixing. Therefore, the
experimentally determined parameters, To and -∆Hrxn, depend on the type of calorimeter,
sample size, sample phase, and scanning rate. To compare the theoretically predicted
values, we chose experimental data from a single reference46 to maintain consistency the
experimental procedure. In this reference46, authors employed a Mettler TA4000 DSC
(0.2 W/gm sensitivity) with DSC25 measuring cell, scanning rate of 4 oC/min to
determine the To and -∆Hrxn values for 19 nitro compounds.
59
3.3 Results and Discussion
3.31 Choice of γp
Assuming a sensitivity of 0.1 oC/min and substituting the experimental onset
temperature for T and BDE for the 19 compounds listed in Table 3.4, we can determine
γp for each compound from Equation (3.9). The average of the γp values calculated for
mono-nitro compounds is 0.67 ± 0.06 and for di-nitro compounds is 0.71 ± 0.06. For a
sensitivity of 0.01 oC/min at To, the average γp value is 0.70 ± 0.06 for mono-nitro and
0.76 ± 0.06 for di-nitro compounds, respectively. Note that increasing the sensitivity of
the apparatus to detect a lower onset temperature increases the γp but the variation in γp
remains the same. Therefore, there is a correlation between the calculated BDE and the
activation energy required to predict the experimentally determined onset temperature,
and irrespective of the sensitivity of the DSC, the γp parameter can be adjusted to
reproduce the experimental data.
3.32 Results
Based on the above discussion we assumed a value of 0.70 and 0.76 for γp for
mono and dinitro compounds, respectively. A FORTRAN program was written to
calculate dT/dt, as given by Equation (3.9), for temperatures starting with 30 oC. The
temperature was increased by 1 oC if the rate of increase of temperature was less than
0.01 oC/min. The temperature at which the gradient of temperature with time increased
at the specified rate of 0.01 oC/min was taken as the onset temperature. The predicted
onset temperatures for 19 different compounds are presented in Table 3.4 with an
average aggregate error of 11% and a bias of -2%. Typical errors associated with the
DSC detected onset temperatures are within ± 5%.
60
B3P86 AM1
Sr. Structure To (oC) BDE To (oC) ? To (oC) error BDE To (oC) * ? To (oC) error
no. Expt.46 (kcal/mol) predicted exp - pre % (kcal/mol) predicted exp- pre %
Using the AM1 scaled descriptors and the above two-parameter correlation, predictions
for onset temperatures for the 19 compounds yielded an average absolute aggregate error
of 7% and a bias of -1%, which is in good agreement with the predictions obtained using
the more expensive B3P86 descriptors. Values of descriptors and predicted onset
temperatures using AM1 are summarized in Appendix C.
* Statistical analysis was performed on 1,5,6,7,11,12,13,14,15,16,17,18,19 compounds in Table 3.7. † Statistical analysis was performed on 2,3,4,8,9,10 compounds in Table 3.7.
76
5. Conclusions and Future Work
Computational screening to reactive hazards presents fundamental challenges in
predicting material properties and is of great interest to industry personnel. The QSPR
approach was also successfully used to correlate impact sensitivities to molecular
descriptors in our research group.60 Such correlations can be refined into a computerized
predictive tool to be run on a desktop computer.
Mary Kay O’Connor Process Safety Center (MKOPSC) is collaborating with
Dow Chemical Company, Midland, MI, and Eastman Kodak Company, Rochester, NY
to access their reactivity databases and develop correlations for a variety of families of
compounds. An initial analysis of available data indicates lack of data on pure
compounds, and computationally amenable compositions. A data set of 48 mono nitro
molecules, with average molecular weight of 234±85 g/mol was used to investigate
possible correlations with molecular descriptors. The energy of reaction yielded an
average value of 85±14 kcal/mol, consistent with results discussed earlier. However, the
onset temperatures did not yield satisfactory correlations with the descriptors discussed
earlier and lack of data on other families of compounds prevents testing of QSPR
descriptors. It is worth pointing out that calorimetric properties can be grouped
according to the functional groups, and values from different sources are summarized in
Table 3.9, 3.10, and 3.11. Based on values in the tables and the earlier classification, it is
possible to perform an easy screening of reactive hazards. For example, if the process
under consideration is ethylene (CH2=CH2) polymerization to form polyethylene, the
expected amount of heat released 20 kcal/mol, as indicated in Table 3.10. Since the
molecular weight of ethylene is 28 g/mol, the energy released per gram is ~ 700 cal,
indicative of a potential reactive hazard. If a molecule has a combination of functional
groups, the total energy released can be assumed to be sum of energy released by
individual functional groups, and the lowest of the onset temperatures among individual
groups can be taken as onset temperature for the combination.
77
Table 3.9. Decomposition Energies for Typical Functional Groups 48
Future experimental work in our research group will focus on building libraries
of data for compounds, which would then be used to develop correlations based on
molecular level descriptors, and development of robust correlations to enable structure-
based predictions. Based on this analysis one can also envision descriptors (or imprints
of energetic materials) that can be developed into a qualitative hazard index.
The discussion so far has focused on characterizing and screening reactive
hazards. But systems where hazards are identified additional considerations and
resources are required. The next chapter focuses on a detailed investigation of
hydroxylamine (HA), as an example of a hazardous system.
81
CHAPTER IV
DETAILED INVESTIGATION OF A REACTIVE SYSTEM
Almost all mid-size and large companies have a reactive hazard management
program to assess potential reactive hazards during storage, transport, and processing of
reactants, intermediates, and products. For evaluating reactive hazards, it is rational to
develop a systematic protocol with a screening step, utilizing available information,
computations and experiments, as discussed earlier, followed by detailed testing. The
classification discussed above can be used as a screen to select the more hazardous
compositions for detailed testing. An overall assessment approach is shown in the
Figure 4.1. Earlier chapters focused on development of tools that can expedite reactive
hazard assessment. This chapter focuses on an investigation of hydroxylamine system, as
an example of highly reactive system.
1. Background
Hydroxylamine (HA), NH2OH, has recently been involved in two major
industrial incidents with disastrous consequences.62,63 Calorimetric studies on aqueous
solutions of HA indicate that it is a highly reactive compound64, but its properties are
Part of this chapter is reprinted with permission from, “Theoretical thermochemistry: ab initio heat of formation for hydroxylamine”, by S.R. Saraf, W.J. Rogers, M. Sam Mannan, M.B. Hall, and L.M. Thomson, J. Phys. Chem., Vol. 107, No. 8, 2003. Copyright 2003 by the name of American Chemical Society (ACS). Part of this chapter is reprinted with permission from “Hydroxylamine production: Will a QRA help you decide?”, by K. Krishna, S.R. Saraf, Y.J. Wang, J.T. Baldwin, W.J. Rogers, J.P. Gupta, and M. Sam Mannan, Reliability Engineering & System Safety, 2003, 81, 215-224,. Copyright 2003 by the name of Elsevier.
82
Figure 4.1. Systematic approach for assessing reactive hazards.
poorly characterized. Pure HA is known to explode at room temperature, and the
decomposition of HA is extremely sensitive to metal contamination.64
2. Ab initio Heat of Formation for HA
For chemicals with validated experimental data, estimations may not be
necessary, but for reactive substances with insufficient experimental data, such as
hydroxylamine, estimation methods are of prime importance.
311+G(2df,p), 6-31+G(3df,2p), 6-311+G(3df,2p), 6-311++G(3df,2p)) including
diffuse91,87 (denoted by “+” for Pople-style) and polarization functions92 (denoted by
85
“d”, “p”, ‘f”, for angular flexibility to represent regions of high electron density among
bonded atoms) were also employed. Finally, the Bond Additivity Correction (BAC)-
MP4 methodology was employed using the parameters listed by Melius and Zachariah93.
Errors in absolute quantities from quantum chemical calculations are often systematic.
To compensate for some of the systematic errors, isodesmic reactions, which conserve
the number of each type of bond in reactants and products, are used to obtain more
accurate heats of formation94. Here, the following isodesmic reactions were employed
for HA:
H2 + NH2OH ? H2O + NH3 (4.1)
H2O + NH2OH ? H2O2 + NH3 (4.2)
To benchmark the computed HA values, the heat of formation for hydrogen peroxide, a
similar species for which reliable experimental data are available, was calculated by the
same methods and with the following isodesmic reaction:
H2 + H2O2 ? 2 H2O (4.3)
The usual procedure for calculating the heat of formation value of an unknown
compound is to combine the heat of reaction obtained from an isodesmic reaction with
the experimental heat of formation values for the known compounds94. The HA heat of
formation using Reactions (4.1), (4.2), and (4.3) were determined from the equations
(4.4), (4.5), and (4.6) respectively using the calculated heat of reaction, ∆HCalcRxn, and
the experimental heats of formation values at 1 atm and 298.17 K for ammonia37,
water37, and hydrogen peroxide38 listed in Table 4.1.
∆Hf, NH2OH = ∆HExptf, NH3 + ∆HExpt
f, H2O - ∆HExptf, H2 - ∆HCalc
Rxn (1) (4.4)
∆Hf, NH2OH = ∆HExptf, NH3 + ∆HExpt
f, H2O2 - ∆HExptf, H2O - ∆HCalc
Rxn (2) (4.5)
86
∆Hf, H2O2 = 2 ∆HExptf, H2O - ∆HExpt
f, H2 - ∆HCalcRxn (3) (4.6)
The choice of isodesmic reaction is important to obtain accurate values. Although there
are 5 single bonds on the reactant side (1 H-H, 1 O-H, 1 N-O, 2 N-H) and on the product
side (3 N-H, 2 O-H) in Reaction (4.1), the N-O bond on the reactant side is not balanced
by a similar σ bond on the product side. Reaction (4.3) is similar to (4.1) in terms of
bond balance with the O-O bond unbalanced on the reactant side. In Reaction (4.2), there
are 6 single bonds on the reactant side (3 O-H, 1 N-O,2 N-H) and on the product side (3
N-H, 2 O-H, 1 O-O), but here the N-O bond is balanced better by the O-O bond on the
product side. A better bond balance should result in a more effective cancellation of
errors, therefore, Reaction (4.2) should yield a more accurate value for ∆HCalcrxn than
Reaction (4.1) at the same level of theory. Thus similar errors are expected in the heat of
formation values calculated using Reactions (4.1) and (4.3), and faster convergence with
increasing level of theory for Reaction (4.2). In addition, agreement between values
obtained from Reactions (4.1) and (4.2) can serve as an indicator that the theory is
adequate to model the system.
Table 4.1: Experimental Heats of Formation (1atm and 298.17 K)
Compound Molecular Formula Heat of Formation (kcal /mol)
Ammoniaa NH3 – 10.98 ± 0.084
Watera H2O – 57.7978 ± 0.0096
Hydrogen Peroxideb H2O2 – 32.58 ± 0.05c
a. Ref. 37. b. Ref. 38. c. Based on the listed experimental errors.
87
2.2 Results and Discussion
Values for the HA heat of formation calculated using the various levels of theory
and basis sets are presented in Table 4.2, and computed N-O bond lengths (HA) and O-O
bond lengths (hydrogen peroxide) are listed in Appendix D.
The Austin Model 1 (AM1) yielded a good prediction for the hydrogen peroxide
heat of formation, but the heat of formation value obtained for HA differed significantly
from the values obtained via ab initio, density functional, or the composite methods.
Semi-empirical methods, like AM1, perform equally well for similar compounds for
which parameters are available. However, in this case, AM1 models the O-O bond in
hydrogen peroxide but does not appear to work well for the N-O bond in HA.
Hartree-Fock (HF) is the lowest level ab initio theory employed in this work. The HF
theory is expected to yield fair to good results, despite the fact that it does not include a
full treatment of electron correlation, because errors are partially cancelled with the use
of isodesmic reactions. Heats of formation calculated with the Hartree-Fock model did
not exhibit consistent improvement with increasing basis sets, but they generally yielded
more consistent results for Reaction (4.2).
The density functional methods, although not truly ab initio, include electron
correlation at only a moderate increase in computing cost, as compared to HF, by using
functionals of electron density. Among the density functional methods, B3P86 yielded
slightly better results for hydrogen peroxide than B3LYP for identical basis sets.
However, even B3P86 with a 5Z basis set has an error of nearly 2 kcal/mol, as compared
to the experimental value, for hydrogen peroxide with Reaction (4.3). Unlike HF theory,
increasing basis functions (cc-pVDZ, cc-pVTZ, cc-pVQZ, and cc-pV5Z) and adding
diffuse functions in the density functional methods leads toward consistent values of the
heat of formation. Similar values were obtained using the 6-311+G (3df, 2p) and 6-31+G
(3df, 2p) basis sets. At the density functional level of theory, there was a significant
88
Table 4.2. Summary of Calculated Heats of Formation (∆Hf)
Hydroxylamine Hydrogen Peroxide
Method Basis Set Heat of formation (kcal/mol)
Difference between rxn (1) and (2)
Heat of formation (kcal/mol)
Reaction (1) Reaction (2) Reaction (3)
AM1 -32.34 -31.31 1.0 -33.61
HF cc-pVDZ -12.14 -12.02 0.1 -32.69
cc-pVTZ -9.76 -12.48 2.7 -29.85
cc-pVQZ -8.83 -13.06 4.3 -28.34
6-31G -10.65a -7.14 3.5 -35.59a
6-31G(d) -16.10 -10.69 5.4 -37.98
6-31+G(d) -17.47 -8.07 9.4 -33.11
6-31G (d,p) -11.81a -12.06 0.3 -32.42a
6-31+G (2df,p) -7.65a -11.71 4.1 -24.93a
B3P86 cc-pVDZ -18.67 -7.79 10.9 -43.45
cc-pVTZ -13.99 -9.31 4.7 -37.57
cc-pVQZ -12.73 -10.08 2.7 -35.03
cc-pV5Z -12.01 -10.39 1.6 -34.20
AUG-cc-pVDZ -10.62 -10.23 0.4 -33.00
AUG-cc-pVTZ -11.83 -10.39 1.4 -34.02
6-311G (d) -21.70 -7.12 14.5 -47.16
6-31+G (3df,2p) -12.63 -10.88 1.8 -34.32
6-311+G (3df,2p) -12.14 -10.89 1.3 -33.83
6-311++G (3df,2p) -12.12 -10.89 1.2 -33.81
B3LYP cc-pVDZ -18.76 -5.24 13.5 -46.10
cc-pVTZ -14.92 -8.38 6.5 -39.12
AUG-cc-pVDZ -10.48 -9.26 1.2 -33.80
AUG-cc-pVTZ -12.18 -9.69 2.5 -35.07
6-311+G(3df,2p) -8.59 -6.17 2.4 -34.99
MP2 cc-pVDZ -14.39 -9.76 4.6 -37.21
cc-pVTZ -10.24 -11.01 0.8 -31.81
89
Table 4.2. (Contd.)
Hydroxylamine Hydrogen Peroxide
Method Basis Set Heat of formation (kcal/mol)
Difference between rxn (1) and (2)
Heat of formation (kcal/mol
Reaction (1) Reaction (2) Reaction (3)
cc-pVQZ -8.61 -12.09 3.5 -29.10
MP3 6-31+G(2df,p) -8.76a - - -28.38a
cc-pVDZ -14.44b -9.82 b 4.6 -37.20 b
cc-pVTZ -10.46 b -10.77 b 0.3 -32.27 b
MP4 6-31+G(2df,p) -11.26 -11.89 0.6 -31.96
MP4(SDT
Q) cc-pVDZ -16.80 b -8.21 b 8.6 -41.17 b
CCSD(T) cc-pVDZ -17.05 -8.11 8.9 -41.52
cc-pVTZ -13.02 c -9.52 c 3.5 -36.07 c
cc-pVQZ -11.56 c -10.61 c 1.0 -33.52 c
QCISD(T) cc-pVDZ -17.08d -8.12d 9.0 -41.54d
BAC-MP4 MP4//HF -12.98 -11.09 1.9 -34.46
G2 -11.78 -11.53 0.3 -32.83
G2MP2 -11.69 -11.67 0.0 -32.60
G3 -11.15 -11.28 0.13 -32.46
G3MP2B3 -11.88 -11.45 0.4 -33.01
G3B3 -11.51 -11.35 0.2 -32.74
CBS-Q -12.18 -11.16 1.0 -33.60
GAe - 4.87 - - -32.50
Experiment
al
-12.0f
-7.9h - - -32.58g
a. Ref. 69 b. Single point energies and thermal corrections for the enthalpies for MP3 (SDTQ) and MP4 (SDTQ) were
calculated at MP2/cc-pVDZ geometry. c. Single point energies and thermal correction for the enthalpy for the CCSD(T)/cc-pVTZ and CCSD(T)/cc-pVQZ
were calculated at CCSD(T)/cc-pVDZ geometry. d. Single point energy and thermal correction for the enthalpy for the QCISD(T)/cc-pVTZ were calculated at
CCSD(T)/cc-pVDZ geometry. e. Group Additvity
f. Based on an indirect calculation as discussed in the text (Ref. 65, 66). g. Ref.37. h. Ref. 70 .
90
change in the calculated heat of formation from the cc-pVDZ to cc-pVTZ basis set. The
magnitudes of the calculated values decreased with Reaction (4.1) as the quality of the
basis set was increased, but they increased for Reaction (4.2). Thus, Reaction (4.1) and
(4.2) approached the basis set limit for the heat of formation from opposite directions.
The composite theories (G2, G3, G2MP2, G3MP2B3, G3B3 and CBS-Q) are
expected to yield the best results since they have been developed to model accurately
thermochemical quantities for small, light-atom, main group molecules. The mean
absolute deviations (MAD) associated with heat of formation value obtained using G2
and G2MP2 theories (with the G2 test set) are 1.2 and 1.6 kcal/mol, respectively95. The
G3 theory is a further improvement over G2 and reduces the MAD to 1.0 kcal/mol76.
The CBS-Q accounts for errors due to basis set truncation by an extrapolation, and the
MAD associated with the method is 1.0 kcal/mol95. The MAD associated for G3,
G3MP2B3, and G3B3, based on heat of formation values for 148 different molecules,
are 0.94, 1.13, and 0.93 kcal/mol, respectively78. All of these composite theories
performed well and predicted accurate energies for hydrogen peroxide.
The MP3 (SDTQ) and MP4 (SDTQ) results were poor for cc-pVDZ, but the
MP3 prediction improved with the cc-pVTZ basis set. CCSD(T)/cc-pVDZ and
QCISD(T)/cc-pVDZ geometries agree well with the experimental values, but realistic
energy predictions were obtained only with the cc-pVQZ basis set.
For Reactions (4.1) and (4.3), as expected, the heat of formation values obtained
for HA and hydrogen peroxide respectively exhibited trends in similar direction for the
various levels of theory and basis sets. For the HA heat of formation values using
Reaction (4.2), there was faster convergence with the basis sets for the same level of
theory. As can be seen from the heat of formation values from Reaction (4.2), accurate
values were obtained at lower levels of theory and with smaller basis sets.
91
2.3 Choice of Best Values
The difference between the values calculated using Reaction (4.1) and (4.2) can
be taken as a guide for selecting theories performing well for the system. The calculated
values, in Table 4.2, that exhibited a difference of 1 kcal/mol or less are shown in bold.
It is worth noting that these theories also predict a reasonable value for hydrogen
peroxide heat of formation (within 1 kcal/mol). However, not all of these methods are
reliable in other respects. Omitted from the final values to be averaged were the less
reliable semi-empirical AM1 predictions, since the predicted value for ∆Hf, NH2OH were
significantly different from the values obtained by other methods. The values obtained
using theories (HF, B3P86, B3LYP) that did not demonstrate an improvement in the
prediction with increasing basis set were also left out. The MP values were also left out
since these values did not exhibit convergence with increases in the basis set or the
perturbation level.
Table 4.3 summarizes the 7 best-predicted heat of formation values and their
averages, where the calculated values using Reactions (4.1) and (4.2) differed by no
more than 1 kcal/mol. The average calculated heat of formation for hydrogen peroxide
was -32.9 kcal/mol with a standard deviation of 0.4 kcal/mol. With Reaction (4.1), an
average value of -11.7 kcal/mol with a standard deviation of 0.3 kcal/mol was
calculated. With the more balanced Reaction (4.2), the average value was -11.4 kcal/mol
with a standard deviation of 0.3 kcal/mol.
92
Table 4.3. Accurate Values for HA Heat of Formation
NH2OH H2O2
Heat of formation (kcal/mol)
Theory Basis set
Mean
Average
Deviation
Rxn (1) Rxn (2) Rxn (3)
G2 1.295 -11.78 -11.53 -32.83
G2MP2 1.695 -11.69 -11.67 -32.60
G3 1.078 -11.15 -11.28 -32.46
G3MP2B3 1.1378 -11.88 -11.45 -33.01
G3B3 0.9378 -11.51 -11.35 -32.74
CBS-Q 1.095 -12.18 -11.16 -33.60
CCSD(T) cc-pVQZ - -11.56 -10.61 -33.52
Average 1.1 -11.7 -11.4 -32.9
St. dev. 0.3 0.3 0.3 0.4
Experimental -12.065 -32.5837
-7.970
The deviations from the average heat of formation for the various methods in
Table 4.3 are shown in Figure 4.2. From this pattern, it is apparent that the deviations
from average for heat of formation values from Reaction (4.1) and (4.3) track each other,
whereas the deviations for Reaction (4.2) do not follow the same trend. The average
calculated value of ∆Hf value for hydrogen peroxide is greater than the experimental
value by 0.3 kcal/mol. Since Reaction (4.1) and (4.3) are expected to yield similar errors,
93
we believe that ∆Hf value obtained using Reaction (4.1) will differ from the true value in
a similar manner and therefore recommend -11.4 kcal/mol for the ∆Hf of NH2OH as the
best estimate from both Reaction (4.1) and (4.2). The mean absolute deviation (MAD)
for each of the methods employed is listed in Table 4.3 and the average MAD value is
approximately 1.1 kcal/mol. However, the HA ∆Hf values are computed from isodesmic
reactions which should yield values with smaller errors, perhaps down to twice the
standard deviation of the various method. Thus the recommended ∆Hf value for HA,
including our precision, judgment of methodology, and accuracy is -11.4 ± 0.6 kcal/mol.
Furthurmore, the agreement and the consistency of the calculated hydrogen peroxide
average value with the experimental value suggests that our calculated average value for
∆Hf of HA is more reliable than the available experimental values, which as discussed in
the introduction, cannot be properly assessed.
-0.80
-0.60
-0.40
-0.20
0.00
0.20
0.40
0.60
0.80
G2 G2MP2 G3 G3MP2B3 G3B3 CBS-Q CCSD(T)
Method
Dev
iati
ons
from
ave
rage
hea
t of
fo
rmat
ion
(kca
l/mol
)
Rxn (1)
Rxn (2)
Rxn (3)
Figure 4.2. Deviations from the average heat of formation values for the
methods employed in Table 4.3.
94
Also, we have illustrated the importance of well-balanced isodesmic reactions for
determining accurate heats of formation, especially at lower levels of theory. Depending
on the level of theory, triple-ζ (6-311G or cc-pVTZ) or larger basis sets are necessary to
predict accurate HA heat of formation values. At all levels of theory the double-ζ (6-31G
or cc-pVDZ ) basis set yielded poor energies, but CCSD(T) and QCISD(T) predicted
accurate geometries in this basis set. The methods employed in obtaining the average
heat of formation value have an absolute accuracy of approximately 1.1 kcal/mol, but the
value obtained using isodesmic reactions is expected to have a smaller error and
therefore 1.1 kcal/mol represents the maximum absolute error in the calculation. As
expected the highly parameterized composite methods (G2, G3, G2MP2, G3MP2B3,
G3B3, CBS-Q) yielded the most accurate values. However, the unparametrized ab intio
CCSD(T)/cc-pVQZ yielded nearly as accurate values and in some cases, depending on
the accuracy needed, density functional methods, MP3, and MP4 may be adequate.
Heat of formation values can also be used for estimating decomposition energies.
If we use the value of –11.4 kcal/mol, CHETAH calculates the maximum heat of
decomposition as –1515 cal/g; thereby indicating a potential reactive hazard. According
to the program HA should decompose with the following stoichiometry:
NH2OH (g) → 0.33 N2 (g) + H2O (g) + 0.33 NH3 (g)
The experimentally observed heat of reaction is around 850 cal/g64, which includes all
thermal effects discussed above. The next section discusses a plausible decomposition
pathway for HA.
95
3. Investigation of Hydroxylamine Runaway Behavior
3.1 Experimental Observations
Pure HA is known to explode at room temperature. Calorimetric studies on
aqueous solutions of HA indicate that it is a highly reactive compound and the
decomposition of HA is extremely sensitive to metal contamination.64 Thermal
instability of hydroxylamine is clear from the APTAC results illustrated in Figure 4.3.
Figure 4.3. APTAC temperature-time profile for 50 wt% HA.
As shown in the above figure a significant reaction for 50 wt% HA is detected at around
120 oC. But when a small quantity of Fe2SO4 solution (8 ppm Fe2+) is added, HA reacts
rapidly at room temperature. Also, the contaminated sample shows a larger exotherm, as
seen from the red curve in the graph. The mechanism of Ha decomposition was
0
50
100
150
200
250
0 500 1000 1500
Time (min)
50 wt% HA + 0.8 ppm Fe2+
50 wt% HA
96
investigated, in presence and absence of metals. The aim of studying elementary
reactions is to understand initiation steps leading to the HA runaway reactions. Such
understanding of elementary reactions will aid in understanding the fundamental
behavior of HA, development of better inhibitors to prevent metal catalysis, and
consequently thermal runaway reactions. The results of our calculations are discussed in
the following section.
3.2 Theoretical Calculations
Runaway reactions, leading to explosion, generally follow a radical mechanism
and the mechanism can be categorized into initiation, propagation, and termination steps.
A key component of such explosive reactions are branching reactions, wherein two or
more radicals are created from a single radical, that accelerate the reaction.40 With the
above considerations, the following rules were observed to generate reaction network for
hydroxylamine decomposition reactions:
1. The first step is cleaving of bonds in the HA molecule. The dimerization of
hydroxylamine was not considered because experimental results do not suggest a
polymerization reaction.
2. There are no ionic reactions.
3. Every specie reacts with every other specie, but for any elementary reaction,
there are no more than two reactants. This is a reasonable assumption since three
or more species coming together is a much rarer event.
4. Experimentally observed products96 or stable species, such as NH3, H2O, N2O do
not participate in the chain propagation step.
5. Solvation effects have been neglected.
The following elementary reactions of hydroxylamine decomposition calculated
using the B3P86/cc-PVDZ level of theory, since good energy values were obtained for
the heat of formation calculations using this theory. Based on earlier discussion, the
97
proposed reactions are divided into three classes – initiation, propagation, and
termination, and the favorable reactions are summarized below. Associated enthalpies in
kcal/mol are indicated by ‘∆’ and products (stable species) are underlined.