SAMPE Journal - March/April 2024 - 43

1. INTRODUCTION
Thermal Protection Materials (TPMs) are
commonly used in the aerospace industry to shield
vehicles and structures from extreme temperature
conditions[1]
. High-temperature composites, and
in particular polymeric-based composites that are
reinforced with carbon fibers, are of interest due to
a combination of excellent properties such as high
strength-to-weight ratio and low ablation; along
with unique structural characteristics that enable
the retention of these properties under extreme
environments[2-5]
.
To achieve these properties, highly controlled
processing conditions in a specific step-by-step
manufacturing approach are often required. This
generally includes a lay-up step; a curing step; a
high-temperature pyrolytic step to convert the
resin phase into an amorphous structure with high
carbon content, followed by several repeats of resin
backfill steps; and finally, a graphitization step to
achieve the desired crystalline structure of carbon
atoms[6]
the pyrolysis process directly affect the extent and
evolution of the degradation reactions. This, in turn,
determines the final thermo-mechanical behavior,
as well as final laminate yield and permeability;
and hence the final properties[2,6]
.
Establishing the relationship between the
pyrolytic processing conditions and microstructure/property
development is critical for
the efficient manufacturing of these materials[5]
.
Typically, a comprehensive testing campaign
is required, yet the scope is often constrained
by time and resource limitations within the
industry. Although several kinetic models exist
in the literature describing the pyrolysis process
of polymeric materials based on the Arrhenius
equation, these models generally account for
the isolated behaviors observed under specific
processing conditions, such as dynamic heating.
This limits their applicability in predicting
outcomes under more complex and industrially
relevant temperature cycles[1-4]
.
This study introduces a novel probabilistic
machine learning (ML) approach that leverages the
underlying physics and established theories of the
pyrolysis process, using Gaussian processes (GPs)
methods[7-9]
2. EXPERIMENTATION
2.1 Sample Preparation
The samples studied in this paper were fabricated
using a proprietary carbon/phenolic resin prepreg
system. [0/90/±45]8S
laminates with approximate
dimensions of 15×15×0.5 cm were fabricated
following
the
. The processing parameters used during
vacuum
bagging
techniques
suggested by the manufacturer, as depicted in
Figure 1. The curing process was performed using
an autoclave under specific pressure conditions,
and the panel was heated from room temperature
to 190ºC and held for an hour according to the cure
cycle also specified by the manufacturer. Upon
curing, density measurements and cross-section
imaging were performed to characterize porosity.
The obtained results were consistent with those
expected. Samples with approximate dimensions
of 0.2×0.1×0.5 cm and initial weights between 8-10
mg were cut out of the panel, dried, and stored in
sealed plastic bags until their analysis.
2.2 Thermogravimetric Analysis
The high-temperature pyrolytic process of the
composite was studied using thermogravimetric
analysis. A TA Instruments
Thermogravimetric
Analyzer
(TGA)
was
to observe the thermal stability of the cured
prepreg samples by monitoring the weight loss
and weight loss rate throughout the degradation
reactions. Experiments were performed under an
inert nitrogen atmosphere to prevent degradation
from oxidative reactions, using open platinum
pans suitable for high-temperature analyses[10,11]
.
. Surrogate ML models are trained using
small datasets from thermogravimetric analysis
tests, transformed into a mathematical domain
based on expressions from pyrolysis kinetics
theories. This paper aims to demonstrate how this
method can be used to accelerate the analysis and
optimization of the pyrolysis process by accurately
predicting the behavior of the material under
various single and double-ramp temperature
cycles. Furthermore, it also aims to illustrate how
the model's confidence can be improved through
an iterative process, guided by ML models, as
opposed to relying on trial-and-error methods.
www. sampe.org
A first round of tests was carried out in which the
samples were heated from room temperature to
1000ºC at constant heating rates of 2, 3, 4, and
5ºC/min and held isothermally for 20 min before
cooling. The final char yield and weight loss rate
for some of the tests with different heating rates
are shown in Figure 2. Specific temperature regions
in which different reactions occur can be seen in
the derivative weight loss curves. All tests showed
three distinctive peaks at temperature ranges of
200-240ºC, 350-400ºC, and 525-550ºC, with slight
variations due to the heating rate.
Overall, the results obtained were consistent
with those expected based on the established
theory for a similar polymeric matrix. Peaks
MARCH APRIL 2024
|
SAMPE JOURNAL | 43
Figure 1. Schematic of the prepreg panel hand lay-up for autoclave curing.
TGA 550
used
http://www.sampe.org
SAMPE Journal - March/April 2024

Table of Contents for the Digital Edition of SAMPE Journal - March/April 2024
