Funding is for Prof. Fernandez-Pello's role as U.S. Co-Investigator for the Japan Aerospace Exploration Agency (JAXA)-sponsored project, “Flammability Limits At Reduced-g Experiment (FLARE)." JAXA International Announcement of Opportunity (AO) to fund experiments to be conducted aboard the Japanese Experiment Module, Kibo, 2012.

The objective of the proposed research program is to continue the experimental study of the flammability of wire materials in space exploration atmospheres and associated computational/theoretical tools to aid interpretation of test results.

Modeling Flame Spread over Insulated Wires using Neural Networks and Genetic Algorithms

The objective of this theoretical task is to use artificial neural networks (ANN) and genetic algorithms (GA) to further understand how certain variables in the problem of flame spread over insulated wires relate to one another and affect the insulation melting and dripping from the flame spread. The information will be used to model both flame spread rate and dripping of insulated electrical wires and to develop predictive equations of the problem as a whole. The equations will be applied to predict burning behavior of wires in normal and microgravity environments. The research primarily supports the above-mentioned Japan Aerospace Exploration Agency (JAXA) program but also could lead to a better determination and ranking of the fire hazard characteristics of potential wire materials to be used in spacecraft for long-term exploration missions. Here a summary of the work conducted during this reporting period is summarized. More detailed description of the work can be found in the references provided.

1. Comprehensive Database To develop an artificial neural network that can predict flame spread rate along electrical wires under various environmental parameters, a comprehensive database of experimental flame spread rate results was first compiled from the current existing work in the field. Altogether, the database consists of approximately 1200 data points, with approximately one third (400) coming from internal experiments and two thirds (800) coming from external sources.

2. Artificial Neural Network Structure and Training To develop the ANN model of this comprehensive database, an ANN was trained to predict the flame spread rate along wires of different sizes and compositions under various ambient conditions. Using the data from the database, a procedure was implemented of data preparation, then training of the ANN, and finally validation of the ANN. The first stage, data preparation, involves transforming the data corresponding to the input parameters, as well as experimentally gathered output data (in this case, flame spread rate), into normalized values. This calculation is carried out by taking each data point of a specific parameter and dividing it by the corresponding mean of that parameter.

Next, half of this data is taken for use in training the ANN, while the remaining half is reserved for validation of the ANN. The basics of the ANN training can first be thought of in terms of a single node in a matrix. Each of these nodes are in a position and take in various inputs, multiplies them by weights, linearly combines those products along with a bias value, and inputs the result into an activation function to determine the node output. This calculation is then carried out for all nodes in the ANN which are typically arranged in layers, so the outputs from every node in one layer become the inputs for each node in the next. Once the calculations have been iterated through the entire ANN, one epoch is considered to have passed. This process can then be repeated for multiple epochs with the weights and biases being adjusted each time through backpropagation to refine the ANN output, leading to more accurate predictions.

The basic structure of the ANN utilized here was created with the open-source Python package, Keras (Chollet, 2015). The input layer consists of 15 nodes, corresponding to the 12 selected input parameters, two of which were vectors with multiple components, with each assigned its own node. Based on the number of input parameters, it was determined that there should be two dense hidden layers with 12 nodes each, and an output layer consisting of only one node, corresponding to the single output parameter of interest, the flame spread rate. The chosen activation function was a hyperbolic tangent, and the weight and bias values were refined over approximately 5,000 epochs.

3. Validation and Training Results After training, validation of the ANN is carried out using the reserved half of the data that was not put through the training process. Because the ANN has never seen this data before, it is able to validate the training of the ANN and show that it is actually able to make predictions and has not just memorized the patterns found in the training data. The results from the validation dataset are obtained by running the corresponding input parameters through the fully trained ANN for one epoch. The results from the ANN’s training show a very strong agreement between the ANN’s predictions and the experimentally measured values, with an average error of 12%. The results of the ANN’s validation, with datasets divided by reference origin, also show strong agreement between the ANN’s predictions and the experimentally measured values, with an average error of 16% for data the ANN had never encountered before. It should be emphasized that experiments with opposed and concurrent flame spread rates, different wire orientations, and varying strengths of gravity are all represented within these ANN predictive results. The initial results from the ANN are exceedingly promising. However, it must be remembered that there are also still improvements that can be made to the ANN to increase the accuracy of its predictions. With further training using a wider variety and more accurate input data, as well as greater insight into influential parameters, predictions may become even better with future iterations of the ANN.

4. Parametric Trends While there is room for improvement in the ANN model, that is not to say that nothing further can be learned from it in its current state. Examining some of the parametric trends predicted by the ANN can give further insight into the flame spread rate along electrical wire insulation problem. This can be done by plotting the results as prediction surfaces in function of different input parameters. As an example, this approach shows a parametric surface that demonstrates the ANN’s predictions for flame spread rate’s dependence on oxygen concentration and ambient pressure. The surface clearly displays the well-known trend for flame spread rate to increase both with increasing oxygen concentration and increasing pressure; however, it also shows that even at extremely low oxygen concentrations or pressures, a high value of the other parameter can still result in a moderately fast flame spread rate. Each of the other parameter surfaces can be examined similarly to gain insight to the flame spread rate over electrical wire problem.

One of the developed surfaces shows the predicted flame spread dependence on ambient oxygen concentration and pressure for a wire with constant 2.9 mm core diameter and 1.2 mm insulation thickness, and another wire with constant 0.64 mm core diameter and 1.7 mm insulation thickness. Another, with constant 0.64 mm core diameter and with 1.7 mm insulation thickness, and varying core thermal conductivities with wire core properties averaged for copper, iron, and nichrome; unless otherwise stated, parameters are held constant with copper core.

5. Neural Networks and Genetic Algorithms using Dimensional Parameters To improve the generality of the predictions and to assist with an equation formulation of the problem as a whole, an additional neural network model was trained using a selection of non-dimensional parameters. To determine the parameters to use in the model, a large set of possible non-dimensional combinations were generated and selected based on the optimal and lowest quantity of inputs. The selection was based on a set of generated neural network models generated by a combination of nondimensional parameters. The highest precision was selected based on predictive/measured deviation by root mean square of the error for a given quantity of inputs.

The results show that the artificial neural network provides good predictive performance with the experimentally measured flame spread rates. The final artificial neural network model used 3 layers, 13 non-dimensionalized input parameters, and up to 10,000 epochs. The selected non-dimensionalized neural network model successfully maintained a low root mean square error of the predictive performance compared to the reported values. This performance was achieved with the more generalized non-dimensional input parameters, while also reducing the over-fitting of the data. The large reduction in the fitting parameters was a result of the reduced number of input parameters. This work will be continued in the extension of the present grant.

Funding is for Prof. Fernandez-Pello's role as U.S. Co-Investigator for the Japan Aerospace Exploration Agency (JAXA)-sponsored project, “Flammability Limits At Reduced-g Experiment (FLARE)." JAXA International Announcement of Opportunity (AO) to fund experiments to be conducted aboard the Japanese Experiment Module, Kibo, 2012.

The objective of the proposed research program is to continue the experimental study of the flammability of wire materials in space exploration atmospheres and associated computational/theoretical tools to aid interpretation of test results.

Research Progress: Modeling Flame Spread and Dripping over Insulated Wires using a Neural Network. The work conducted during this reporting period provides a coherent understanding of flame spread parametric trends and associated fire safety issues in electrical systems for space applications. This understanding was obtained through use of an artificial neural network (ANN) that was trained to predict the flame spread rate along “laboratory” wires of different sizes and compositions [copper, nichrome, iron, and stainless-steel tube cores and HDPE (high density polyethylene), LDPE (low density polyethylene), and ETFE (ethylene tetrafluoroethylene) insulation sheaths] and exposed to different ambient conditions (varying flows, pressure, oxygen concentration, orientation, and gravitational strength). For these predictions, a comprehensive database of 1200 data points was created by incorporating flame spread rate results from both in-house experiments (400 data points) as well external experiments from other sources (800 data points). The predictions from the ANN showed that it is possible to merge together various data sets, including results from horizontal, inclined, vertical, and microgravity experiments, and obtain unified predictive results. While these initial results are very encouraging with an overall average error rate of 14%, they also show that future improvements to the ANN could still be made to increase prediction accuracy.

Summary of Results.

Training Data and Validation: The results from the ANN’s training showed a very strong agreement between the ANN’s predictions and the experimentally measured values, with an average error of 14%. However, it is expected to see good agreement here, as this is the data that was used to train the ANN. The true indication of the ANN’s performance is revealed by the validation results, which provide the results of the ANN’s validation with datasets divided by reference origin. Again, strong agreement was observed between the ANN’s predictions and the experimentally measured values, with an average error of 16% for data the ANN had never encountered before. Such strong agreement gives the indication that the selected input parameters of core and insulation densities, thermal conductivities, specific heats, and cross-sectional areas, oxygen concentration, pressure, flow velocity, and gravitational strength are all highly important to the wire flame spread problem. However, these results also show that there is great room for improvement in the ANN predictions.

Parametric Trends: While there is room for improvement in the ANN model, that is not to say that nothing further can be learned from it in its current state. Examining some of the parametric trends predicted by the ANN can give further insight into the flame spread rate along electrical wire insulation problem. Because of the previous focus on spacecraft applications as well as the current focus on terrestrial transportation and structural applications, several parametric trends were examined. For the more spacecraft-related applications as well as limited terrestrial applications a parametric surface was obtained which demonstrates the ANN’s predictions for flame spread rate’s dependence on oxygen concentration and ambient pressure, spanning both the oxygen concentration and pressure ranges of the gathered dataset. The surface clearly displays the well-known trend for flame spread rate to increase both with increasing oxygen concentration and increasing pressure. However, it also shows further nuances of the problem that have not previously been well explored, such as the response of flame spread rate to high oxygen concentrations at low pressures as well as its response to high pressures at low oxygen concentrations. The ANN’s predictions show that these parameters somewhat balance each other out and that even at extremely low oxygen concentrations or pressures, a high value of the other parameter can still result in a moderately fast flame spread rate. Aiming for further relevance to terrestrial transportation and structural applications, another parametric surface was obtained that demonstrates the ANN’s predictions for flame spread rate’s dependence on wire core cross-sectional area and axial flow velocity, spanning both the wire core diameter and axial flow velocity ranges of the gathered dataset with all other variables held at the same previous constants and this time with 21% oxygen concentration and 101 kPa (14.6 psi) ambient pressure. This surface shows that there is much more variation in flame spread rate when varying core cross-sectional area and axial flow velocity, suggesting that these parameters may be highly important for assessing fire safety. Another interesting feature that is found in the surface is the steep drop off in flame spread rate when the cross-sections area of the core approaches ~10 mm2, with some slight variation depending on the axial flow velocity. This dramatic decrease in flame spread rate may illustrate a critical core diameter, where the wire core transitions from being a source conducting heat toward the flame to a sink drawing heat out of it. Interestingly, this effect does not seem to be nearly as present in the opposed flow (negative axial flow velocity) regime.

Future Work: The initial results from the ANN have shown to be very promising, and it is believed that the ANN could be used for much more in-depth analyses in the future. To begin with, only a very few number of flame spread rate parametric surfaces were examined in this study. Analysis of all the other possible flame spread rate parametric surfaces could bring so much further insight to the flame spread rate along electrical wire problem and potentially reveal results or nuances that have previously gone undiscovered. A sensitivity analysis examining each of the parameters affecting the flame spread rate along electrical wire insulation problem could also be performed using the predictions of the ANN. Such an analysis could be an immensely useful tool for shedding further light onto which parameters in the problem are the most important and influential when in comes to flame spread rate and assessing fire safety. Finally, the ANN could be expanded to include further parameters. As mentioned previously, some variables such as electric fields or electric currents and external radiation were excluded from this analysis. However, adapting the comprehensive database and the ANN to account for these variables could make the ANN predictions even more comprehensive and relevant to the current applications as well as additional ones.

Funding is for Prof. Fernandez-Pello's role as U.S. Co-Investigator for the Japan Aerospace Exploration Agency (JAXA)-sponsored project, “Flammability Limits At Reduced-g Experiment (FLARE)." JAXA International Announcement of Opportunity (AO) to fund experiments to be conducted aboard the Japanese Experiment Module, Kibo, 2012.

The objective of the proposed research program is to continue the experimental study of the flammability of wire materials in space exploration atmospheres and associated computational/theoretical tools to aid interpretation of test results.

A program to study the flammability characteristics of electrical wire insulations that will complement the existing JAXA project entitled “Fundamental Research on International Standard of Fire Safety in Space” (FLARE) is being carried out at University of California Berkeley (UCB) under NASA sponsorship. The final objective of the project is to provide an assessment of wire materials flammability in the conditions expected in “Space Exploration Atmospheres” (SEA), i.e., low velocity flow, external radiant heating, reduced ambient pressure, elevated oxygen concentration, and microgravity. The near-term objective of the research is to develop and conduct tests in a ground-based FLARE/UCB apparatus that reflects the environments that are expected in space-based facilities, and that supports the International Space Station (ISS)/Kibo experiments to be conducted under the FLARE project. The apparatus is used to study the flammability of electrical wires, specifically ignition and flame spread. Tests are being conducted with laboratory type wires consisting of a conducting core and PE insulation. A summary of the research conducted during this period is reported here.

The research has one primary goal: to understand the changes in electrical wire flammability in the environment expected in space-based facilities, particularly those at exploration atmospheres. To date, flammability testing has focused mostly on atmospheric conditions. Work conducted at NASA Glenn Research Center (GRC) and at the combustion laboratories at UC Berkeley indicate that combustible materials are in some cases more flammable in SEA atmospheres than in Earth atmospheres. Also, if the insulation material melts during the wire burning, major differences in flammability may occur between normal and micro-gravity, because the molten material will not drip in microgravity. Thus, considerable work remains to be done to understand how the new spacecraft cabin environment influences material flammability, particularly that of wires.

The overall project experimental methodology consists of obtaining flame spread flammability characteristics of electrical wires and wire insulation materials (primarily those being studied in Hokkaido University) under external radiation and in SEA environments. Experiments were conducted primarily in normal gravity to observe the effect of melting and dripping on the flame spread over wire insulation. Effects, such as gas flow velocity and direction, external radiation, ambient oxygen concentration and pressure, were analyzed for different laboratory type wires. The research primarily support the above mentioned JAXA program but also could lead to better determination and ranking of the fire hazard characteristics of potential wire materials to be used in spacecraft for long term exploration missions.

Research Progress:

A summary of the research progress during this reporting period is presented in this section. The summary is divided into five sections, corresponding to different aspects of the work conducted during this period.

1. Effect of Flow Velocity on Flame Spread along Insulated Electrical Wires. The work reported in this section was presented at the 11th U. S. National Combustion Meeting, Western States Section of the Combustion Institute, March 24–27, 2019, Pasadena, CA.

The goal of this study was to examine how the burning, in terms of flame spread rate is affected by the flow conditions (opposed/concurrent and flow speed) and characteristics of the wire (insulation thickness, metal core size, and material) for relatively small to intermediate sized wires. The mass loss due to dripping was also recorded and assessed. Then, these results were combined with others’ results and compared against one another to determine larger trends that could not be observed from the scope of this study alone.

“Laboratory” wire samples with cores of either solid copper (2.5 mm or 0.64 mm diameters) or stainless-steel tubing (2.413 mm outer-diameter) with surrounding low density polyethylene (LDPE) insulation sheaths (4 mm outer-diameter) were burned subject to various airflow conditions. Concurrent, opposed, and no flow flame spread rate as well as mass loss due to dripping were determined. The flame spread rate was found to vary linearly with the airflow velocity, increasing for concurrent flame spread and decreasing for opposed flame spread, and the mass loss due to dripping was found to remain approximately constant for all airflow velocities. When compared to the literature, there was not a clear trend between flame spread rate and airflow velocity; however, a very clear relationship between flame spread rate and insulation thickness was observed, where, as the insulation thickness decreased, flame spread rate increased. When comparing mass loss due to insulation dripping from the current study with the literature, it was found that core diameter as well as heat transfer properties had the largest influence on insulation dripping, with larger cores and more conductive cores exhibiting fewer dripping effects and smaller cores dipping in excess. These experiments and comparisons with prior studies provide further understanding of the complex flame spread mechanisms along electrical wiring and serve as a baseline for a larger set of ongoing experiments in a 1g environment to study the effect of lower ambient pressures and higher oxygen concentrations on flame spread along electrical wires.

2. Effect of Ambient Pressure on the Piloted Ignition and Subsequent Flame Spread Across Simulated Electrical Wires. The work presented in this section was presented at the 2019 Fall Meeting of the Western States Section of the Combustion Institute, October 14-15, 2019, Albuquerque, NM.

The goal of the present study is to investigate whether utilizing different lengths of igniter exposure times to ignite various types of “laboratory” wire samples subject to multiple pressure environments and exposed to a constant opposed airflow will influence flame spread results. These results are relevant for comparison with future similar experiments planned in the International Space Station.

Due to electrical wires being potential sources of fire ignition in spacecrafts and with the cabin environments of NASA’s next generation of spacecrafts planned to operate under reduced pressure and increased oxygen concentration conditions, it is highly important to understand the burning behavior of electrical wires in their operating environments. To begin understanding the potential effect these environmental conditions may have on material flammability, experiments examining ignition delay and flame spread rate in both an atmospheric and low-pressure, 60 kPa environments were performed. In these experiments, “laboratory” wire samples containing cores of various materials, including copper, nichrome, and stainless steel, where surrounded with either 4 mm- or 3 mm-outer diameter LDPE insulation sheaths and ignited with an electrically-heated nichrome coil placed around one end of the wire. Experiments utilizing various lengths of time for which the igniter was applied were performed in both the described environments with horizontally aligned wire samples and opposing airflows of 10 cm/s.

In the environment exposed to atmospheric pressure, it was found that for wire samples with thin copper rod cores or other less conductive cores (nichrome rods and stainless-steel tubes), the length of igniter exposure had practically no effect on the flame spread rate. For the more conductive wire samples containing thicker copper rod cores, an effect was observed in which longer lengths of exposure to the igniter produced faster flame spread. Repeating these experiments in the low-pressure environment caused delays in ignition and enhanced the effect of igniter exposure length on flame spread rate. As in the atmospheric pressure environment, tests in the low-pressure environment showed samples with cores considered here to have low conductivity (thin copper rods, nichrome rods, and stainless-steel tubes) had negligible changes in flame spread rate as the length of igniter exposure increased. The remaining sample types, having thicker copper rod cores and being more conductive, showed similar trends to one another of drastic increase in flame spread rate with increased exposure to the igniter.

While further research and analysis are needed to further quantify the effect of igniter exposure length on flame spread rate, the current results clearly show that in environments at and below atmospheric pressure, an effect is observed for certain wire types that are highly conductive where increased igniter exposure times result in faster flame spread. It is important to fully understand these results to develop a ignition method that allows for consistent results across environments with various pressures in order to more accurately describe the flammability and flame spread of materials subject to these conditions, especially because these results are relevant for comparison with future similar experiments planned in the International Space Station.

3. Effect of Reduced Ambient Pressures and Opposed Airflows on the Flame Spread and Dripping of LDPE Insulated Copper Wires. The work reported in this section has been published online ahead of print in the Fire Safety Journal (see Bibliography section).

The goal of the present study was to examine the combined effect of sub-atmospheric, ambient pressure, and opposed airflows on flame spread over wires with a high conductivity core such as copper. The insulation dripping off these wires was also analyzed under these conditions, since it has been shown that it affects the rate of flame spread and burning of insulated wires. The results of the present work are also relevant for comparison with future experiments planned in the International Space Station (ISS) with similar wire, which is the primary objective of this work.

The combined effect of reducing the ambient pressure and increasing the opposed flow speed on horizontal flame spread and dripping of copper-cored, LDPE-insulated wires was examined through experiments to increase understanding of the fire hazard electrical wires pose in spacecraft environments, and to provide data for comparison with future microgravity experiments. Results showed that the flame spread rate as well as the molten, burning insulation drip frequency were found to decrease both with decreasing pressure as well as increasing opposed flow speeds. Contrarily, it was found that the total mass dripped increased both with decreasing pressure and increasing opposed flow speeds. It is thought that these results are due to the wire core acting as a heat sink and drawing a significant amount of heat out of the flame, which affects both the rate of flame spread and the rate of insulation burning. Comparison with results from other studies with wires of different core material or dimensions show that the effect of the environmental parameters on the flame spread and mass burning of insulated wires depends strongly on the core conductivity as well as core and insulation diameters. Consequently, data obtained from specific wire tests should not be extended to other wires without justification.

4. Concurrent Flame Spread over LDPE Insulated Copper Wires in Reduced Ambient Pressures. The work presented in this section has been published in Fire Technology.

The goal of the present study is to examine the combined effect of sub- atmospheric or ambient pressures and low air flow speeds on concurrent flame spread over copper wires. The insulation dripping off these wires was also analyzed under these conditions, since it has been shown that it affects the rate of flame spread and burning of insulated wires. Furthermore, since dripping does not occur in the absence of gravity, these results are relevant for comparison with future similar experiments planned in the International Space Station (ISS) [R. Friedman, “Fire Safety in the Low-Gravity Spacecraft Environment,” Denver, Colorado, 1999].

The combined effect of reducing the ambient pressure and increasing the flow speed on horizontal concurrent flame spread and dripping of copper-cored, LDPE-insulated wires was examined through experiments to increase understanding of the fire hazard electrical wires pose in spacecraft environments. Results showed that the flame spread rate as well as the molten, burning insulation drip frequency both decrease with decreasing pressure as well as decreasing flow speeds. As well, it was found that the total mass dripped increased with decreasing pressure. It is thought that these results are due to variations in the heat transfer to the insulation from the flame as well as from the insulation to the environment via the wire core acting as a heat sink and drawing a significant amount of heat out of the flame, which affects both the rate of concurrent flame spread and the rate of insulation burning. Comparison with results from other studies with wires of different core material or dimensions show that the effect of the environmental parameters on the flame spread and mass burning of insulated wires depends strongly on the core conductivity as well as core and insulation diameters. Consequently, data obtained from specific wire tests should not be extended to other wires without physical justification.

5. Modeling Flame Spread over Insulated Wires using Neural Network and Genetic Algorithm. The objective of this task is to use neural network to further understand how certain variables in the present problem relate to one another and affect dripping/flame spread. This information will be used to be able to model/predict both flame spread rate and dripping. In addition, it will be used to determine predictive equation of the problem as a whole.

Funding is for Prof. Fernandez-Pello's role as U.S. Co-Investigator for the Japan Aerospace Exploration Agency (JAXA)-sponsored project, “Flammability Limits At Reduced-g Experiment (FLARE)." JAXA International Announcement of Opportunity (AO) to fund experiments to be conducted aboard the Japanese Experiment Module, Kibo, 2012.

The objective of the proposed research program is to continue the experimental study of the flammability of wire materials in space exploration atmospheres and associated computational/theoretical tools to aid interpretation of test results.

Note is continuation of "Fundamental Research on International Standard of Fire Safety in Space - Subteam 2: Wire Combustion with External Radiation in Support of the JAXA Project Fundamental Research on International Standard of Fire Safety in Space," grant NNX14AF01G, with the same principal investigator Prof. Carlos Fernandez-Pello. See that project for previous reporting.