Download as pdf or txt
Download as pdf or txt
You are on page 1of 23

COMPARISON OF MACHINE LEARNING MODELS FOR

CLASSIFICATION OF ACOUSTIC FIRE


EXTINGUISHING STATUS

MICRO PROJECT REPORT

submitted by

JENNIFER THOMAS
M23CHIS13
to

APJ Abdul Kalam Technological University


in partial fulfillment of the requirements for the award of
M. Tech Degree in Industrial Safety and Engineering

DEPARTMENT OF CHEMICAL ENGINEERING

TKM COLLEGE OF ENGINEERING, KOLLAM

DECEMBER, 2023
ABSTRACT

Fire extinction is a critical challenge facing society. Recently, sound wave-based fire
extinguishing systems have emerged as an innovative technology to suppress flames. However,
optimizing the performance of these systems requires robust analytics. This study evaluates
five machine learning models - K-Nearest Neighbours (KNN), Support Vector Machine
(SVM), Decision Tree, Random Forest, and Naive Bayes - on a dataset of 17,442 experiments
from a sound wave fire extinguishing system. The goal is to accurately predict flame extinction
versus non-extinction. The dataset contains features on fuel type, flame size, sound decibel
level, frequency, airflow, and distance. The models are evaluated using accuracy score. This
analysis provides insight into the optimal machine learning approach for developing intelligent
decision support systems that can enhance the efficiency and capabilities of sound wave-based
fire suppression technology. The final model has potential real-world application for predicting
and preventing fire damage.

Keywords: Fire classification, sound wave extinction, machine learning, KNN, SVM,
Decision Tree, Random Forest, Naive Bayes

i
TABLE OF CONTENTS

1. INTRODUCTION ................................................................................................................. 1

2. LITERATURE REVIEW ....................................................................................................... 2

3. OBJECTIVES ........................................................................................................................ 6

4. METHODOLOGY ................................................................................................................ 7

4.1 Data Acquisition ............................................................................................................... 7

4.2 K-Nearest Neighbours ...................................................................................................... 7

4.3 Support Vector Machine ................................................................................................... 8

4.4 Decision Tree.................................................................................................................... 8

4.5 Navie Bayes...................................................................................................................... 9

4.6 Random Forest ................................................................................................................. 9

4.7 Accuracy Score ................................................................................................................. 9

4.8 Confusion Matrix ............................................................................................................. 9

5. RESULTS............................................................................................................................. 11

5.1 Data Analysis.................................................................................................................. 11

5.2 Accuracy Score ............................................................................................................... 13

5.3 Confusion Matrix ........................................................................................................... 13

5.4 Correlation Matrix .......................................................................................................... 15

6. CONCLUSIONS.................................................................................................................. 16

REFERENCES ........................................................................................................................ 17

ii
LIST OF FIGURES

Figure 1 Count of each fuel type plot ...................................................................................... 11


Figure 2 Relationship between distance and airflow by status plot ......................................... 12
Figure 3 Distribution plot......................................................................................................... 12
Figure 4 Confusion matrix ....................................................................................................... 14
Figure 5 Correlation matrix ..................................................................................................... 15

iii
LIST OF TABLES

Table 1 Accuracy score ............................................................................................................ 13

iv
Comparison of Machine Learning Models for Classification of Acoustic Fire Extinguishing Status

1. INTRODUCTION

One of the most destructive and hazardous forces on Earth is fire, which can quickly cause
significant harm to infrastructure and human life. Creating efficient fire suppression systems is
consequently crucial for safeguarding property and public safety. Sound wave or acoustic fire
extinguishing technologies have recently become popular as a creative way to identify and put
out fires before they get out of control. The way these devices function is by producing
particular sound frequencies that interfere with combustion and put out the fire.

To maximise the effectiveness of acoustic fire extinguishing, however, sophisticated analytics


and intelligence are needed to forecast the circumstances in which sound waves will effectively
put out the fire. A promising range of methods for creating data-driven models that can
precisely categorise the results of acoustic extinguishing on the basis of variables like as fuel
kind, flame size, sound intensity, frequency, airflow, and distance is offered by machine
learning. The capacity to forecast fire suppression outcomes with accuracy can lead to more
informed control decisions and increased productivity.

This study uses a dataset of 17,442 experiments from an acoustic fire extinguishing system to
compare the classification accuracy of five well-known machine learning algorithms: K-
Nearest Neighbours, Support Vector Machine, Decision Tree, Random Forest, and Naive
Bayes. Finding the best modelling strategy to forecast whether or not sound waves would
effectively put out the fire is the aim. A number of indicators are used to assess the models'
performance, including accuracy scores. The goal of this study is to establish an intelligence
foundation that can be used to develop intelligent decision support systems for next-generation
fire suppression technologies by determining which machine learning technique can most
reliably classify acoustic extinguishing results on this dataset.

1
Dept of Chemical Engineering Safety Management
Comparison of Machine Learning Models for Classification of Acoustic Fire Extinguishing Status

2. LITERATURE REVIEW

Murat Koklu et al. explores the use of machine learning to the optimisation of a new sound
wave fire-extinguishing system. The researchers used five different machine learning
algorithms: Artificial Neural Networks (ANNs), K-Nearest Neighbours (kNN), Random
Forest, Stacking (an ensemble combining ANN, kNN, and Random Forest), and Deep Neural
Networks (DNNs). The researchers leveraged a rich dataset of 17,442 experiments
encompassing fuel type, flame size, sound parameters, and fire outcomes. In order to optimise
the accuracy of flame extinction or non-extinction predictions, the study carefully assessed
every model using a rigorous 10-fold cross-validation procedure. The outcomes were
remarkable: with an amazing 97.06% classification accuracy, the Stacking ensemble was
crowned the winner, closely followed by Random Forest with 96.58% and ANN with 96.03%.
Although DNNs and kNNs demonstrated commendable accuracy (94.88% and 92.62%), the
ensemble method outperformed it by a substantial margin. This research demonstrates the
enormous potential of machine learning to improve the effectiveness of sound wave fire
extinguishing technologies, opening the door to more precise and effective fire suppression
systems that could reduce damage and save lives.

Yavuz Selim Taspinar et al. explores the optimisation of a revolutionary sound wave fire
extinguishing system through the use of rule-based machine learning. A system aimed at early
flame extinction was built by the researchers in recognition of the enormous threat posed by
fire, and a comprehensive dataset including 17,442 experiments was used to assess the system's
efficacy. They used three popular rule-based techniques, CN2 Rule, Decision Tree, and ANFIS
(Adaptive-Network Based Fuzzy Inference Systems), focusing on features such as fuel type,
flame size, acoustic parameters, and airflow. The inquiry was further enhanced by statistical
analysis employing Box Plot, Scatter Plot, and Correlation Analysis. Remarkably, the selected
algorithms produced outstanding outcomes: ANFIS classified data with 94.5% accuracy,
followed by CN2 Rule at 99.91% and Decision Tree at 97.28%. These encouraging results
demonstrate how rule-based machine learning may improve sound wave fire extinguishing
technology, opening the door to quicker and more accurate fire response times and ultimately
better protection of people and property.

Maila R. Angeles et al. investigated the possibility of acoustic fire extinguishers—a greener
substitute—for chemical-free, efficient fire suppression is investigated in this study. By
2
Dept of Chemical Engineering Safety Management
Comparison of Machine Learning Models for Classification of Acoustic Fire Extinguishing Status

examining a bigger frequency range (14-90 Hz) and higher wattage to target larger fires, it
expands on previous research. According to user assessments, the device is deemed suitable as
a substitute fire extinguishing instrument (scoring 3.8 out of 5) while lacking an automatic
response feature. This study advances the continuous creation of novel firefighting
technologies that could have a big impact on user friendliness and environmental safety,
especially in emergency scenarios where minors might have to handle the equipment. But more
research into power, design, and frequency spectrum optimisation could improve this
promising technology's efficacy and breadth of applications. Further investigation into user
testing beyond acceptability measures, such as comparisons with conventional extinguishers
concerning cost-effectiveness, safety concerns, and suppression speed, would offer a more
thorough grasp of the potential benefits and drawbacks of acoustic fire extinguishers.

Caiyi Xiong et al. explores the possibility of using acoustic waves to put out firebrand-attached
flames while they are in flight, keeping in mind the crucial role these flames play in the
development of wildfires. In order to achieve extinction, the study investigates the link between
sound pressure, frequency, and firebrand velocity using an adjustable pendulum system and a
standardised wooden firebrand model. The results show that extinguishing requires a minimum
sound pressure requirement, with small increases linked to higher frequencies. Interestingly, as
firebrand velocity increases, the minimum pressure for extinction lowers dramatically,
indicating a combined influence of motion and acoustic oscillation in upsetting flame stability.
This finding is further corroborated by the quantification of the extinction limit using a
distinctive Damköhler number that connects fuel residence duration and flame chemistry.
Overall, by utilising the targeted destabilisation effects of acoustic waves, this study offers a
viable path for firebrand flame reduction in the wildland-urban interface (WUI), opening the
door to a potentially life-saving and environmentally beneficial method of wildfire suppression.

Stefan Ivanov et al. investigated a combination system that makes use of acoustic waves for
flame extinction and Deep Neural Networks (DNNs) for video-based fire detection, this work
innovates in both areas. The system detects fires in video streams with an inexpensive hardware
configuration that includes a Movidius stick for DNN acceleration. A specially made acoustic
extinguisher that uses a sinusoidal sound modulated by a triangle waveform is activated upon
detection. This special design makes sure that there is enough sound pressure to put out flames
up to 130 cm away, which is a long length for this kind of technology. This strategy has a

3
Dept of Chemical Engineering Safety Management
Comparison of Machine Learning Models for Classification of Acoustic Fire Extinguishing Status

number of benefits. In comparison to traditional techniques, DNN-based detection offers


quicker and more precise fire identification, which could reduce reaction times and damage.
The acoustic extinguisher is safe for sensitive surroundings and the environment because it
doesn't use water or dangerous chemicals. Furthermore, the novel acoustic design and cost-
effective hardware point to high applicability and scalability. To analyse the system's
effectiveness in various settings, address potential environmental factors influencing wave
propagation, and determine the extinguishing technology's dependability for a range of fire
types and sizes, more research is necessary. Despite these difficulties, the combined strategy
offers a viable path forward for improving the detection and suppression of fires, which could
result in safer and more long-lasting solutions.

Sourav Kumar Bhoi et al. studied the current fire detection technologies to properly mitigate
the enormous devastation caused by this persistent menace. Since fire is the most common
accidental cause of death, it is imperative to do research into novel detection strategies. In order
to address this issue, this study examines FireDS-IoT, an Internet of Things (IoT) based system
that makes use of machine learning and sensor data in concert. Four sensors—MQ-135 (CO2),
MQ-2 (smog), MQ-7 (CO), and DHT-11 (temperature)—embedded with an Arduino are used
by FireDS-IoT to collect accurate environmental data. In order to distinguish between three
states—fire, no fire, and possibly fire—this data is then loaded into Python decision tree and
K-Nearest Neighbours (K-NN) algorithms. By diverging from conventional binary detection
techniques, this innovative method enables a more sophisticated comprehension of the
surroundings and may even enable proactive measures in ambiguous circumstances. The
algorithms in the study's experimental setting are trained using recorded scenarios, which has
produced encouraging results. K-NN outperformed the decision tree with an accuracy of
93.15% as opposed to 89.25%. This demonstrates how effective and dependable K-NN can be
as a fire detection technique inside the FireDS-IoT architecture. When the system detects a fire,
it sends an emergency message to a registered mobile number, allowing for a prompt reaction
that could save lives. However, additional study is necessary to address any potential
drawbacks. Subsequent research endeavours may involve augmenting the dataset and
examining its applicability to diverse real-world situations. Furthermore, investigating
ensemble approaches using K-NN and additional algorithms may improve robustness and
accuracy. Through the resolution of these issues and the expansion of the encouraging base
established by FireDS-IoT, further research can lead to more comprehensive and effective fire
detection systems, which will ultimately result in a safer future for everybody.
4
Dept of Chemical Engineering Safety Management
Comparison of Machine Learning Models for Classification of Acoustic Fire Extinguishing Status

Paweł Niegodajew et al. explores the intriguing field of acoustic wave-mediated flame
extinction, providing insightful information about a viable fire suppression alternative. The
researchers examined variables like speaker power, extinction pressure, frequency, burner
power, and distance to the flame using a lab setup that produced horizontal sonic waves. Their
main findings show an interesting relationship between extinction difficulty and frequency:
lower frequencies were more successful in putting out the fire. This implies that the flame's
basic dynamics are influenced by the acoustic waves' deeper interactions with it. Remarkably,
the extinction pressure was less affected by the fuel load, suggesting that the technique might
be used to a wider range of fuel types. The researchers used a Schlieren apparatus to visualise
the flame's behaviour under sonic impact. This method, which is based on fluctuations in
refractive index, catches the interaction between the flame and the acoustic waves quite well.
According to their findings, fluctuating disturbances and a constant mean flow of sound
produce extinction. The flame is eventually put out by this complex interaction that upsets its
stability. This work opens the door for more investigation and advancement in the exciting
topic of acoustic flame extinction by making a substantial contribution to our understanding of
the phenomenon. Future research could examine how well this method works in real-world fire
conditions, with larger flames, and with alternative flame geometries. Further research into the
precise mechanisms underlying the observed frequency dependency and the roles played by
disturbances and mean flow could provide more profound understanding and possibly result in
improved acoustic fire suppression devices.

5
Dept of Chemical Engineering Safety Management
Comparison of Machine Learning Models for Classification of Acoustic Fire Extinguishing Status

3. OBJECTIVES

• Apply machine learning models like KNN, Naive Bayes, random forest, decision tree,
SVM to predict fuel flame extinguishing status

This involves training and testing several different machine learning classification models on
data related to fuel flame status to see which one can most accurately predict whether a flame
is extinguished or not. You will apply supervised learning techniques like K-Nearest
Neighbors, Naive Bayes, Random Forest, Decision Trees, and Support Vector Machines.

• Compare performance of models to determine most accurate one

After training the models, you will evaluate and compare their performance using metrics like
accuracy, precision, recall, F1-score. This will show which model makes the most accurate
predictions for the extinguishing status. The best performing model can be selected as the final
model.

• Identify key parameters that indicate extinguishing status

Analyze the most important input features in the best model to understand which parameters
of the data are most indicative of determining the extinguishing status. For example, variables
like flame temperature, flame height, fuel flow rate may be key factors. Identifying these key
parameters provides insight into the prediction.

6
Dept of Chemical Engineering Safety Management
Comparison of Machine Learning Models for Classification of Acoustic Fire Extinguishing Status

4. METHODOLOGY

4.1 Data Acquisition


Data Source: The dataset used in this study was obtained from the Kaggle repository,
specifically the "Acoustic Extinguisher Fire Dataset" by Murat Koklu
(https://www.kaggle.com/datasets/muratkokludataset/acoustic-extinguisher-fire-dataset).

Data Description: This dataset comprises 17,442 rows and 7 columns, encapsulating extensive
information about fire extinguishing experiments conducted using a sound wave system. The
features include:

• Fuel Type: The type of fuel used in the experiment (e.g., Gasoline, Kerosene, Thinner,
LPG).
• Flame Size: The size of the flame in centimetres.
• Decibel (dB): The sound wave intensity in decibels.
• Frequency: The frequency of the sound wave in Hertz.
• Airflow: The airflow rate in meters per second.
• Distance: The distance between the sound source and the flame in centimetres.
• Extinguished (Target Variable): A binary variable indicating whether the flame was
successfully extinguished (1) or not (0).

Data Splitting: Training data contains 75% of the dataset and testing data contains 5% of the
dataset.

4.2 K-Nearest Neighbours


KNN (K-Nearest Neighbours) closely examines the connections between the training data's
characteristics and extinction status during training. In the training data, KNN finds its k nearest
neighbours (most comparable cases) when faced with a novel, unseen fire scenario. It then uses
the most frequent result among those neighbours to estimate the extinction status for the current
situation, thereby using experience to make educated predictions about future fires.

7
Dept of Chemical Engineering Safety Management
Comparison of Machine Learning Models for Classification of Acoustic Fire Extinguishing Status

4.3 Support Vector Machine


The data points indicating extinguished and non-extinguished flames are successfully separated
using a hyperplane created using SVM (Support Vector Machine). In order to do this, it
maximises the margin, or the separation between the closest data points from each class and
the hyperplane. Because of its emphasis on maximising the margin, SVM is more resilient to
outliers and improves generalisation to new data.

• Principal Benefits:

Effectiveness in high-dimensional spaces: SVM is appropriate for the multidimensional


character of this dataset since it can handle complex decision boundaries.

Robustness against outliers: By emphasising margin maximisation, it becomes less susceptible


to outliers and lessens their effect on the correctness of the model.

Excellent generalisation performance: SVM frequently performs well when applied to


previously unseen data, which is essential for practical fire extinguishing applications.

• Particular Things to Think About with This Dataset:

Kernel Selection: Based on the properties of the data and any potential non-linear relationships,
the best kernel (such as a linear, polynomial, or radial basis function) will be selected.

Hyperparameter tuning: To maximise the performance of the model, parameters such as the
kernel coefficient and regularisation parameter will be fine-tuned.

4.4 Decision Tree


Decision trees provide an organised method for identifying patterns in the dataset. They build
a framework that resembles a tree, with branches standing in for choices made in response to
feature values. The data is divided by each node in the tree according to a distinct attribute,
which eventually results in leaf nodes that indicate the anticipated class (extinguished or not
extinguished). To train the model, the most informative features are chosen iteratively to
produce splits that effectively divide the data into different classes. This procedure keeps on
until an appropriate stopping criterion—like a maximum depth or a minimum quantity of
samples at a node—is satisfied. A clear and understandable model for predicting flame
extinction based on the discovered correlations between attributes is offered by the decision
tree that results.

8
Dept of Chemical Engineering Safety Management
Comparison of Machine Learning Models for Classification of Acoustic Fire Extinguishing Status

4.5 Navie Bayes


Using probability and the presumption of feature independence, the simple but powerful
classification method Naive Bayes generates predictions. Its foundation is the Bayes Theorem,
which determines how likely it is that an object would fit into a particular class depending on
its characteristics. In jobs where features aren't tightly connected, such text classification, spam
filtering, sentiment analysis, and other areas, it frequently performs remarkably well, despite
its "naive" assumption. It is a well-liked option in many machine learning applications because
of its ease of use, speed, and capacity to handle both numerical and categorical data.

4.6 Random Forest


Similar to a group of professionals brainstorming to get the best prediction, Random Forest It
creates a "forest" of many decision trees, each of which is trained using a distinct random subset
of the features and data. The final forecast is the most popular vote, with each tree casting a
vote for the class it predicts. Compared to individual decision trees, Random Forest is less
prone to overfitting and more accurate because of its collaborative nature.

4.7 Accuracy Score


The accuracy score is the main metric used in this study to assess how well the model predicts
flame extinction. It is determined by dividing the total number of predictions by the percentage
of accurate predictions (both extinguished and not extinguished). This simple statistic makes it
easy to assess how well the whole model can classify the results of flame extinction. A high
accuracy score in the context of fire suppression indicates dependable model performance,
boosting trust in the model's practical applicability.

4.8 Confusion Matrix


Confusion matrices are an important tool in this work for assessing how well machine learning
models classify the results of flame extinction. By contrasting the model's predicted
classifications with the actual results, a tabular representation known as a confusion matrix
helps to visualise the model's performance. Beyond only accuracy, this matrix reveals trends in
both accurate and inaccurate forecasts.

For this dataset, the confusion matrix will have four key values:

9
Dept of Chemical Engineering Safety Management
Comparison of Machine Learning Models for Classification of Acoustic Fire Extinguishing Status

• True Positives (TP): Correctly predicted flame extinction.


• True Negatives (TN): Correctly predicted no flame extinction.
• False Positives (FP): Incorrectly predicted flame extinction (Type I error).
• False Negatives (FN): Incorrectly predicted no flame extinction (Type II error).

Through a thorough analysis of these numbers, we are able to evaluate not just the overall
accuracy but also the model's propensity to commit particular kinds of errors, providing us with
a more comprehensive picture of the model's performance and areas for development in this
particular context.

10
Dept of Chemical Engineering Safety Management
Comparison of Machine Learning Models for Classification of Acoustic Fire Extinguishing Status

5. RESULTS

5.1 Data Analysis


Count of Each Fuel Type: Figure 1 displays the frequency of each fuel type within the dataset.
It helps to understand that gasoline, kerosene, LPG are greater than 5000 and thinner around
2000.

Figure 1 Count of each fuel type plot

Relationship between Distance and Airflow by Status: Figure 2 shows the relationship between
distance and airflow, with the data points coloured by status. This helps us to understand that
when the air flow is higher and distance is smaller it creates more extinction state of the fire.

11
Dept of Chemical Engineering Safety Management
Comparison of Machine Learning Models for Classification of Acoustic Fire Extinguishing Status

Figure 2 Relationship between distance and airflow by status plot

Distribution plot: Figure 3 shows the distribution of DISTANCE, DECIBEL, AIRFLOW and
FREQUENCY.

Figure 3 Distribution plot

12
Dept of Chemical Engineering Safety Management
Comparison of Machine Learning Models for Classification of Acoustic Fire Extinguishing Status

5.2 Accuracy Score

Table 1 Accuracy score

Model KNN SVM Decision Naïve Bayes Random


Tree Forest
Accuracy
94.81% 94.13% 96.84% 87.27% 96.38%
Score

Decision Tree Outperforms: With an accuracy score of 96.84%, the Decision Tree model
outperformed Random Forest, which came in second with 96.38%.

KNN and SVM Perform Well: With accuracy scores of 94.81% and 94.13%, respectively, KNN
and SVM both showed strong performance.

Naive Bayes Underperforms: With an accuracy of 87.27%, Naive Bayes is the least accurate
model and may not be the best choice for this dataset.

5.3 Confusion Matrix


False Positives: Of the three models, Decision Tree had the fewest false positives (62),
suggesting a decreased propensity to incorrectly forecast flame extinction when it hadn't
happened.

False Negatives: Random Forest had the fewest false negatives (88), indicating a decreased
likelihood of mis-predicting the actual extinction of flames.

Decision Tree: High accuracy was attained, but it may be prone to overfitting, which could
limit its ability to generalise to new data.

In addition to performing effectively, Random Forest frequently provides superior


generalisation over individual Decision Trees.

Naive Bayes: Its poorer performance may be due to its inability to capture intricate correlations
between features.

13
Dept of Chemical Engineering Safety Management
Comparison of Machine Learning Models for Classification of Acoustic Fire Extinguishing Status

Figure 4 Confusion matrix

14
Dept of Chemical Engineering Safety Management
Comparison of Machine Learning Models for Classification of Acoustic Fire Extinguishing Status

5.4 Correlation Matrix


Figure 5 shows the correlation of different parameters with the STATUS of the fire. The highest
correlation is showed by the AIRFLOW followed by DISTANCE and FREQUENCY.

Figure 5 Correlation matrix

15
Dept of Chemical Engineering Safety Management
Comparison of Machine Learning Models for Classification of Acoustic Fire Extinguishing Status

6. CONCLUSIONS

• Among the models tested, the Decision Tree model outperformed Random Forest (96.38%),
KNN (94.81%), SVM (94.13%), and Naive Bayes (87.27%) with the greatest accuracy of
96.84%. This suggests that Decision Trees may successfully identify patterns in the data to
forecast the state of a flame that has been extinguished.
• The extinguishing state was shown to be most correlated with the distance between the
sound source and flame, airflow rate, and sound frequency. This sheds light on the crucial
variables for acoustic wave-based flame suppression prediction.
• Additionally, KNN and SVM models did fairly well, achieving accuracy rates greater than
94%. They are adaptable solutions for this kind of classification challenge because of their
non-parametric nature.
• The findings show that machine learning methods such as Random Forest, Decision Trees,
KNN, and SVM may be used to accurately categorise the effects of suppressing acoustic
fires based on variables such as airflow, frequency, and distance.

16
Dept of Chemical Engineering Safety Management
Comparison of Machine Learning Models for Classification of Acoustic Fire Extinguishing Status

REFERENCES

1. Koklu, M., & Taspinar, Y. S. (2021). Determining the extinguishing status of fuel flames
with sound wave by machine learning methods. IEEE Access, 9, 86207–86216.
https://doi.org/10.1109/access.2021.3088612
2. Ivanov, S., Stankov, S., Wilk-Jakubowski, J., & Stawczyk, P. (2021). The using of deep
neural networks and acoustic waves modulated by triangular waveform for extinguishing
fires. New Approaches for Multidimensional Signal Processing, 207–218.
https://doi.org/10.1007/978-981-33-4676-5_16
3. Taspinar, Y. S., Koklu, M., & Altin, M. (2021). Classification of flame extinction based
on acoustic oscillations using artificial intelligence methods. Case Studies in Thermal
Engineering, 28, 101561. https://doi.org/10.1016/j.csite.2021.101561
4. Xiong, C., Liu, Y., Xu, C., & Huang, X. (2020). Acoustical extinction of flame on moving
firebrand for the fire protection in Wildland–Urban Interface. Fire Technology, 57(3),
1365–1380. https://doi.org/10.1007/s10694-020-01059-w
5. Angeles, M. R., Rañada, J. V., Lopez, D. C., Apilado, J. R., Carlos, Ma. S., David, F. J.,
Escario, H. J., Rolloda, A. I., & Villanueva, I. C. (2020a). Development of variable
acoustic soundwave for fire prevention. Lecture Notes in Electrical Engineering, 331–339.
https://doi.org/10.1007/978-981-15-0058-9_32
6. Angeles, M. R., Rañada, J. V., Lopez, D. C., Apilado, J. R., Carlos, Ma. S., David, F. J.,
Escario, H. J., Rolloda, A. I., & Villanueva, I. C. (2020b). Development of variable
acoustic soundwave for fire prevention. Lecture Notes in Electrical Engineering, 331–339.
https://doi.org/10.1007/978-981-15-0058-9_32
7. Bhoi, S. K., Panda, S. K., Padhi, B. N., Swain, M. K., Hembram, B., Mishra, D., Mallick,
C., Singh, M., & Khilar, P. M. (2018). Fireds-IOT: A fire detection system for smart home
based on IOT Data Analytics. 2018 International Conference on Information Technology
(ICIT). https://doi.org/10.1109/icit.2018.00042
8. Niegodajew, P., Łukasiak, K., Radomiak, H., Musiał, D., Zajemska, M., Poskart, A., &
Gruszka, K. (2018). Application of acoustic oscillations in quenching of gas burner flame.
Combustion and Flame, 194, 245–249.
https://doi.org/10.1016/j.combustflame.2018.05.007
9. Friedman, A. N., & Stoliarov, S. I. (2017). Acoustic extinction of laminar line-flames. Fire
Safety Journal, 93, 102–113. https://doi.org/10.1016/j.firesaf.2017.09.002

17
Dept of Chemical Engineering Safety Management
Comparison of Machine Learning Models for Classification of Acoustic Fire Extinguishing Status

10. Hou, S.-S., Chung, D.-H., & Lin, T.-H. (2015). Experimental and numerical investigation
of jet flow and flames with acoustic modulation. International Journal of Heat and Mass
Transfer, 83, 562–574. https://doi.org/10.1016/j.ijheatmasstransfer.2014.12.047

18
Dept of Chemical Engineering Safety Management

You might also like