AIR FORCE INSTITUTE OF TECHNOLOGY KADUNA

SCHOOL OF POST GRADUATE STUDIES

DEPARTMENT OF ELECTRICAL ELECTRONICS ENGINEERING

Re vie w o f ma c h in e l e a r n in g me t h o d s in
s o u n d a n o ma l y d e t e c t io n ; r e l a t e it t o
t r a in a n d r a il wa y.

RESEARCH METHODOLOGY PRESENTATION

BY
ADEBISI RICHARD RASHEED
MURKTAR NASIR
ADEKOGBA OPEYEMI OLAJUMOKE
 SCOPE

INTRODUCTION
 RELATED WORKS
 PRESENT WORK
 CONCLUSION
 INTRODUCTION
An anomaly is a pattern in data that does not conform to a well-defined notion of normal
behavior. An anomaly has a statistical meaning, where some event occurs very infrequently
(rare event). Anomalous Sound Event Detection (SED) offers novel options to improve audio-
based man-machine interaction in many useful applications in trains and railways such as fault
detection, surveillance system and safety monitoring on railway stations.

 Anomaly detection techniques are categorized into three:

• Signal Processing based method
• Deep learning method
• Machine learning method
 However, our focus is on Machine Learning ML method.
 Machine learning (ML) is the process of estimating unknown dependencies or structures
in a system using a limited number of observations or datasets.
 Machine learning uses two types of techniques:
• Supervised Learning, which trains a model on known input and output data so that it can
predict future outputs. Here, the training sets are fully labelled. Examples are F1-score,
Precision, Evaluation Metric etc.
• Unsupervised Learning, which finds hidden patterns or intrinsic structures in input data.
Examples are Area under ROC curve etc.

 Basic ML methods are: decision trees, Bayes network, Artificial Neural Network, Support
Vector Machines, K-nearest Neighbors.
 In the United States (as a case study), the railroad transportation system was first
considered in 1826, making it one of the oldest transportation systems.
 Today, the industry is categorized as industry 4.0 due to its dependence on technological
advances to enhance the quality of services and utilization of resources.
 In developed country, The railroad transportation is more widely used than any other
mode of transportation.
 Due to the several assets incorporated with the rail systems, the Federal Railroad
Administration (FRA) provides operative thresholds that guide the operation of the railway.
 As such, rail track alone takes about 55% of the total maintenance cost even though it can be
minimized by adequate optimal maintenance scheduling.
 Despite the enormous spending, track-related accidents have continued to be a significant
source of transportation casualties over the past years;
 In 2021, according to FRA, the total accident recorded was 7823, with track failure
contributing the larger portion (42%). In order to mitigate these losses, there is a need to
explore different approaches to how maintenance data are acquired processed and
implemented.
 Factors, such as cracks, ballast issues, rail discontinuity, loose nuts and bolts, burnt wheels,
super elevation, and misalignment occurs on rails.
 Consequently, delayed detection can pose grave danger and threats to the safe operation of
rail transport.
 The traditional procedure of manually inspecting the rail track is both inefficient and prone to
human error and biases. However Machine learning techniques has the potential to
distinguish and recognize sounds based on pitch, energy, sound entropy, and spectral
analysis.
 Hence, this review leverages machine learning methods to detect and classify faults in railway
tracks.
Review of related works

 Main keywords for this review are: Machine Learning methods; Sound anomaly detection
and Fault Detection on railways; were used to survey published papers over a five (5) year
period.
 Because this is a widely researched and dynamic area, the review focused on a relatively
short time period and encompassed the most recent sources to ensure the analysis
conducted was novel.
 According to the search request, Google scholar returned 54 results from which fifteen
publications were selected.
 The surveyed articles were selected according to the direction of the literature review and
the indicated criteria (application area, significance of achievements, reliability and
feasibility of results).
 Related works…
 In recent times, the rapid advancement in the railroad industry has shown the impact of the science
of big data analytics, covering insight extraction, cost-saving, and increasing transportation efficiency.
In the cause of this review, there are many methods of sound anomaly detection using machine
learning and other methodology; which are:

 1. Bearing Fault Detection for Railway Traction Motors through Leakage Current (2021): In this
paper, a bearing fault detection method for a railway traction motor through leakage currents was
proposed. The proposed detection method combines octave band analysis and machine learning. The
abnormality simulation experiments with an inner race fault bearing are conducted and the
effectiveness of the proposed method is verified. From the experiments, it is confirmed that the
proposed method can detect failures of railway traction system well at specific conditions and
leakage currents have potentials to be used for bearing fault detection.
Related works…
 2. Applications of Machine Learning For machine Fault Diagnosis (2018): Intelligent fault
diagnosis (IFD) refers to applications of machine learning theories to diagnose machine faults.
 Although, advanced signal processing methods help to ensure which types of faults or where the
faults happened in machines. However, these methods greatly rely on the specialized knowledge
that maintainers mostly lack. Furthermore, the diagnosis results by signal processing methods are
too specialized to be understood by the machine users.
 Therefore, modern industrial applications prefer the fault diagnosis methods that could
automatically recognize the health states of machines.
 IFD diagnosis procedure includes three steps, i.e., data collection, artificial feature extraction, and
health state recognition
 IFD diagnosis procedure steps…

 Artificial feature extraction and Embedded methods are used to integrate the feature
selections into training the classifiers.
 However, the major limitation of technology is its requirement to have a large database of
various normal and abnormal data for machine recognition. Also when the train is at maximum
speed, the fault recognition error rate is high.
3. Condition Monitoring for Railway Point Machines (RPMs) Based on Sound Analysis
and Support Vector Machine:

 RPMs are one of the key equipment’s in the railway system to switch different routes for the
trains. Condition monitoring for RPMs is a vital measure to keep train operation safe and reliable.
 In this paper, a novel intelligent condition monitoring method for RPMs based on sound analysis
was proposed.
 Time-domain and frequency-domain features are obtained and normalized, using z-score
standardization method to eliminate the influences of different dimensions. Binary particle
swarm optimization (BPSO) is utilized to select the most significant discrimination feature subset.
 The effects of the selected optimal features are verified using Support vector machine (SVM), 1-
Nearest neighbor (1NN), Random forest (RF), and Naive Bayes (NB).
 Experiment results indicate SVM performs best on identification accuracy and computing cost
compared with the other three classifiers.
 Limitations: At present, most studies on fault diagnosis for RPMs are based on motor electrical
signals. However, electrical signals are not easy to acquire, and the collection process may bring
interference to the RPMs which is unwanted addition noise.
4. Rail track condition monitoring: a review on deep learning approaches.
 In this paper, Railway track inspection was classified into three main steps.
 The first step is the pre-processing of the ’wav’ files for eliminating undesired sounds. Secondly,
feature extraction is performed by using spectrograms. Then the spectrograms are generated on-the-
fly to make the system more flexible. Lastly, the classification model is trained for railway track fault
detection.
 In this research, three different deep learning models are tested for their performance. The research
focuses on convolutional 1D, convolutional 2D, and long short-term memory (LSTM).
 The dataset used in this research is a balanced one with three classes including normal, super
elevation, and wheel burnt.
 The first step of preprocessing is to remove the noise of air, rain, and all the dead space in the audio
by using a signal envelope function with a threshold of 100 Hz. After cleaning all audio files, they are
then sampled to 16 kHz and then split into 1.7 s, 3.4 s, and 8.5 s durations for the experiments.
 Limitations: Further review show that there is no enhancing of dataset in different terrains and they
should incorporate other sensors such as a gyroscope, a seismic sensor, and an optical sensor to
further improve the performance and robustness.
 Furthermore, in future, the cart will be capable of recording the geographical location for each audio
recording and the location of the track fault will be provided along with the fault type.
 PRESENT WORK

 Rail System Anomaly Sound Detection through Machine and Deep

Learning Applications:
 This review explores building machine and deep learning models
with the help of acoustic transducer to detect anomalies in a rail
SCADA system through the usage of network traffic data.
CONCLUSION
References
1. Yao, D., Sun, Q., Yang, J., Liu, H., & Zhang, J. (2020). Railway Fastener Fault Diagnosis Based on Generative
Adversarial Network and Residual Network Model. Shock and Vibration, 2020.
https://doi.org/10.1155/2020/8823050
2. Vileiniskis, M.; Remenyte-Prescott, R.; Rama, D. A fault detection method for railway point systems. Proc. Inst.
Mech. Eng. F J. Rail Rapid Transit 2018.
3. Ardakani, H.D.; Lucas, C.; Siegel, D.; Chang, S.; Dersin, P.; Bonnet, B.; Lee, J. PHM for railway system—A case
study on the health assessment of the point machines. In Proceedings of the 2012 IEEE Conference on
Prognostics and Health Management (PHM), Denver, CO, USA, 18–21 June 2012; pp. 1–5.
4. Bolbolamiri, N.; Sanai, M.S.; Mirabadi, A. Time-domain stator current condition monitoring: Analyzing point
failures detection by Kolmogorov-Smirnov (K-S) test. Int. J. Electr. Comput. Energ. Electron. Commun. Eng. 2012,
6, 587–592.
5. Silmon, J.; Roberts, C. Improving railway switch system reliability with innovative condition monitoring
algorithms. Proc. Inst. Mech. Eng. F J. Rail Rapid Transit 2010, 224, 293–302.
6. Cristianini, N.; Shawe-Taylor, J. An Introduction to Support Vector Machines and Other Kernel-Based Learning
Methods; Cambridge University Press: Cambridge, UK, 2000.
7. Shanthi, T.S.; Lingam, C. Review of feature extraction techniques in automatic speech recognition. Int. J. Sci.
8. Eng. Technol. 2013, 2, 479–484.
9. 51. Shen, P.; Changjun, Z.; Chen, X. Automatic speech emotion recognition using support vector machine. In
Proceedings of the International Conference on Electronic & Mechanical Engineering and Information
Technology, Harbin, China, 12–14 August 2011; pp. 621–625.
Thanks for listening…