217

3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Abril 2020

ADHESION LEVEL IDENTIFICATION IN WHEEL-RAIL

CONTACT USING DEEP NEURAL NETWORKS

Sanaullah Mehran Ujjan

ME Scholar, NCRA- Condition Monitoring Lab.

Mehran University of Engineering and Technology. Jamshoro, (Pakistan).

E-mail: mehranujjan44@gmail.com ORCID: https://orcid.org/0000-0003-1879-8754

Imtiaz Hussain Kalwar

Head of Department, Electrical Engineering.

DHA Sua University. Karachi, (Pakistan).

E-mail: imtiaz.hussain@dsu.edu.pk ORCID: https://orcid.org/0000-0002-7947-9178

Bhawani Shankar Chowdhry

Professor Emeritus.

Mehran University of Engineering and Technology. Jamshroo, (Pakistan).

E-mail: bhawani.chowdhry@faculty.muet.edu.pk ORCID: https://orcid.org/0000-0002-4340-9602

Tayab Din Memon

Chairman, Department of Electronics.

Mehran University of Engineering and Technology. Jamshoro, (Pakistan).

E-mail: tayabdin82@gmail.com ORCID: https://orcid.org/0000-0001-8122-5647

Dileep Kumar Soother

Research Assistant, NCRA- Condition Monitoring Lab.

Mehran University of Engineering and Technology. Jamshoro, (Pakistan).

E-mail: dileepkalani1994@gmail.com ORCID: https://orcid.org/0000-0002-6211-1078

Recepción:

29/01/2020

Aceptación:

17/04/2020

Publicación:

30/04/2020

Citación sugerida Suggested citation

Ujjan, S. M., Kalwar, I. H., Chowdhry, B. S., Memon, T. D., y Soother, D.K. (2020). Adhesion level

identication in wheel-rail contact using deep neural networks. 3C Tecnología. Glosas de innovación aplicadas

a la pyme. Edición Especial, Abril 2020, 217-231. http://doi.org/10.17993/3ctecno.2020.specialissue5.

217-231

218

http://doi.org/10.17993/3ctecno.2020.specialissue5. 217-231

3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Abril 2020

ABSTRACT

Robust and accurate adhesion level identication is crucial for proper operation of railway

vehicle. It is necessary for braking and traction forces characterization, development of

maintenance strategies, wheel-rail wear predictions and development of robust onboard

health monitoring systems. Adhesion being the function of many uncertain parameters is

dicult to model, whereas data driven algorithms such as Deep Neural networks (DNNs)

are very good at mapping a nonlinear function from cause to eect. In this research a solid

axle Wheel-set was modeled along with dierent adhesion conditions and a dataset was

prepared for the training of DNNs in Python. Furthermore, it explored the potential of

DNNs and various data driven algorithms on our noisy sequential dataset for classication

task and achieved 91% accuracy in identication of adhesion condition with our nal

model.

KEYWORDS

Wheel-rail contact, Adhesion, Solid-axle Wheel-set, Deep learning, Neural networks,

Classication, Time-series data.

219

http://doi.org/10.17993/3ctecno.2020.specialissue5. 217-231

3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Abril 2020

1. INTRODUCTION

Adhesion level identication is an important task for proper operation of a railway vehicle

as traction eort is a direct function of adhesion coecient. Hence dierent magnitude

of tracking and braking forces is required in dierent contact conditions, making adhesion

level identication crucial for tracking and braking forces characterization (Spiryagin et al.,

2014). Both overestimation or under estimation can cause either derailment or rapid noise

at wheel rail interface (Olofsson, 2009).

Adhesion coecient quantitatively relates ratio of traction eort to the normal load on the

wheel and qualitatively represents contact condition between wheel and rail.

Many cases of derailment have been reported in Pakistan (“A Timeline of Neglect: Train

Incidents in Pakistan”, 2019) due to various causes but with a suitable mechanism of

adhesion level identication and adhesion control at least one factor (over/under estimation

of adhesion condition) can be tackled down.

Adhesion problem has sought attention of many researchers and many solutions ranging

from mathematical control theory to statistical and genetic algorithms (Bibi, Chowdry, &

Shah, 2018) have been proposed and applied (Shrestha, Wu, & Spiryagin, 2019). Although,

here we compare two of them, model based and data driven work on adhesion, but later

being rarely applied we aim to express the potential of data driven models in railway

domain.

1.1. MODEL BASED WORK

Several attempts to correctly estimate the adhesion have been reported in recent years

related to model based approaches (Shrestha et al., 2019; Malvezzi et al., 2013; Hussain,

2012; Ward et al., 2012) where a mathematical relation between excitation and response of

the system is constructed. Adhesion has been expressed as a function of slip velocity and

acceleration by Carl and Brook (1985), and Malvezzi et al. (2013). Function of longitudinal

vehicle velocity and pressure in contact zone by Spiryagin et al. (2016), function of lateral

creep force and yaw movement by Ward et al. (2012).

220

http://doi.org/10.17993/3ctecno.2020.specialissue5. 217-231

3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Abril 2020

However modelling a physical phenomenon such as adhesion with too many uncertain

or even unknown factors is a time taking task. As any mathematical model inherently has

the tendency to ignore some parameters in the process but ignoring adhesion eective

parameters for the sake of simplicity and robustness of the model can cause diculties

especially when the model is later on used for complex tasks e.g. development of a Real-

time on-board health monitoring system in locomotives (Shah et al., 2020).

Nonetheless, with the current state of knowledge about adhesion it remains a riddle

to be solved. With the beginning of Era of Big data and Deep learning models whose

implementations in many domains yielded state of the art results. In 2018 Rail safety and

standards board announced 300,000 euros for encouraging the researchers to develop data

driven solutions for solving the adhesion riddle.

1.2. DATA DRIVEN WORK

Only few examples of implementation of data driven solutions for solving adhesion have

been reported up till now, Castillo et al. (2016) have used Articial Neural Networks (ANNs)

to estimate adhesion states in an ABS system. Li, Feng, and Wei (2015) have used optimized

Recursive ANN with the focus on optimizing the utilization rate of adhesion available.

Gajdar, Rudas, and Suda (1997) have used classical BP Neural network to estimate friction

coecient of the rail wheel which is mathematically dierent but closely associated with

adhesion coecient. Data is extracted through a simulated wheel rail condition. Malvezzi

et al. (2013) have used Neural networks in combination with an adhesion model to identify

adhesion coecient. Experimental data is acquired during tracking and braking tests.

Zhang et al. (2017) have used a deep auto sparse encoder to estimate adhesion status of a

locomotive however the parameters chosen for training the Autoencoder are not directly

measurable and have been acquired through an SMC estimator, so the accuracy of the data

driven model heavily depends upon the SMC observer. From the above mentioned data

driven approaches are experimentally veried and show viable results (Castillo et al., 2016;

Li et al., 2015). However, very few approaches solely focus on adhesion identication and

ignore the complexities associated with it, also any data driven algorithm should be able to

identify adhesion conditions from directly measurable parameters rather than relying on

estimated parameters (Zhang et al., 2017; Gajdar et al., 1997).

221

http://doi.org/10.17993/3ctecno.2020.specialissue5. 217-231

3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Abril 2020

In this research, a data driven method based on DNNs is employed, which solely focuses

on adhesion condition identication. Main objective of this study has been to produce a

self-reliant data driven solution to the adhesion riddle. Data driven model should predict

adhesion condition from noisy sensor data, without relying on estimation theory to get

clearer data at their input. This will enhance possibility of integration of this model in an

onboard condition monitoring system.

Rest of the paper proceeds as follows. Section 2 and 3 describe the simulation of the wheel-

rail interaction and dierent adhesion conditions, Section 4 includes data recording and

dataset preparation, Section 5 explains the process of implementation of DNNs, Section

6 compares and discusses the results obtained with dierent data driven algorithms, and

Section 7 concludes the paper.

2. SIMULATING WHEEL-RAIL INTERACTION

In order to acquire data for the training of DNNs, a Wheel-rail contact model which

manifests the running of a wheel on a track in dierent adhesion conditions was needed

so a nonlinear model of a solid axle Wheel-set established in Hussain, Mei, and Ritchings

(2013) was simulated in MATLAB/Simulink.

Figure 1 shows diagram of the modeled solid-axle Wheel-set, F

and F

are longitudinal

creep forces, w

and w

are angular velocities of right and left wheel respectively and V is

longitudinal velocity of the Wheelset under the eect of creep forces.

Figure 1. Solid Axle Wheel-set on a straight track.

Below are the equations of motions of this Wheel-set model.

222

http://doi.org/10.17993/3ctecno.2020.specialissue5. 217-231

3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Abril 2020

(1)

(2)

(3)

(4)

Here θ

is angle of twist when both wheels have dierent angular velocities, where as F

, F

creep forces of both wheels and can be calculated as:

(5)

(6)

Where F

and F

are total creep forces of left and right wheel calculated from adhesion

coecient µ and normal load N.

(7)

(8)

µr and µl are adhesion coecients representing the dierent adhesion levels, these

coecients are calculated by Polach model (Polach, 2005).

3. SIMULATING DIFFERENT ADHESION CONDITIONS

In order to manifest the dierent adhesion conditions or adhesion coecient values in (8)

(9) on which the Wheel-set model was to run, we used Polach model to generate dierent

adhesion conditions (Polach, 2005), Figure 2 shows dierent adhesion conditions on which

the simulations are carried out.

223

http://doi.org/10.17993/3ctecno.2020.specialissue5. 217-231

3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Abril 2020

Figure 2. Adhesion curves from Polach Model.

Here are the equations used in simulating dierent adhesion conditions based on Polach

model (Polach, 2005).

(9)

(10)

(11)

In (9) µ is adhesion coecient and A, B are curve tuning parameters whereas V

is creep

velocity, In (10) F is creep force, Q is wheel load, and ε is gradient of tangential stress in the

area of adhesion which is calculated in (11) from a, b half axes of contact ellipses, contact

sheer stiness C, and total creep s.

4. DATA RECORDING AND PREPARTION OF DATASET

On each of the contact conditions manifested by Polach model, Wheel-set was run for

5 minutes and parameters which are practically measurable through sensors namely,

longitudinal velocity V, angular velocity of both wheels and the integration of the dierence

between them θ

was generated using simulations in presence of constant track disturbance.

224

http://doi.org/10.17993/3ctecno.2020.specialissue5. 217-231

3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Abril 2020

Figure 3 shows the one sample of theta recorded during 0.63 seconds of simulation on 7

dierent adhesion conditions, length of the sample 0.63 seconds is chosen with a trial and

error process, lower sample size being more preferable as it would increase the response

time of the DNN, however 0.63 was identied as limit without deteriorating performance

of the model, magnitude of θ

seems to lower with decrease in adhesion, but frequency

seems to increase. However signal’s characteristics are not part of discussion here as data

driven algorithms will be feed individual values of and should classify and identify them as

dierent signals

Figure 2. Single Sample of θ

on different Adhesion conditions.

Keeping in mind the robustness required in the model, recorded parameters were analyzed

and θ

was chosen as the parameter to train the DNNs as it is a scaled and meaningful

representation of angular velocities of both wheels, hence reducing the no of features and

simplifying the process of convergence of DNN.

Recorded sequence data was converted into a comma separated le (CSV) and preprocessed

using Python programming in order to be used for training the DNN, Sequences of

recorded over dierent adhesion conditions were spilt on equal interval of 0.63 seconds,

labeled and stacked into a matrix of n x m shape as shown below:

225

http://doi.org/10.17993/3ctecno.2020.specialissue5. 217-231

3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Abril 2020

(12)

Here n=100 no of features in a single sample of theta for 0.63 seconds, where m= 3290

total no of examples from all classes, 470 from each class or adhesion condition. Last row

represents labels associated with each class or adhesion condition [0, 6]. Data matrix was

shued and spilt using 70/30 rule into Train set which was to be used for training the DNN

and Test set on which the validation of the trained model was done.

5. DNN IMPLEMENTATION

DNN was implemented using the Tensorow and Keras, High End APIs of Python used

for implementation of Deep learning models. We went through various Hyper parameters

congurations of the DNN, no of neurons in each layer, no of layers, activation functions,

loss functions, regularization parameters, and optimizers. Figure 4 shows the conguration

of the model which exhibited the highest validation accuracy and less generalization error.

Figure 2. Block diagram of DNN.

The four densely-connected (each input is connected to every neuron in succeeding hidden

layer) each with 32 neurons with a relu (rectifying linear unit) activation were used. Each

neuron represents a matrix multiplication and summation operation, every neuron

superscript l representing the no of layer and n is number of neuron are calculated as in

(13) and then relu activation function is applied as in (14).

226

http://doi.org/10.17993/3ctecno.2020.specialissue5. 217-231

3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Abril 2020

(13)

(14)

In (13) w

represents weights of the NN which are randomly initialized from Gaussian

distribution and are optimized over iterations using an optimization function reported in

Kingma and Ba (2015). Equation (15) shows the way Adam optimizer way of updating

its weight, it can be thought of a combination between stochastic gradient descent and

RMSprop (Root Mean Square proportion).

(15)

Here η is the step size like learning rate,

and are exponentials of moving averages of

weights. Loss function used here is categorical cross entropy which is one of the most used

ways of loss calculation in multiclass classication problems. Softmax at the output is a stack

of multiple sigmoid activation functions in order to calculate the probability of each class.

6. RESULTS AND DISSCUSSION

Having formulated problem as a multiclass classication one, multiple data driven algorithms

were tried, Table 1 shows validation accuracy achieved by dierent data driven algorithms.

Table 1. Accuracy of different data driven algorithms on our dataset.

Algorithm Accuracy

Decision Tree 80%

Logistic Regression 19%

K-nearest Neighbours 62%

Support Vector Machine-Polynomial 71%

Support Vector Machine-Guassian 35%

Deep Neural Network (DNN/MLP) 91%

Table 1 shows that DNN outperformed traditional data driven methods by a large margin

on our sequence classication task and achieved reasonable accuracy to work with. Keeping

in mind the future real time implementation, and reasonable accuracy achieved on a simpler

227

http://doi.org/10.17993/3ctecno.2020.specialissue5. 217-231

3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Abril 2020

model other Deep Models such as Convolutional and Recurrent Neural Networks are not

compared here as those may be bulky and an overkill in terms of computational cost of

those models.

Confusion matrix in Figure 5 shows the accuracy achieved by DNN on individual classes

(adhesion conditions) and Table 2 show accuracy as well as precision and recall of the DNN

achieved by model.

Figure 2. Confusion Matrix (DNN).

Table 1. Classication report (DNN).

Precision Recall F1-score

Cd (Dry) 0.82 0.89 0.85

Cw (Wet) 0.93 0.99 0.95

Cm (Medium) 0.95 0.84 0.89

Cl (Low) 1.00 0.99 0.99

Cvl (Very low) 0.94 0.95 0.94

Cvvl (Very low) 0.87 0.83 0.85

Ce (Extremely low) 0.88 0.92 0.90

Accuracy 0.91

Macro average 0.91 0.91 0.91

Weighted average 0.91 0.91 0.91

228

http://doi.org/10.17993/3ctecno.2020.specialissue5. 217-231

3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Abril 2020

7. CONCLUSION AND FUTURE WORK

This paper demonstrates the potential of data driven model a Deep Neural Network solely

focusing on identication of contact condition or adhesion condition. It was observed in

the results that deep neural networks performed well on the task of inferring adhesion

condition from directly measurable parameters and achieved 91% accuracy on data.

However experimental data collection in railway is expensive and open access datasets are

a rarity, providing labels to that data is another problem in its entirety. Point to the future

is to attempt further on adhesion riddle using data driven model that may overcome the

hurdles of less availability of data and should be able to work in an unsupervised setting.

ACKNOWLEDGMENT

We acknowledge the support of the ‘Haptics, Human Robotics, and Condition Monitoring

Lab’ established in Mehran University of Engineering and Technology, Jamshoro under the

umbrella of National Center of Robotics and Automation funded by the Higher Education

Commission (HEC), Pakistan.

REFERENCES

A Timeline of Neglect: Train Incidents in Pakistan. (2019, October 31). The

Express Tribune. https://tribune.com.pk/story/2090607/1-timeline-neglect-

train-incidents-pakistan/?__cf_chl_jschl_tk__=67b51aba1d9fc784047d7a9762

c146c04cc3533c-1580365929-0-AcF2YBpMeT3YhfRtd7XjbqIHW9ee6w56y

PvQP0YDpiMvDN3xSyqEQ9ABBwqUJBVwBWTQvnuWfHkSJxYVI48aCK-

CgNuuVfT1N3OQ

Bibi, R., Chowdry, B. S., & Shah, R. A. (2018). PSO Based Localization of Multiple

Mobile Robots Emplying LEGO EV3. In 2018 International Conference on Computing,

Mathematics and Engineering Technologies (iCoMET), Sukkur, Pakistan. https://doi.

org/10.1109/ICOMET.2018.8346452

Carl, B., & Brook, S. (1985). United States patent. Geothermics, 14(4), 595–599. https://

doi.org/10.1016/0375-6505(85)90011-2

229

http://doi.org/10.17993/3ctecno.2020.specialissue5. 217-231

3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Abril 2020

Castillo, J. J., Cabrera, J. A., Guerra, A. J., & Simón, A. (2016). A Novel Electrohydraulic

Brake System with Tire-Road Friction Estimation and Continuous Brake Pressure

Control. IEEE Transactions on Industrial Electronics, 63(3), 1863–1875. https://doi.

org/10.1109/TIE.2015.2494041

Gajdar, T., Rudas, I., & Suda, Y. (1997). Neural Network based estimation of friction

coecient of wheel and rail. In Proceedings of IEEE International Conference on Intelligent

Engineering Systems, Proceedings, Budapest, Hungary, 315–318. https://doi.org/10.1109/

ines.1997.632437

Hussain, I. (2012). Multiple model based real time estimation of wheel-rail contact conditions. (PhD

thesis). University of Salford. http://usir.salford.ac.uk/id/eprint/38094/

Hussain, I., Mei, T. X., & Ritchings, R. T. (2013). Estimation of wheel-rail contact

conditions and adhesion using the multiple model approach. Vehicle System Dynamics,

51(1), 32–53. https://doi.org/10.1080/00423114.2012.708759

Kingma, D. P., & Ba, J. L. (2015). Adam: A method for stochastic optimization. In 3rd

International Conference on Learning Representations, ICLR 2015 - Conference Track Proceedings,

San Diego, 1–15. https://arxiv.org/abs/1412.6980

Li, N., Feng, X., & Wei, X. (2015). Optimized adhesion control of locomotive airbrake

based on GSA-RNN. In 7th International Conference on Intelligent Human-Machine Systems and

Cybernetics, Hangzhou, China, 2, 157–161. https://doi.org/10.1109/IHMSC.2015.222

Malvezzi, M., Pugi, L., Papini, S., Rindi, A., & Toni, P. (2013). Identication of a

wheel-rail adhesion coecient from experimental data during braking tests. Proceedings

of the Institution of Mechanical Engineers, Part F: Journal of Rail and Rapid Transit, 227(2),

128–139. https://doi.org/10.1177/0954409712458490

Olofsson, U. (2009). Adhesion and friction modication. In Wheel-Rail Interface Handbook.

Woodhead Publishing Limited. https://doi.org/10.1533/9781845696788.1.510

Polach, O. (2005). Creep forces in simulations of traction vehicles running on adhesion

limit. Wear, 258(7–8), 992–1000. https://doi.org/10.1016/j.wear.2004.03.046

230

http://doi.org/10.17993/3ctecno.2020.specialissue5. 217-231

3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Abril 2020

Shah, A. A., Chowdhry, B. S., Memon, T. D., & Kalwar, I. H. (2020). Real Time

Identication of Railway Track Surface Faults using Canny Edge Detector and 2D

Discrete Wavelet Transform. Annals of Emerging Technologies in Computing (AETiC), 4(2),

53–60. https://doi.org/10.33166/AETiC.2020.02.005

Shrestha, S., Wu, Q., & Spiryagin, M. (2019). Review of adhesion estimation approaches

for rail vehicles. International Journal of Rail Transportation, 7(2), 79–102. https://doi.org

/10.1080/23248378.2018.1513344

Spiryagin, M., Cole, C., Sun, Y. Q., McClanachan, M., Spiryagin, V., & McSweeney,

T. (2014). Design and simulation of rail vehicles. CRC Press. https://doi.org/10.1201/

b17029

Spiryagin, M., Wolfs, P. J., Cole, C., Spiryagin, V., Sun, Y. Q., & Mcsweeney, T.

(2016). Theoretical investigation of the eect of rail cleaning by wheels on locomotive

tractive eort. In CORE 2016, Maintaining the Momentum, Conference on Railway Excellence,

Melbourne, Victoria. https://trid.trb.org/view/1468498

Ward, C. P., Goodall, R. M., Dixon, R., & Charles, G. A. (2012). Adhesion estimation

at the wheel-rail interface using advanced model-based ltering. International Journal

Of Vehicle Mechanics and Mobility, 50(12), 1797–1816. https://doi.org/10.1080/00423

114.2012.707782

Zhang, C., Sun, J., He, J., & Liu, L. (2017). Online Estimation of the Adhesion Coecient

and Its Derivative Based on the Cascading SMC Observer. Journal of Sensors, 2017,

Article ID 8419295. https://doi.org/10.1155/2017/8419295

231

http://doi.org/10.17993/3ctecno.2020.specialissue5. 217-231

3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue Abril 2020