DEVELOPMENT OF BINARY CLASSIFIER FOR Railway TRACK SURFACE FAULTs DETECTION USING SOLAR POWERED VEHICLE
Railways serve as the common mode of traveling as it is fast and secure way of transportation. It is considered to be the backbone of any nation. But like all the good things in the World, railways do require maintenance. Due to lack of which Pakistan had marred over 100 of train accidents, most of
2025-06-28 16:32:04 - Adil Khan
DEVELOPMENT OF BINARY CLASSIFIER FOR Railway TRACK SURFACE FAULTs DETECTION USING SOLAR POWERED VEHICLE
Project Area of Specialization RoboticsProject SummaryRailways serve as the common mode of traveling as it is fast and secure way of transportation. It is considered to be the backbone of any nation. But like all the good things in the World, railways do require maintenance. Due to lack of which Pakistan had marred over 100 of train accidents, most of which were because of the train derailments. Train derailments usually occur owing to number of reasons. Majority of these accidents are due track surface based faults that are hard to identify using traditional visual inspection techniques. For it, more modern state of the art algorithms of artificial intelligence are required. Therefore, we propose development of a solar powered portable track recording vehicle that will apply logistic regression and image processing methodologies for the identification of surface based track faults. The vehicle will be fully electrical and will have processing power of Jetson Nano for the identification of the track fault. Moreover, it will be autonomous and wirelessly controlled using a cloud network.
The developed prototype will be programmed by Python which is open source programming language and will implement APIs like Tensorflow and OpenCV for fault classification method. The reason for using binary classification which is also known as logistic regression is due to its fast and precise decision making capabilities which as suggested by the former studies is more than 80% efficient.
Project ObjectivesThe project is aimed for the development of cost efficient and precise prototype for the inspection of the track faults that have the potential to cause train derailment related accidents. This will be a huge breakthrough for the Pakistan Railways (PR) because they are lacking behind drive by inspection methodologies. It will be a great entry to the existing PR’s assets because of the following objectives:
- To make a dataset of railway track faults in collaboration with PR.
- Development of a solar powered track recording vehicle (TRV).
- Controlling of the TRV.
- Development of a novel rail wheel design that is capable enough to identify track surface based faults like squats and turn out frogs.
- Interfacing and programming of camera modules using Jetson Nano.
- Capturing of a live video with the minimal latency.
- Development and deployment of Neural Network Algorithm (Binary Classifier) for the railway track surface faults detection by training from the collected dataset.
- Testing of the algorithm.
- 3D printing of special type of camera holders.
- Fulfillment of SDG 7, 9, 11, 13.
- To integrate this system with the existing technologies that PR possess.
- To achieve track surface based faults detection accuracy of more than 80%.
- To record the identified damages on a database.
The project implementation plan will be divided into 2 phases that are hardware and software. The hardware part will include:
- Development of the novel rail wheel design.
- Development of the novel camera holder design that will have the minimum ratio of light interference and reflection.
- Design and development of the solar powered TRV structure.
- Interfacing of the camera modules and Jetson nano with the TRV circuitry.
Whereas for the software part will include:
- Collection of the surface based track faults dataset in collaboration with the Pakistan Railways.
- Installing and learning of the tensorflow and Opencv APIs in the Jetson Nano.
- Development of the binary classifier.
- Testing of the algorithm.
- Measuring the accuracy of the binary classifier.
- Implementation of the developed algorithm on the live video feed.
To understand the working of the software and hardware for the classification of the surface based track faults is demonstrated in the block diagram as shown below:
The project benefits are mentioned as below:
- Most of the drive by inspection methodologies that the developed countries like China and Japan use involves the implementation of the inertial measurement units (IMUs). These IMUs are inefficient when it comes to the detection of the surface based track faults. These surface based track faults play a pivotal role in the rail corrugation which results in the train derailment accidents. So, with the development of this TRV, such surface based faults will be determined using a binary classifier.
- As in the traditional trains and the motorized Track Recording Vehicles (TRV), the rail wheels are huge in size therefore; it is hard for these TRVs to diagnose the surface based track faults. So in response to that our prototype proposes a unique rail wheel design that will precisely analyze the surface based track faults.
- We had collaboration with the Pakistan Railways (PR) in which we had discussion about their techniques and methodologies for the detection of the track damage. We were surprised to know that, they still use the traditional visual inspection techniques for identification of the track faults. So therefore, our proposed system will substitute their old techniques with an expected efficiency of over 80%.
- 100% of the equipment that the PR uses, is diesel operated. Which is huge drawback as it is not environmental friendly. To mitigate to this issue considering the ongoing global warming crises, our proposed system will be solar powered.
- Most of the AI based algorithms are not fast on portable processing units like Jetson Nano or Raspberry PI because they required processing power unlike binary classifier. As binary classifier works on simple logistic regression based approach, it is both fast and accurate as well as it requires less processing speed.
- This project has a potential to be part of the smart cities alternatives that the current government is taking in order to digitized Pakistan.
- Moreover, the developed prototype fulfills the sustainability development goals 7, 9, 11 and 13.
The final product will be divided into two parts:
Hardware:
In the hardware, the developed prototype will have:
- A broad gauge of 1.676m which is the standard for the rail vehicles.
- Electrical motor of 120rpm, 12V and 2A are proposed of high torque for driving the vehicle.
- In order to make the proposed TRV portable, a strong metal frame with less density will be designed which will have position for the placement of the rail wheels.
- The rail wheel will have width of 9cm as 7cm are of the rail wheels will be on the track whereas; the 2cm will be part of the flange.
- The motor connections will be made with relay module. This will later be connected with a 12V, 7AH battery.
- The battery will be charged using two of the 24W solar plates.
- Cameras will be fixed in a special cavity shaped frame that will have a square opening of 7cm covering the track on the top view. The frame will be 3D printed and will be mounted on the extreme end of the TRV.
Software:
Whereas, the software part will consist of:
- Installation of the basic API in the Jetson Nano Board.
- Testing of the camera module in the speed of 40km/h.
- Collecting the dataset of railway track faults as well as of healthy rail track.
- 500 Images of each healthy and faulty track will be distributed in the two folders which will later be read by the algorithm.
- Training of the algorithm.
- Testing of the algorithm.
- Measuring the accuracy of the algorithm.
- Making the algorithm to work on the live feed.
| Item Name | Type | No. of Units | Per Unit Cost (in Rs) | Total (in Rs) |
|---|---|---|---|---|
| Total in (Rs) | 80000 | |||
| Raspberry pi 3b+ | Equipment | 1 | 8500 | 8500 |
| Jetson Nano | Equipment | 1 | 20000 | 20000 |
| Camera module V2 | Equipment | 3 | 4500 | 13500 |
| Solar Panel | Equipment | 1 | 13000 | 13000 |
| Camera Holder | Equipment | 1 | 6000 | 6000 |
| LCD monitor 15 inch | Equipment | 1 | 5000 | 5000 |
| Memory card 32 gb | Equipment | 2 | 2000 | 4000 |
| Dc battery 12V | Miscellaneous | 1 | 1200 | 1200 |
| Keyboard +mouse combo | Miscellaneous | 1 | 1490 | 1490 |
| BreadBoard | Miscellaneous | 1 | 150 | 150 |
| HDMI | Miscellaneous | 1 | 180 | 180 |
| Memory card reader | Miscellaneous | 1 | 2400 | 2400 |
| Jumper wires | Miscellaneous | 78 | 10 | 780 |
| 1 channel relay timmer | Miscellaneous | 3 | 460 | 1380 |
| Small alarm siren electric speaker | Miscellaneous | 2 | 490 | 980 |
| VeroBoard | Miscellaneous | 2 | 140 | 280 |
| Glue gun | Miscellaneous | 1 | 300 | 300 |
| Hot melt glue rod transparent | Miscellaneous | 1 | 500 | 500 |
| Soldering wire | Miscellaneous | 2 | 180 | 360 |