PRE2016 3 Groep3: Difference between revisions

From Control Systems Technology Group
Jump to navigation Jump to search
No edit summary
Line 47: Line 47:
** Removing rust, using a laser
** Removing rust, using a laser


=== Side Objectives ===
==== Side Objectives ====
* Not have much wear of itself on the tracks
* Not have much wear of itself on the tracks
* Charge in front of the trains for optimal use
* Charge in front of the trains for optimal use

Revision as of 21:32, 3 April 2017

Group 3: Railway Maintenance Robots

  • 0902228 | Lindsey van der Aalst
  • 0938349 | Thomas Bastiaansen
  • 0948949 | Micha van den Herik
  • 0939318 | Tim van Leuveren
  • 0855969 | Job van der Velde
  • 0941574 | Floris van der Velden

Abstract

Introduction

Delays with the trains are a common complaint of most people, and the company Dutch Railways (‘Nederlandse Spoorwegen’) takes a lot of the blame. Some of the delays are caused by small objects positioned on the tracks or the condition of the railway tracks by itself. As a result, the train’s stopping distance increases by big margin. For this problem, a small robot is designed to minimize these problems. It will check the tracks for snow and leaves and use laser technology to free the tracks of these things. Not only that, it also detects the wear and will ultimately also maintain the condition of the tracks.

USE Aspects

User: NS

Could be an expert knowing all the in’s and out’s of the machine. But in general, it is the NS themselves. Efficiency is important for the User. The machine will need to have the ability to move at the same speed as standard NS trains and be able to remove obstacles, leaves and snow when needed, as well as detect any wear on tracks and railway switches. The machine should not conflict to much with the current situation. The Netherlands already has one of the most tight packed schedules in the world, with single delays often causing a chain of delays. The machine should work between (or outside) this schedule, else it will not have any benefit. The main purpose of the machine is to prevent delays and when it is not able to fit in the current schedule, it will only cause more delays. The machine should be easily operable. However, since not everyone has to use this machine, easily operable is not high on the priority list.

Society: Train travelers

Delays can occur due to many reasons, for example tracks that are in need of reparation, or bad weather conditions. Train travelers want to get from point A to B as quickly as possible, delays don't add to the train traveling experience. By the use of an automation machine, which can detect and remove obstacles that cause delays, train travelers can get from A to B more quickly. Time always translates to money, and for all three USE aspects money is on the priority list.

Enterprise: ProRail

As mentioned in User, efficiency is important for both the User and Enterprise. The enterprise is also held partly responsible for the delays and thus they would like to prevent them as much as possible. Also, the Enterprise want the machine to be most profitable as possible. The cost of the machine is then also desired to be as low as possible, while still doing its tasks. It should be reliable because failure can lead to even larger delays or train accidents, which in turn lead to larger costs. It has to be cheaper than the ways used currently or it should weigh up to the costs of the delays, else it is not profitable investing in it. Most of the arguments mentioned in User and Enterprise will overlap. In our case, we will be more focusing on the Entrepreneurial side of ProRail.

Our focus

Our main focus for this project is on the Enterprise, ProRail, and a little bit on the User, NS, since these two have quite some things in common. To us, the most important aspects are efficiency, reliability and costs of the machine and these aspects go best with the Enterprise. We would like the machine to be reliable and efficient, while keeping the costs as low as possible. Our focus lies here because there are already some systems that are able to do (part) of the jobs we want to achieve. But we would like to combine them and make them better. And for our product to be of any interest to the Enterprise, the costs must be low. At least lower than what is currently spent on these activities. But we will not only focus on production costs, also, the maybe a bit more transparent, indirect costs of the machine. Like for example when the machine is broken and thus non-operable, it will cost money. If the machine is slow, it will cost money. These 'costs' are taken into account under the aspects reliability and efficiency respectively.

Objectives

  • Functions at the same time as other trains are in use (same speed as the trains)
  • Detection wear of tracks
    • Rust
    • Cracks (ultrasonic?)
    • Dimensions & shape
  • Maintenance of tracks;
    • Removing snow, using a laser
    • Removing leaves, using a laser and compressed air/shovel
    • Removing rust, using a laser

Side Objectives

  • Not have much wear of itself on the tracks
  • Charge in front of the trains for optimal use
  • Modular ‘carts’ -> different equipment for different tasks
  • Additional detection: Condition of welds, fasteners, sleepers and ballast, temperature of railway
  • Possible detection of railway track geometry using gyroscope. (heavy maintenance required for readjusting railway track geometry)

Extensions

  • Good for the climate and environment.

Approach

  • The focus lies on the User and the Enterprise, which are the NS and ProRail, respectively. Especially the Enterprise aspects are important for this system. This means that the system needs to be efficient, sustainable and that the production costs need to be as low as possible while still remaining quality.
  • Research has to be carried out about state-of-the-art technology. For example, one of the recent developments in railway technology is a laser which can remove leaves from the railway tracks [1]. We will also implement this technique into our system. Also, currently a monitor has been developed to check the condition of the tracks [2]. This technique is used to measure the cross section without contact. This technique could possibly be used for our system.
  • A literature study will make clear if our idea is really innovative and unique. We will also do research about how the system needs to be designed, what the most efficient form is, how it needs to be loaded, etc.

Planning

https://drive.google.com/open?id=0B8ju55_U5nZ4LWRLalRCbkMxVlU

The above link guides you to our Gantt chart, which has been made with the help of the program "Microsoft Project". We have divided our plan into research, prototype and requirements, and deliverables. First, we’ve described the milestones and the date on which they have to be accomplished. After that, we’ve split up these milestones into different tasks and allocated people to these different tasks, as can be seen in our Gantt chart.

Our planning can also be seen below, where the milestones are boldfaced:

Gantt chart (Job + Lindsey)

Planning & task division 15-02

Explanation focus enterprise (Floris)

Current expenses maintenance & disturbances (Floris)

Define exact problems & suggested solutions (Micha)

USE aspects & problem definition 20-02

Technology competitors in maintenance & problem detection (Tim)

State of the art technology for robot design (Job & Thomas)

Own approach railway switch maintenance (Tim)

State of the art 20-02

Interview NS 22-02 (Floris & Job)

Define prototype & specifications 20-02

Define requirements & specifications 28-02

Propulsion system (Floris & Tim)

Sensors (Job & Micha)

Maintenance equipment (Job & Micha)

Power supply & charging (Floris & Tim)

Railway coupling (Thomas & Lindsey)

Communication to head station (Thomas & Lindsey)

Interface (Thomas & Lindsey)

3D model (Floris & Job)

Localisation of self and trains and other robots (Micha & Lindsey)

Weather forecast integration (Tim & Thomas)

Sensor feedback integration (Tim & Thomas)

Hardware & software 14-03

Summary of expenses (Floris & Lindsey)

Budget 17-03

Functionality Robot (Job & Tim)

Use of robot (Micha & Thomas)

Future extensions/upgrades (Micha & Thomas)

Functionality, use and future adaptation 27-03

3D Model (Floris & Job)

Reflection USE aspects (Thomas & Lindsey)

Deliverables 30-03

Final presentation/demonstration 03-04

Peer review

State of the Art

Disturbances

There are many different causes for the disturbances in the Dutch railway system. From january 2011 until february 2017, 15110 disturbances were reported. Shown in Graph 1, the most common disturbances are accordingly [3] :

  1. Faulty train (2279 disturbances, 15,1%)
  2. Signal interference (1640 disturbances, 10,9%)
  3. Railway switch failure (1593 disturbances, 10,5%)
  4. Collision with a person (1498 disturbances, 9,9%)
  5. Repair work (691 disturbances, 4,6%)
  6. Previous disturbance (529 disturbances, 3,5%)
  7. Signal and handle failure (450 disturbances, 3%)
  8. Signal and railway switch failure (390 disturbances, 2,6%)
  9. Power outage (388 disturbances, 2,6%)
  10. Level crossing failure (320 disturbances, 2,1%)
  11. Miscellaneous (5346 disturbances, 35.3%)

The miscellaneous disturbances consist of both the weather and external factors, which consist of rare disturbances such as theft or vandalism of the copper in the railway tracks, people or animals close to the railway tracks or roadside fires [4] . Concerning the weather, it can have a big impact on the train schedule as the different seasons in the Netherlands all influence the schedule.

On the one hand, there is the turbulent weather in the fall and winter. Leaves are a well-known problem in this time of year. But actually, the main problem is not the leaves, but actually the smoothness of the railway track. The leaves and rain together results in a mush, which makes the tracks more slippery, which on its turn the grip of the train wheels decreases. As a result, the circular shape of the wheels changes and they need to be repaired. In addition, the stopping distance increases exponentially, which has to be accounted for [5] . Snow and ice, next to the slipperiness of the railway track, also cause the railway switches to freeze or get blocked by the snow and ice.

Not only the cold, also hot temperatures can have impact on the railway tracks. Due to the increase in temperature, the steel stretches which causes the tracks to bend. The railways are then unusable to be driven over by a train.

With the railway maintenance robots, the disturbances concerning the railway switches, the weather and a part of the repair work are planned to be solved. These three different disturbances cover a notable part of the total disturbances in the Dutch railway system. Assuming a high efficiency, the railway maintenance robots could potentially prevent a great part of these disturbances, resulting in thousands less disturbances over the researched period.

Detection

Detection is a great part of the railway maintenance robots. Therefore, the railway maintenance robots need to be equipped with numerous sensors in order to determine things of and on the railway tracks. These include the leaves, rust and snow, the profile of the tracks, and the temperature.

Detection and removal of leaves, rust and snow

A visual representation of the laser removal system

One of the aspects of the railway robot is to perform maintenance on the tracks. The focus herein lies with removing rust, snow and leaves. All of these tasks are possible with state of the art lasers. According to Oliver Smith [6] leaves on railway tracks alone are cause for 5800 hours of delay per year for the British National Rail. A special microwave ray has already been found to be effective in removing wet leaves from the tracks, but further research on using lasers for this purpose is still being conducted and is estimated to be even better.

In the car industry a handheld 1000-watt rust removal laser is already available [7] . This laser is able to remove rust, dirt, coatings and paint in mere seconds. The laser works by adding its energy to the dirt/rust layer, which evaporates, while the base material reflects and spreads most of this energy, thus remaining unaffected [8] [9]. The laser can be optimised and tuned for specific base surfaces, to further ensure its safety.

Ice and snow can also be removed by adding laser energy. The patent of Roger and Rose Vega describes an ice removal system for airplanes [10] . The laser vaporizes the ice by moving slowly over the covered surface, thereby re-exposing it.

These three different laser technologies could be combined for all three maintenance purposes since the basic principle for removing the unwanted substance is the same, after which it could be mounted on a railway robot.

Profile detection

A problem with the railway can also be that is shifts in the cobblestones on which the track is placed. The track can be shifted into the cobblestones, resulting in a height difference between the two tracks. Currently, to detect whether or not the railway has shifted, is detected by a railway constructors themselves when checking the normal maintenance planning. This can be done quicker and more efficiently than what the current plan of action. With the use of a gyroscope inside of the monitor machine, the angle of the train can be measured. Also with the help of a device which measures velocity, the position of the vehicle along the track can be determined with an analog to digital converter. Moreover, a whole digital implementation can be made of the track (and if done accordingly, compared to what the original geometry of the tracks has to be. This process can save time, since it is all done digitally instead with the use of humans on only small portions of the track. The vehicle which will detect the geometry can move at higher speeds, and process the data immediately. Comparing the processed date digitally will also make the comparison more accurate than what humans can make of certain parts of railway tracks [11]. An improvement of what the can be digitally implemented of the geometry of the total railway, the same type of measurement can be used to measure the wear and profile of the railway. The old method involved physical contact with the railway and was only able to measure the geometry and undulation of the railway (should a certain threshold be achieved, maintenance workers will have a closer look on the railway). With the new laser method these same parameters, as well as more important ones such as wear and profile of the railway, can be measured. The new innovative approach used, is based on image analysis and processing to reconstruct the whole track profile digitally (just like the geometry measurement). The railways reflects light back into cameras which can detect lasers and can internally process this data. The data will then be converted to a 3D projected image of the track. Using this technique, no extra wear will be made to the railway while measuring the wear. Moreover, the measurements can be done more quickly, since all the data is processed while the vehicle is moving over the track. Using a high-performance architecture, a big amount of information can be processed in a smart and fast method, since it is not possible to constantly store all the images and process them offline (for example with the use of pipelining and parallelism). Also, the use of high-level image analysis avoids the need for continuous and accurate alignment of the monitoring system with the track. The image processing method can be designed in such a way, that it can self-align itself (for example with the combination of the gyroscope as mentioned above) [12].

An example of where profile detection is already used, is the RailMonitor. To detect the wear of the railways the railmonitor will be implemented on the railway maintenance robot. The railmonitor is a mobile measuring system which can measure the cross direction profile of the railway tracks [13]. This system uses a laser for the measurements and stores these measurements internally. Some special software will then compare the measurements with the references for cross direction profile and draw conclusions accordingly. The results are also shown immediately on a screen on the device itself. For the implementation of this system on the railway maintenance robot some features will be improved. Such as the communication of the measurement result immediately to the headquarters instead of storing it in the device itself. Also the screen will be unneeded.

Temperature

This technique is especially useful during wintertime because these switches causes many disturbances during wintertime. The switches can freeze and can become clogged. To prevent the railway switches to freeze and become clogged there is a heating system built into the switches which can heat the switch when the temperatures drop below zero degrees Celsius. However this system does not always work and this problem is hard to detect in time. Therefore the railway maintenance robots need to be equipped with an infrared camera, in order to detect the temperature difference between the switches and the straight parts of the railway tracks.

Drone Tests [14] [15]

Railway switches as seen through a thermal camera

In order to check the heating of the railway switches, ProRail currently uses unmanned helicopters equipped with an infrared camera. This camera can detect whether the heating system is operating or not.

In this picture it is seen that the heating system is working properly. The infrared images provide ProRail with information over the switches and ProRail can act accordingly. The railway maintenance robot will use this technique for which the drones are used now. In order to detect the temperature difference, the infrared camera needs to be placed on a certain height to take proper pictures.

Felix

Felix is the first mobile robot for inspecting railway switches [16]. This robot is equipped with profilometers which create a 3D reconstruction of the inspected switch. This is a useful robot to increase the reliability of the railway switches but can only be used for inspecting these switches. The railway maintenance robot can do this either and can be deployed for other tasks such as cleaning the railway. It can also inspect the railway itself along with the railway switches.


ProRail’s ways to tackle autumn

To add something to the maintenance work of ProRail, we first need to find out what ProRail already do against the bad autumn weather. A big problems is the leaves on the tracks. These leaves get squished when a train rides over them and create a mousse which causes the trains to lose grip. This causes the trains to slide over the rails for up to 800 extra meters [17] . ProRail tackles this by cutting the trees around the tracks short and thus preventing leaves from falling onto the tracks.

Also, the use their own developed gel, Sandite [18] . Sandite is a mixture of sand, metal particles and potato starch. This gel roughens the rails which makes it unable for new leaves to attach to the rails and it also breaks down the already present mousse of crushed leaves. Another way they try to prevent delays is to change the train schedule by reducing the amount of trains on the tracks. This way, it is possible for trains to travel at a slower speed and take more time accelerating and decelerating.

ProRail also did tests with both laser systems and water spraying systems to see whether these could help removed leaves from the tracks. The water spraying system included a water pipe mounted into the rail beams, which dropped small drops of water into the tracks every so often. The results of the test were successful; the tracks turned out to be less slippery [19] . This technique still has to be implemented in the Netherlands.

The laser technique was developed in collaboration with the Technical University of Delft. The lasers, mounted underneath the train, heat up the tracks. This causes leaves and other dirt not to attach to the tracks [20] . The power in the laser beam is carefully monitored, as this is important for the protection of the rails. They say the power is chosen with such care that they can write on the head of a match without it igniting [21] . Also the wavelength has been carefully selected to be 1064 nanometres. This turned out to be the perfect wavelength for the job, as it only absorbs leaves and other organic materials, leaving the metal underneath unaffected. [22]

"The technique works with a neodymium yttrium aluminium garnet of Nd:YAG laser, which produces 2 kilowatts of infrared radiation in 25.000 pulses per second. The pulses instantly heat the leaves to 5.000 degrees making it simply pop off the rails. A test by Network Rail in the United Kingdom revealed it was successful up until 65 km/hour.” [23] .

The technique was such a success that the machine continued under the name Laser Railhead Cleaner in collaboration with the NS, ProRail and Strukton Rail.

Current focus of ProRail

To find out what will be important properties for our robot we have to find out which aspects ProRail prioritizes. Part of this can be found already and for the other part we will try to contact ProRail themselves to hear from them personally what they focus on.

As they mention themselves in their year overview for 2015, they care a lot about efficiency and low costs, both in management and maintenance. For each decision, they take social, environmental, and economical aspects into account. And, as they state themselves, they try to be as transparent as possible and incorporate views from all their stakeholders [24].

Also, because they are under supervision by the Dutch government, the Ministry of Infrastructure and Environment monitors the focus of ProRail, sending them a yearly ‘priority letter’. This letter gives ProRail the main outline the government wants them to pay extra attention to.

Sensors

Localisation & Communication

Localisation

There is already a system that reports the position of the trains to a Radio Block Centre (RBC). This system is the European Train Control System (ETCS), which is part of the European Rail Traffic Management System (ERTMS). Trains in Europe are carried out with GSM-R (Global System for Mobile communications – Railways), which has, amongst other functions, the ability to report the position of the trains, for example after a specific time interval or after passing a specific location, but also on request. On top of reporting the position, it also reports things like the estimated speed, the direction of train movement and train integrity information to the RBC [25].

Thus, our robot can be integrated into this system by implementing GSM-R on it. This systems makes sure that the robot doesn’t collide with other trains or robots and it also keeps track of on which parts of the railways our robot has already been [26] .

Weather

The above specified system could be extended by a weather forecast system. This means that information about the weather will be send to the robot via the GSM-R, in the same way as data about the position of other trains and robots will be send to the robot. By involving a weather forecast system into the ETCS, the robot knows where to scan for leaves and snow and, if present, remove the leaves and the snow.

Contact with the NS

Floris and Job attended a lunchlecture given by the NS on Wednesday 22nd of February. During this lecture mostly information about working with the NS was given, since this was the goal of the lecture. But at the end, we asked one of the employees for more information about the laser removal system. We knew the NS had done tests with this system previously, and we wanted to learn from their results. The employee present did not participate in this test and we got his information to get in contact with him. He was willing to ask his colleagues and thus we received an email address from someone who knew more about the results of the laser project.

Week 4

Requirements

The costs (including both money and time) of the making and operating of the robot must be lower than the costs for the current maintenance operations and the costs for delays (Enterprise).


Communication

The robot must be integrated to the GSM-R (Global System for Mobile communications – Railways) network (User).

Information about the weather must be acquired with the help of the Internet of Things and sent to the robot through the GSM-R (User).

The robot must be able to react on the information it acquires through the GSM-R (User).

The robot must be able to react on the information it acquires through the sensors (User).

Laser Maintenance

The laser maintenance system must be able to operate at 160 km/h (User, Society, Enterprise).

The laser maintenance system must be able to remove snow and leaves on a certain railway track by passing only once (User, Enterprise).

The laser maintenance system must not present danger to bystanders (Society).

Sensors

The sensors have to be able to operate at full speed (160 km/h maximum) (User, Society, Enterprise).

The sensors have to process the data such that the laser can remove obstacles in time (User, Society, Enterprise).

The sensors do not have to be sensitive to weather (Enterprise).

The sensors have to not consume more power than delivered (Enterprise).

Power supply and charging

The robot must work on supplied voltage by the catenary (Enterprise).

The power consumption must be lower than the power received from the catenary (Enterprise).

The robot must have small battery to overcome gaps in the catenary (Enterprise).

The diesel-generator cart must produce enough power to keep the robot running (Enterprise).

Communication

Current train network:


The robot must be integrated with the GSM-R network.

See Localisation (week 3).

GSM-R arranges, amongst other things, data communication between the train and the Radio Block Centre (RBC). When the train or the robot passes a balise (an electronic beacon or transponder placed between the rails of a railway), information about the train or robot will be send to the RBC. This information includes the speed and location of the train or robot. The RBC can also send information to the train or robot, for example permission to enter the next track [27].


Weather forecast integration and sensor feedback integration:


Information about the weather must be acquired with the help of the Internet of Things and sent to the robot through the GSM-R.

See Weather (week 3).

Information about the weather will be sent to the robot via the GSM-R, in the same way as data about the position of other trains and robots will be send to the robot. However, the information about the weather must be taken from the internet, which means that the Internet of Things must be involved in order to gain information about the weather.

The robot must be able to react on the information it acquires through the GSM-R.

The robot receives information about the weather and information about other trains and robots through the GSM-R.

-Weather

See Weather (week 3).

By involving a weather forecast system into the GSM-R, the robot knows where to scan for wet leaves and snow/ice and, if present, remove the wet leaves and the snow/ice with laser technology.

-Information about other trains and robots

Information about other robots can be used to optimize the efficiency of the robot. For example, if a robot receives the information that another robot has just scanned the tracks on which the robot is driving for rust, it would be inefficient to scan these tracks again for rust.

Since the robots will be coupled to trains during the day, the robots themselves don’t have to react on delays or malfunctions about other trains in order to prevent collisions. Namely, the trains to which the robots are coupled will already receive this information and their schedule will be adjusted to this. It follows then that the schedule of the robots will also be adjusted, since they are coupled to the trains.

During the night, the robots will work autonomously and they will not be coupled to trains, which means that they have to receive information about the location and speed of other robots to prevent collision and react on this.


The robot must be able to react on the information it acquires through the sensors.

It is already described above how the robot will react on the information about wet leaves and snow/ice. This means there are three types of information left that the robot can receive through sensors, namely information about rust, wear or flaws in geometry. If there is rust detected, laser technology will be activated in order to remove this rust. If wear or flaws in geometry are detected, the robot must report this to the RBC via the GSM-R. The RBC can then decide to take action to solve this.

Autonomous Behaviour

The most important aspect of a robot is its ability to perform tasks autonomously. For the railway maintenance robot, this behaviour manifests itself in two main tasks: the day shift and the night shift.

Day shift

During the day shift the robot must be able to couple and uncouple itself to and from trains in order to scan the tracks or perform maintenance at high speed. It should also be able to determine the track sections that are to be scanned and in what order, according to a schedule. The removal of rust, snow or leaves should be done based on known information from previous scans, a signal from the head station that maintenance is required and where, or because of direct input from the sensors on the robot itself.

Night shift

The night shift is meant for problem detection and maintenance that cannot be performed at the velocity of a train. Again, the robot must be able to determine by itself which sections to scan or perform maintenance upon. Apart from that, it should be able to couple itself to the diesel generator if no catenary is available as a power supply. Finally, it should place itself in logical location to couple to a train in time for the day shift.

Sensors

The purpose of this section is to do research on whether it is possible to detect different kind of obstacles (wet leaves, snow and ice etc.) and to tell each of those obstacles apart (if necessary). Next whether it is possible to detect certain obstacles, rust/wear and the contour of the railway at high velocities. At last how much power will the sensors consume?


Types of sensors

Before we can look at how each sensor can operate, each sensor has to be defined beforehand. This subsection will give a more detailed approach to how the sensors will operate and what kind of different sensors have to be implemented into the vehicle.


Obstacle detection

The method the sensor uses to detect any obstacle on the rails is mounted in front of the vehicle with a predetermined field of view of the track [28]. The sensor produces at least one signal representative of a section of the track ahead of the vehicle. Next, an obstacle detection device attached to the sensor will process the at last one signal it received from the sensor. When the sensor detects a discontinuity in the track, it will send a signal to the next process of the obstacle detection. This next device in the process will determine whether or not the obstacle on the track is able to be removed by the integrated laser in the vehicle. More can be read about the integrated laser in the section [Insert section here].

Wear/Contour detection

For the wear and contour detection/measurement, other sensors have to be used, since the use of the sensor mounted to the front of the vehicle is not able to detect the wear of the railway in any way. A smaller type sensor in the form of a CCD (Charge-coupled device) camera coupled with a laser pointed at the rails will be used to get a measurement of the contour and wear of the rails it is pointed at. CCD cameras have become a major technology for digital imaging in recent years. The camera will acquire a local digital image of the track underneath the railway of a resolution of 512x512 pixels. To achieve optimum observation, two of these CCD cameras will have to be placed in the vehicle. The CCD camera will observe a laser plane, reflected by the railway. The sensor will be equipped with software to determine whether or not the railway is damaged and needs any repairs. The sensor is also able to look at the surface of the railway (width and bulging of the railway). The same software will also be able to determine whether or not the railway will need any more maintenance. The signals this sensor picks up of the wear and surface of the railway can be stored and send to a main storage for further processing or simply storing the data.

Geometry detection

To measure the geometry of the rails, the use of a gyroscope is ideal to measure whether the rails has moved itself into the stones. With the further use of a velocity measurement device (a wheel tachometer for), the position along the track is determined. The combined combination of the position and angle of the train along the track can give us a virtual ‘picture’ of the laying of the track on the stones it has been build on with an analogue to digital converter. Moreover, a whole digital implementation can be made of the track and (if done accordingly) compared to what the original geometry of the tracks has to be.


Vehicle speed

In this subsection, it is discussed if it is possible for each of the sensors to be able to operate at the same speed as trains in the Netherlands operate at (maximum of 160 km/h) as stated in the requirements of the sensors. Again, each of the different sensors will be discussed separately.

Obstacle detection

For the first discussed sensor above, to detect any obstacles on the track in front of the robot with a camera mounted on the front of the vehicle, the author’s of the paper claim that it should give real time feedback back to the driver of the train. From this claim, one can make an educated guess that the sensor has to be able to work at full speed. If this is not the case, the person driving the train would not be able to receive any feedback from the vehicle moving in front of the train.

Wear/contour detection

For the detection of the wear and contour of the railway with the use of a laser and CCD-camera, testing of the sensor as discussed in the paper was only done on the Milan underground for over a year. Note that this is not on any outdoor rail track as can be seen from the map [29]. This would factor out any weather or other outside variables to the tests. Nevertheless, the author states that the simulations that have been performed on the sensor to mimic the lightning of daylight that are typical for a moving carriage, show still attractive results and show the efficiency and the effectiveness of the proposed approach. This state-of-the-art technology still needs more thorough testing to verify the results of the simulations, but in theory the sensors should be able to work on full-moving carriage and passenger trains, with any weather, light and oscillating train tracks variables.

Geometry detection

For the last sensor to measure the geometry of the railway, only the sensor that has to keep track of the position of the train has to be taken into account. The gyroscope simply gives certain values of the geometry of the track and the velocity measurement device has to take samples of the gyroscope at a high enough rate to create samples worth comparing to the original geometry of the railway. The author of the patent implemented a wheel tachometer and accelerometer that should be able to measure the velocity and take sample of the gyroscope at high velocities (the exact values are not given, but it is stated that velocities of trains can be easily obtained).


Power consumption

A requirement we also give to the sensors is that they have to not consume more power than delivered for obvious reasons. Again, each sensor will be separately discussed to what amount of power each sensor may consume. Note that ‘may’ is stated by the amount of power consumed. This is due to the fact that the real amount of power used is not stated in any of the papers nor in the patents. In order to get some values of use, an estimation is done on each of the different components used in the sensors. This estimation can vary with respect to the power consumed by the sensors discussed in the papers and patents.

Power supply and charging

To be cost efficient, power consumption must be critically designed.


A bridge near Meppel (The Netherlands) with a gap in the catenary, the trains photograph is still raised.[30]

The robot must work on supplied voltage by the catenary

This is the way power is received so it is important the robot can work with these currents. Too high currents cause the wires inside the components to burn, destroying the robots functions. Too low currents cause the components to not function properly. An easy fix if this either is too high would be to introduce an internal transformer to lower the current to the desired strength.

The power consumption must be lower than the power received from the catenary

This means the robot may never use more power than what can be pulled from the existing overhead power network. If the robot uses too much power, other users of the network may experience difficulties. So the power consumption must be carefully monitored to prevent the robot or other users from running low on power.

The robot must have small battery to overcome gaps in the catenary

As there are sometimes gaps in the catenary, a battery must be included to overcome these small gaps, as the robot would otherwise stop working. A gap in the vatenary can be seen on the right, where a bridge near Meppel (The Netherlands) is depicted. As can be seen the pantograph is raised. This battery should have a capacity of at least 5 minutes work and can be charged back up when reconnected to the catenary. For tracks where the battery will not suffice, a diesel-generator cart will be added to the robot to power it.

The diesel-generator cart must produce enough power to keep the robot running

Quite similar to the catenary requirement mentioned earlier. The extra diesel-generator cart must also be able to produce enough power to keep the robot running.

Design

Drawings

Sideview of T.R.A.M.S. and the diesel generator
3D view of T.R.A.M.S.

Calculation

For determining the maximum length of the robot the position of the train driver is needed.

Sideview of T.R.A.M.S. and the diesel generator

With this picture the viewing angle of the train driver can be determined. With this angle the maximum length of the robot can be determined.

Sideview of T.R.A.M.S. and the diesel generator

So the maximum length of the robot is about 13 meters. A good length for the robot to operate well and for all the equipment to be installed is 5.5 meters.

Week 5

Budget

Current expenses of ProRail

Before we can make a budget of the robot, we will estimate the current expenses of ProRail on the trouble caused by weather conditions (wet leaves and snow) and on detection and maintenance of rust, geometry and wear of the railway tracks.

In 2015, ProRail has spent €1.098 million on maintenance and management of the tracks (see Current Expenses in Week 2). Of course, not all of this money is used for the trouble caused by weather conditions and for the detection and maintenance of rust, geometry and wear of the tracks.

In week 2, we investigated that 4,6% of the disturbances will be due to repair work. Furthermore, 35.3% of the disturbances are miscellaneous. The disturbances due to the weather fall into this category. A rough estimate is that 3% of the disturbances will be due to the weather. We have contacted ProRail about this in order to get a specific number, but we haven’t received an answer yet. We hope to receive one as quick as possible, but until then we will work with the estimated percentages.

As stated before, 3% of the disturbances will be due to the weather according to our estimations. Therefore, we will estimate that ProRail spends 3% of the total budget on finding and implementing solutions for problems caused by the weather. This means that ProRail can spend almost €33 million on problems with leaves and snow. In England, around €17 million (15 million GBP) per year is spend on the problems caused by leaves [31]. Since this is about half of the budget for problems with leaves and snow, this finding supports our estimation.

So, in total, 7,6% of the disturbances are caused by problems that the robot could solve. Since the total budget for maintenance and management of the tracks is €1.098 million, we will estimate that ProRail reserves 7,6% of €1.098 million for the trouble caused by wet leaves and snow and for the detection and maintenance of rust, geometry and wear of the railway tracks, thus the budget of ProRail for this will be around €83 million per year. However, our robot will not be able to fix all maintenance problems. Therefore, we have chosen to split this budget in two and estimate €41,5 million per year as a budget for ProRail.

Our goal is to keep the costs of our robot lower than this amount.


Costs of the robot

In total, ProRail has 7.021 km railway tracks [32]. Most of the trains drive between 5 A.M. and 1 A.M., which means they drive for 20 hours during the day. The average speed of a train is 140 km/h, which means that, theoretically, one robot would need 7021/140=50 hours to check the railway tracks once. From this we can conclude that theoretically 3 robots would be necessary to check all the railway tracks during the day. Of course, this is the ideal case in which each robot drives on each trajectory exactly once. Since there are three main stations in the Netherlands, we could indeed take a robot per main station. However, our calculations are based on the ideal case. Therefore, we estimate that two robots per main station would be enough. This means that there are 6 robots necessary in total.

According to Adapt Laser Systems, the selling price of the CL 1000 laser system (1000W), which is the one we need, starts at €446.761,- (480.000 USD) per laser [33]. There are two lasers necessary for each robot, which means that the costs of the lasers will be €893.522,- for each robot.

A camera which resembles the one that we are planning to use, is the Highspeed camera PCE-TC 225. The price of this camera is €4.495,- netto and €5.438,- bruto [34].

Alternative Maintenance Systems

There are several alternatives to laser cleaning the railway tracks, such as high pressure water jets and industrial power brushes. The water jet uses up to 2500 bar to remove snow, leaves and rust from metal objects. However, if compared to the laser removal system, it requires a steady supply of water, uses an equal or larger amount of electricity, takes up the space of half a shipping container and is comparable in price [35]. An industrial power brush system would have lower acquisition costs, but would require higher maintenance costs because of the brushes that need to be replaced due to wear. The operating speed of the brushes would also not reach 160 km/h [36]. On top of this, the laser maintenance system does not damage the railway tracks, in contrast with the aforementioned alternatives.


Laser Specifications

The main specifications of the laser maintenance system are [37][38]:

  • 1000 W
  • 3x 480 V
  • 250000 pulses/second
  • 800 kW pulse

The current operating speed of a laser leave removal system, as tested by the NS, lies between 65 and 80 km/h [39]


Batteries vs External power source

First, the use of internal power storage (batteries) was taken into consideration, so the robot could completely function on its own without the need of external energy sources. It was calculated how much energy needed to be stored based on our requirements. Following our requirements, the robot needed to work at least 8 hours. This meant maintaining a speed of 80 km/h and using the sensors and lasers for 8 hours.

Before calculating the total energy consumption, the available batteries were examined. One of the most high end batteries currently available is the Tesla Powerwall with a capacity of 14kWh for a price of €6.300 [40].

First, we determined what the costs of maintaining a speed of 160 km/h are. These costs are largely caused by resistances. The power consumption can therefore be estimated by calculating the total resistance of the robot. The main resistance the robot faces is air resistance, with also small amount of rolling resistance and gradient resistance. For our case, we can assume gradient resistance to be of such little influence that it can be regarded as zero. This because the Netherlands is mostly flat and the height difference the robot will most often face is a maximum of around 5 meters.

Air resistance can be calculated using the following formula:

FAir.png

Here, rho is the density of air at sea level, v is the current speed, Cd is the drag coefficient and A is the front surface area. The density of air at 20 degrees Celsius and standard pressure is 1.2041 kg/m^3. Cd is often determined by experiments and examples can be seen below. For our robot, we will choose the drag coefficient of a half-sphere - 0.42 - since it will most likely have such a shape. The frontal area is equal to a guess for now, we took a look at a list of frontal areas of cars, and chose the car with the largest frontal area; the Dodge Ram 1500 QC. This car has a frontal area of 35.1 sqft [41].

Examples of drag coefficients [42]


Rolling resistance can be calculated using the formula:

FRolling.png

With Cr the rolling resistance coefficient and N the normal force. The normal force is equal to the mass times the gravitational constant 9.81 and the rolling resistance coefficient for train wheels on rails is equal to 0.001 [43] [44] . No matter what the mass is, the rolling force is 100 times smaller than the air resistance, so the rolling resistance can therefore be neglected or if it is decided to include it, the mass has no big influence on the total resistance.

The total resistance is equal to the sum of all resistances and can be calculated using the following formula:

FTotal.png

To convert the total resistance to power the formula is used:

FPower.png

In which v is the current speed, equal to 80 km/h. These values give an outcome of 89.8 kWh just to overcome resistances. This means even more power is needed to accelerate or to use the equipment on board.

From these numbers it was concluded that it was impossible to have our robot solely powered by internal power storage and thus an external power source was needed.

Memory storage of the sensors

An important aspect of the sensors is the memory space required to store the data the sensors pick up. Since the railway in the Netherlands consists of quite some distance, approximately 7021 km of railway, a lot of data is required to store the benchmark for the sensors. This benchmark is needed for the data picked up by the sensors to have a meaning. Without this benchmark, the data of the sensors could not be interpreted correctly. In this section, the possible total amount of data needed for this benchmark is discussed as well as some possible methods to tackle the problem of memory storage on the vehicle itself.


Obstacle detection

Regarding the obstacle detection, the chosen ICCD camera[45] used by the sensor is capable of reaching 100 billion frames per second (fps). This amount of fps is only possible to achieve at an extremely low resolution. This camera has also listed two other specifications on its datasheet. Namely:

  • Standard resolution: 782 x 582 pixels
  • High resolution: 1360 x 1024 pixels

In order to choose the right resolution, we have to have a look at the required fps the ICCD camera has to achieve. We assumed that a leaf is approximately 10 centimetres big. A train drives at maximum speed 160 km/h, that is 44.4 m/s, or 4444 cm/s. If we want to have a good sampling rate, we want to sample at least every 10 cm of the railway track. This would result in a sampling rate of 4444/10 = 444 frames per second (fps). For easier calculations and since the sampling rate has to be of at leas 444 fps, 500 fps will be used in the following memory calculations.

The datasheet states that the camera has a dynamic range of 12 bit (optionally 14, but this is when one decides to use the UV-splitting optic-mechanic on the camera. This will greatly decrease the system sensitivity and is unwanted for our use). To calculate the amount of storage needed for one single frame, we can use the following formula:

Memory space = Amount of pixels * Dynamic range [46]

Using this formula, the amount of memory per frame can be calculated. The memory space required for high-resolution will be approximately (12*1360*1024)/8 = 3.6MB per frame. Using the standard settings on the device, a memory usage of (12*782*582)/8 = 0.6MB per frame is obtained. Should the device run on approximately 500 fps to detect obstacles on the railway, 500*3.6MB = 1.8GB (for high resolution) or 500*0.6MB = 300MB (for standard resolution) of data is needed for each single second of the camera to store its data. We stated that our robots would be able to ride for 20 hours consecutively at a time or 20*3600 = 72000 seconds. Thus the maximum amount of memory space needed to store all of the train tracks in the Netherlands 1.8GB*72000 = 130 TB is needed for the high-resolution frames and 0.3GB*72000 = 21.5 TB is needed for the standard resolution. For 2 cameras in use, it can then be concluded that the maximum amount of storage needed is 260 TB for high resolution and 43 TB for low resolution for the obstacle detection sensor.


Wear and contour detection

Regarding the contour and wear detection of the railway, the usage of CCD camera’s will be used and create internally a picture of 512x512 resolution picture of the contour. The paper states that the system uses up about 1.7 GB/s for 2 cameras, 4 cameras in total will be used for the measurements of both sides of the railway, resulting in a total usage of 3.4 GB/s. using the same calculation as for the previously used sensor, 72000 * 3.4 GB/s = 244 TB of maximum storage regarding the contour and wear detection of the measurement vehicle. A small note of this sensor as well is that the sampling rate of the contour and wear detection is at a lower rate than the obstacle detection, at 200 samples per second, because of the processing power of this sensor is lower than the obstacle detection. However, the lower sampling speed does not provide any more errors and can still be used with high accuracy.


Geometry detection

At last, the geometry detection will be discussed regarding the memory storage.which is the least memory heavy sensor of all of the used sensor. This is due to the fact that the gyroscope does not have to use pictures to obtain the right values, only numbers. Neither does the sampling speed have to be as high as the obstacle detection, nor as high of the contour and wear detection. A good sampling rate would be of taking every half a meter a sample of the gyroscope. This would result in taking approximately 100 samples per second and a total of 7.200.000 samples. The range of the height difference in railway tracks in the Netherlands is approximately 75 meters (as read from [47]). If we want to have a resolution of 10 centimetres, we can calculate the number of bits needed to store the digital value of the analogue converted value. Amount of bits needed = 75/0.1 = 750 bits = 10 bits per sample. The total amount of memory needed to then store all of the measured data is (7.200.000*10)/8 = 9 GB of data.


Conclusion total memory

The total amount of data required to store the maximum amount of data of the sensors will add up to:

  • High resolution: 375 TB
  • Standard resolution: 265 TB

This amount of data is only at the maximum amount of stored data. This means that all of the processed data will have to be stored in the permanent memory. This will definitely not be the case should the robots be used. The memory that will be stored will only be off a very low margin, approximately 5% at maximum of what will be measured. This results in a storage capacity of approximately:

  • High resolution: 20 TB
  • Standard resolution: 15 TB


Implementation


Obstacle & contour and wear detection

For the obstacle and contour and wear detection sensors, it is important to have enough processing power to process the data which the sensors provide. The amount of data that needs to be processed will be of approximately 5GB/s (by adding the amount of data needed of the previous subsections). What kind of processing computer will be needed in our measurement robot, will be discussed in the next section {add section}. After this data has been processed, only the data will be stored that stands out from the benchmark tests. This outstanding data will be stored in the permanent memory of the robot, while the other non-outstanding data will be overwritten by newly sensed data, or simply erased. Should the robot come to stop at one of the main stations, where it will upload it’s permanent data to the global data cloud. Further data processing in the cloud (mainly data sorting) will be performed.


Geometry detection

Regarding the geometry detection, all of the data will benchmark data will have to be stored on the permanent memory of the robot itself. Each different point in the railway tracks will have a different angle (railway turn) and a height (railway not sinking in the stones). Each half a meter, a data point will be measured, for all of the data to be stored 185 GB will be needed. This amount of data is not much at all compared to the total amount of data needed by the other sensors. It can be concluded that the amount of maximum data needed to be stored, can all be stored on the robot itself. Also taking into account that the robot does not have to ride all the amount alone, for a total of 44 hours but only for 20 hours a day at maximum. This would result in the robot only needing a maximum of 85 GB of added storage, should all the data measured need to be stored. Moreover, not all of the measured data needs to be stored. Only the data that is different compared to the benchmark needs to be stored. To get easier values, it would be ideal for the robot to be able to store approximately 100 GB of data regarding the permanent memory and approximately 5 GB of temporary memory (RAM).


Conclusion implementation

Previously idea is to have the processing done on the robot itself and each stop upload it to some cloud. When at the stop, also have to robot download new data from the cloud to process the measurements of the upcoming railway track. However, after doing more research only a small amount of data is needed to benchmark the obstacle, wear and contour detection sensors, with a respectively low amount of data needed for the geometry detection. From this, we concluded that it is possible for the benchmark data to be stored on the robot itself and need a high-speed computer to compute the big amounts of data provided by the sensors. The supercomputer used in our robot will be discussed in the section 'Proccesing data'.


External power supply

As the external power sources, 2 methods are used. Firstly, a standard pantograph will be fitted on top of the robot to retrieve power from the catenary already in place. Secondly, a special diesel-generator cart can be hooked to the robot to provide electricity when there is no catenary.

In the Netherlands, two systems for the catenary are used: a voltage of 1,5 kV or 25 kV. The 1,5 kV system supports a speed up to 160 km/h and is used in the main part of the Netherlands. The NS only has trains capable of handling 1,5 kV. The 25 kV system support speeds up to 200 km/h [48]. This system is only used on the Betuweroute and the HSL. These are both tracks the NS doesn’t ride on. ProRail is currently thinking about upgrading the catenary to 3,0 kV, both to supply trains with more power and to minimize losses [49].

In 2016, the NS used an average of 71,9 Wh per travelled kilometer compared to an average of 71,0 Wh in 2015 [50]. In combination with the 18,5 billion traveled kilometers in 2015, this equates to a total energy consumption of 13,1 billion kWh [51]. This is to prove that the energy consumption of our robots do not have a significant influence on the energy consumption of the NS, and thus the capacity of the system should not be a big problem.

The robot will be using a standard pantograph to receive the power from the catenary. This is used by all the trains and thus will also work for our robot.

Secondly, for the diesel generator, the WhisperPower WS-Q 20 Mobile will be used. This generator delivers a power output of 20 kW, which would in theory be enough for our robot. The propulsion of the robot will take up a little over 11 kW, the laser will consume 2*1000 W and the rest will be used for the cameras, the computer and the charging of the battery.

Dimensions and laser movement system

The length of the robot has a maximum of 11 meter. This was determined with the viewing angle of the train driver and the height of the sitting position of the train driver. The height of the sitting position is between 2.6 and 3 meters. This was asked to a real train driver. So the maximum height of the robot is 2.6 meters. But when the robot is actually 2.6 meters high, the train drivers visibility is not optimal. So it was decided that the robot will have a maximum height of 2.25 meters so that the visibility of the train driver will not be significantly influenced. There is one drawback, which is that the train driver cannot see the tracks directly in front of him, but all the important signs are placed alongside the tracks. So the train driver can still see the signs. This 2.25 meters is when the connection with the catenary is folded in. The actual robot will have a length of 5.5 meters. This length was chosen so that the robot will not be too long. So that the train drivers visibility will remain pretty good. This length is needed for all the equipment to be placed such as the laser units, sensors and driving mechanism.

The robot should be able to operate in both ways. So a rail system underneath the robot will be designed for the lasers and sensors to move from side to side. This will keep the costs of the robot a bit lower so the costs of the maintenance and robots overall will be kept low. This rail will span the entire length of the robot. The lasers are placed in front of the front wheels. So when the robot will be operating in the other direction the lasers will move to the other side of the robot. The rail system has some drawbacks. It will not be as precise as it has to be. For the lasers to operate well, a high accuracy is needed. The lasers have to be focused very precise on the tracks. Another drawback of the rail system is that the lasers are very sensitive for vibrations. Another solution for this problem is that the upper part of the robot, so everything except the driving mechanism, can turn around. This will then lock in place so that it cannot move anymore. This solution will prevent the vibrations so the lasers can operate properly.

The length of the robot for now is 5.5 meters. This may not be the length that is used in the final model because the train driver cannot see directly in front of him. This might be a problem but not for the signals the train driver has to see. These signals are placed always next to the tracks. The robot is not wider as the railway tracks are so the train driver can still see them.

Week 6

Budget (continued)

The spokesman of ProRail states that every piece of railway tracks will be checked at least once a month for wear [52]. Therefore, we determine that our robots need to check every piece of railway tracks at least once a month for wear, geometry flaws, and rust. As stated before, the detection of wear, geometry flaws, and the detection and removing of rust happens at night. During the night (from 1 A.M to 5 A.M), the robots execute this detection and removing. There are around 30*4=120 hours in which the robots can drive during the night each month. Since they drive at 80 km/h during the night, one robot can check 80*120=9.600 km of railway tracks in one month. In total, there is 7.021 km of railway tracks in The Netherlands. Therefore, one robot would be enough to do the detection at night.

However, when there lies snow on the railway tracks, this must be removed as soon as possible. As stated in week 5 in the Budget section, we would theoretically need 3 robots to check all the railway tracks during the day in the ideal case. As a buffer, we have chosen 6 robots in total (see Week 5).

Cleaning laser system & diode

The lifespan of the diode of the laser system is 10.000 hours [53]. In the worst case scenario, a robot drives 20 hours a day. The same robot cannot drive the full day and night, because the robot needs time to up- and download data.

This means that the robot could drive at least 10000/20=500 days before the diode needs to be replaced. Compared to the price of the laser system, the price of the diode is almost negligible. We estimate the price of the diode at €3.253,- (3500 USD) [54].

We estimate the lifespan of the camera and the sensors at 10 years [55].

Furthermore, according to our estimations the whole laser system can run for at least 80.000 hours if we replace the diode every 10.000 hours. Therefore the two cleaning laser systems must be replaced every 80000/20/365 = 11 year. As can be read above, the two laser systems together cost €893.522,-. Thus, 1/11 * 893522 = €81.229,- will be spend on the laser systems of one robot per year.

Sensors

As reported in the Sensors section in week 4, there are three types of sensors that we want to use. Obstacle detection and geometry require both one sensor. Wear/contour detection requires two sensors. Therefore, there are four sensors necessary in total. Unfortunately, we couldn’t find the exact prices of these sensors. Therefore, we will estimate that each sensor costs the same as the Highspeed camera PCE-TC 225 (see Week 5), which is around €5.000,-. Thus, we will estimate the costs of the sensors to be 4*5000=€20.000,-.

Battery

An example of a battery that we can use is the Tesla Powerwall, as was already stated before. The price of this battery is €6.300,-. The warranty for this battery is 10 years [56], therefore we take 10 years as a lifespan for this battery. Hence, 6300/10 = €630,- will be spend for one battery per year per robot.

Scharfenberg coupler

For the Scharfenberg coupler (see Modular system & Coupling), €100.000,- of the budget will be available. The estimated lifespan of this coupler is the same as the estimated lifespan of a train, thus around 40 years [57]. Each robot needs a coupler and also each diesel generator needs two couplers. Since there are 3 diesel generators and 6 robots, we will need 12 Scharfenberg couplers in total.

Diesel generator

The WhisperPower WS-Q 20 Mobile will be used as diesel generator [58]. Since the diesel generator is barely used, we expect it to have a lifespan of at least 10 years for our purposes. We would like a diesel generator at each main station in order to have one available when needed, which means that there will be 3 diesel generators in total. We have contacted WhisperPower for a price indication of their generator and the corresponding components and from them we have received the following quotation:

Generator1.PNG Generator2.PNG

Of course, the diesel generator can’t drive on its own, which means that a bogie is necessary for this. This bogie doesn’t need any propulsion, in contrast with the bogie for our robot. Thus, it is quite a simple bogie, which will cost €5.000,- at the most.

CPU

The price of the CPU would be at most €10.000,- per robot, but probably less. However, to make sure that the actual price of the CPU will suit our budget, we’ll take €10.000,- as an estimation for the price of the CPU.

The average lifespan of a CPU is 20-30 years [59]. However, the CPU in our robot will be heavily used. Therefore, we take a lifespan of 10 years.


Turntable mechanism

For the turntable mechanism, we have contacted International Bearing Services BV. From them, we have received a quotation:

IBS offerte.PNG

According to this quotation the costs of the turntable mechanism are €1.458,- (including btw). The expected life span of this mechanism is around 10 years.


Bogie

The purchase price of the bogie that will be necessary to build one robot is estimated at €200.000,- by us. This bogie can have a lifespan of 30 to 50 years [60]. Therefore, we will make use of a lifespan of 40 years in our calculations. However, it requires some maintenance to reach this amount of years. We will assign €10.000 a year for the maintenance for each bogie.

Connection to the catenary

The price of the connection to the catenary, which is needed for the robots to act autonomously at night, would be at most €10.000,- per robot. The estimated lifespan of this connection is at least 10 years.

Personnel (station chief)

Since society isn’t completely ready for automatisation [61], we think that it’s better to let one person keep an eye on these robots from the main station of communication. A station chief earns around €2.303,- per month [62]. So, a station chief earns around €27.636,- a year.

Building costs

An engineer earns around €54.203,- per year [63]. We suspect 6 engineers to work at the production of one robot for a full year to finish it.

Additional costs

The parts of the robot that are mentioned above are the most expensive parts. Of course, there are way more parts and stuff necessary. We will determine an extreme upper limit for the budget of the additional costs. We will expect to spend no more than €100.000,- per robot per year on things that aren’t mentioned above.

Buffer

As a buffer we take €100.000,- per robot in case something goes wrong.

Budget

Budget

Comparison

As mentioned before, we have estimated that ProRail spends around €41,5 million per year on trouble caused by wet leaves and snow and on the detection and maintenance of rust, geometry and wear of the railway tracks. Our goal of this project is to spend way less per year on the robots that could solve the same problems. As can be seen in our budget, we have succeeded in this. Our budget shows that there is €3.932.394,- per year necessary for the robots. This is only 9,48% of the estimated budget that ProRail spends on the same problems.

Battery

Because of part of rails without catenary, it is necessary for our robot to have a small internal battery. This battery must be able to run the complete train for 5 minutes. As a big part of the energy is consumed by maintaining its speed, the battery should have a capacity of at least 10 kWh. An option for such a battery would be the Tesla Powerwall mentioned before, costing €6.300.

Turntable Mechanism

In order to save costs on equipment a system is required that rotates the top part of the robot, so the lasers are orientated to the front of the robot and the coupling to the rear, without turning the bogie on the tracks. This could also be achieved by designing the robot to be symmetrical, however, it is much more expensive to apply a second set of lasers, sensors and coupling to the robot, than installing a turntable mechanism.

The mechanism consists of two parts, a heavy duty slewing ring bearing and an electric motor, that turn the top part of the robot. This part of the robot includes the following systems and their respective weight:

  • Two lasers ~ 930 kg
  • Two cameras ~ 70 kg
  • CPU & harddrives ~ 50 kg
  • Two TESLA powerwalls ~ 240 kg
  • Coupling ~ 200 kg
  • Additional sensors, body and equipment ~ 100 kg


The total weight is estimated to be 1800 kilograms. The requirements for the turntable mechanism is to turn this mass, with a length of 5.5 meters, 180 degrees within half a minute. To calculate the required torque, power and rotational speed of the turntable, the top part of the robot has been modeled as a rotating arm with length 2.75 meters, and a mass of 2000 kilograms on its end. According to the calculations in appendix A, the required torque is 212 Nm, the power 22 W and the rotational speed 10 rpm. Since there is a gear ratio of between 1 to 10 and 1 to 20 between the internal gear of the slewing ring and the electric motor, the required torque for the motor is a factor 10 to 20 lower, while the rotational speed is a factor 10 to 20 higher.


Modular system & Coupling

A modular cart system is applied to the robot. The maintenance robots can then be linked to the train, with the same using the same mechanism as normal trains. There are two different carts; the main cart, which includes all the sensors and lasers needed for maintenance. Also, a propulsion system with a telescopic catenary system for which the cart receives its needed power when it is not connected to the train itself. This cart will operate in mostly in front of the train, where it is coupled with the Scharfenberg coupler. Its catenary system is folded in and not used, as the cart receives its power with the coupling. After the train schedule, thus at night, the cart will separate itself from the train, extend the connection to the catenary and work on its own. Vice versa, it can also connect itself to the train, after its nocturnal shift.

However, at some tracks in the Netherlands, a catenary system is not present, for which the cart can not use the catenary to receive its power. Therefore, when the cart is not connected to the train, thus at night, a second cart is needed, consisting only of a diesel engine with a generator. This cart is connected in the same way as the main cart would be connected to the train; using the Scharfenberg coupler.

The operation of the maintenance during the day and the night is as follows:

Day, with catenary: connected to train, power through coupling, train receives power with catenary

Day, without catenary: connected to train, power through coupling, train receives power with diesel generator

Night, with catenary: autonomous, power through catenary

Night: without catenary: autonomous, power through separate generator cart

The Scharfenberg coupler is used to transmit power through the carts. This is only needed when the cart is connected to the train. Here, the cart is not connected to the catenary, as this obstructs the view of the machinist. The manufacturer of these couplers, Voith GmbH, is contacted. Through that company, it has been confirmed that, in order for the coupling to work, an electric head needs to be applied, which is also called an “E-coupler”. This E-coupler is available in several types with specific pin lay-outs for different trains. This also means that the robot can only be connected to a single type of train with the E-coupler in place. In addition, the transmitted power through the coupler differs as well. It has been assumed that the power through the coupling is enough to fully power the robot.

At all the railway tracks in the Netherlands, the so-called sprinters will be regarded as they run over the most railway tracks. In comparison, the bigger trains, intercities, will only run over the tracks between the big cities. The sprinter most commonly used in the Netherlands is the Sprinter Lighttrain. This train is relatively new and manufactured by Bombardier Transportation, a Canadian company, in cooperation with Siemens[64][65]. For this Sprinter, it is assumed that a general coupler is used, the Type 10[66]. This coupler is used around the world in high-speed applications on, applied to high-speed trains since 2002. On this coupler, an electric head is applied in order to transmit power through the coupler. Accordingly, this coupler is put on one end of the robot, attached to the turntable mechanism. In this way, the robot can always connect to a train or an extra modular cart, dependent of the rotation of the turntable mechanism.

In addition, the coupler is put on one side of the modular diesel generator, in order to connect it to the main robot. The diesel generator does not use the turntable mechanism, for which it needs two couplers on either side.

Processing power

The processor used in the robot, needs to be able to compute a lot of data in a very short amount of time. However, each sensor will have a different hierarchy in the processing ‘waiting line’ (also known as ‘memory hierarchy’ in general). To get the best possible, it would be ideal that each of the sensor’s data can be stored and processed all at the same time. This is somewhat achievable with the use of multiple cores in the Central Processing Unit (CPU). Note that it says somewhat, since completely having each sensor’s data being processed on each core independently, since the processor will not become twice as powerful by adding double the amount of cores. Furthermore, each sensor has a different amount of data load for the processor, resulting the the different core per sensor not being achievable. However, using good algorithms for the CPU regarding memory hierarchy, it is possible to have different cores overlap in processing the data for each sensor.

It has been shortly discussed that we want a multi-core processor with a high enough clockspeed and processing power to process all the data the sensors provide to the CPU. Fortunately, the sensors we use, will output only data that is already viable for comparison regarding the obstacle detection as well as the wear and contour detection. This would not require our own CPU to also process the raw data of these sensors. The geometry detection will, however, have raw data output. But since these are simply numbers representing the angle and height of the railway at different points on the railway, this raw data already has a meaning in of itself. The question now remains, what is kind of processor will be needed in order to suit the requirements of the robot?

The difficult aspect of precisely determine the processor that will suit the needs of our robot is all the different specifications regarding the lasers. Specifications as the clock frequency, the amount of cache memory, the number of cores/threads (where the cores are the physically implemented cores on the CPU and the threads are the virtual cores in the CPU). Since the fact that CPUs are not made for specific tasks like our implementation, the general implementation is difficult to compare to the needs of our robot. We cannot simply specify certain minimum aspects the processor has to have in order to be able to operate in our robot. It is however simple to say that the higher the values, the better for our robot. This is however not reliable to say, thus further research has to be made.

A good method to start, is to determine the amount of time the processor has to process the data before the obstacle reaches the laser. The amount of space in between the laser and obstacle detection unit is approximately 1 meter. With the train moving at 140 km/h would result in approximately 26 ms. However, the camera will be running at 500 fps, which would result in 2 ms per frame to be processed, which will be used as minimum amount of time for the processor to process the incoming data of the obstacle sensors.

Next, would it be more efficient for the robot to have different processors working on different processes. The data sensed by the obstacle detection sensors only has that small amount of time to process the incoming data. The data of the contour and wear detection, as well as the geometry detection will not directly need the processed data to have immediate feedback. It would make sense to have two different processors that suit their needs. This would also create more air to breathe for the processor which needs it feedback to go directly to the laser.

A good start for the obstacle detection processor, is to have a look at the different type of processors and what kind of purpose they serve [1]. The types of processors that would suit the need of the obstacle detection can be the Xeon E generation processors. Even though the website only describe the E3 generation processors, the E5 processor generation has been released as well and will be well suited for our obstacle detection data processing. One of the most important aspects is the amount of cache ram and clock frequency of the processor. The big amount of cache storage will be perfect for the different frames that can be stored in the same cache. While being processed they can be overwritten by new frames afterwards. The other aspect, clock frequency, is also important. The higher the frequency, the faster the processor will be able to deal with incoming information. One last note on this processor as well is that it has multiple cores. These cores work in parallel to each other and possibly can all be used to process different frames at the same time. Each single frame will take up approximately 3.6 MB of space and multiple different frames can be processed after one another without the processor having to wait to finish it first picture.

For the geometry and wear and contour detections sensors, not direct input is required from the processed data of these sensors. Moreover, the processed data of the wear and contour detection will need to be virtualised, something the E5 generation is not capable of doing. The E3 generation processors on the other hand are capable of virtualising the processed data and are capable of storing an virtualised image of the raw data provided by the sensor. This should be smartly implemented into the processor, such that the code running on the processor will only make a virtualised picture of the railway when the measurement is off of the benchmark measurements. The simple numbers provided by the geometry sensors can directly be stored when necessary.

Positioning

An important aspect of the realisation of our robot is the placement of the different applications of our robot. We have had multiple possibilities of where to place certain sensors, how many wheels we want to use and where we want to position our lasers (e.g. in front or in the back of our robot). The final design we chose to implement is having the obstacle detection all the way in the front, a meter after that the laser removal component just in front of the bogie. After the bogie the contour and wear detection sensor will be attached, pointing at the railway. Inside the machine itself, the geometry detection sensor is placed.

The placement of the laser and sensors underneath the body of the robot has been discussed multiple times. A switch between contour and wear detection has been discussed. The reason why we would switch these two is because of the time we would generate by placing the laser removal more to the back of the robot. This would come into conflict with other components of the robot. Firstly, the contour and wear detection would then not be able to determine any good values for the measurements should there be any obstacles on the railway track. The obstacles on the railway have to be removed at first before any valuable data can be achieved. Second of all, the laser removal has now been placed after the wheels of the robot. The robot will then not be able to remove any obstacles preventing efficient movement, resulting in having lost one of our biggest goals we set in the first place. Moreover, the time that we generated is of no use, since the robot still be restricted to the time of each frame captured by the camera of the obstacle detection sensor (2ms). Moving it more to the back gives the stored data in the processor more time to ‘make it’ to the laser component, which is not necessary.

Not only the positioning of the components on the robot has been discussed, also simple aspects, easily overlooked have been discussed. Aspects like how many wheels do we want to implement on our robot? The first idea was to implement 4 wheels and make a design like a car, with the idea behind this design being for more stability. After doing more research in this aspect of the robot, conclusions were drawn that a lot of designing of the bogie can be discarded should be want to use a premade bogie. The premade bogie already consists of implemented wheels with brakes and serves as a base to build the rest of our robot on. To not oversize our robot, we were left no choice but to take the smallest available premade bogie of 3.5 meters long. This premade bogie comes with only two wheels attached to the bogie. The previous thought of designs of the robot could be discarded should the premade bogie want to be used. Since no real knowledge was present on how to accurately make a bogie, the premade bogie was chosen to be implemented into our design.


Week 7

Budget

Current Expenses

It is hard to determine the exact cost of the trouble caused by weather conditions, because total amounts spent on maintenance are given but specific amounts, like the ‘cost’ of snow, is not to be found anywhere. Also, these costs are hard to estimate. These costs are not linear compared to the occurrences, meaning each disturbance costs the same amount of money.

The total maintenance costs can be found in the year overview of ProRail, and because ProRail is under partial supervision of the Dutch government, their year overview is public and can be found on their website. The year overview for 2016 has not been published yet, so all data used is from 2015 [24].

In 2015, ProRail received €1.098 million to spend on maintenance and management of tracks, this is a little bit more compared to 2014. They spent €950 million. From this money, ProRail spent €139 million on large-scale maintenance and €269 million on small-scale maintenance. Large-scale maintenance is the maintenance needed to ensure reliability and quality in the medium-long to long term. This includes for example polishing the tracks or preparing the tracks for the winter season. Small-scale maintenance includes all the maintenance needed to ensure availability and safety, as well as incidental maintenance. This is more short-term maintenance. Examples of these are inspections or replacing of (small) components.

ProRail also spent €154 million on managements. Their year overview states this was largely used on ICT services, of which some are used to detect problems in advance. To us, the large-scale maintenance costs are of most use, since they cover the weather conditions. The management costs could be of some use but are not our priority. When our machine can detect problems on rails in advance, some ICT systems will not be needed anymore. This can be a huge cost saver, because in 2015, ProRail invested €60 million these ICT systems to prevent disturbances.


Business Plan

Comparing our budget to the estimated costs of ProRail shows that our idea would really be a big improvement, especially for Enterprise. Therefore, we are going to investigate what would be necessary if we would continue our ideas of the robot.

Thus, in this section we will make a business plan for our robot in which we specify important research that need to be carried out and milestones that need to be reached.

First, we would need to talk to ProRail and the NS to see if they would be interested in such a robot. If they see potential in our idea, we can make a time planning in which we take the following things into account:

Important research:

Actual overview of all part- and construction costs

From a lot of products we were able to acquire a quotation or a price indication. However, there are still a few products of which we had to make an estimation of the price or use the price of a likewise product. If we want to continue with the robot, we need the exact prices of all products and of the building costs. Constructing a prototype would be a good way to determine the exact costs of the robot.

Maximum power transmission through the coupler

For now, it is still unsure how much electricity can be transmitted through the train coupler. It is assumed to be enough for the needs of the robot, however we would like to have a formal confirmation. If it turns out not enough power can be transmitted, a new solution has to be found to supply the robot with enough power.

Adding the robot to the GSM-R network

During this project, we have assumed that we can implement GSM-R on the robot in order to add the robot to the European Railway Traffic Management System (ERTMS). Research is necessary to find out if this is indeed possible and to find out how this can be arranged.

Review and support current design

The current design is not very well thought of. More time was spent on the specifications of the internal parts than on the exterior. In the future, we could focus more on the exterior design of the robot. If the dimensions of the needed equipment are known, the final dimensions of the robot itself can be determined. This also includes aerodynamic optimisation and user-friendly contours.

Gaining trust in the robot from User, Society, and Enterprise

As stated before, a lot of people are having problems trusting robots to handle work correctly. Therefore, we would need to convince User (NS), Society (train travelers), and Enterprise (ProRail) that this robot is really an improvement by showing the budget and that it is completely safe (for example, by telling them that the laser isn’t harmful for bystanders). Also, the fact that there is a shift in employment must be shared. This means that there might be people who will (partially) lose their job, for example the people who execute the removal of rust, wet leaves, and snow, but new jobs will also be created. For example, there are engineers necessary to do the research and construction of the robot and there is a station chief on the head station necessary (in the beginning) to keep an eye on the robot.

Make the robot universal for all NS trains

The only train we focussed on for now is the Sprinter SLT. The robot has the same coupler as this train, but for making the robot universal the robot needs a coupler which can couple to all the trains used by the NS. Not only should the robot be able to couple, the coupler should also be able to transmit enough electricity to power all the equipment.

New name

Find a new name (Track Rat).

Software

For the propulsion and speed control of the robot a computer is needed. This computer needs software to keep track of the speed of the robot. It also needs to communicate with the GSM-R network to keep track of the other trains operating on the railways. The software also needs to be able to react to speed limits and lower the speed if needed when checking the wear of the railway tracks.


Milestones:

Meeting with ProRail / NS to discuss the idea

If we eventually produce our robot, ProRail and NS should be convinced that the robot profitable and can work efficiently. When ProRail and NS are convinced we can discuss if the idea of implementing the robot in the GSM-R network can actually work. Also when we want to produce the robot we need to know the precise budget of ProRail to determine how efficient the robot will actually be. Maybe ProRail or NS will come with some improvements.

Prototype

Before we can make the actual robot, we first need to make a prototype. This prototype can also help to give ProRail and the NS a better view of our idea. Furthermore, this robot can help to detect problems.

Testing the prototype

In order to detect problems, the prototype needs to be tested of course. Also, we would for example be able to test the maximum speed and the propulsion system.

Discussion

In this section it will be discussed what ideas have changed during the previous weeks and why.

Turntable mechanism

In the beginning, we wanted to make the cart symmetrical in order to let the robot operate in both directions. However, we encountered a problem, because the lasers and sensors couldn’t work in both directions. Since the lasers are very expensive, it also isn’t an option to put four lasers on one cart. Therefore, we decided to implement a turntable mechanism on the robot. This was a much cheaper option.

Electric motors

In the beginning we wanted to have the electric motors separately, but later we have chosen to put the electric motors into the robot. This saves an extra cart and it also allows the robot to work autonomously.

Bogie

The robot had to be around 5,5 metres long. However, the smallest bogie (including electric motors) was already 3,5 metres long. At first, we wanted two bogies, one in the front and one in the back. However, this would make the cart way too long and thus it would have a negative influence on the sight of the machinist. Thus, we have decided to only use one bogie for our robot.

Catenary

If it wasn’t possible to transmit power through the coupler, we would have needed a connection to the catenary even when the robot was coupled with the train. However, this connection would be right in front of the machinist. Therefore, we decided to make a connection to the catenary which would go under the window, through which the machinist looks, and then upwards to the catenary. Luckily, we received an email from Voith GmbH that it was possible to transmit power through the coupler, so we don’t need this connection to the catenary during the day.

Week 8

Appendix A

To calculate the required torque, power and rotational speed of the turntable mechanism, formulas (1) to (3) are used.

P = 𝛕∙𝝎 (1) 𝛕 = I∙𝛂 (2) 𝝎 = 𝛉/t (3)

Here, P is power in [W], 𝛕 is torque in [Nm], 𝝎 is rotational velocity in [rad/s], I is the moment of inertia in [kg∙m2], 𝛂 is the rotational acceleration in [rad/s2], 𝛉 is the angle in [rad] and t is time in [s].

To calculate the moment of inertia and the rotational acceleration, formulas (4) and (5) are used.

I = m∙r2 (4) 𝛂 = 2𝛉/t2 (5)

Here, m is the mass in [kg] and r is the length of the arm in [m].

The torque is calculated by combining the mass moment of inertia and the rotational acceleration, as shown in formula (6). This acceleration is over 15 seconds because half the turn is used for accelerating and the other half is used for decelerating.

𝛕 = m∙r2∙2𝛉/t2 = 2000∙2.752∙2∙(𝝅/2)/152 = 212 Nm (6)

The rotational velocity is calculated in formula (7) and translated to rpm in formula (8).

𝝎 = 𝝅/30 ~ 0.1 rad/s (7) 𝝎 = 0.1∙60/(2𝝅) ~ 10 rpm (8)

The required power for the turntable mechanism can be calculated according to formula (9).

P = 212∙0.104 ~ 22 W (9)

Sources

  1. Smith, O. (2016). No more commuter misery? Trains fight leaves with lasers. Retrieved from http://www.thememo.com/2016/09/12/train-leaves-leaf-zapping-trains-rail-safety-and-standards-board-are-arming-up-with-microwaves-and-lasers/
  2. ETS SPOOR B.V. (n.d.). Railmonitor. Retrieved from http://www.etsspoor.nl/producten/meetapparatuur/railmonitor/
  3. Statistics about railway disturbances in the Netherlands. Retrieved from https://www.rijdendetreinen.nl/statistieken/
  4. Different kind of disturbances around railway tracks. Retrieved from https://www.prorail.nl/reizigers/storingen-op-het-spoor/
  5. Article about leaves on the railway tracks. Retrieved from http://www.metronieuws.nl/nieuws/binnenland/2016/11/die-rot-blaadjes-op-het-spoor-waarom-doen-ze-niets/
  6. Smith, O. (2016). No more commuter misery? Trains fight leaves with lasers. Retrieved from http://www.thememo.com/2016/09/12/train-leaves-leaf-zapping-trains-rail-safety-and-standards-board-are-arming-up-with-microwaves-and-lasers/
  7. Sorokanich, B. (2016). This Hand-Held Laser Makes Rust Literally Evaporate. Retrieved from http://www.roadandtrack.com/car-culture/classic-cars/videos/a30597/best-rust-remover-laser/
  8. P-Laser (2017). Laser cleaning applications. Retrieved from http://www.p-laser.com/applications_detail.aspx?AGUID=1f846979-8fea-4745-bcea-663800c027e5&LGUID=8565a502-c109-43ef-b1a1-dfba5f3edbf6/
  9. Schreiner, M. et all (2008) Handbook on the Use of Lasers in Conservation and Conservation Science. Retrieved from http://www.science4heritage.org/COSTG7/booklet/
  10. Vega, R. et all (1990). Laser ice removal system. Retrieved from https://www.google.com/patents/US4900891/
  11. Paper geometry railways. Retrieved from http://crema.di.unimi.it/~fscotti/ita/pdf/Scotti02.pdf/
  12. Paper wear railways. Retrieved from https://docs.google.com/viewer?url=patentimages.storage.googleapis.com/pdfs/US6218961.pdf/
  13. Railway monitor. Retrieved from http://www.etsspoor.nl/producten/meetapparatuur/railmonitor/
  14. Drones with infrared cameras 1. Retrieved from https://www.prorail.nl/nieuws/proef-met-drones-controleren-wisselverwarming-met-infraroodcamera-s/
  15. Drones with infrared cameras 2. Retrieved from https://tweakers.net/nieuws/86694/prorail-zet-drones-in-om-verwarming-van-wissels-te-controleren.html/
  16. Railway switches inspection robot Felix. Retrieved from http://research.loccioni.com/en/robotics/felix/
  17. Tackling a leafy issue, Railway Technology, retrieved from http://www.railway-technology.com/features/feature1457/
  18. Herfst, ProRail, retrieved from https://www.prorail.nl/reizigers/weersinvloeden/herfst
  19. Test met watersproeiers tegen gladheid door blaadjes op spoor geslaagd, Spoor Pro, 26 march 2015, retrieved from http://www.spoorpro.nl/spoorbouw/2015/03/26/test-met-watersproeiers-tegen-gladheid-op-spoor-geslaagd/
  20. Nu al herfst?, ProRail, 14 august 2014, retrieved from https://www.prorail.nl/reizigers/nieuws/nu-al-herfst
  21. Bladeren op het spoor weglaseren, Nemo Kennislink, 20 august 2014, retrieved from https://www.nemokennislink.nl/publicaties/bladeren-op-het-spoor-weglaseren
  22. With Laser Railhead Cleaner your train will never be late because of leaves in its way, Wonderful Engineering, retrieve from http://wonderfulengineering.com/with-laser-railhead-cleaner-your-train-will-never-be-late-because-of-leaves-in-its-way/
  23. Zapping the leaves from the rails, Delta, 8 september 2014, retrieved from http://delta.tudelft.nl/artikel/zapping-the-leaves-from-the-rails/28710
  24. 24.0 24.1 Jaarverslag 2015, ProRail. Retrieved from http://www.jaarverslagprorail.nl/FbContent.ashx/pub_1000/Downloads/ProRail-jaarverslag-2015.pdf/
  25. Europees Spoorwegbureau. System Requirements Specification (SUBSET-026). Paragraph 3.6.5. Retrieved from http://www.era.europa.eu/Document-Register/Pages/Set-3-System-Requirements-Specification.aspx
  26. European Rail Traffic Management System (ERTMS). Retrieved from http://www.railway-technology.com/projects/european-rail-traffic-management-system-ertms/
  27. Clear CinCom. GSM-R: What Is It, And Why Does It Matter? Retrieved from http://gsmr-info.com/
  28. United States Patent, Obstacle Detection System, December 2000. Retrieved from https://docs.google.com/viewer?url=patentimages.storage.googleapis.com/pdfs/US6163755.pdf
  29. Underground train map of Milan. Retrieved from http://italy4.me/wp-content/uploads/2015/01/karta-milan-metro.pdf
  30. Bridge by Meppel (Netherlands) without catenary, retrieved from https://en.wikipedia.org/wiki/Overhead_line
  31. Tackling a Leafy Issue. Retrieved from http://www.railway-technology.com/features/feature1457
  32. ProRail in cijfers. Retrieved from https://www.prorail.nl/over-prorail/wat-doet-prorail/prorail-in-cijfers
  33. Berke, C. (2016). Laser Rust Removal: Too Good To Be true? Retrieved from http://shortshift.co/laser-rust-removal/
  34. Highspeed camera PCE-TC 225. Retrieved from https://www.pce-instruments.com/dutch/index.htm?id=google-nl&_artnr=3650570&_p=5438.95&_pmode=0&_pbexkey=74&_date=20170310000501&_pbhash=b6e2287ec9f5de8edebe8d8b4d2a1a4eed2f956ee5b8aa4175be524fd9261b0d&gclid=CjwKEAjwzKPGBRCS55Oe46q9hCkSJAAMvVuM4Fza-mwEB0kj6YoZD2SDZ1Y4IfdoLMq0Rj4KzlL5IBoCPhTw_wcB
  35. Falch (2017). cont jet 125 2500-26-0-e. Retrieved from http://www.falch.com/products/g588.html
  36. Baveck Machines (2017). Rail infrastructure - Rust removing brush. Retrieved from http://www.baveck.nl/en/rail-infrastructure/24-rust-removing-brush/
  37. P-Laser (2017). Laser cleaning applications. Retrieved from http://www.p-laser.com/applications_detail.aspx?AGUID=1f846979-8fea-4745-bcea-663800c027e5&LGUID=8565a502-c109-43ef-b1a1-dfba5f3edbf6
  38. Adapt Laser Systems (2017). Lasersystem CL 1000 Most Powerful hand held laser on the market. Retrieved from http://adapt-laser.com/ourproducts.php?id=21
  39. Gray, R. (2015) End of the line for autumn train delays? Dutch begin trials to zap leaf litter from the tracks with LASERS. Retrieved from http://www.dailymail.co.uk/sciencetech/article-2860418/End-line-autumn-train-delays-Dutch-begin-trials-zap-leaf-litter-tracks-LASERS.html
  40. Tesla Powerwall, retrieved from https://www.tesla.com/nl_NL/powerwall
  41. Examples of frontal areas of cars, retrieved from http://ecomodder.com/wiki/index.php/Vehicle_Coefficient_of_Drag_List
  42. Examples of drag coefficients, retrieved from http://www.brighthubengineering.com/hydraulics-civil-engineering/58434-drag-force-for-fluid-flow-past-an-immersed-object/#imgn_2
  43. Examples of rolling resistance coefficients, retrieved from https://en.wikipedia.org/wiki/Rolling_resistance
  44. More examples of different rolling resistance coefficients, retrieved from http://auto.howstuffworks.com/tire4.html
  45. Used ICCD camera, retrieved from http://www.stanfordcomputeroptics.com/products/iccd-framing-camera.html#features
  46. Calculations storage data, retrieved from http://eckstein.rutgers.edu/mis/handouts/storage-calcs.pdf
  47. Height differences in the Netherlands, retrieved from http://www.geo-informatie.nl/geodesk/hgt-300dpi.pdf
  48. Information about the caternary, retrieved from http://energievoorziening.funnymedia.nl/bovenleiding%20nieuw/bovenleiding.htm
  49. Meer spanning op het spoor?, ProRail, 8 january 2014, retrieved from https://www.prorail.nl/nieuws/meer-spanning-op-het-spoor
  50. Jaarverslag NS 2016, retrieved from http://www.nsjaarverslag.nl/jaarverslag-2016/energie3/a1128_Energie
  51. Mobiliteitsbeeld 2016, Rijksoverheid, retrieved from http://www.nsjaarverslag.nl/jaarverslag-2016/energie3/a1128_Energie
  52. Onderzoek naar plotselinge slijtage op spoor. Retrieved from http://nieuws.nl/algemeen/20130226/onderzoek-naar-plotselinge-slijtage-op-spoor/
  53. Adapt Laser Systems (2017). Lasersystem CL 1000 Most Powerful hand held laser on the market. Retrieved from http://adapt-laser.com/ourproducts.php?id=21
  54. 400W 808nm CW High Power Laser Diode Stack for Heat Treatment, Very Powerful ! Retrieved from http://www.ebay.com/itm/400W-808nm-CW-High-Power-Laser-Diode-Stack-for-Heat-Treatment-Very-Powerful-/270918323182/
  55. Gampat, C. (2015). The Lifespan of a Difital Camera Is Way Too Short. Retrieved from http://www.thephoblographer.com/2015/09/11/the-lifespan-of-a-digital-camera-is-way-too-short/
  56. Tesla Powerwall, retrieved from https://www.tesla.com/nl_NL/powerwall
  57. SpoorPro: Steeds vaker tekort treinonderdelen. Retrieved from http://www.treinreiziger.nl/spoorpro-steeds-vaker-tekort-treinonderdelen/
  58. W-SQ 20 Mobile. Retrieved from http://www.whisperpower.com/nl/4/5/37/producten/generatoren-1500-1800-rpm/w-sq-20-mobile.html
  59. What is a CPUs Lifespan? Retrieved from http://www.tomshardware.co.uk/answers/id-2567579/cpus-lifespan.html
  60. SpoorPro: Steeds vaker tekort treinonderdelen. Retrieved from http://www.treinreiziger.nl/spoorpro-steeds-vaker-tekort-treinonderdelen/
  61. Shewan, D. (2017). Robots will destroy our jobs – and we’re not ready for it. Retrieved from https://www.theguardian.com/technology/2017/jan/11/robots-jobs-employees-artificial-intelligence
  62. Loonwijzer. Retrieved from https://www.loonwijzer.nl/home/salaris/salarischeck?job-id=5112020000000#/
  63. van de Velde, P. (2014). Hoogste salarissen na studie techniek. Retrieved from http://www.telegraaf.nl/dft/geld/werk-inkomen/22218328/__Hoogste_salaris_na_techniek__.html
  64. Sprinter manufactured by Bombardier Technologies. Retrieved from http://www.bombardier.com/content/bombardiercom/en/transportation/projects/project.sprinter-netherlands.html?f-region=all&show-by-page=50&page=1&f-country=nl&f-segment=all&f-type=all&f-name=all
  65. Sprinter manufactured by Bombardier Technologies. Retrieved from https://nl.wikipedia.org/wiki/Sprinter_Lighttrain#cite_note-2
  66. Type 10 Scharfenberger Coupler. Retrieved from http://voith.com/en/products-services/power-transmission/scharfenberg-couplers/scharfenberg-couplers-railcars-14310.html