Robot maintenance and railway inspection robots – interview with Andrew Starr

Claire chatted to Andrew Starr from Cranfield University all about robot maintenance and railway inspection robots.

Andrew Starr holds a PhD in condition monitoring of robotic production systems and is a Chartered Engineer. His career started with British Aerospace, before working for the University of Manchester, University of Huddersfield, and the University of Hertfordshire. At Cranfield University he is Professor of Maintenance Systems, and former Head of the Centre for Life-cycle Engineering and Management. His research is in monitoring of mechanical systems with special interests in railways, aero gears, sensor systems, and maintenance optimisation through data fusion.

In this episode
00:00 Intro
00:30 Welcome
01:51 Robot maintenance
17:59 Railway inspection robots
33:16 Thanks for listening

Find out more
Andrew: https://www.cranfield.ac.uk/people/professor-andrew-starr-695715
Robot Talk Podcast: https://www.robottalk.org

#Robot #Robotics #Railway #Transport #Maintenance #Inspection #Industry

Robot talk is the podcast that sits down with robot enthusiasts from around the world and ask them the questions you always wanted answered like how do you monitor the robots and how does that thing Work hello everyone welcome to robot talk I’m your host CLA Asher and this week we’re returning to the industrial side of Robotics to find out more about how robots are used for inspection and maintenance and also how we can monitor the robots themselves to improve precision and

Quality before that I’d like to remind you about our competition to win your very own robot talk t-shirt our winner for February was Frank fer and we’ll be selecting another lucky winner at the end of March for more information about the competition and how to enter check out robot talk.org

And don’t forget that you can send in your questions for my future guests either by contacting us on social media we’re at robot talkpod on all the socials or you can submit questions via our website by going to robot talk.org so with all that said let’s get on with this week’s

Interview I recently had a great conversation with Andrew star from Cranfield University all about robot maintenance and Railway inspection robots today I’m talking to Andrew star a professor of Maintenance systems at crownfield University whose work focuses on monitoring of mechanical systems hi Andrew great to have you on the

Podcast hello CLA and thank you for inviting me you’re working on robotics for industrial maintenance so to get us started can you give a bit of an overview of some of the different ways that robotics is being applied in this area robotics has been used in the manufacturing industry for decades now

And uh what’s been interesting over the last 10 years or so is the move into more um mobile Robotics and mobile manipulators that is Vehicles which are a combination of some kind of unmanned ground vehicle with a a robot on top and uh it’s of great interest to those who have highrisk

Environments for people to do really detailed tasks so some examples of those are uh nuclear industry um Railway um some aspects of Aerospace and uh really the applications are ramping up very rapidly at the moment so this is the I can never remember the 3DS was it dangerous difficult dangerous and dirty

Tasks certainly uh with the partners that we work with um those are some of the things that come up um and and often the people who are asked to do these kind of tasks are also there at night um consider for example in something like inspection of a railway

Tunnel it’s a horrible place to be it’s dark and it’s dripping with water and it’s got to be at a time when the trains don’t run so uh people are doing that during the night and uh it really is a risky dangerous place to be because it’s

Uh slippy the trains aren’t there to run you over but uh it’s a pretty rotten job so if we’re able to move those jobs into um closely supervised situations where people are remote uh either in a truck or a pter cabin or um even better in a comfortable office and even better

Than that if uh there’s more autonomy in the task then the tasks can be supervised during the day or even for part-time workers as opposed to people who are there on the spot yeah absolutely absolutely and although your work now uses robots to monitor different mechanical systems um you

Actually began your career working on monitoring the robots themselves um specifically robots on car assembly lines so what challenges in this area have now been solved um and which ones remain yeah I was privileged to start um my research career with a group at Manchester University who were a spin out

Company and uh sly no longer in existence but the the company that I worked with for my PhD was Ford and uh I worked on four different sites two in Germany two in the UK and there were lots of robots in use doing things like welding um component

Handling and application of some uh specialist materials like sealant around windscreens for example um one of the problems basically a robot is supposed to get something from A to B um it’s just a manipulator and uh one of the problems that turned up was that they didn’t get

From A to B at all um or alternatively when they got to B they were quite inaccurate and that’s really difficult to measure it’s not just at Arms Reach you’re um dealing with a machine that is um quite quick and heavy and of often handling tools which you wouldn’t put in

The hands of a human operator because they’re too big and heavy and they’ve got no guards on because they don’t need it so the machines are in cages so uh the result for example of an imprecise weld is that uh either the the welding gun isn’t on the edge of say a

Car body around a doorway or a windscreen misses it completely or it’s on the edge in which case it leaves ragy Edge um or it might um strike the um the high quality bit of the car body that that would be um visible afterwards so

Some of those could be covered up with a rubber seal or something so they might pass but quite a lot would fail and the errors of end positioning were at least 10 millimeters in some cases wow so that’s really difficult to uh pick up optically so um it was being picked up

By quality control so um the the uh spin out company at Manchester woles and maintenance had been involved in dealing with some of the welding Transformers at first and they were getting too hot and they were using a technology that had only just come out of the military which

Was infrared Imaging um it really cost an arm and a leg at that point and it was all displayed in monochrome but it could detect blocked welding Transformers which had got dirt in them and it could detect the the over temperature in these so having seen that

I extended this idea to using video to measure the position and to detect the repeatability even with the limited number of pixels that we had on video cameras at the time I wouldn’t say it was very portable but uh with a tripod you could get the camera static and with

A trolly load of Gears I mean batteries desktop computers um 50 m extension cables um you needed all of that to get on the production line but then you could actually detect the um the movement of the robot itself in practice so those were some of the uh

The early tests but later on I met probably about 10 years ago a company in the Midlands who refurbishes industrial robots and ask them about some of the faults that had led to this imprecision which basically was um either poor tuning in the control circuitry or um

Damage to some of the um closed loop sensing um or mechanical forward Loop components which would have things like backlash in Gears and shafts and bearings and damage to Motors and so on and essentially they they told me all all of that been fixed however they still had things that they needed to

Replace before robots could be returned into the secondhand Market they weren’t just uh resprayed yellow or or orange or green depending on the manufacturer um one of the problems that they still encounter is multicore cables so we need to get quite High currents to the end of some end factors

Like welders okay and and they Bend continuously backwards forwards backwards forwards and as a consequence those uh turn into the same phenomenon as you get using a piece of wire wool imagine you’ve got a a neat new um wire wool scourer or something and after a few

Uses it’s now a kind of prickly pad which goes through your rubber gloves and uh and scratches everything and it’s it’s is an end of life well copper braided cable does this and it even punctures its own insulators yeah that’s not good so these are the real life things

About uh machines that as they come fresh out of their box or their crate um appear to be perfect brand new um we can’t think of them degrading but my job my whole career has been about predicting when and how will this degrade and how can we detect it or even

Better prevent it so that it doesn’t break down in uh ways that are either dangerous or produce loss of production or destroy things around them and cause expensive repairs yeah so I’ve heard about this concept of predictive maintenance um would robots these days kind of I guess they’d have a lot of

Senses or whatever built into them to try and detect this rather than just waiting for a regular scheduled inspection or whatever yes to a degree um so if it’s possible to work out um that a component degrades in a a pretty regular manner then what we do is redesign it

And make sure it doesn’t do that if you can’t redesign it then you build in something that detects it and um for a long time now those approaches have been called a canary after detecting carbon monoxide and a mine and um those could be quite simple things like going over temperature for example

Um everybody’s got these in um things like Central hitting and others are more complex because in robotics you’re often looking for acceleration up to a speed some kind of continuous speed or path and then deceleration down to zero and we want some kind of static point and that’s quite hard to achieve

Perfectly so as a consequence we have to deal with things like overshoot of electromagnetic and and uh components with mass that they all don’t quite behave as you want them to they don’t accelerate properly and then they don’t stop properly and then the in particular things like a continuous path is really

Hard to achieve and the more complex it is the more um Motors and axes are accelerating and decelerating to to maintain something that we think is quite a uh a simple kind of task to program so detecting all those some of those are monitored um but essentially

They’re a closed loop system and um I detected some of this problem early on in some of our Automotive work when the Fitters who were highly skilled technicians at um repairing and controlling and um managing the robots said oh yes one of the problems with um our gearboxes and they were backlash free

Designs really clever um design but the U the robot doesn’t know where the gearbox is the Fitters have discovered that if you gave the robot a hearty shove you could get it to go around one notch on its gear and the robot didn’t know because it’s just a

Dumb machine and uh depending on where the uh the rotary encoder is to pick up that axis typically they are very close to the motors on on robots because it’s easy to assemble them there but that means that the end point is really at the end of all the SLO in

The gear boxes and the shafts and the bearings and um the the programmers have to just assume that it’s in the right place which uh it rarely is so um all of those things are kind of expected but there’s things that happen which are not expected as well and um you’ve seen um

Some control systems in the Aerospace industry recently doing some things that we really didn’t expect them to do with some really serious outcomes and um hence being able to uh to catch the unexpected is something that’s uh really important where um safety is concerned and um productivity and uh damage to the Capital

Equipment yeah the uh the unknown unknowns are always the challenging part very much so yes and uh the the uncertainties and the unknowns are um a real big issue and uh something that um I do regularly with our researchers uh particularly new PhD students is to think over how we handle that and

There’s huge enthusiasm for datab based solutions to this these days that the artificial intelligence or the Deep learning will find everything and uh I always challenge them to to think about well okay but can you pick up um something like corrosion or delamination of some surfaces metallic surfaces are really important

For keep keeping things rotating or or sliding and uh when they get damaged they they don’t progress in a nice smooth manner they Splinter off they they create spalls on a surface or um material floating around in lubricants appears as uh as bits and we count them

And um if if there’s something that parts company with its surface and ends up in a lubricant then it doesn’t arrive in ones and twos it arrives in thousands and it arrives in lumps and clots so actually data oriented Solutions aren’t that great at um predicting this but the

Reason I can predict that it might happen is that there’s an awful lot that we know about this there’s some Physics there’s some experience we’ve got good analytical models of the way these things behave so one of my big Ambitions at the moment is to try and get a a kind of

Unification of um models which are physics-based um experience-based um we had expert systems back in the 80s and 90s nobody talks about them anymore but some of them were really clever because they captured that expertise and today we can run the whole lot in parallel so if I

Get some kind of probability out of um a data oriented solution and then the expertise said oh yeah okay we’ve seen this before in this kind of pattern we’ve never seen this before um so it needs investigating further with more data more frequently and perhaps call in

People then we can get much more robust systems and uh that I think is quite important rather than uh nailing our colors simply to one M yeah yeah definitely I mean I think it’s there’s something sort of tempting and pleasing about the idea of being able to just

Automate everything and have it just you know do its own thing and leave it alone but the you can never really replace the expert of of humans yes and increasingly we’ve worked in regulated environments and uh regulations have a a distinct purpose so um you know we’ve

Heard about more regulation in food in Pharmaceuticals um lots of things in in daily life and uh I’ve done quite a lot of work with the transport industry and it’s not surprising that things in aviation and Railway transport really have important rules in place and it

Would be very foolish to try and train systems from scratch without embedding those rules yeah and there are things about um speed and force and uh checking and double cheing um and trying to deal with um the the human factors and the known degradations that uh we have with materials components and Systems that leads quite nicely into the next topic I wanted to talk about which was um an autonomous Railway inspection and repair robot you’ve been working on in collaboration with network rail um so when I think about this kind of task I think you know it seems like it brings

Together a lot of different challenges um from computer vision all the way to like dealing with weather um so can you tell me about this robot and some of its features yeah um we’re on our third generation of way robots now and um although uh we have built some uh for

Some specific tasks that there are others in the industry which are dealing with some specific areas of uh repair and uh job control um we started about five years ago with a demonstration as part of the European shift rail program which um despite that political maneuver that we

Remember um we were still in shifter Rail and um Network rail in particular was committed to a number of projects within this really big um billion Euro program and part of the area that we got involved with was about command and control systems okay the company had committed to a physical

Demonstration of jobs coming from a computer system which then needed to be transmitted to an autonomous vehicle which would then go and do its job and then report back um what it had done and to report that the job was finished or needed further action and to go back to

The the initiating um command system to to deal with that so getting from a command system a fairly conventional idea of um an industrial job control system all the way down to um physical demonstration requires some transmission of messages and deciding what’s in those messages but when they’re received

Starting the vehicle in in some kind of search pattern to go and find its location and what the um incident is that needs dealing with then some kind of action and uh finally those steps in Reverse once it knows it’s done it so in the first robot we looked for simulated faults which

Basically were a red feature and the whole thing was done at scale um we didn’t do it a tiny scale we did at a medium kind of scale um at about on10th scale vehicle which is approximately 5 in gauge and uh that means you can buy

The track you can buy some um components to run on it and uh we automated the vehicle uh with an Arduino but it also had a robot on board and the robot on the Arduino continuously argued about who was in charge uh but it did the job and it went

Around several exhibitions showing the demonstration that it could find the Target and uh if it was on the left rail then it um instructed the robots to do it job on the uh the red Target on the left Rail and then uh if it was on the alternative rail then then it could

Demonstrate that it it had multiple responses to to its inputs um worked very well except when we had to do a demonstration on a red carpet and then of course it picked up inputs everywhere these These are the the kind of practical concerns that one

Has to deal with yeah uh later on and just as we hit 2020 and the pandemic uh we took delivery of of a clear path robotics warthog which is one of the larger uh autonomous ground vehicles and uh this could do a lot more for us because it um

Carries a package of instrumentation for navigation and for um more precise localization and situational awareness so for example U GPS for location and navigation but uh liar and some additional sensing that we put on to detect things close up and um you mentioned cameras multiple cameras for um detecting different parts of the

View and one of them we fitted on the end of a robot arm as well on top of the vehicle um so that uh it could position the camera in different places but of course at the beginning of the pandemic everybody had to go home and we had hardly anybody in the

Laboratory what I think I might have done in retrospect is to change the wheels on the vehicle to railway Wheels but at the time we wanted to make sure that this rather expensive vehicle that didn’t belong to us um was completely um uh that any changes could be completely

Undone uh so we designed and built a trolley for the the ugv to run on and uh it used well continues to use um rollers which transmit the power from the tires of the unmanned ground vehicle the warthog um to the track through Railway Wheels through through some belt

Drives so that’s a little bit clunky however it enabled the thing to propel itself up and down the railway we had to learn how to uh get the vehicle on and off its trolley the TR we can manage with about four people quite easily but the warthog weighs about 200 kg so you

Don’t lift it your drive it it’s very agile uh but we had to get it up ramps onto the trolley but once we’d done that we were able to do a series of increasingly complex tests uh we put a piece of Railway track on um a corner of the runway at

Cranfield Cranfield has its own airport so there’s fair amount of space which is a real asset compared to a City Center University yeah so we were able to use this piece of track without having to ask anybody and if you want to go and close a railway that that is quite an

Important uh uh sequence of events to do as it’s possible to do it for for for maintenance and for changing track but uh having your own piece of track makes a big difference to to getting things ready and uh testing out all the subsystems and we needed to test things

Because um for example um GPS is fine outdoors with plenty of sky but there’s about 10% of the Railway where it doesn’t work the the reason for that is things like buildings and tunnels and cables um the tiny signals that you’re looking for from the satellites are easily disrupted by Elric magnetics so

Um Faraday cages are no good um Canyons between buildings and um cuttings um on the railway they’re no good so we really had to have um additional um packages of instrumentation like encoders uh initial measurement unit and others all of which were coming together in a data Fusion algorithm to uh estimate positions

Between GPS fixes the GPS guys also will happily sell you real time nematics rtk which gets your Precision down to 20 mm but on linear assets like Railways or highways or pipelines or electrical distribution you’d have to have a repeater every 200 meters so it’s an absolute non-starter you need something that’s more

Universal right having tested some of these subsystems um on our own site we then did a number of demonstrations Heritage Railways Heritage Railways are real regulated Railways but they’re much easier to close down than the main line and as a consequence we were able to have a whole Railway at our disposal and

Uh we’ve been to two different places the Northampton and Lamport Railway which is quite close to us and uh the Eckles born Valley Railway up near Derby and uh both of these offered us excellent facilities and uh in in brief when you’re trying to do um these kind

Of um outside activities you need to get people there you need to park them up you need to feed them um and of course you need a railway but plenty of Railway about um but um having all the other stuff nearby and um having good welfare

For the team is important I think um you also hinted at the range of skills that we needed to be able to do this so um part of my role is to bring together a team of people who can deal with the whole range of tasks that we need for

Building some of the robotics from scratch modifying things modifying commercial systems and that involves the mechanics the electrics the Computing the instrumentation and then there’s some uh systems where frankly we don’t know what we’re going to get until we get get it and hence I need really quick thinkers

Who are willing to get their hands dirty and uh fortunately at Cranfield University I have those in abundance um because we’re able to work in not just our permanent postdoc team but also um some very keen PhD students and master students I don’t have any undergraduates

At Cranfield uh if I did they would be involved too and we we’ve done some some great team efforts fantastic I mean it sounds like a really interesting and fun project to work on um especially the inclusion of the opportunity to work with a model Railways which is always fun absolutely

Um well we started um at relatively small scale um not least of which um we did a number of exhibitions in the UK um and around Europe and having things at scale means that it could be packed in a a big kind of flight case and uh shipped

Out I mean it’s a van load rather than yeah rather than a suitcase full um but you can show a lot of um of activity and a lot of proof of concept at scale but then the railway Partners wanted things showing in in real life so having built

The ugv version we then went on to um convert a more proprietary rail trolley now of them have petrol or diesel engines but there’s a new version from a company in the UK that has electric propulsion but it really is a dumb track trolley its main control um other than

Peeping its horn it main control drives it away from you right and uh you can load it up with uh up to a ton of materials and tools and uh if you want it to drive in the other direction it literally unscrew the hand control put

It in the other end of the vehicle and then drive back always driving away from you so um without any technical drawings because we were not allowed access to them we needed to get into the circuitry that dealt with the propulsion and the braking the warning lights and the

Warning Sounders and uh we put in interfaces to uh control all of those with suitable controllers and amplifiers and uh then the top layer of computing is a very sophisticated um high-powered um bit of portable Computing with lots of different inputs and outputs and then we retrofitted all

The instrumentation that we’d had um specified on the warthog and uh put that in from scratch into the uh the proprietary track trolley so that takes the form essentially of a um 4T Square base and any payload can be applied to that and uh back in May last year we

Demonstrated a u an Ultrasonics payload from a well-known Railway manufacturer mature technology but it uh it needed to be tested in this um robot control environment so we did that at uh worksworks born Valley Railway and uh you mentioned things like waterproofing earlier you’ve never seen so much rain in two

Days it absolutely tipped down and in the um in the gaps between the rain well we did operate during the rain everything was waterproof it’s got to be in this country well yeah um hot and cold and waterproofing uh vibration and dust these are normal things that um the

Operational equipment has to deal with and um after we’ve finished doing proof of concept then the next stage is to think about how do we get this into the supply chain um and uh that needs a good specification and so proof of of specification is is a lot of the job

That we’re doing um but yes getting everybody up there in their um uh protective equipment everybody knows what PPE is is these days but in in that case it was orange gear um plenty of warm layers and gloves um hard hats and in terms of um welfare

On some of these sites we need to get gazebos for shelter for some of the equipment but mainly for people and plenty of flasks of hot water for tea and coffee and batches of sandwiches in because you’ve got to get real people to do these real jobs and to look after

Them while we’re out on side so much more to it than just pure robotics ABS well to some extent by that point the robots is the the robot is the easy bid Andrew it’s been great chatting to you today I’ve been speaking to Andrew star a professor of Maintenance systems

At crownfield University thank you thanks CLA thanks for listening if you enjoy robot talk please do share the podcast subscribe and leave a review to let us know what you think and make sure you check out our social media channels to see a photo of the latest version of the Railway inspection and repair robot that Andrew

Is working on we’re at robot talkpod on all the socials next week I’ll be talking to Patricia Shaw from abis withth University all about home assistance robots and robot Learning and Development until then I’ve been CLA Asher and this has been robot talk robot talk is brought to you by the

Hamlin Center Imperial College London