This is a really good video about how the accessibility of building automation with tools such as Raspberry Pi and Python is so much greater than people realise. This most interesting point it towards the end, this was built in-house by an employee, they didn't go out and bring in an automation consultancy, they used their own experience and enabled their own employees to build it.
Transcript of relevant part:
> The funny thing is, people too often jump for the consultant and to go find the 200 quid an hour amazing mega person.
> But you have universities, you have young talent bubbling up, you have talent inside your own staff.
> You have got to innovate. If all you do is just chuck money at consultants you're not going to be beating the competition.
> You've got to come up with new ideas and suck ideas from different places.
> And this is just one example where one of our staff unlocked huge potential in our organisation.
I have given talks to factory managers about finding that one or two employees that know Arduino/Raspberry for these projects. Every factory has such persons. They would love to do it!
The reception on this is mostly neutral or negative from management. When you identify these guys, turns out they have great hardware or programming skills, but no integration or architecture concepts. The existing system (ERP, MES) owners also like to put a lot of red-tape around anything unconventional connecting to their systems.
Managers also are held accountable for the project, even if the costs are low. They prefer to spend a lot and have a consultancy to blame on failure, then their internal employee project they selected.
What consultants do is they help the company navigate within it's own bureaucracy and processes. Even to implement the Raspberry solution.
I work in this area and know the mindset of ERP and MES folks. It's great for long-term volume production of stuff that's very well established, but makes innovation really, really intractable.
That said, lots of people focus on how inexpensive things like the rpi are, but whether the unit costs $35 or $3500 makes a very small difference in the overall cost of the project. I am not totally against rpi's for factory stuff, as long as the folks who implement using those can slow their roll and think some more about the lifetime of their automation project and it's future maintenance.
There are PLC's that can run FOR DECADES without problems in a factory environment. Will an arduino/rpi be able to do that? Will someone 15 years from now know even know what an rpi is or be able to buy one?
I see so much stuff reinvented using raspberry pi, arduino and sparkfun sensors simply because people have exposure to those and don't know what a PLC is. And PLCs don't even have to be expensive, a $200 one from automation direct would often be enough.
PLCs are often a pain to integrate with multiple systems and are not nearly as versatile and simple as an internet connected pi running python, for certain desired tasks.
Please tell me which PLC you can get for $200 that runs python, can query some REST APIs, query and write to some databases, communicate with several other types of PLC and equipment, check a half dozen types of local sensors, and perform complex operations without purchasing a software subscription or having to be programmed in ladder logic.
Obviously they are suitable to different tasks, but there's a reason someone might reach for a PI to integrate new functionality, including on top of an existing PLC. They're tremendously handy in this environment, not just used because people don't know better.
I mostly agree with you and perhaps this was the wrong article to put my comment on because I think in this case they did well with raspberry pi.
> without purchasing a software subscription or having to be programmed in ladder logic
I am thinking of cases where a small PLC, some industrial sensors and ladder logic is a good, reliable and easy solution, but someone reaches for a raspberry pi and python scripts because that's what they've been exposed to.
I agree adding something on top of an existing PLC can be great too. In my first job nearly 20 years ago, I did machine automation and we always added an industrial mini ITX PC to our PLC panels. This cheaply added datalogging (an advertised feature for the customer) and remote programming capability (usually an unadvertised feature that made us look really impressive when we could fix something immediately without going to the very remote site). Depending on what was available on site, we could connect remotely by the site's internet, by a modem/landline, or by a cellular radio (configured as a modem, as this predated reasonable prices for data plans). Sometimes the PLCs were from the 80s and none of the competitors would quote anything except tearing everything out and starting over, but would could find a way to add to the old system.
Connecting remotely by some official upgrade from the PLC manufacturer was usually technically possible, but impossible in practice to get authorization to gain access to the network, and get things configured to allow access to the PLC from offsite. But dialing up to our own installed modem bypassed all of that, and since a PC is versatile, there was always some way to bridge the PLC's programming interface to the PC.
I recall that we had to go out of our way to find PCs that didn't look like PCs and were DIN-rail mounted. Anything that looked like a PC would eventually get messed with by someone.
WAGO pfc controllers series can do everything you mention. They are "just" Linux boxes with a fully featured integrated I/O bus (with any input/output type you can imagine). They can be programmed in pretty much any language and you can deploy with docker.
That said. Really large factories use DCS. (Distributed Control System) they are horribly expensive but integrates everything in one system. Controller programming, Human machine interfaces, MES and even ERP. Best known to me is ABB 800xA and Siemens PCS7.
I think is a bit nuanced than that and there is a clear reason why a manager might go for a contractor instead of in house.
A good manager will have a clear understanding of his slice of the company, what work needs done, what are the hoops, what problems can arise, how the work needs done. Same manager most likely has no understanding what it means to build, deploy and maintain a hardware + software thing. And since we are talking about a good manager, they are aware of that. Hence they call for a contractor that will build, deploy and maintain said thing.
And lets face it, will these tinkering workers actually able to deliver a robust system? You already mentioned the lack of integration or architecture concepts. The red tape might be there to prevent data corruption. What user interface are they going to use? Fragile touchscreen or a big button to mark the end of the step (eg, after the worker finished mounting the mud guards). Will that button/touch screen survive the number of mud guards that need to be mounted every year? Or that prosaic issue: where do you store the spares? etc etc There is a reason why this kind of innovation, like automatic testing, deploying etc, happen in software companies, building, deploying and maintaining software is the core competency.
I don't doubt there are happy cases where the tinkerer is more than a tinkerer, the video we are commenting on proves that.
On the other hand, this sort of thing should be encouraged. Have the workers come up with POCs, then pass that to the expert contractor to help with the extra bits. When the startup I work for decided to pivot, I suggested to build a framework to help with automation: have a server that boots remote thin clients and have a basic client that you can plop all kinds of peripherals in. I had to use cases in mind: automation for small factories and bar/restaurant management. We decided to count trees instead.
In my (limited) experience this type of thing is very common on the design side but much more rare in manufacturing. An arduino/raspberry pi lets an ME/EE with limited software skills get a lot of test automation done.
Last big company I worked for had a large automation team where everyone did a lot of consultant managing and zero engineering. The culture clash was rough during design handoffs. They’d come back to design after their three station monstrosity with 8 motors, hundreds of pounds of steel, and 12 pneumatic lines was failing to do the one simple task it had been over designed for. Another more specific highlight was getting a quote of around $50k to add an ambient temp sensor to one station.
Yes to all of this. It's something British businesses have traditionally been bad at, recognizing and developing the skills and knowledge of their staff, but it looks like Brompton are setting an extremely good example here.
I suspect they're also small enough that this kind of thing is easier to do without politicizing within the organisation; you can have individual line workers talk to the CEO. It's also an unusual corner of skilled "mass-custom" manufacture: rather than turning out millions of identical bikes, they turn out thousands of bikes to millions of slightly different specification combinations.
I am really tired of how bad Raspberry Pi are. It is basically overpriced hobby board. I would not recommend building any sort of industrial solution on top.
USB stack uses a lot of CPU, a lot hardware is missing and is software emulated. Some USB 1.1 keyboards do not work at all (2.0 are fine).
Constant issues with power hungry peripherals. Unstable USB connections. WiFi has tiny antenna and is prone to interference...
Software "support" is limited to some strange Debian version. Video playback acceleration works, but good luck dealing with outdated packages. I tried to use Pi for router, and included NetworkManager could not even handle WPA2 (only WPA1). No hardware acceleration for encryption, goodbye disk encryption or SSH...
And it is bloody expensive, for 160 euro (4GB, PI 4), I can get much better x86 mini computer, that runs any mainstream distro, has way more stable hardware, and comparable power consumption. Pi 4 B has Broadcom BCM2711 CPU that is 28 nm...
>I am really tired of how bad Raspberry Pi are. It is basically overpriced hobby board. I would not recommend building any sort of industrial solution on top.
But Raspberry Pi is intended to be a hobby board. Nobody claimed it is of industrial level and strength. Because some people try to use it instead of more robust and expensive solutions, sometimes with success, it doesn't mean it is the answer for everything.
And yet it's astonishingly successful, because it's a de-facto standard. Getting an industrial PC with comparable IO is quite tricky. Although in this instance I don't think they're using the IO, just a screen.
> for 160 euro (4GB, PI 4)
I see the scalping prices are still in play because it's so popular.
Well, in fact that's not the price for an RPi 4 4G, that I think it's around 70-80. But the low stock is used by some people to list insane prices.
Also, NetworkManager supports even WPA3 as a client, but I don't think it's the best option for building a router. It's better suited for a regular client. Anyway, if you were building a router in a small board like the RPi4, I'd recommend you to use OpenWRT instead of Debian. Also, if you want a "more standard" distro, just use Armbian. It's just plain Debian with ARM support.
Yes, the RPi is scarce, expensive and not super powerful, but also its target is not being a wireless router at all: only one interface, no external antennas... It's the wrong board for that.
Get a GL.iNET board with OpenWRT or a x86 multi interface miniPC for that.
> low stock is used by some people to list insane prices
Another reason not to use it!
> NetworkManager supports even WPA3
I used it as a hotspot. Armbian has some weird obsolete NetworkManager, that does not work with WPA2. OpenWRD does not support other functions I need.
And I did not use buildin networking managment, because it keept switching network interface names (multiple WIFI and LAN cards over USB). I can bind NetworkManager config to MAC and skip interface names completely.
Perfectly valid reason! I in fact am buying Orange Pi SBCs this days
> I used it as a hotspot. Armbian has some weird obsolete NetworkManager, that does not work with WPA2. OpenWRD does not support other functions I need.
I don't follow that. Armbian has the same package versions as Debian. I just checked on an Armbian Bullseye I have in an Armv7 board and it has support for WPA2 Personal, WPA2 Enterprise, and WPA3. If you want to keep a regular Linux distribution probably your best bet is using something like 'hostapd'
> And I did not use buildin networking managment, because it keept switching network interface names (multiple WIFI and LAN cards over USB). I can bind NetworkManager config to MAC and skip interface names completely.
You can set those names static easily with systemd. Create a file for every nic like '/etc/systemd/network/10-eth10.link' and fill it like:
Well, it depends for whom. Also, you need adequate software and hardware. About wireless, the chipset in an RPi4 is designed for clients, not APs, and its capabilities on having multiple clients are quite limited:
Routers like the Omnia Turris, Mikrotik, PC Engines... use specific mini PCIe cards that can be as expensive as a full Raspberry Pi 4, just for the wireless.
Intel NUC Kit NUC7CJYHN Barebone (Celeron Dual Core-J4005) barebone is sold for 112 euro. Add RAM and SSD...
> PS: PIs are usually a lot cheaper.
No they are not. Maybe a few years ago. But PIs use obsolete processor, with limited manufacturing capacity. I do not believe they will get any cheaper any time soon.
That's a barebone kit without HDD or RAM... (so stack an extra 100 bucks on there, and if you're paying 100 for a pi, it already includes an SD card for storage). So no, it's WAY more expensive.
Not to mention you're going to be looking at quite a bit of fiddling to get any sort of linux distro playing nicely on it, so from a tooling side, costs will be much higher.
It's also literally 5 times bigger than a pi in a standard enclosure, so enclosure costs and space costs.
Plus, it idles at the same power (~10 watts) that the pi pulls during max load. It maxes out at ~70 watts draw. 10 watts over 5 years = $35. 70 watts over 5 years = $250.
Plus - I refuse to touch celeron with a 20ft pole. They're consistently one of the worst performing processors I've ever had the displeasure of using. You call the pi's hardware old, but the celeron in the model you linked is 5+ years old now, and it's the shittiest of the shitty models that intel sells.
You can hate on the pi all you want - I tend to agree, they're a bit overpriced for what you get at the best of times, and when they're both expensive and hard to get... alternatives are a great option. But that hideous thing is not the alternative you should be suggesting.
8 GB DDR4 is 20 euro. Another 20 for basic SSD. Raspberry PI I found does not have a case, no CPU heatsink, no power supply, no SD card... And running system out of writeable SD card is another joke!
Not sure how I would "fiddle" with installing Linux distro. There is nothing out of spec, very common intel hardware. With PI3 and Raspbian, my keyboard did not work, because Pi does not support USB 1.1 correctly!
This system idles around 5 watts, depending on peripherals. Max consumption for PI4 is 15 watts.
Celeron in this system is a good CPU for small servers. Look at Serve The Home reviews. But main reason are peripherals, you get Intel chipset, Intel network card, Intel USB and so on.
You do not have to deal with shitty chinese chipset from TVs! It does not even implement spec correctly! Initial revision of PI4 would not even power up, because most basic USBC power delivery (single resistor) was botched!
This really makes me mad. The idea that you can buy a bunch of terminals and, Voila!!, you've got an awesome production control system is utter crap.
I've worked on systems like this and it takes a LOT OF SOFTWARE to drive those Pi's. If those folks spent less that a hundred grand on that, I'm really sad about how badly they screwed their programmers. If they do not have full time staff supporting it, I'm really sad for their factory staff because the stuff is getting worse every day.
The fantasy that software is a secondary consideration ("it cost pennies!!"), almost free is infuriating.
On the one hand, I totally agree. On the other, the project as described is well within reach of a small team (1-5 people) of programmers that could pull it off in 3-12 months for under $100k (maybe even closer to $10k-$20k).
The way the guy was talking, it sounds like he had one of his employees who also happened to be studying CS whip up something that works while still paying base salary rates. If so, this means the owner got an insane deal by paying a full time programmer worth maybe $100k-$200k at a $30k rate.
In the end, I think you're right. The guy probably got an insane deal because he convinced one of his employees to do free labor. I do wish projects like that weren't so expensive. I'll also point out that for a savvy owner who has some technical experience, this is also a very feasible (relatively cheap) project.
If I was buying this solution and you told me you would do it for $20k, I would not consider your proposal for one more minute. I have written things like this. A decently robust solution takes a lot more than that. If you are doing a burdened rate of $100/hour, that's 200 hours, five person weeks. That's barely enough time to get a working MVP, let alone test thoroughly and deploy. Let alone implement management tools to input the orders and the screens for the workers to access and update, let alone the reporting and archiving. Let alone maintenance and support.
And, if you are paying $100/hour burdened, you are hiring people for $50/hour or less. Either you are hiring newbies, in that case, who is doing the design, or you found somebody desperate and are taking advantage.
Which is to say, the assertion of 1-5, $10-20k makes me mad, too, as another form of devaluing the profession of software development.
I should have specified in more detail, my apologies. The number, $10k-$20k was in hardware costs, separate from labor costs. So, potentially $10k-$20k in addition to whatever labor costs there are.
I didn't say a thing about being mad that they had a solution. I am mad at the guy for claiming that this tremendously useful tool could be had for "pennies".
I will grant you this. A hacker, meaning a teen in a basement making it work in the minimum way, could create a thing like this.
I'm not that. The person who did this for that cycle company is not that (because that low-class hacker couldn't procure and implement dozens of pi's, or the networking, or the training). And, if he or she could, the company would be nuts to base their business on it.
IT systems like this are hard and expensive. Suggesting otherwise is, I think, irresponsible. Someone that sees that video, believes it's just "pennies" and goes to his boss, promising to accomplish something like that is going to fail and be harmed.
Perhaps a good idea to keep in mind this is a Brit speaking, and we tend to have a lot of understatement in our communication. I interpreted 'pennies' as 'a good bit less than if we'd been stupidly-overcharged by the mercenary consultants he refers to'. I don't interpret 'pennies' as one of their employees getting screwed over. I mean, they may well might of, or they may well have had a significant wage increase and new clout within the company... .
I worry about this. It compromises the brand if they are willing to cheap out on automation hardware at this level. What else are they cheaping out on?
Point being there's almost absolutely no reason why they should have used a raspberry Pi in this environment. They have poor power management, EMI resilience, terrible enclosures as a rule and are somewhat prone to fucking out spontaneously with things like SD corruption. Your warranty and service options are basically the trash can. If a station goes down, what happens to your line? The entire cost-benefit analysis is probably wrong here. If it were me I'd buy one of the cheap Lenovo industrial mini PCs and get 5y NBD warranty, actually tested Ubuntu support, without all the other issues and buy an extra one.
They have the cash sitting around to do this properly.
Incidentally also none of this is really special. We were doing exactly this with recycled PC desktops 25 years ago running stuff in VB6 on windows NT on much larger and higher revenue and audit-heavy engineering lines.
The software is a separate issue. Buying that in is usually expensive and problem-ridden.
I think a more positive take away is that this is a relatively small traditional manufacturing business with a few hundred employees that managed to automate a lot of their workflows at a small captial cost using off the shelf components and a willingness to take a stab at skilling up their own staff to build a solution to solve their usecase.
The alternative is hiring a consultancy that may or may not actually deliver anything worthwhile at the cost of many thousands a day and leaving you with a solution that you cannot maintain because you completely lack the internal compentence as a consequence of outsourcing all of the development in the first place. More small/medium sized businesses should take risks like this.
We should honestly be questioning, esp in the UK, the degree to which we depend upon consultants for seemlingly everything and how little businesses and government are willing to invest in taking small bets on their own employees over hiring frankly overpriced consultants from Accenture or McKinsey who are often expensive suits backed by a bunch of very recent graduates with little or no experience.
But a part that's missing is the reason WHY companies often prefer to hire inefficient outside consultants that will under deliver, over promise, and often leave you with a half functional system at the end.
It's for class politics and control reasons that managers and executives don't want to invest in on-site talent to expand their skills. They want disposable labor even if it costs them more money in the end, because it's a decision motivated by their power interests. They don't want to depend on on-site talent, they don't want to promote you, they don't want to pay you what you are worth, they don't want your talents to develop.
They want to pay an outside agency whatever amount of money for the promise of a system, because even if the consultant fails they have fulfilled their capitalist due diligence and did things the right way, rather than having had local employees that are crucial to you that you invest in do work for you.
If you take issue with this characterization, I'd ask you why do many people here find it notable that this company had an existing employee develop a system rather than just dragging in the consultants.
I don't disagree that there might be a class issue involved esp in UK workplaces, but there is also this mystery that it seems like everyone knows that the big consultancies fail to deliver much more often than than they succeed by everyone except for the people in management with the budgets to hire these people.
They are parasites on both the private sector, public sectors, and even government policy development. I actually suspect its corruption more than classism.
I am speaking as someone that used to be one of those pink face graduates behind the expensive suits that was horrified at how hollowed out so many of our publically traded companies and government are.
What it shows is just one thing: that having a 'call for pricing' blurb on your website, which is pretty much the norm for anything beyond the most trivial industrial automation hardware is a showstopper for many people. They just want to solve their problem, preferably today but tomorrow will do. Not go through the endless filters-of-revenue-extraction that the regular industry purchase trajectory will subject you to.
Exactly the same thing happened in the 80's with personal computers. Suddenly you didn't need to wait for the accounting department to write their purchase order after 'their' sales department had - reluctantly - come to an agreement with 'your' bean counters about what the budget was and how you were going to have to do with less of what you needed but more than what they really would have liked to give you.
A RasPi fits within any discretionary spending budget, there likely isn't even a write-off on them, it's just booked as costs in the year it is expended. And associated bits and bobs for sensors and actuators will be bolted on as needed.
In the end it may not be the best or the cheapest solution but "now" beats "someday" in many industries. I personally see this as exactly the reason why Brompton is the breakout success that it is: they show adaptability rather than to play by the book.
Well it's not really though. You just buy it from Farnell or RS here in the UK and it arrives the next day. I mean you literally just buy an Omron PLC and the ethernet board and RTFM much like you have to do with the Pi.
Your point about the cost is right but you have to ask where the cost is: some guy spending two months working out how to interface a Pi to something or this being canned and your ask being process modelling. At the same time that guy is then upskilled in fairly generic automation which can be used at other levels across the business. It's very easy to write off unproductive uses of employee time.
Brompton is a success but I'm not sure why. The bikes are by far the most horrible thing I've ridden other than the Sinclair A-bike. I'd rather walk.
Omron doesn't offer an $80 PLC with gpio that runs python and can directly integrate with your ERP and other software like REST / SQL that anyone could teach themselves if they know python. If it takes two months to create a fully developed local, customized visual build out and assembly accounting software stack that also interfaces with torque sensors and serialization like this one, why is that any kind of a loss? How is integrating a pi with other systems any less valuable of a skill that isn't applicable across other sides of the business, compared to the very narrow task of integrating slow moving PLC hardware that has clunky ass proprietary software most of the time?
I don't know if you've worked in industry, but companies actually won't invest in their local employee to get trained on 'industry standard' automation hardware. They contract that out. So you have no one on site who knows how anything works, half the time. The choices are often "pay s ton and have all the traditional investment of time in a contractor setting up PLC hardware and finding external software integrators to do all the other actual useful stuff you want, or just don't do the project because all of that is expensive and resource intensive."
There is no way a Pi with GPIO should be anywhere near any industrial equipment.
I have worked in the industry and it has been a mix of in-house and hired cabinet chimps.
If you have the time and money to work out how to interface a Pi to GPIO and write a software front end, you have the time and money to buy a PLC and integrate that with a COTS HMI front end.
> There is no way a Pi with GPIO should be anywhere near any industrial equipment
That's a pretty strong unqualified statement. I disagree. There are lots of ways in which a little Unix SBC can find a role in industry, lots of that is non critical but very useful, such as impromptu data logging and visualization. I agree that a Pi is far from ideal but but the time you have an industrial computer with the same specs as a Pi you're looking at a massive increase in cost and that may mean that the whole thing gets canned. Something is better than nothing. The 'interface' to a Pi is usually a solved problem in that you can just attach a bunch of pre-cooked stuff to it. And some of the industrial Pi models are surprisingly well hardened. There are literally 10's of thousands of Pi's chugging away happily in various guises in all kinds of machinery. I have worked 'in industry' as well and I don't have any holy houses. It either works, or it doesn't and then it gets replaced.
And so on. These range from 'just pretty packaging' to 'opto isolated IO + stable built in supplies'. Once you have stable power, isolated IO and good storage (all three are known causes for problems when you take stuff into harsh environments, and the Pi is pretty critical when it comes to dealing with these) you are still lightyears ahead of where you'd be with anything the competition fields with respect to available interfaces, development environments, display options etc etc.
Compared to your average Siemens, Omron, Allan Bradley, ABB or whatever floats your boat's toolchain the Pi rocks, you have access to pretty much all of Linux. The biggest problem with Pi's is availability (but even there: all manufacturers have had their issues on that front too).
Anyway, long story short: let people do what they think is right, if they get it to work reliably and affordably then so much the better. Industrial automation hardware has always been ridiculously expensive and underpowered for what it offered, some disruption certainly won't harm them.
Brompton is a success because it solves an actual problem that people have: last mile transport in a way that actually works, does not require batteries, is reliable, reasonably quick (I have one and once you're used to it you can make pretty good speed on them) and it folds up to the smallest package on the market that is still practical. You can take it with you on public transport here in NL and it doesn't classify as a bike but as a carry on which means no extra arrangements, just walk onto the train and unfold at your destination. It takes 30 seconds and even though they are a bit heavy they are well designed.
Raspberry Pis are awesome. I would buy 5 for various personal projects right now, if they were in stock. It's not just post-COVID supply chain issues either. They have had several big gaps where they were basically unavailable, starting with the very first Pi 1. I would be very hesitant to build a business solution on them today given that I don't know where I would get more in the future.
I love mine too, but what really transformed everything was converting it to an electric bike (via the expensive-but-effective Swytch pack). Now I can easily go to-and-from work, or up a bigass hill, with zero sweat.
Transcript of relevant part:
> The funny thing is, people too often jump for the consultant and to go find the 200 quid an hour amazing mega person.
> But you have universities, you have young talent bubbling up, you have talent inside your own staff.
> You have got to innovate. If all you do is just chuck money at consultants you're not going to be beating the competition.
> You've got to come up with new ideas and suck ideas from different places.
> And this is just one example where one of our staff unlocked huge potential in our organisation.