6 ms·

What techies keep getting wrong about industrial automation

37 points by wolframhempel 3y ago

typhonic 3y ago
Twenty years is a mighty short time period for expecting a huge paradigm shift in industry. In 1973 I was touring industrial plants and got to see an automated retrieval and delivery system operating in a plant warehouse. (It might have been the Copolymer plant in Baton Rouge - hard to remember.) So even 50 years may be a bit too short. Looking back a few hundred years it is easy to see that we have made progress. In the mining industry, for example, the need for large haul trucks is reduced by long conveyors. The longest I have seen is two miles. Those conveyors are barely attended to by operators and their operation is monitored by control systems connected wirelessly and reliably. The real leap will come, not from increased automated operation, but from automated maintenance. That won't happen until well after IOT is more fully implemented. By that, I don't mean more devices connected to the Internet. In my opinion, IOT's potential lies in the future of the "Things" making decisions based on information from the connected devices.
- HeyLaughingBoy 3y ago
  I can agree with that. It was about fifteen years ago that a colleague (he was a reliability expert) and I did a presentation on using remote monitoring (I'm not sure that the IoT terminology was widespread at the time) to perform predictive diagnostics on our products in the field. And it was probably another five years before that when I led the first team at the company to implement the remote monitoring project that his data was based on. I don't work there anymore but I'd bet that not much has changed since then. I was lead developer on another IoT project about 3 years ago, and besides using cell modems instead of Ethernet, because the machines in this case were in geographically remote regions, the state of the art had barely changed.
  - DanielHB 3y ago
    I started working in the IoT field a few years ago as well and I am surprised by how hard it is to get an IoT project going compared to other cloud computing stuff...
    - airbreather 3y ago
      A lot of people think that Industrial Automation is subset of IT, but there is a big difference between IT and OT and where are only just beginning to see any real convergence emerging.
random3 3y ago
I'm not following this line of reasoning >. It can transport a payload of 363 tons and costs around five million - which comes out to around 14k per ton of payload. Did the unnamed author divide the price of the truck to the weight of a single load? If so, ignoring that it's not specifying a distance and hence the 1:2 cost ratio may be highly inaccurate, why? Is that some industry standard? It would be weird not taking into account amoritization cost/period too. Finally, in conclusion, what do "techies" get wrong?
- airbreather 3y ago
  If by techies they refer to IT, then just about everything. I have seen many, many cases over the decades where OT was confused for IT by management and the extensive OT considerations were ignored or had to be re-learned when an IT approach was taken to OT problems. Often the IT based solution in OT world never ended up working at all and had to be discarded, or re-engineered from the ground up. Just because you have boxes with screens connected by blue cables, history has proven many times that this does not make IT the best people to architect and implement your OT system, by a long, long way.
- ArekDymalski 3y ago
  > Finally, in conclusion, what do "techies" get wrong? I read the answer as "there are several factors and broader context besides direct productivity gains that need to be taken into account when considering automation". And to be honest some techies do indeed forget about them. Elon Musk seems to be the the most striking example of "it's just an engineering problem" mindset.
- karmakaze 3y ago
  I don't know either but "[Hivekit] is trying to facilitate this."
- nyrikki 3y ago
  They chose the Caterpillar 797F, which is their largest offering. Not sure how they even think that the hoe that is loading them will also be able to load on-highway trucks. It would be like loading water glasses by tipping a 55gal barrel. The operator inefficiency of dumping into 20 trucks for an equivalent of one would reduce the time the hoe operator could actually mine. As they are obviously claiming to solve things for very different use cases it looks like they have practically zero knowledge about their customers needs.
3seashells 3y ago
[dead]
nyrikki 3y ago
> So, while various companies have set out to automate the giant dump trucks, having a continuous chain of smaller, self-driving trucks using standard parts and without the need for a driver cabin would be much more efficient So, to anyone with dirt experience, hivekit just called themselves out as being the techies that don't get it. On-road trucks are fragile and require highly attentive drivers to run without destroying the trucks off-road. Everything breaks in the dirt, and almost all mining operations need very nuanced knowledge to conduct safely, for people or machines. > Areas, where traffic is highly structured and controlled. That isn't mining or the oil patch. You are dealing with mud, dropped rocks that will take out brake chambers, continually changing conditions etc... The Moravec Paradox hits all these attempts, because people without domain knowledge under estimate the skills required. While minimizing operator count is useful, this isn't Minecraft. It takes a lot of skill and being adaptive changing conditions to even different with an excavator. Some mines with consistent material properties and safe conditions do automate digging up a stope, but most mining requires expertise and informed judgement. Even in the oilfield, drilling is a high skilled job, were you have to adopt to changing conditions to avoid breaking the drill stem, which is expensive to fish and fix. Even robots in warehouses are a challenge and that is far closer to the authors claimed conditions.
- atoav 3y ago
  Playing the devils advocate here, beancounters would now say that wherever any true disruption happened there were domain experts that told them it couldn't be done. And they are not completely wrong. If you asked any decent professional whether their work could be replaced by automation, they would probably give you reasonable sounding arguments why it is not that simple. But as a techie that has on the ground experience with sensors, electronics, computer vision, programming and grew up close to mines I think it is key to realize that there are hard problems and there are hard problems. And mining automation is the latter.
  - nyrikki 3y ago
    In mining, the bean counters would use child and forced labor if they still could. But truck shovel routing is a version of VRP, trucks and shovels need to be matched or the approximations rise into the double digits of inefficiency. If you look at the powder river area where they use the model they mentioned, truck shovel is for removing the spoil on top, and thus is a dynamic environment. The actual coal is moved by conveyer belts at the mines I have been to. VRP is nontrivial and the bean counters know this. Note this fairly readable paper and the complexity of 2VRP vs 3VRP vs kVRP. https://www.sciencedirect.com/science/article/pii/S0166218X04002860 https://www.sciencedirect.com/science/article/pii/S0166218X0... If they have some new method, it won't apply across surface, sub-surface, and oil at the same time unless P=NP.
c_o_n_v_e_x 3y ago
>What techies keep getting wrong about industrial automation 1. Internet connections can be some combination of slow, unreliable, intermittent, or expensive. They also may be completely unavailable, although this is changing in the era of Starlink. 2a. Consumer grade compute hardware dies a quick death in harsh environments. 2b. Consumer grade compute hardware typically has more compute power than industrial rated computers. I've seen a handful software engineering teams have to drastically rebuild software because compute resource constraints were not realized until it was time to deploy software at the edge. 3. You are not allowed to push updates to your software willy nilly. 4. Service tech callout and installation costs are a big consideration for remote assets (oil wells, pipeline compressor stations, etc.), particularly at scale.

37 points by wolframhempel 3y ago

Twenty years is a mighty short time period for expecting a huge paradigm shift in industry. In 1973 I was touring industrial plants and got to see an automated retrieval and delivery system operating in a plant warehouse. (It might have been the Copolymer plant in Baton Rouge - hard to remember.) So even 50 years may be a bit too short. Looking back a few hundred years it is easy to see that we have made progress. In the mining industry, for example, the need for large haul trucks is reduced by long conveyors. The longest I have seen is two miles. Those conveyors are barely attended to by operators and their operation is monitored by control systems connected wirelessly and reliably. The real leap will come, not from increased automated operation, but from automated maintenance. That won't happen until well after IOT is more fully implemented. By that, I don't mean more devices connected to the Internet. In my opinion, IOT's potential lies in the future of the "Things" making decisions based on information from the connected devices.

HeyLaughingBoy 3y ago
I can agree with that. It was about fifteen years ago that a colleague (he was a reliability expert) and I did a presentation on using remote monitoring (I'm not sure that the IoT terminology was widespread at the time) to perform predictive diagnostics on our products in the field. And it was probably another five years before that when I led the first team at the company to implement the remote monitoring project that his data was based on. I don't work there anymore but I'd bet that not much has changed since then. I was lead developer on another IoT project about 3 years ago, and besides using cell modems instead of Ethernet, because the machines in this case were in geographically remote regions, the state of the art had barely changed.
- DanielHB 3y ago
  I started working in the IoT field a few years ago as well and I am surprised by how hard it is to get an IoT project going compared to other cloud computing stuff...
  - airbreather 3y ago
    A lot of people think that Industrial Automation is subset of IT, but there is a big difference between IT and OT and where are only just beginning to see any real convergence emerging.

random3 3y ago

I'm not following this line of reasoning >. It can transport a payload of 363 tons and costs around five million - which comes out to around 14k per ton of payload. Did the unnamed author divide the price of the truck to the weight of a single load? If so, ignoring that it's not specifying a distance and hence the 1:2 cost ratio may be highly inaccurate, why? Is that some industry standard? It would be weird not taking into account amoritization cost/period too. Finally, in conclusion, what do "techies" get wrong?

airbreather 3y ago
If by techies they refer to IT, then just about everything. I have seen many, many cases over the decades where OT was confused for IT by management and the extensive OT considerations were ignored or had to be re-learned when an IT approach was taken to OT problems. Often the IT based solution in OT world never ended up working at all and had to be discarded, or re-engineered from the ground up. Just because you have boxes with screens connected by blue cables, history has proven many times that this does not make IT the best people to architect and implement your OT system, by a long, long way.
ArekDymalski 3y ago
> Finally, in conclusion, what do "techies" get wrong? I read the answer as "there are several factors and broader context besides direct productivity gains that need to be taken into account when considering automation". And to be honest some techies do indeed forget about them. Elon Musk seems to be the the most striking example of "it's just an engineering problem" mindset.
karmakaze 3y ago
I don't know either but "[Hivekit] is trying to facilitate this."
nyrikki 3y ago
They chose the Caterpillar 797F, which is their largest offering. Not sure how they even think that the hoe that is loading them will also be able to load on-highway trucks. It would be like loading water glasses by tipping a 55gal barrel. The operator inefficiency of dumping into 20 trucks for an equivalent of one would reduce the time the hoe operator could actually mine. As they are obviously claiming to solve things for very different use cases it looks like they have practically zero knowledge about their customers needs.

3seashells 3y ago

[dead]

nyrikki 3y ago

> So, while various companies have set out to automate the giant dump trucks, having a continuous chain of smaller, self-driving trucks using standard parts and without the need for a driver cabin would be much more efficient So, to anyone with dirt experience, hivekit just called themselves out as being the techies that don't get it. On-road trucks are fragile and require highly attentive drivers to run without destroying the trucks off-road. Everything breaks in the dirt, and almost all mining operations need very nuanced knowledge to conduct safely, for people or machines. > Areas, where traffic is highly structured and controlled. That isn't mining or the oil patch. You are dealing with mud, dropped rocks that will take out brake chambers, continually changing conditions etc... The Moravec Paradox hits all these attempts, because people without domain knowledge under estimate the skills required. While minimizing operator count is useful, this isn't Minecraft. It takes a lot of skill and being adaptive changing conditions to even different with an excavator. Some mines with consistent material properties and safe conditions do automate digging up a stope, but most mining requires expertise and informed judgement. Even in the oilfield, drilling is a high skilled job, were you have to adopt to changing conditions to avoid breaking the drill stem, which is expensive to fish and fix. Even robots in warehouses are a challenge and that is far closer to the authors claimed conditions.

atoav 3y ago
Playing the devils advocate here, beancounters would now say that wherever any true disruption happened there were domain experts that told them it couldn't be done. And they are not completely wrong. If you asked any decent professional whether their work could be replaced by automation, they would probably give you reasonable sounding arguments why it is not that simple. But as a techie that has on the ground experience with sensors, electronics, computer vision, programming and grew up close to mines I think it is key to realize that there are hard problems and there are hard problems. And mining automation is the latter.
- nyrikki 3y ago
  In mining, the bean counters would use child and forced labor if they still could. But truck shovel routing is a version of VRP, trucks and shovels need to be matched or the approximations rise into the double digits of inefficiency. If you look at the powder river area where they use the model they mentioned, truck shovel is for removing the spoil on top, and thus is a dynamic environment. The actual coal is moved by conveyer belts at the mines I have been to. VRP is nontrivial and the bean counters know this. Note this fairly readable paper and the complexity of 2VRP vs 3VRP vs kVRP. https://www.sciencedirect.com/science/article/pii/S0166218X04002860 https://www.sciencedirect.com/science/article/pii/S0166218X0... If they have some new method, it won't apply across surface, sub-surface, and oil at the same time unless P=NP.

c_o_n_v_e_x 3y ago

>What techies keep getting wrong about industrial automation 1. Internet connections can be some combination of slow, unreliable, intermittent, or expensive. They also may be completely unavailable, although this is changing in the era of Starlink. 2a. Consumer grade compute hardware dies a quick death in harsh environments. 2b. Consumer grade compute hardware typically has more compute power than industrial rated computers. I've seen a handful software engineering teams have to drastically rebuild software because compute resource constraints were not realized until it was time to deploy software at the edge. 3. You are not allowed to push updates to your software willy nilly. 4. Service tech callout and installation costs are a big consideration for remote assets (oil wells, pipeline compressor stations, etc.), particularly at scale.