Edge computing, BigData, what to achieve...

userxyz

Member
Join Date
May 2002
Location
any
Posts
2,768
Hi,

As we all know, Siemens is hitting us with an overflood of marketing about Mindsphere, Industry 4.0, Edge, etc...
I am here in the automobile industry where the management is talking about implementing a Bigdata server with Edge computing and process mining... I'm not sure everything is clear to them what to achieve...

First things first...
To my understanding, Mindsphere and Edge Computing are compareable but the main difference is that edge is a software to run on a local bigdata server while mindsphere is on a internet cloud where you pay for datapoints.

Today we have OEE implemented here, we have Scada systems and we have a proficy historian with web based reports. Our tools are really powerful allready. I'm not sure what added value edge computing can create here...

When you look at a factory where you have several equal machines that do the same and run batches, or a factory where you have a lot of processes in parallel with batching an process steps... then I can understand that AI (edge) can find patterns for making the production more efficient and better, the availability can be optimised by such tools I think...
But when you like at a Production line in a car factory where the factor availability isn't there because everything is in one sequence... then I have my doubts about what AI could improve.

But, I am looking for answers, so that's the reason for my post here :)

Kind regards,
G
 
Last edited:
I have to wonder how much the marketing group has added the latest buzzwords to already mature products in order to give them extra life. I''m sure there are some new things, but having been in the auto industry approaching 40 years the real valuable tools that I've implemented and the plant has actually used numbered just a few.

Andon is a must have. Typically driving bingo boards hanging in the aisles or feeding control rooms.

FIS (Factory Information System) is very common (albeit with different names) but requires a infrastructure of servers to store a lot of data. Typically archiving fault occurrences. The maintenance folks gets a Pareto chart of fault occurences and/or blocked/starved times so they know where to focus their efforts.

The last couple of years has seen a big push for part tracking and operation enabling on the machining side. If the company has already implemented it in assembly, its a short reach for manufacturing. Individual components are immediately laser engraved with 2D bar coded serial numbers and every operation queries the server to make sure that indivdual part has previous operations completed and then sends any measurables (significant characteristics) to the server once the operation is complete.

So, can AI help here? I guess someone could make a case for it but its like anything else, will someone actually use it.
 
From my perspective, 90% of what someone trying to sell you Big Data or Edge Computing is already possible. I'm honestly not sure if they marketing guys know that or not.

I think Edge Computing vs Big Data in the cloud are different but complimentary things. The way I've always seen it is that edge computing allows for local data processing, with only the results of the processing need to actually be uploaded to the cloud. Could be something like monitoring alarms locally at the Edge, and then only sending a summary up to the cloud.

I think the big improvement in "Edge Computing" over traditional methods is that one of the goals is to be able to centrally manage the configurations, and update them automatically. You make your configuration in the cloud, and then it does the processing locally. In reality, this could be totally possible with a traditional PLC solution, but I've never seen anybody do it.

I haven't really seen one Big Data application that couldn't have been done similarly with existing SCADA software, the difference is just in how easy different parts of the application are.

I think one of the places that AI is being promised is in quality and maintenance improvements. It can monitor what is actually being produced to be able to predict defects or downtime before they happen. In reality, though, these types of things can mostly already be done in PLCs/SCADA, and the "AI" they are promising is super duper basic. It isn't "smart", it is either just creating alarms when some configured value crosses a threshold, or MAYBE performing some statistics.
 
I'm not sure either of you intended to do this, but the examples you both gave of useful tools were for looking at things you knew were there (process optimization or directed maintenance effort). I think what the marketing pencil-necks are trying to push is that, if you have billions of data points from every device in your plant, there WILL BE software available at some point that can find correlations no one could have ever even imagine existing. I don't think that software is out there yet but you need to have the infrastructure in place to use it when it gets here. So the early push is on connectivity. Later it will all be about analytics.

Keith
 
5%

To me it looks like a tool that can optimise a production, but maybe for only 5% improvement. When you look at OEE, I think that would be at least 30%. The percentages are just guesses, but you get my point here :). In general it all comes down to the user. Yes you can create standards in software and hardware, yes you can create data models, yes you can build OEE and reports, but the people must use the information, change things, evaluate, etc...
I think Edge and other tools will improve, also the data formats will improve, etc etc... but will this improve production... maybe in 20 years, but not today.


It is all hype. We/our customers have been doing this since the 1990s.
 
OK, just to give a different view. Big Data isn't about the relationships that you know exist, is about the ones you haven't even thought of. You feed it a load of data, you don't tell it what to look for, but it finds relationships. Yes there is a load of hype, but don't dismiss it. The tools to do 'Big Data' have only recently become available, you need a lot of computing power and huge amounts of storage.
 
Oee

Hi,

Yeah you are right, but I think there are a lot of other tools (like OEE) that have to be processed upfront before searching to optimise that last 5%. I think that an installation needs to run well before you can do edge computing, too much failure data in a BigData would be a mess. But I agree that it can add value. The question I have to add here... will we need BigData specialists in the future ?

Kind regards,
Combo

OK, just to give a different view. Big Data isn't about the relationships that you know exist, is about the ones you haven't even thought of. You feed it a load of data, you don't tell it what to look for, but it finds relationships. Yes there is a load of hype, but don't dismiss it. The tools to do 'Big Data' have only recently become available, you need a lot of computing power and huge amounts of storage.
 
I am just working from a couple of TV programs I have seen, so there is no great insight to be had from me. So these are very much my un-researched opinions, I don't normally like typing about something I am not reasonably certain about, but what the heck.

will we need BigData specialists in the future ?
In the instances I saw they passed the raw data to another company that had the specialist tools and computing time. But I did a quick web search and there are now open source BigData tools. Once you have results you need the expertise to translate them in to things that management can understand. So, simple graphs of complicated data.

too much failure data in a BigData would be a mess
I wouldn't want to feed it bad data, but good data about when things go wrong is still very valid, and as equally important as when things go right. With BigData more data is always better. You wouldn't feed it data about one machine, you would have data for the whole production facility.
 
Originally posted by Combo:

Yeah you are right, but I think there are a lot of other tools (like OEE) that have to be processed upfront before searching to optimise that last 5%.

Specific to the Big Data aspect, I think you are limiting your application scope too much. The real power of Big Data (if it works, and I'm not convinced it will in manufacturing) is it will find correlations you haven't thought of yet. Granted, many will match up with things you already know and have tools to address. It's the ones you DON'T know about that potentially hold the real power.

What I'm waiting for is when the results of a data analysis seem to indicate that an aspect or two of Industry 4.0 are just absolute nonsense. That will be a fun fight to watch unfold.

Keith
 
:)

Specific to the Big Data aspect, I think you are limiting your application scope too much. The real power of Big Data (if it works, and I'm not convinced it will in manufacturing) is it will find correlations you haven't thought of yet. Granted, many will match up with things you already know and have tools to address. It's the ones you DON'T know about that potentially hold the real power.

What I'm waiting for is when the results of a data analysis seem to indicate that an aspect or two of Industry 4.0 are just absolute nonsense. That will be a fun fight to watch unfold.

Keith


:cry:🍻
 

Similar Topics

Hi, I need to deploy, in relatively near future, one project which main purpose is to test Edge computing capabilities, understand pros and cons...
Replies
13
Views
6,411
I'm trying to build my Classic Step 7 programming skills this weekend. I get stuck on little things that are not covered in YouTube tutorials. I'm...
Replies
7
Views
248
Hello, I have been tasked with adding some analog signals for display and alarm setup in some old Schneider Electric HMIGTO HMI-panels. I have...
Replies
4
Views
128
Good afternoon everyone, could someone guide me on how to program a falling edge in a wave micro control using CWDesigner? It is to reset a...
Replies
2
Views
120
Can anyone share a way to make a button on the HMI that hits the "Ack Page" in an Alarm and Event Summary Window? or better yet - a plc tag...
Replies
5
Views
246
Back
Top Bottom