United Manufacturing Hub

<@795725198980677734> <@335006809716686851>

some community generated content here to get things started

https://learn.umh.app/course/landing-the-umh-in-your-company-what-sysadmins-will-find-irresistible/

https://learn.umh.app/course/simplifying-tag-browsing-in-grafana/

https://learn.umh.app/course/bringing-iiot-to-life-inside-maykers-demo-factory/

https://learn.umh.app/course/building-a-packml-simulator-dashboard/

I will start! 🔥 
See here: TLT Automation! Frontend mostly build in Node-red, backend fully UMH based!

Costumer: https://tltautomation.com/
TLT automation is a machine builder that builds serie machines for the picking & processing of (white) mushrooms. 
Endcostumers with UMH deployments: https://www.muzzroom.com/ (Belgium) & https://www.walshmushrooms.com/contact/Golden_Mushrooms.html (Ireland) (New deployments in Belgium, Ireland and Germany coming in 2025)
Reference of previous project: https://www.siemens.com/global/en/company/stories/industry/factory-automation/handling-robotics-digital-twin-tlt-automation.html?linkId=300000021839099

great case! any chances sharing how the dashboarding looks and some implementation insights like how the data is extracted from the PLCs, how the UMH is hosted, etc?

Yes, i was planning to share that as well! 🙂

Here's a sneak peak but more to come!
https://cdn.discordapp.com/attachments/1344336987171196929/1344591475446382614/image.png?ex=67c177e5&is=67c02665&hm=e594fc9deaa09a3d6aa85615db58ce39ea56bc74b35cdeff651c72514add4bc7&
https://cdn.discordapp.com/attachments/1344336987171196929/1344591475937247232/image.png?ex=67c177e5&is=67c02665&hm=0802846bd136626643cb91fd8dc3a87cf9a7bc8140c52a6eb9c9585e28be5bfd&
https://cdn.discordapp.com/attachments/1344336987171196929/1344591476415270973/image.png?ex=67c177e5&is=67c02665&hm=625b0f09ad4bd0972ecbde98c9fa9e84008a3427928b280bc9de471bf34fe180&

So let me (finally) follow up on this! 😄 

Let's start with a factory layout (of Muzzroom in Belgium; Ireland is a little different)

As our core machinery we have our picking equipment:
https://cdn.discordapp.com/attachments/1344336987171196929/1346429612045111346/image.png?ex=67c827ca&is=67c6d64a&hm=4bf0d6d8e2da50205925a4fad2dc20caccd6a9d6583647b064ff540869bd9a55&

These machines (6) can be physically seen in the Siemens video.
These all have the exact same internals (mechanics, electricals, IP adressing,...)

This makes our setup very scalable. We can collect the exact same data across all our current deployed machinery and 1:1 compare them.

As  brain these machines have a Siemens Open controller. This is an IPC that has hypervisor on it. This has 2 "VMs"; a (weird IoT) Windows 11 and a virtual PLC. 

On this virtual PLC we have our control software and exposing the data over opcua
on the Windows VM we have a nodered instance (communication with some external equipment over S7) and a benthos instance reading the opcua broker of the PLC and sending it to a broker on the production network

this setup we have 6 times. (12 more of these machines to come in 2025 on this site)

This is the (relevant part) of the production network:


https://cdn.discordapp.com/attachments/1344336987171196929/1346494887100420106/image.png?ex=67c86495&is=67c71315&hm=9b8763e9436ec698b536340ba3502b6cadd07c7cfe8145dfeab6186c35669eb8&

So we have a production network with our (basic) networking
As main server we have a nuc with an UPS

This runs Proxmox with 2 VMs
1: UMH stack on Ubuntu desktop LTS

I know it's not perfect but this is how it (currently) works: 
1. Benthosses publish to docker hiveMQ-CE broker
2. Nodered subscribes to the full docker broker
3. Nodered checks and adds umh topic/message structure if lacking (_historian & timestamp_ms ) 
4. nodered publishes on umh hivemq broker
5. data goes thru UMH to timescale/grafana

For the rest we have some dashboards shown across the factory and some nodered flows running

main production dashboard:
https://cdn.discordapp.com/attachments/1344336987171196929/1346498807587799121/image.png?ex=67c8683c&is=67c716bc&hm=a8a6f6e8c7061596d6e4e3d1bd4b3c0369c425f67369981a7c29246743060afe&

each growing room has a dashboard like this:
https://cdn.discordapp.com/attachments/1344336987171196929/1346499030456209458/image.png?ex=67c86871&is=67c716f1&hm=9b6af5cb23803f2b63c9ee01d486f126739ce4a343e088b363ce59c2ade96783&

we also have a small MES integration application:

here the operator has an ipad attached to the wall and fills in finished pallets.
When he presses print a label is printed for that pallet which he sticks to the pallet.

These printed labels are scanned on arrival in the next plant and integrated in their MES
https://cdn.discordapp.com/attachments/1344336987171196929/1346499710109876224/image.png?ex=67c86913&is=67c71793&hm=7da7609610e993f7733c5a3476c3aada4a3a8f39d8e981dbd1743d303596fb0d&
https://cdn.discordapp.com/attachments/1344336987171196929/1346499710461935626/image.png?ex=67c86913&is=67c71793&hm=31c580a33907563b2fd7012bf2c35f244b5fd7d03e74326eaf267d42d8e2fa4b&

a finished label looks like this:
https://cdn.discordapp.com/attachments/1344336987171196929/1346500264969900253/image.png?ex=67c86997&is=67c71817&hm=35f89ddcca7b8c5462897a6cbb7de6092e7f8598dc4a4f8ab087660f843c1c77&

when the truck drives moves a stack of finished pallets he clicks "MOVE TO TRANSPORT" and then a consignment note gets printed
https://cdn.discordapp.com/attachments/1344336987171196929/1346500554892906567/image.png?ex=67c869dd&is=67c7185d&hm=fcf066bc9f9aa7c0d21771f416b747902f7feb12be14be61d4ca00536568f506&

We also log this production info and report it on a daily and monthly basis to the plant manager


https://cdn.discordapp.com/attachments/1344336987171196929/1346501092371857418/image.png?ex=67c86a5d&is=67c718dd&hm=103bbbd8579070ec66802d66d013aab928a533dcf2070ba70ec4fa5713e8a69a&
https://cdn.discordapp.com/attachments/1344336987171196929/1346501092803874837/image.png?ex=67c86a5d&is=67c718dd&hm=e68d49af78939779ee5a06397201c610b7f314732a0a6be1bd5b48bd250f0858&

We also have some more in depth analysis (mails & dashboarding) on machine performance (which i can not share)

We also send a similar mail on departure of the truck, this way the receiving factory knows when & what will arrive (this way they can prepare some space/work more efficient)

interesting. How do you handle the dashboard with erp integration? It looks like it is done in nodered? Do you have a separate nodered instance for each screen? Native nodered does not support multiusers so it can be a problem with multiple users using the same dashbord if you have inputs, since the inputs are the same on all screens.

All those things are done with nodered indeed. (UMH is used as historian only)

We have one nodered instance per plant & costumer (so one for Muzzroom)

this is in dashboard 2:
https://cdn.discordapp.com/attachments/1344336987171196929/1346509325861912659/image.png?ex=67c87208&is=67c72088&hm=04afea2c3a3b42656c950773189a9e6ce6f8c45dad51313c45f5657699a39966&

this is in dashboard 1:
https://cdn.discordapp.com/attachments/1344336987171196929/1346509414886150265/image.png?ex=67c8721d&is=67c7209d&hm=3ef3896c98e948f67437137544b147b33c5ddb4d6e41e735ca4a40cc3aa0a73e&

we have not faced issues yet with inputs from multiple screens

instead we want this:
for the pallet application we have an ipad for the operators at the factory floor (with the webpage open) and an ipad for the truck drivers at the logistics area (with the same webpage open)

So as shown most of our interesting things happen within the factory

outside the factory we have a free cloudbased hivemq broker with similar dashboards (so the office in the next village can keep track of production in the facotry)

We are currently working on 2 more "digitalisation" projects in this plant:
- NATting of our OEM machines to the production network 
- Machine learning project (study on how env parameters influence growing & quality of the mushrooms)

upcoming: As our machine networks contain camera's, drives,... that we want to monitor as an OEM we require access to the machines. Currently we do this via Teamviewer
but to make this better we want to nat specific devices to the production network. This way we can monitor all devices/machines at once instead of 1 by 1

If you have for example a text input on a dashboard, the writing in the textfield will reflect on all dashboards.. The same if you have a checkbutton or similar. All users will see the same thing which can be a problem.. I also don't like that you don't have a login so you can have different users using different screens, and also to log which user sends which command.

For Nat i could send you the settings i used for ubiquiti edge router to NAT and firewall protect the S7 data we collect.

That's a fair point, i've been thinking about this too
to fix this i would use the free version of flowfuse to add the user management. They have some good tools for that

my OEM will prefer to use Siemens hardware (discounts & vendor relationship...) 
but thanks! 😄

upcoming project 2: 
we scan each mushroom with a Lidar camera before its boxed. 
We score it based on multiple parameters (color, diameter, height,...)
each parameter we score 0 to 10 and we make a total score based on these. 
These scores are getting saved in the UMH historian
We call these scoring parameters

Next to that we save all growing parameters per mushroom (who picked it, what time, what was the in & outside temperature at that moment,...)
We call these growing parameters. These get saved as well.

in our project we're searching for correlations between both. 

Here the UMH historian will prove its value! 🙂

this project is still in research and supported by IFM

(as i dont have the AI / data analytics skills to make this happen alone)

thanks for sharing the insights! 
is that correct that for now you don't have much a use of the historian's data? right now it's just being collected for further usage? (AI, etc.) Also do you implement (or plan to) any classic MES functionality like OEE, shift scheduling/production planning or any leverage any data from from ERP?

I could not convince my costumer yet that proper historian functionallity is useful. We do have a few grafana dashboards but they are not used often

but they prefer realtime data in nodered over the historical data in grafana

Our advanced analysis dashboards (which i can not share) have OEE/KPI calculations on our machinery. (both historical & live)

we dont have shift/production planning in this software (its done by another tool i didnt work on)

nice! would it be OK for you if we were to do a blog article out of this? a community spotlight?

For me that's fine 
i will have to check with TLT as well

but feel free to get started with it, they will probably be okay with it

TLT Automation

(more is coming; still working on something else atm)

@everyone it took a while but i extended the TLT thread by a bit

other end costumer other implementation requirements/dashboarding/...

So our costumer in Ireland has just one machine (scaling next year to 10)
For now we're just testing his specific requirements (special cardboard boxes instead of regular plastic ones), herefor we visualize some extra machine parameters instead of regular production parameters
https://cdn.discordapp.com/attachments/1346510636028264489/1346793979697692702/image.png?ex=67c97b22&is=67c829a2&hm=884838ed3434e3084d4645065e836c63ede069579264e42f9baf9c880aab2ef9&

As this is (currently) just is a stand alone application we dont have an advanced setup yet

so we have an UMH instance but just acting as historian without any fancy interfacing with other systems

Also the datacollection/visualization is quite simple

we subscribe to every tag we want, use a big switch and visualize them
nothing fancy
https://cdn.discordapp.com/attachments/1346510636028264489/1346794583815753779/image.png?ex=67c97bb3&is=67c82a33&hm=b6aba7fa7296738eda3334c0011a215f00964e6bfc0996356bd1589bffe3822d&

our main KPI in this test setup is the boxfilling (which is only at 50%) 
so we need to optimize the machine mechanicly to be able to handle the special cardboard boxes better before we go to full production.

TLT AUTOMATION (IN IRELAND)

Impressive! Do you have some screenshots/architectural layouts/further info... you can share?

Sure! So we have basically split it up into 2 parts. The Home Assistant system runs on a Raspberry Pi, where various staff members (some handy 20-somethings) have access on an do a bunch of projects like adding sensors and experiment with other ESP32 gear.

Then we publish to the MQTT broker of UMH, where the business owner and I then process that data into UMH. We built 3 custom data bridges, for sales orders, work orders and issue tracking. We then query the data from various data sources in the company via API's , plus sensor data from the sensors on Home Assistant to create our own mini-MES solution.
https://cdn.discordapp.com/attachments/1348606708540248208/1348654974124556348/image.png?ex=67d04052&is=67ceeed2&hm=d78b491f0dabc47401aaa9a0a759deca50f63bef8370d5f1e600c3939659b4fc&

(Ridder iQ is the ERP system)
https://cdn.discordapp.com/attachments/1348606708540248208/1348655123412418620/image.png?ex=67d04076&is=67ceeef6&hm=7c3019aa34d40f1fb21dc69f770f3da7bde23783b1ddfb461e1b2631fc7e2adb&

So if there is a rework order at a specific machine, that qualifies as important (ERP data), then we show what issue is connected to that, if it's running, and we activate a light in the shopfloor indicating the operators a rework order has to be completed ASAP (or is running already)

The main idea is that we allow the staff of the company to create their own solutions with whatever they want, we just subscribe and publish to MQTT brokers and then do a bunch of enrichment of data from ERP and MES software already in the factory.

If you have any further questions, just let me know 🙂


https://cdn.discordapp.com/attachments/1348606708540248208/1348655873727139983/image.png?ex=67d04129&is=67ceefa9&hm=b168a4759659afdd625685057b6daa51874f6d175c803f53301f2b2f2b11c4be&

This runs at the lasercutter workplace, to monitor for rework orders
https://cdn.discordapp.com/attachments/1348606708540248208/1348656389345775626/image.png?ex=67d041a4&is=67cef024&hm=d655d0b7f65491db7ade4524350b120c1de018cdf40c7005b312e615f70126c4&

I think the most awesome thing is since we completed it, the client themselves built like another 5 Grafana Dashboards of all kinds of combinations of datasets.

how do they make them? they write the timescale queries themself?

Honestly, https://chatgpt.com/g/g-TwQAiOL8n-grafana-master lol! they just have access to the TimescaleDB for column infos and then dump that into this tool

Same for NodeRED, 80% of the flows is just copy paste with https://chatgpt.com/g/g-adwIUhhSi-node-red-expert

Only thing they cannot do themselves is bridges and we agreed to not create a any umh/v1/ topics wihout first discussing it for overview

interesting. Why do you use homeassistant ?

My client already had it with some sensors on it. The home assistant runs on a raspberry on a firewalled network where they tinker with, without having access to other business software systems. UMH itself does has access to the full network and all business systems (API of ERP, MES etc.)

So we decided to just leave that and let it publish to HiveMQ in UMH (and also let UMH publish back). In that way, we only had to open one port and device on the firewall.

That was one of the cool things, that they already had some of these tinker projects on the shopfloor. So we just tapped into that. We ourselves did not had to touch any sensors or anything like that.

Just wanted to mention again that it's a PoC, so we wanted to change nothing, just build on top of what we had 🙂

Next, we probably will look into connecting a welding robot and a lasercutter over OPC-UA, but for the lasercutter the machine vendors wants like 10K per device to activate the module 🤮

then why activate the module? is there another way? other protocol and convert it to opcua/mqtt/...?

Interesting. At home i am running homeassistant in the same kubernetes cluster as umh and i am using zigbee2mqtt and zwave2mqtt to handle my devices. This sends all sensor data direct to the umh mqtt broker and homeassistant subscribes to that topic.. This way homeassistant is somewhat standalone but in the umh i can easily access all mqtt data and push the interesting data to historian.  if you don't want to use HA you could use only zigbee2mqtt and zwave2mqtt to handle all sensors and publish to the broker..

are you guys okay with using zigbee & zwave in a industrial environment? what sensoring are we talking?

For any thing related to PoC, to land and expand, I'm okay for anything. A PoC shouldn't be mission critical anyway. Long term, I would not know. I'd probably opt for a more industrial focussed solution. Same applies for HA, it's fine for a tiny shop, but I wouldn't use it at scale. However, I now landed it, and that's the most important in this phase! 🙂

zigbee is used in for example schneiders industrial wireless buttons so  it should not be a problem. zwave is even better i would say, it has better range and follows a strict standard. My home system with zwave has been running without any problems at all for about 5 years now.

I also started now with making various views and screens with NocoDB

Here's the full ERD also visible, quite cool
https://cdn.discordapp.com/attachments/1348606708540248208/1350200275562004592/puT8DbrPvIv8QEdmAAAAAElFTkSuQmCC.png?ex=67d5df7f&is=67d48dff&hm=678f9e5fd3bc94e724924d4cf295704640893fe176bcf2026389bcb6009fde70&