Found this
https://play.google.com/store/apps/details?id=com.vaigaiflow.bacnetmodbustoolkit
And a brief explanation here:
https://www.hvac-talk.com/posts/27007718/
I never used it. But putting it out there for people to play with. Lemme know if you like it.
If nothing else, having someone see you scan BACnet with your phone and go wha?!?! Is awesome.
The recording is live! The panel discussion at the Niagara Summit 2026 on AI in real buildings has just been posted. This is how to view it:
Go here:
https://share.vidyard.com/watch/SaMeBb9HUM24YZS26RYDBL
Password is: NS2026
the title is: AI in Real Buildings: What's Working, What's Risky, & What's Hype
I attempted to bring a real world and practical perspective to the discussion. I hope the community finds it useful.
My view is that Generative AI (LLMs broadly speaking) will see modest *productive* usage in the short term. The challenges are a lack of solid training data in an industry (that is unlike IT) with poor documentation and many 'unadvertised features' installed in infinitely unique buildings. Add to this, LLM technology is relatively new, the gains from the use of LLMs are hard to quantify in the economy more broadly as well as in the built environment, the risk (financially and reputationally) of physical damage, and the relative immaturity of the technology.... It means that although the potential is real, that potential is being realized unevenly right now.
Contrast that with Machine Learning. It is a 15+ year old technology that is far more mature. It is discrete and easier to put guardrails on and implement cleanly. It trains on much less data, and the results of mechanical optimization are easy to measure and verify in real dollars. In my mind, Machine Learning makes MUCH more sense for implementation in the near term.
I flesh out these concepts in the presentation.
Thanks to the MANY who came up and introduced themselves at the summit during the session and afterwards. Also thanks to Ruben Yanez-deLeon for moderating. I look forward to seeing what he does in the coming years. He is one to watch. Thanks for the co-presenters for allowing me to share the stage.
I am the ugly guy on the left. 😊
Has anyone used this? I am curious to know how it worked in the wild. I found some documentation on it, and I was wondering how it worked, if it was useful, - basically tribal knowledge concerning this.
Called an 'AutotunerPID' - documentation from 1991 - so not knew.
***
'Description
'PID Loop function (Long)
' AutoPID() is a function providing cruise-control for process loops.
' AutoPID() automatically selects and tunes the Proportional,
' Integral and Derivative control loop constants and time frames.
' AutoPID() returns a value from -1.0 to 1.0 which is the magnitude of
' change to be made to the controlled Output Variable (-100% to +100%).
'EndDescription
'Properties
'Name:AutoPID:True
'Type:InfinityFunction:False
'EndProperties
'Code :
'PID_tuner_NC() PID loop tuner with no comments 02-Oct-91
****
For AutoTunerPID Wizard keep in mind that it is diff erent than PID or often referred to as BPID. There are 3 types of PIDs
***
Specification for PID Tuning function. (AutoTuner or Tuning Wizard)
Original Issue 2, 2nd October 1991.
The PID tuning function is designed to be run once on a system to ascertain the required values of Proportional, Integral and derivative factors (Kp, Ki, Kd), when complete the function is normally to be replaced by the standard PID function as it will run faster with less system overhead. However when tuning is complete the PID tuning function will continue to control the plant using the standard PID function. 16. To begin tuning the function should be enabled after being previously disabled or alternatively element [1] of the passed data array may be set to 0. Element [1] of the passed data array may be used to determine tuning completion when it reaches a value of 5. 19. The format of the function will be as follows:
PIDtuner (Ena,Type,Cv,SP,Kp,Ki,Kd,Db,OP,Lag,Dev,Pd,Ta)
***
My personal favorite session was the last one I attended about Division 25. extremely practical.
Retrofitting Data Center Legacy Infrastructure for the AI Era
Uptime is king. So retrofits and upgrades need to have no or little downtime that centers on stats that are important to the customer is looking for.
The GPUs respond very quickly. This might mean that your chillers are starting and stopping a lot to try to match the output as it varies. Instead, design with a buffer tank to absorb the changes and run with more than one chiller on but at a lower level. Then systems are not cycling completely off and wearing the equipment out. Instead the chiller are usually running but they can ramp up and down to meet the demand, with a little bit of flexibility that the buffer tank allows for.
A BRICK ontology allows for defining the systems for relationships and standards to allow. For scaling. AI optimization allows for energy saving measures. Uses ML with a 5-15 minute adjustments of setpoints only. Safety interlock always active, and Niagara enforcing the limits on the AI. Data center construction operates on tight deadlines. 18mos? never, you will get a 6 mo deadline and have it moved up 3 Mos.
Deploying Niagara in Mission-Critical Environments
You cannot focus on your scope, you need to think of this as a team sport, and the goal has to be a successful outcome. Collaboration saves a lot of time and hassle.
Architecture principles are simplicity, resilience, redundancy, and removing any single point of failure. Spend extra time on design to look at the details. Copy paste is dangerous when portions are not double checked to ensure that they are meeting the requirements for this particular system - close enough is not good enough. Saving $10 on a part could cost 100k in lost time, engineering, rework and other issues. So it is better to get it right early and recheck often as you go to catch problems.
If an integrator is called to do a job and is not involved in the design phase early - instead they are called to bid and are just given a scope and told do it and we gave you 4 weeks... Don't be on that job. Walk away, it is not laid out right.
Lessons Learned from Intelligent Building Projects - Div 25 Intent vs Outcomes
Division 25 us specification. Guideline 25 is developed by SI and owner that shows how you will comply with spec 25. Often out spec 25 and div 25 both into the contract docs.
The key to getting a quality install means the owner needs to be involved early, write it into division 25. Otherwise, some technology may get adjusted or value engineered out to save cap ex but that will cost several times more money in OpEx.
ASHRAE is working on integrations for buildings standard- it will be coming soon.
Packaged controls are cheap and quick, but inflexible. Sometimes you want to leave them on, like a chiller. But if you have packaged controls, Div25 can specify the need to include software tools. Or to get a sequence to put on a set of a prints. The university likes SI mounted controls. 25 will say we want 3rd party controls, and that will push into Div23 where the spec will need to tell the mech contractor to buy with 3rd party co tails to meet Div25, and if these don't match it is an RFI to reconcile them.
Not many engineering firms understand Div25. 10 firms in the US tops. Instead an owner will need to build it themselves. Then educate your engineers.
Often there are areas with a low bid requirement that might replace a good installation with a cheap one. To avoid this write in a special requirement that says there is an interview process where the company will need to send a copy of the programming and then read it back to me. Find out how familiar are they with the software and have them submit samples of projects done.
Or in critical environment where downtime is measured in millions per minute they will need to show how they will meet the program spec and make it work in the form of a programmed ck triller as part of the bid contract.
Becoming more common that the MSIs will need to carry cyber security ins. ASHRAE will be making it a standard and they are working on it.
Cybersecurity: Is Your Organization Next?
They showed a provisioning job for distributing certificates to the user trust store using provisioning.
They showed creating a self signed CA and CSR for signing your code and program objects. Starting in 4.15 passwords are Hashed, Salted, and encrypted. This will carry over into N5. Salting is important. 4.15 includes several new options in the station transfer and passphrase area to include making the station compatible with revs before 4.15. Without this 4.14 and earlier workbenches.
Also there is a syslog integration to plug Niagara into existing SOC tools and monitoring. It is located in the platform service under the station.
Host header validation whitelists access requests to ensure they are coming from known specific locations, limiting the attack surface.
Fox and Web connections are being deprecated, to leave TLS only as a more secure access. FoxC was demonstrated. The cool thing is that it makes a cloud connection that is encrypted and secure with extremely minimal effort. There will be a good video for all of this when the recordings are released. They say 2 to 3 weeks.
The annual ra ra session is this morning. Everyone picks a theme song to walk up to the stage with. Kinda silly.
Introduced the concept of a 'premium workbench'. They talked a little about what it will have... but not on what it costs...
Niagara says they spend 40% of their R&D on cyber security. kind of an interesting statistic.
they spoke about secure by default - the settings come out of the box set to the most secure settings. And a security dashboard that analyzes the modules, and other included systems to bring visibility to the security posture of your system as currently configured. The old security thing was deprecated in JAVA, So this is not the same as the one in N4, it replaces it. Niagara says it is better.
New 3D models for the Px graphics. Also out of the box templates for graphics.
Ask to Niagara will have an AI assistant that is trained on the Niagara docs and processes. It is embedded in the existing help. Want to allow for cloud storage access within (Premium) Workbench. Again, no costs mentioned. gotta follow up on that.
There is a new MCP service that is coming out. That is the Model Context Protocol. For the ability to train models and agents from Niagara using MCP so you can tell them not only what to do, but what NOT to do.
Planning to support BACnet rev 23 as of N5.2
Niagara Data Service allows for a cloud based server with a cloud based data lake that uses an API to write to Niagara.
Announcing a new cloud based connection feature FOXC - pronounced FOXY.... hehe. Allows cloud connection initiated in minutes. Requires a cloud SMA
New NRIO. ALL 24V power. Drop in replacement. Encrypted comm channels between the JACE and the IO, so available on N5, not N4. Can mix IO34 and NRIO34 and 16 both on an N5 station. Has universal IO that are all software configurable as any input or output.
Edge10was discontinued, and will be rereleased as Edge18. Will take the new NRIO modules on the Edge not the JACE.
Developer drop will be in May, SDK initialized. Bringing Linux support and other things for developers. July is beta testing time for N5. September will allow early access for building owners.
Datacenter spend in the BAS space was 1/10 of the spend in commercial spaces 10 years ago. 3 years ago it was 20% of commercial spend. As of now, the spending is about equal, and will grow rapidly. Cloud used to be storage for pictures, now it is compute. So market will grow fast.
Distech is releasing a PLC line. ... will use the Codesys programming tool. i need more details on that.