Nexus #18 (4/14/2020)
Announcing Nexus Pro; Analytics from cradle to grave; AI to control buildings; and a(nother) discussion on defining "analytics"
👋 Welcome to Nexus, a newsletter for smart people applying smart building technology—written by James Dice.
If you’ve been forwarded this email, you can sign up for your subscription here:
If you have thoughts, questions, ideas, or tips: hit reply, I’d love your feedback!
And by the way: if you missed last week’s edition, you can find it here.
Disclaimer: James is a researcher at the National Renewable Energy Laboratory (NREL). All opinions expressed via Nexus emails, podcasts, or on the website belong solely to James. No resources from NREL are used to support Nexus. NREL does not endorse or support any aspect of Nexus.
Here’s an outline of this week’s newsletter:
⚡Nexus Pro launches in May
🤔 On my mind this week
🧰 Analytics from Cradle to Grave, the new way to commission buildings
🤖Using AI to Optimize the Flow of Energy, my summary of a whitepaper by AI startup Brainbox AI
1. ⚡ Nexus Pro launches in May
Many of you game-changers have asked how you can support Nexus and get more involved in the growing community. Now there’s a way to do so.
In May, I will be launching a Nexus Pro membership option. Just like when the newsletter began, this is an experiment, and I think you’re going to love what we’re putting together. Initially, in addition to the free newsletter and podcast, here’s what members will get:
weekly deep dives (like this one) featuring the latest trends and podcast show notes;
access to community discussions including “ask me anything” sessions and virtual happy hours with industry leaders;
access to Nexus vendor landscape and the site Archives.
As a token of my appreciation, those who sign up before May 1st get a 20% early adopter discount—forever.
Click here to redeem:
Note: We will also have special rates for groups, students with ‘.edu’ email addresses, and those in need due to the COVID-19 pandemic. Email James for details at firstname.lastname@example.org.
2. 🤔 On my mind this week
Like all of you, I’m continuing to track the impact of COVID-19 on our industry. Just like I did in the first Nexus Deep Dive, I’ll continue to share my thoughts here as I have them. If you’re looking for the signal in the noise, here’s the best COVID-19 content I’ve seen this week:
COVID-19’s Impact on BAS Contractors (Automated Buildings), paired with Construction Project Cancellations Rose Sharply Last Week (Globest)
As usual, please add any great stuff you’ve seen to the comments.
Oh, and one more thing I’m thinking about: the Nexus podcast is now on Spotify. Episode 3 with Nick Gayewski, CEO of KGS Buildings, is coming soon.
3. 👨💻The latest LinkedIn discussion:
I’ve heard a lot of questions and confusion lately about how to define “analytics for buildings”.
The confusion is understandable—it often seems like everyone has a different definition. That’s why I originally wrote this essay on defining Energy Management Information Systems.
One particular point of confusion is whether Building Automation Systems (BAS) are types of analytics software. This gets more confusing as analytics firms add advanced supervisory control capabilities.
What do you tell people when they confuse BAS and analytics?
4. 🧰 Automating functional testing
+ Analytics from Cradle to Grave (ASHRAE Journal; Altura Associates)—Anytime I’m helping develop specifications for a new building or retrofit, I find myself sharing this article with my team. If you aren’t incorporating automated functional tests into your controls specifications, sequences of operations, and commissioning process, I suggest you give this a read. Here’s a summary:
The old way to do functional testing:
(…) testing forces equipment into the test conditions to then observe and record equipment response. Test execution details are documented on written or electronic forms while a data logger or automation system records time-series trend- data for specific data points. These trends are then post- analyzed and performance is compared to the expected response to determine acceptance.
The forced conditions can be achieved through the use of automation system overrides by the controls programmer (equipment schedule adjustment, speed/ position commands, temperature/pressure sensor overrides) or field interventions (occupant thermostat resets, in-hand valve/window/sash position adjustments, etc.). Active testing is generally required to comprehensively test the SOO in a timely manner.
The new way, with analytics included:
With (analytics including in the commissioning process), the step-by-step test procedures for each piece of equipment are still included. However, the execution of these tests is transformed from a manually executed, time-intensive process to an automated process that can run a virtually limitless sequence of override steps on multiple equipment at once, without the need for human intervention. This process unlocks an unparalleled degree of scalability, repeatability, and schedule flexibility in the testing regime that is impossible with a traditional approach.
I’m thinking about doing a Deep Dive on this topic. Let me know in the comments if you like that idea. Are you a leader in this practice? Hit reply and let me know if you’d like to be featured in the Deep Dive.
5. 🤖 Better HVAC control w/ deep learning
+ Using AI to Optimize the Flow of Energy (Brainbox AI, whitepaper)—This whitepaper is a great primer on how Brainbox AI—and AI in general—can provide a new type of supervisory control for HVAC using deep learning and parameter optimization techniques. I’m assuming many of you don’t have time to read these sorts of white papers, so here’s a summary…
It begins by acknowledging, as we’ve discussed before, that modern buildings need more than static control sequences:
we need to think about energy in terms of flow and modulate equipment performance dynamically in response to the ways in which the internal and external environments change over time
The paper then outlines the three-step process used by Brainbox’s AI: predicting, comparing, acting, and learning from the flow of energy using Deep Learning.
☝️Let’s pause right there… here’s a quick intro video to deep learning if you’re new to the concept.
Okay, back to energy flows...
Every built environment has its own unique set of habits and behaviors that affect the river of energy that flows through it. And, these characteristic behaviors are expressed in a collection of thermal energy equations, which, once calculated, will not change over time for a building unless major renovations are carried out.
Deep learning involves developing a multi-layered brain—referred to as an artificial neural network— that can perform a specific task, like recognize a human face in an image. The process of deep learning works by breaking down this larger task into many smaller tasks, which each layer of the neural network is then trained to perform. Once a layer of the brain has completed its isolated task, for example, establish the outline of the face, it then transforms the image into a new version and sends it to the next layer to be processed.
For buildings, instead of establishing characteristics of a face, the layers of the neural network are finding what Brainbox calls the Leak Rate for each zone:
According to the laws of physics, thermal energy will always work to reach an equilibrium, so energy will always flow across a differential to balance the amount of energy on either side.
There are two critical factors to consider when calculating leak rate. The first is that every barrier has a unique leak rate that is dependent on its material composition and on how the building itself was built. Therefore, two identical windows installed in two different buildings will have different leak rates. The other factor is that leak rates are dynamic and vary according to changing conditions, including occupancy and weather, which means that the flow of energy in a zone is constantly changing over time.
If energy is leaking out of a zone, then maintaining that zone at the derived temperature requires supplying thermal energy to that zone at the same rate at which it is leaving.
Once the leak rate is calculated and predicted, the building’s efficiency at meeting that leak rate is easy to calculate.
The power-to-thermal load relationship refers to how efficiently the engine in an HVAC system converts power into thermal energy that it then delivers to the various zones in a building.
Then the system can adjust parameters to optimize that efficiency:
While the efficiency of the engine is determined to a large extent by the equipment that was purchased and how that equipment was installed, operators are able to adjust various settings on the engine in order to optimize its performance. For example, in addition to being able to open or close dampers, operators can also adjust the engine to serve more or less air/minute and to slow down the production of heat. Playing with these settings makes it possible to fine tune the engine and find the ideal configuration—or combination of settings—for achieving the most efficient power-to-thermal load relationship for a given zone.
And this same optimization can be done to predict and optimize for future conditions:
maintaining balanced thermal equilibrium in a zone in the most efficient and effective way possible—which means ensuring occupant comfort, saving money, and reducing carbon footprint—requires calculating the ideal engine configuration under all possible conditions for each zone within a building.
This prediction is done without the typical full energy or semantic model that is standard practice today:
the AI would not be concerned with the kind of HVAC system installed within the building; it would simply look at the available data and calculate the ideal configuration for the engine from an efficiency perspective for all zones under all possible given conditions.
While you could construct a simulation to model a building’s behaviour, the process is significantly more time consuming and less accurate than using deep learning for optimization. In comparison to a 40-50% accuracy rate when using simulations to predict building behavior, our research shows that AI is far more accurate.
☝️Another pause… It seems like Brainbox’s perspective represents one of two camps that are forming:
those who propose using parameter optimization backed by AI as described here
those who propose using a digital twin to simulate the future and alternative control sequences
Using deep learning, it is possible to predict what will happen in a zone in the future by first calculating what the leak rate will be and then using that projected leak rate to predict the temperature.
With the predicted leak rate, the AI can then test and find the optimal sequence of operations:
When the AI engine, based off of its predictions, sees an unwanted event in the future that will disrupt the balanced thermal equilibrium in a zone, the engine begins testing various micromodifications to the HVAC system to figure out the effects that different combinations of adjustments will have on that zone’s future conditions. The AI engine then uses the results of these tests to evaluate which combination of modifications to which pieces of equipment will, over time, most effectively eliminate the unwanted event. Once the right course of action is determined, algorithms then instruct the HVAC control system to make the modifications to the selected components to get the zone to the desired condition.
For example, let us consider a situation in which the engine predicts that the temperature on a particular floor in a building will be too high in three hours. In response to this prediction, the neural network runs models on different micro-modifications to the HVAC system to see the impact that these possible modifications would have on the predicted temperature increase. Using this information, the AI decides on the best course of action—for instance, lowering the blinds and turning on the air conditioning—that will bring the floor to where it needs to be in order to ensure that the rise in temperature in three hours does not occur. It then instructs the HVAC control system in real time to make the desired modifications in order to create the right conditions on the floor.
And the optimization includes more than just zone temperature. It can consider and optimize for all of the changing and pertinent variables required for our buildings to interact with each other and the electric grid of the future.
Our solution performs these evaluations and modifications in real time for all the zones in a building based on the variables selected for each zone. In this way, by controlling the future in each zone, the aggregate result is that it is able to optimize the energy flow in the entire building without any human intervention. In addition to working on multiple variables for the different zones in a building, our AI engine can also look at energy spending and carbon emissions as part of its decision-making process when trying to figure out the best course of action for eliminating unwanted events in the future
OK, that’s all for this week—thanks for reading Nexus!
If you have thoughts on this week’s edition, let us know in the comments.
And don’t forget about the Nexus Pro early adopter discount: