Lifetime Learning for AI Everywhere

For decades, from even before we called everything the IoT (Internet of Things), maintenance has been the barrier to digital sensing and operating of the physical world. Wired sensors were reliable, but expensive to install, and often an esthetic nightmare once installed. With self-power, sensors became cheap enough to put everywhere, but faced a new challenge—maintenance.

For a long time, deployments were limited by battery life. Many initiatives were short lived, running until the batteries wore out. Changing the batteries was expensive, sometime more than the initial installation. Committed organizations developed scheduled battery changes to control costs, just as they had done before for re-lamping projects.

We solved that problem by making sensors so cheap we could just leave them and install replacements. Or we (notably members of the EnOcean Alliance) tuned communications to be so light-weight that in situ energy harvesting could keep systems working.

Now we face yet another maintenance challenge, that of intelligence management.

Today’s sensors have become smarter, sometimes referred to by the indeterminate name “edge devices”. Sensors and Edge Devices likely transmitted more than 20 zettabytes of data for central storage last year, although there are no firm estimates on 2017 data gathering. With that much data being stored, the communications requirement was easily in yottabytes.

This much data creates a new challenge. The IoT not only requires that we get actionable information that matters, but that we get it before it is too late to matter. There is too much data and too many situations to rely on timely central decisions.

The enable drinking this firehose of data, we are starting to rely on sips at the edge. Edge Devices are making the initial decisions as to what data means, and what data needs to be brought into the middle. Local decisions are made faster, without interference from temporary high priorities elsewhere in the IoT. For all but the simplest scenarios, this model requires learning at the edges. There are large open source libraries now of Artificial Intelligence (AI) code for Raspberry Pi and Arduino.

This presents a new maintenance problem, managing and updating AI routines and algorithms.

The big software companies are preparing the tools we will need. Thousands of AI systems in each building will require tools to manage the rapid evolution algorithms. New algorithms will require managed roll-outs Rapid evolution forces diversity of algorithm and information as systems will change far faster than their installed life. Oracle is pushing GraphPipe, an open source software project for efficiently deploying and managing AI models at scale. Microsoft is right there with them, with large platform management announcements expected this Fall.

The problem of managing intelligence in millions of devices is solved already, before most people know they have the problem.

In the last year Pi architecture devices have blown right past the $40 and even $20 price points, with full systems expected for $7 and perhaps $4. Arduino platforms not only run open source Linux and Android, but with open source hardware offer potential easy integration directly onto integrated specialty hardware components.

The barriers to fully intelligent small systems across every aspect of buildings are falling even faster than pioneers such as Alper Üzmezler and the Project Sandstar for smart controls have publicly projected. It is my personal belief that while full platforms such as the Pi have higher initial costs, in part because they include a GPU not needed to manage a display, that this means they are pre-adapted for high speed signal processing. This will not play out slowly.

Leaving IB-CON with Microgrids and DC Distribution in mind

I write (and post,you have to be amazed at the technology we so take for granted) this on a plane flying away from a great IB-Con, the REALCOM Intelligent Buildings Conference. It is a trade show like no other, with deep involvement of both the technology leaders in Real Estate, i.e. CIOs and CTOs of the largest REITs and those that would sell to them. The panel discussions were embued with new issues tied to deep adoption of IT not only into the corporate operations of estate, but into the operations of the hidden systems within.

Deep analytics and deep security concerning embedded systems, BAS and others were recurring themes this year. There were frank discussions of using the BAS to get to corporate information, and of using hacks to destroy building internal operations. There was just enough White Hat “think like a hacker” to keep the talks interesting.

But what really stands out at REALCOMM in the focus on emerging technologies. Jim Young and Howard Berger have a genuine interest in start-ups, identifying the ones that could do a lot of good, and helping them to meet their early hurdles. New companies may get coached on messaging and presentation. They go out of their way to introduce potential risk-takers with the new technologies. I have even listened in as companies just out of angel funding get coached through their next steps. The unseen services these two provide are immense.

On the other side, they create a real community among the technologists on the ownership side of real estate. Some come back year after year to challenge each other with the changing world of real estate. I have written here before of the challenges of setting up start-up office for millennials, of coffee shops and food trucks replacing the in-house conference rooms and in-house sandwich shop.

Some of these owners have set up their own coaching for new tenants, helping them with marketing, and financial planning, and other topics the young founder of a new venture may not know. At one level, this is raw self-interest, for a tenant that goes out of business is a tenant that breaks his lease. But at another level, and I think a truer level, it is a commitment to helping other flourish, so long as they learn and work hard, so that we all flourish. And I think this commitment and community starts with Jim and Howard.

My most immediate concerns this year were microgrids and semantic frameworks, as well as the Energy Mashup Lab. These topics are no surprise to my regular readers.

A moderated a microgrid session with CleanSpark and Stem, two technical companies with quite different focuses. Because another vendor, an early start-up, dropped out, I expanded my own comments on personal microgrids. What was remarkable was how each participant agreed on the big issues, the big benefits, and the driving forces. As an industry, microgrids are now know where they are going. Years ago, I moderated similar sessions, and the speakers were coming out of the labs, with vision, but not yet much delivery. Today, either of them, and maybe a dozen more vendors, can deliver systems out of the box.

Those systems are quite different though. They share a commonality of benefits: lasting reduction of energy risk, capabilities to work with real energy markets to reduce costs, a capability of consuming local storage for local purposes rather than the dead end of net metering, and privacy and security for the building and its occupants. The prices are coming down, leading to three-to-five year ROIs on pure energy costs without pricing the other elements. The risk is now low. The question is now moving toward “Does a microgrid make sense in this state with these regulations?”…and regulatory frameworks are starting to predominate. Keep an eye on these technologies, because if you have a site with greater than average price risk, or reliability risk, or security risk, you should be considering a microgrid now.

At the end of the day, I finished in a discussion of low voltage DC lighting. Again, long-time readers know I have been enthused by this technology for five years. It is now coming to market (LumenCache) with standard parts, standard high-performance LEDs, modular component s anyone can install and maintain. I hope to learn more about this company and its products in the weeks ahead.

Which makes me look ahead. Is it time, at last, for the eMerge alliance, and for DC-based distribution inside the building to come to the fore? Storage (batteries) are DC. Solar PEV is DC. Digital electronics and LEDs are DC. With less need for heat shields and conversion, LEDs are cheaper, safer, and more reliable. Without the need to convert from DC to Ac to DC (storage) and from DC to AC to DC (storage to use), there is a 30% “free” increase in efficiency. With enough distributed energy generation, DC power, as Edison thought it should be, may be back.

That’s all for now. I’m tired and travelling.

Privacy, Self Defense, and Smart Energy

I spent some time last week down a country road, watching the local power. I watched three phases that were greatly out of balance. I observed trapezoidal wave forms. We could see the home appliances of everyone else on the road, as they each turned on and off.

Together, we watched the power coming into his lab. They were his neighbors, and he knew them from observation. He could relate when they changed their appliances, and how they lived their lives. He could tell from the patterns how they affected the shared local electric distribution circuit. There were some especially odd patterns, second level harmonics that caused some unusual recurring spikes. It wouldn’t be hard, simple machine learning, really, to learn these special patterns for some types of equipment, and then to search for them.

This is all so much easier than it was even a couple years ago. Affordable gigahertz sampling is no longer cost prohibitive. Industrial espionage can be done from across the street. Soon private detectives will be able to read the activities in houses from down the street, using only a power connection, pattern matching against an on-line database, and a little creativity. Your house and business is now an open book, with or without the participation of utilities.

This technology was not built to look out, however. Monitoring and analyzing the distribution feed is a mere side effect of the system I was checking out. The purpose of these systems is not to spy on the distribution system, but to defend against the distribution system. What we could see on the samples is also felt by the building.

The purpose of this monitoring is to fix the power inside. Each phase of power is simultaneous corrected to near ideal wave forms. The effects inside the building are extraordinary. When supplied with an ideal power wave, electric motors become audibly quieter. While that alone makes an industrial space pleasanter, it reflects an underlying reduction in vibration and in generated heat. At the same time, the motor begins to operate at its faceplate output.

This is what I mean by defense against the distribution system. Excess vibration, and the associated noise and heat, are caused by the noise on the electrical supply, by wave forms that are less than the ideal. Traditional power conditioning systems often create trapezoidal or triangular wave forms—they may protect from spikes and sags, while they increase wear and tear. It’s too early to predict how much ideal power forms will extend the life of equipment, but reduced noise and reduced heat are strong benefits on their own.

While one can hear the change in motor operation, florescent lights and digital equipment benefit as well. Long time readers of this blog know that my house is beset by something that causes even my incandescent lights to fail in clusters. Having watched the power on this nearby local distribution loop, it seems likely that I have seen the answer, even while all parameters are “within spec for home distribution.”

The plan of course, is for the local distribution to get worse. While we watched, we saw changes to power on the entire loop when the charging of a single neighbor’s electric vehicle began. Even the best solar panel installations affect these wave forms, and most installations are far from the best. The effects not only damage neighbor’s equipment, but they may increase metered power use for those neighbors as well.

Defense from the grid, especially from the smart grid is an important new market. Distributed energy resources are in all our future, and they make such defense more important.

An allied outcome of this defense is that the view from the outside is obscured. The systems behind the power controller cannot be inspected as we inspected the neighbors. Unbalanced power use, that is, power unevenly spread across the three power phases is balanced on the outside of the controller. Power factor is optimized. This ideal power load reduces metered power, often substantially. The operation of individual motors and digital systems looks from the supply-side as a single ideal consumer. Energy-use privacy is protected and restored.

As consumers, we don’t yet know how to think about and use this kind of product. As smart energy, distributed energy resources, and electric vehicles become more widely deployed, we will want to learn.

The Taxonomies of oBIX

Note: Niels Bohr famously observed that prediction is very difficult, especially about the future. Getting down into the technical weeds of a specification that is not yet complete is also difficult. I received numerous requests to explain how Haystack fits into future versions of oBIX. OBIX is a specification whose development is in mid-flight. OBIX 1.1 comes out for its first public review in July. The enterprise wrapper for oBIX, aka oBIX 2.0 is months away. Perhaps some readers here will join and help us get to the final form faster.

OBIX does 1.1 not require or support Haystack. OBIX 1.1 will not even mention haystack, except, perhaps, as an example. OBIX 1.1 will be able to provide metadata for any point. That metadata may be drawn from any formal or informal taxonomy. oBIX 1.1 does not define how taxonomies are applied to an oBIX server. Haystack is useful taxonomy of growing popularity that can be used to provide metadata about any oBIX point.

Haystack is a taxonomy that describes a lightweight building information model (Slim BIM) for BAS systems. Haystack tags are unique in that they were developed as a folksonomy, i.e., through an informal consensus among users. Haystack advocates may point out that all the formal taxonomies once created to classify internet searches were beaten by the automatically generated folksonomy at the heart of the Google search engines. Traditional large BIM models provide taxonomies developed through formal processes and often mandated by national agencies; metadata in oBIX can be the entry point into Big BIM. OBIX is taxonomy agnostic, and can support both, or either.

Interactions with an oBIX server begin by entering the “lobby” and asking for information about the system. One of the new inquiries in 1.1 will be “Which meta-information standards do you support?” A valid answer is “None”. For backward compatibility, an error message, from an oBIX 1.0 server that does not understand the question must be interpreted as answering “None”. If the oBIX server supports one or more meta-information standards, it will name them. We have not spent much time on the Lobby inquiries yet, but I think this answer should include a local tag, a URI for each taxonomy, and an optional URL for queries based on that taxonomy. Those queries are a subject for oBIX 2.x.

Under oBIX 1.1, a client can query a point for its metadata. The oBIX server returns a collection, with each element including a tag identifying the element’s taxonomy, and the metadata information. If some of that metadata is based on Haystack, then the returned metadata information may include one or more Haystack Tags. The same set may include elements drawn from other taxonomies. It is not hard to imagine a single BAS gateway that supports a Haystack, EMIX (Energy Market Information Exchange), Tenant Information, and situation awareness / security.

There are many taxonomies for building systems already in wide use. Walmart and Target, two companies that have unusually complete construction and commissioning specifications, have long mandated the use of specific tagging standards. The Intelligent Kitchen standard, promulgated by McDonald’s could specify a meta-information specification. Many use oBIX to interact with control systems that have nothing to do with BAS. Groups such as OPC, used widely in industrial scenarios, have their own taxonomies. SensorML, a standard developed by the Open Geospatial Consortium (OGC) is widely used for scientific observations and for situation awareness; SensorML provides a taxonomy that can easily be applied to oBIX points.

Every taxonomy is the outward manifestation of an information model. Haystack assigns responsibility for assembling a building’s specific model to the client. The client must assemble the sum of all the tags, and follow all the references, to create a coherent model of the systems exposed. There will be many incomplete models generated from BAS gateways that are badly integrated or commissioned. To enable a client to query the model directly, the server itself must have a model. Model-based queries are part of oBIX 2.x and have no place in oBIX 1.x.

Not all BAS systems need to or will incorporate model service or even meta-information. It is easy to imagine an information appliance that acts as the model holder for an underlying metadata-free [BACnet] system. Such a system would provide direct access to the points in the underlying system, and offer up the meta-information provided by the taxonomy. There might be advantages to setting these up as audit-servers unable to interfere with the underlying control operations. A standards-based BIM server, serving up BIMSie, may be an example that brings such systems into conformance with DOD and EU expectations without requiring re-development of the underlying control protocols.

We should resist the impulse to develop the one, true, absolute application model for all time, and baking the taxonomy that represents that model into every low level protocol everywhere. What we should do, is develop standard lamina, layered information models that live outside the work of an individual integrator, but provide higher level access that increases the value of the initial integration.

Consider a microgrid consisting of a green building, and an oBIX serving using Haystack to describe its underlying systems. Alongside could be an oBIX server managing solar generation, and another managing private wind farm. The oBIX gateway to these distributed energy resources could support SensorML-derived tags, useful to describe the weather and environmental data gathering that best predict energy generation. All three systems could also support the EMIX taxonomy to describe the energy supplied as well as the energy used in the green building.

oBIX works with collections of points named Contracts. Within the simpler taxonomies, one can imagine building a contract to include all points with a given tag. A more interesting query might leverage the model in the taxonomy; for Haystack, this might include all temperature sensors on Air Handlers with a relation to a given chilled water loop. Some queries will not be answerable from a single interface. An external BIM server might be the appropriate way to build a query against a more complex taxonomy. Such queries are out of scope for oBIX 1.x; we intend to define a model for such queries within oBIX 2.0.

The most interesting contracts will be built from querying two or more taxonomies at the same time. Look to a generic query language for both intra- and cross-taxonomy contracts in oBIX 2.x. We have some ideas on how to do this already, but that is much, much deeper in the weeds then I want to go at this time.