Data Streams + Tableau

We’ve begun preliminary scrubbing of the data and have created a rough skeleton in Tableau for our cloud sourced environmental data to feed into. Ryan will be culling his energy data as well and we will be able to integrate it into our dashboard to display multiple variables at once. The goal is to have a web (browser) based dashboard display that will link to data in the cloud and update as the fields are repopulated. This web dashboard can then be plugged into a project site, our DLR Group home page, or a client website.

Data Collection

With our software now in place, we are collecting information on several sets of workstation energy-related data.  To reiterate, after careful examination, Michael and I determined that collecting it every second or even every minute might be a bit excessive and not provide any relevant information that every 15 to 30 minutes couldn’t.  Although our devices give a live feedback, we’re trying to determine whether that is as important as collecting and storing large amounts of data that is useful to a client over and extended period of time.

For example, would you tell a client that at any particular second such as 12:55:03 they used X amount of energy and at 12:55:06 they used the same amount?  No!  We imagine they are looking for big picture items.  Energy spikes.  Vampire loads.  Load balancing.  Have you ever imagined having a circuit panel that is smart?  It would be something to help re-evaluate energy efficiency and reconstruct it to predict energy usage?  The below graph represents twin energy spikes and also the monthly low.  How can we design a system that avoids unwanted “H” spikes and maintains lower “L” points.  This graph demonstrates that the building users are generating an overall upstream use of their energy.  That is bad.  Without Data Streams, this path could continue onward without reflecting the increased energy usage back to the occupants.  Let’s try to grab those “H” levels and bring them down, helping to change the trajectory of our energy use graph.

Data Streams Graph

I mentioned earlier that we have several sets of workstations (by employee) energy usages.  Let’s start to examine the most curious one.  Conference Room Guy.  The conference room isn’t an employee of DLR Group, however he uses energy just like a person would.  Conf. Room Guy has his own room, monitor (HDTV), laptop and controls.  Let’s investigate what he uses when not in use versus when in use.  Cue random numbers on excel spreadsheet.

Excel Image

So I’ve taken the liberty to manually evaluate what all these numbers mean. I’ll have to admit this sooner or later, but my background is actually in engineering.  Architectural Engineering class of ’01 with an emphasis in mechanical design.  How I ended up a licensed architect is another story.  So excel and five decimal digits is a familiar scene to me.  Numbers are really the stars here.  What do we see in Conference Room Guy’s stats that are easy picking?  I’ve noticed that the energy use stats are three times higher for about 90 minutes at a time.  What that tells me is that when the conference room is in use, we’re pulling nearly THREE times the wattage as when it is not in use.  That is PLUG-LOAD levels ONLY.  This data goes back as far as January and its easy to see when we use the conference room.  So I know what each month uses for each workstation on a consistent hourly basis based on month.

Wouldn’t it be interesting to see if a computer could predict hour by hour usage based on 10 years of conference room guy experience? Yes, we’re trying to get there and the start of every great journey begins with the first step.

Today is that first step.


Even Edison had setbacks

So for full disclosure we did say we’d post our triumphs and failures.  Today I discovered a failure.  The energy monitoring devices have a touch sensitive on/off switch.  (Don’t ask me why an outlet needs a touchscreen?!?) but if you have something dangling from your monitor that can carry static electricity.  That can be enough to switch it off.  So you could be modeling a whole bunch in Revit and then <Boom> monitors and computer are suddenly off.


Fortunately I NOW have a backup plan.  An APC battery backup plan!

Battery Backup

Live Motes!

Sensor Motes are up and running in two locations at the Seattle Office – Cafe and Large Conference room. Two more sensor motes are in production to measure open office environments as well. We will continue to monitor and troubleshoot the motes over the next several weeks until the bugs are ironed out.

Data bias

Researchers use plenty of statistical methods and mathematics, in isolation or combination, to turn data into a prediction.  What starts as collection, quickly turns to analysis.  The inevitable “what can we do with it?” and “why are we doing it?”

Ironically a little data coding can answer these questions.


<Dataset1 = N+1)


Do you know why you are doing this?

If Yes, goto <code1>

If No, Stop <end>


dir <b>

Do you know what to do with the information?

If yes, goto <code2>

If No, stop <end>


dir <b>

Display results for dataset(N+1)

<goto formula>

Usually you have to go through the process four or Five Times before you asked why enough to get to a credible answer.  A technique originally developed by Sakichi Toyoda.  Name sound familiar? The process was included with the Toyota Motor Coproration during the evolution of its manufacturing methodologies.  The beauty lies in the simplicity.  It has no bias correlation.  Data bias has the capacity to enrich or destroy entire libraries of information related to your research.  The road to discovery is almost always led by un-corrupted data delivery methods.  i.e. no human interference to slant it one direction or another.  If you ask your data, “Hey data, I want you to do “this” for me.” Ultimately it can be geared towards a bias to collect the information in a particular way to suit your need.  However, this isn’t truly accurate.

Research begins with a hypothesis, not an answer.  It is easy to confuse the two, especially when you are desperate for an answer.  Hypothesis should lead to a tenable theory, which should lead to an answer or another question.  The conceptual framework is the analytic tool used to make distinctions and organize ideas to get to the results. Read “Good to Great” by Jim Collins describing the differences between a Hedgehog and Fox as a way to organize a principle to view the world.  Once you have the framework in place, you can begin to set categories of results in place.  This will help define the ultimate goals of your data collecting.


For Data Streams, that is a series of building related data sets that will help design professionals determine outcomes before they happen and if caught early enough, corrected before the final design goes out.  Imagine if you will, a simple rectangular office layout.  Cubes on one side, big layout table in the center and project viewing wall on the other.  The design team promotes that clients should walk on the viewing wall and staff should walk on the cube side.  The center table divides the room equally in half.  Which way do I turn if I’m the client?

There is no realistic answer to this because of so many variables.  Personal preferences, time of day, mood, level of interest, and so on.  Let’s try to measure what we can do to change that outcome.  Could we simulate a different organization of the room, by say, moving the center table slightly closer to the cubicals? In theory, that SHOULD define the space so that clients know that the wider side is for them. Is that what will happen? What influences their decisions and what happens when they make it through the end of the table, stop at a cubicle and decide to walk out of the office on the now narrower cubicle side?

Yes, that takes time.  But if you don’t find time to do it right the first time, when will you find time to do it over again?

The video below demonstrates multi-point infrared motion tracking connected to a computer that allows for deliberate motion sensing of occupants.  In this instance I’m experimenting with the Leap Motion controller.



Social Media

Stay up-to-date with our latest findings and progress