EMAIL INTERVIEW  Anwer Bashi & Ken Sinclair

Anwer Bashi
Senior Research Engineer
Computrols, Inc.

Fuzzy Logic

Fuzzy logic is a proven AI technology for mapping human knowledge into computers and creating sophisticated control strategies using human-understandable logic. As building automation technology gets more complex and interconnected, and as more is expected from building engineers, artificial intelligence will have to play a greater role in building automation systems.

[an error occurred while processing this directive]

Sinclair:  What is your background in fuzzy logic and building automation?

Bashi:  My bachelors degree in engineering was focused on controls and cybernetics (cybernetics is related to artificial intelligence and the man-machine interface), my masters was in estimation, and my Ph.D. coursework focused on hybrid and multiple-model estimation. I started working at Computrols doing building automation in 1999.  I am currently employed as the Senior Research Engineer at Computrols.

Sinclair:  Your article generated significant interest here on AutomatedBuildings.com, did you receive any feedback?

Bashi:  The feedback was mostly very positive, but I also received an interesting and somewhat amusing email from someone who seemed to be upset that I was trying to mathematically explain away the meaning of life. Just in case anyone else thought this, I want to make it clear that I understand that fuzzy logic is math, and is obviously aimed at solving math problems. “This statement is false” is semantic nonsense, but it is probably the most famous and simple example of an expression that is contradictory using bi-valued logic, but solvable using a multi-valued logic (i.e., fuzzy logic). So, for the record, I don't think that fuzzy logic replaces philosophy or spirituality any more than algebra does.

Another interesting email I received was from someone who had considerable experience in the building automation industry. He wrote that fuzzy logic has been explored in a limited fashion by his former employer (one of the major players in the building automation industry), but had eventually been rejected.

Read the article - Artificial Intelligence in Building Automation: Fuzzy Logic - published July 2006

Sinclair:  If it has been tried before, why do you think it failed?

Bashi:  Obviously, it's hard to say without the details, but adding a fuzzy logic variable or statement without sufficient support does very little to help in building automation.

What I mean by that is that there are many support technologies required to make it truly useful. For example, the logic should be customizable in an easy to understand language that already has as much relevant information programmed into it as possible. Even better, it should be easy to select a control strategy from a list of strategies prepared by experts, and stagger them in a way that does not cause one strategy to fight another.

Also, fuzzy logic is only one tool in many that are required to enable smarter building automation. A fuzzy logic statement cannot tell how similarly a set of air handling units or valves are operating without being able to quantify the operation using a system model of some sort (interested readers can Google “system identification” or “probability distribution functions”). Also, failure modes can be more easily detected if the fuzzy reasoning can make use of probability theory (“Bayesian Reasoning” or “Multiple Model Estimation”). I didn't touch on these topics in the article since it would have made it less readable and significantly longer. Of course, the person programming the system should not have to know how these things work, just that AHU11 is exhibiting behavior that places it outside of the 99% confidence interval of normal operation.

Finally, if system information on how the points relate is not automatically provided, it would place a burden on the system programmer to fill in the blanks. A VAV controller is often connected downstream from an AHU. If this information is made automatically available to the system, then a single logic statement can be written that applies to all VAVs, backing off the air pressure in a duct if every one else connected to the same duct is at minimum cooling (for example).

[an error occurred while processing this directive] Sinclair:  Could fuzzy logic or artificial intelligence be used in ways other than control or fault detection?

Bashi:  Absolutely. It could be used, for example, in reporting.

It is possible to calculate a number called a “surprise factor” when looking at statistical data. Basically, it is how far measured data differs from the expected, so for example, if an AHU were operating somewhat differently from the others, it would be surprising.

Another number is a “novelty factor”. This is how a conclusion reached by analyzing a subset of the data differs from conclusions reached by analyzing the rest. For example, if the general consensus is that heating strips should not come on until the VAV box has backed off to minimum CFM, and a small group of VAV boxes are breaking this general rule, then it can be considered novel. I know this seems to be saying the same thing as in the last paragraph, but a “novelty factor” corresponds to a new rule that can be derived from some of the data that is unexpected when looking at the rest of the data, while the “surprise factor” can only really be applied to mechanical system operation.

So imagine being able to run a report where you can tell the computer that, in addition to the other standard metrics you might want to see in a report, you also want to see the 20 most interesting things about your building. Some of them may be of no consequence or not directly relevant (many more overtimes were scheduled this month than usual), some may be interesting (while cooling tower #1 is meeting setpoint, the system transfer function has suddenly changed) and some may be very significant (more energy than normal is being used by chiller #3 per tonne of cooling).

The math to do this sort of analysis has already been developed. Marketing companies know, for example, that people are more likely to buy beer when purchasing diapers (for some reason), and so the beer aisle will be placed on the way to the diapers aisle. It is just a matter of taking known math and applying it to building automation.

Incidentally, the amount of computation required to do some of this analysis is phenomenal and may require many hours to run, depending on the amount and scope of data.

Sinclair:  What are some challenges of using AI in building automation and how do you plan on dealing with them?

Bashi:  I think that probably the greatest challenge will come from the lack of information. While it is obviously impossible to include every piece of information in a system, it is possible to create the BAS so that as much information as possible is added effortlessly when being initially programmed.

For example, most variable air volume boxes in a building will be downstream from an air handling unit and be very similar in make and purpose. Rather than taking a bottom-up approach, where controllers are added to the database, and points programmed and named, imagine going with a top-down approach.

Let's say we want to add 8 VAV's to AHU11. We select from a list of VAV makes or profiles, and choose which expert systems we would like them to support (expert systems in Computrols-speak are combinations of Fuzzy Logic statements and other AI constructs). The VAV's can be added all at once, set up hierarchically downstream from AHU11, with each point programmed with the correct profile (setpoint, space temperature, supply temperature, actuator, etc.). Each profile comes with default programming such as PID's, alarms, binary lock-outs, and logic.

Since the VAV's and AHU's know where they are in the buildings mechanical architecture, the binary lockouts can automatically reference the appropriate AHU, and the AHU knows which VAV's to use for throttling pressure or adjusting temperature for optimal environmental control and energy savings. Also, if multiple units share the same profile, it is also safe to assume that they should behave in a similar fashion. This allows fault detection and diagnosis by allowing the aggregation of more data.

Of course, it must be possible for the building engineers to modify the VAV templates and move the assigned points around since some installations may not perfectly match the provided templates, or some engineers will want to add novel functionality to the system.

Sinclair:  How do you think this would impact the working life of building engineers in the near future?

Bashi:  Artificial Intelligence in building automation is not one of those technologies that would reduce the number of engineers required to run a building, but rather one that will make the engineers more efficient and effective.

The building engineer would still need to oil a stuck valve, but he would know much sooner that it was stuck. System-wide optimization could allow for better energy savings, and it would be easier to perform a high-level audit of the health of a building (with an intelligent summary available which goes into as much detail as desired).

[an error occurred while processing this directive] Sinclair:  You say in the article that “fuzzy logic may indeed be the next frontier in building automation”. Could you elaborate.

Bashi:  While I think that anyone trying to predict something about the future should be willing to be wrong about half the time, I also think there are some indicators that the use of artificial intelligence in building automation (for which fuzzy logic is an important tool) is likely to increase.

Complexity:
As systems get more complex or more is demanded from a building automation system, it becomes increasingly harder to ignore their interconnectedness. And as computers get more intimately involved in optimizing a controls strategy (for example, picking the optimal chill-water temperature to run in a building), we need to have a better way to transfer human knowledge and intent to the computer. Fuzzy logic is perfect for translating human into computer knowledge.

Social Factors:
Another trend we are seeing is that those in the automation industry are being required to provide more information to others involved in the IT or management departments. AI-like techniques could provide a much more useful report than statistics alone.

Market Forces:
Energy efficiency is getting more important as fuel prices rise. While there may be some fluctuation in daily energy prices, there is little to indicate that the trend will reverse and energy will get cheaper in the future. This is, of course, barring new miracle technologies such as cold-fusion or anti-matter reactors.
This means that the best building automation systems would need to be able to optimize the building with a level of intelligence and control greater than that currently available.

Technological Developments:
Significant computational horse-power is required for many AI techniques. As micro-controllers become faster and cheaper, this becomes less of an issue.

Sinclair:  Could you very briefly summarize what you would want readers to take away from your article.

Bashi:  Fuzzy logic is a proven AI technology for mapping human knowledge into computers and creating sophisticated control strategies using human-understandable logic. As building automation technology gets more complex and interconnected, and as more is expected from building engineers, artificial intelligence will have to play a greater role in building automation systems.

[an error occurred while processing this directive]
[Click Banner To Learn More]

[Home Page]  [The Automator]  [About]  [Subscribe ]  [Contact Us]