Predictive Maintenance
Guests: Tim Darrah – Intelligent Systems and Rick Bennet – National Facilities Solutions
00:00 – Intro
10:48 – What is predictive maintenance and how does it work?
13:27 – What are the benefits of predictive maintenance?
19:20 – Are fault detection and predictive maintenance the same thing?
19:46 – Many facility managers have concerns about this being a replacement for their services. What would you tell someone with this concern?
24:10 – What does it look like to implement a solution like this?
30:48 – How do you evaluate a service partner in the predictive maintenance field?
36:56 – Can the time horizon of the notifications be adjusted to meet the needs of your customer?
39:31 – Tim discusses the false positive rate of predictive maintenance programs.
42:16 – Rick discusses the importance of establishing baseline metrics before implementing predictive maintenance program.
Helpful links on the topic:
IAQ Tools for Schools: Preventive Maintenance Guide
Condition Based Maintenance Plus (DoD Guide Book)
How AI can save money, spare lives, and reduce downtime
NAVSEA’s Condition Based Maintenance Plus (CBM+) Initiative
Full Transcript
[00:00:02.760] – Scott Holstein
Hello, everyone! Welcome back to the Building Technology podcast. This is Scott Holstein, and today I’m joined by Tim Darrah of Intelligent Systems and Rick Bennett of National Facility solutions. Rick has over 25 years of experience in the HVAC industry as a service technician, supervisor, instructor, designer, LAN commissioning engineer. His areas of expertise include high and low-pressure chillers, large LX systems, DDC controls, Pneumatics Hydronics, air makeup, computer room systems, process equipment, steam, high and low-pressure and frequency drives. Tim Darrah, on the other hand, served seven years in the US Army as an airborne infantryman and later an Avionics technician.
[00:00:51.620]
In 2017, Tim graduated from Tennessee State University with a Bachelor of Science in Computer science. Prior to entering the PhD program at Vanderbilt University, his focus at Vanderbilt was on extending the current state of the art in health management technologies for cyber-physical systems to include system-level prognostics, predictive, maintenance and decision-making. In 2019, Tim started his company Intelligent Systems and has been working with National Facility solutions to develop their flagship product, Facility Coach, which gives facility managers the predictive edge when it comes to maintenance and indoor air quality.
[00:01:33.520]
In 2020, he was awarded the prestigious NASA Fellowship to support his thesis research and is working towards embedding the technology behind Facility Coach into i-performance a National Facility solutions product so that he can focus on finishing his degree. Thank you both for joining us today. How are you guys doing?
[00:01:54.270] – Rick Bennet
Doing well. Thank you.
[00:01:55.650] – Tim Darrah
Doing well.
[00:01:56.810] – Scott Holstein
Good. Good. So Rick, starting off with you, 25 years of experience in the HVAC industry, you probably feel like you’ve seen it all, but tell us a little bit about how you got started and how you’ve seen things change over the years.
[00:02:12.120] – Rick Bennet
Thank you for having me. Yeah. 25 years. Quite some time.
[00:02:18.580]
Yeah. I mean, once I think I’ve seen it all, certainly see something that reminds me that I have not bas started off early in my career as an installation mechanic. Lan progress through the years, up through the trade, through various means of experience and education. I was a field technician at one point in my life and worked on installation and repair and rebuilds a centrical chiller. I worked in DDC Controls, large boiler plant then ended up going to school to get my engineering degree, which led me into a different phase of the business.
[00:03:10.440]
Collectively, I like to think of myself or describe myself as a Swiss Army knife with all the different experiences being able to have the experience of installing, service, and installation of the equipment and then with the engineering, my education has really made me a well rounded, focused individual. So that leads well into, you know, the new sector that I moved over a decade ago, which is the commissioning side of the consulting world. So collectively, again, really good experiences and this industry just changes so fast. It’s good to have that background to follow up on.
[00:03:57.740] – Scott Holstein
Yeah, absolutely. And I think that pretty much anybody you work with probably appreciates the fact that you have that kind of field experience. One of the things that I love talking about is our CEO started off as an electrical technician at Computrols 10, 12, years ago, something like that. And it just gives you a much different perspective.
[00:04:20.200]
So. Tim, reading through your here, you’re building a very interesting resume for yourself. You’ve done a lot of different things. Not a lot of people have NASA their resume. So that’s extremely interesting to me. But tell us a little bit about how you progressed into this facility coach product and what’s brought you to the HVAC and facilities industry?
[00:04:47.760] – Tim Darrah
Really started out as a couple of my buddies commiserating in the bar about their problems managing mold outbreaks. And one of his facilities, actually, several of the facilities. And they looked at me and they said, you’re a smart guy, you know, how to program and build sensors and stuff like that. Is there a solution to this? And I said, really, there’s a solution to everything. It’s a matter of do we have the technology available to us to implement it.
[00:05:22.460]
And so I kind of went down a side tangent on that and built a prototype system and put it up in my house, in that case, dealing with mold, the parameters, really our temperature and humidity. And there’s a lot of research out there that talks about the growth zones for different mold strains, given temperature and humidity. And so I said, it looks like we can get you some type of early warning detector when conditions are conducive mold growth. So we kind of have to be careful with the language that we use.
[00:06:04.150]
And they said, great, you know, let’s push this further. And so we started a company and then at Vanderbilt University, I went through the proper channels in terms of the technology transfer office and then going through some of the entrepreneurial boot camps, if you will, at the Wondery, which is vendor built center for Innovation. And then that led to a small grant that got us at a conference. And that led to some other connections. And then we ended up applying for and receiving a small National Science Foundation grant, to further go out and perform what’s called customer discovery. And at this time, we’re really starting to focus on mold growth and indoor air quality. We haven’t really, at this time, gotten very much into the maintenance aspect of things. But through the NSFI core program, that’s when we realized that we needed to make a slight pivot here, because really, what’s going to drive the pen on the paper signing the check is the bottom dollar. And at this time, this was pre-COVID, so indoor air quality wasn’t really pushing that button very much.
[00:07:26.040]
But, hey, if you can tell me when my unit is going to go out before it goes out. That’s revolutionary. And so we started to look at that in more detail. And that’s where this idea of facility coach came in. So when you have access to the data and when you have the skill set of a machine learning engineer such as myself or many, many other individuals out there who are in varus computer science programs, you have to have these skills. Then you could start to use this data, apply your knowledge, then come up with some useful algorithms that might say, hey, the performance of this air handling unit has degraded, and then you bring in some domain experts, because, as you mentioned, I’m not an HVAC expert. Rick here is you bring in these experts with domain knowledge that and now you say, okay, what do we do? What do we do when the data is telling us this stat, when Rick might say, well, this means the coil probably needs a clean or will this means the blower motor is about to go out. So once you start tying these pieces together this package, then you could stat start delivering valuable solutions to other customers. And that kind of circles back to your question.
[00:09:00.530]
I got roped into this because of my expertise when it comes to data analysis and machine learning and focus, as you mentioned earlier on cyber-physical systems. And that’s literally anything that has a wire going to it a line of code written for and interfaces with the environment. So that’s definitely the mechanical system inside of a facility falls into that stat category. And that’s how I ended up working on this stuff.
[00:09:29.000]
The topic for today is predictive maintenance. actually talking to one of my co-workers today, and I was like, recording a podcast on predictive maintenance like, I thought it was preventive me. So I’m like, well, it’s both. So I’m excited to get into the topic, but before doing that, I know, Tim, you just touched on a little bit there, but is there anything else that we should know about this facility code product that you all are working on?
[00:09:56.200]
It’s early stages right now, and hopefully, within the next six to twelve months, it will be embedded into the product that National Facility Solutions uses for their customers. And we’re in the early stages, and we are looking transition certain things from preventive maintenance to predictive maintenance.
[00:10:23.500] – Scott Holstein
Well, with that said, let’s go ahead and get into the topic today. And I always try to start at the most basic level. I don’t want the first question to yield an answer that’s going to go over any facility manager’s head. Can you guys just start off by talking about what predictive maintenance is and how it works?
[00:10:46.260] – Rick Bennet
Sure. So I will take the first shot at that here. And I kind of put in my simplistic terms. But for me, predicting maintenance is an asset management practice of repairing an asset or piece of equipment before fails based on data received about it. And for me, it’s the third phase of asset management, and the first phase of that management is corrective. Maintenance when repairs are made after a problem or failure occurs, preventive maintenance is scheduled repairs made based on experience and predictive maintenance repairs made because data for an asset indicates that a failure is imminent in, you know, for me, when I think about that, I think about driving down the road and my “check engine light” comes on my car so that’s too late. Right? So there’s something already wrong. So when I’m looking at this, it is. Hey, you know, let’s get to that problem before that light comes on.
[00:11:50.830] – Scott Holstein
I like that answer. Tim, anything to add to that?
[00:11:53.660] – Tim Darrah
Yeah, that was well said and the only thing I’ll really add to that is the real key difference as Rick was mentioning is that between preventive maintenance and predictive maintenance. Predictive maintenance, you’re using historical data in real-time data the system as it’s operating, to drive the maintenance process. And so whereas preventive maintenance, as Rick mentioned, you’re using data from the manufacturer’s sheets or best practices in the form of checklists.
[00:12:32.720] – Scott Holstein
I was actually reading an article recently on the difference between building automation system alarm, predictive maintenance notifications and the way they described it, like they said, it’s really the predictive maintenance notifications are. You don’t end up getting those alarms because alarms and building automation are set up for when something in the system goes so far out of parameters that it’s not an emergency necessarily. But it is something that you’re raising a red flag on and saying this needs to be addressed. rather than waiting until the equipment degrades, the point of getting that alarm you’re actually identifying ahead of time, which is just staying one step ahead.
[00:13:20.680]
Now that we have an idea of how predictive maintenance works, what are some of the benefits of predictive maintenance?
[00:13:26.570] – Rick Bennet
Yeah, so some of the benefits predictive maintenance. And there’s a lot so, you know, definitely temp jump in and help me out here. But, you know, from my perspective, avoiding downtime and improve productivity is one also extending life of assets and the ferment of new purchases. There’s also reduced cost and complexity of repairs. mitigate additional and related damage of certain types of equipment, you know, say, like pumps that may have vibration that turns into bearing failure. And another one that would really be relative today is to meet regulatory standards and compliance.
[00:14:16.080]
Tim mentioned indoor air quality, which is very relative today as well. It also helps manage spare parts, materials and inventory. And ultimately, I think we all probably agree. That would also boost the bottom line help some energy savings. And again, some money on repairs and things of that nature as well.
[00:14:41.780] – Tim Darrah
Yeah, just to go off that right now, it’s somewhat hard to find solid statistics on, the cost-benefit analysis or the savings of a predictive maintenance program versus a preventive maintenance program because it’s so new. But there’s a lot of great information from the Department of Energy, the US Navy, and they were actually really the pioneers of preventive and predictive maintenance. If you start looking through the literature and going back the EPA, they’re another great reputable source, and all these organizations, they’re unbiased.
[00:15:28.840]
They don’t have any sway, one way or another, and they echo some of the things that Rick just said. You can save 25 to 30 PRN in your maintenance costs, provide a four times to a ten return on investment just by having the accurate information that you need that you get with a predictive maintenance program and, this enables you to more effectively allocate your staff and other resources. As Rick mentioned, parts ordering. It’s no good when you have a critical failure in one of your buildings and you don’t have the part on hand.
[00:16:09.270]
But this warehouse that’s 79 miles away has it and now you’re paying expedited shipping and you’re paying perhaps off-hours labor costs, which is definitely more than what you’d be paying if you were able to put a guy on that job at ten a Mr. If you had the information ahead of time, traditional practices and several organizations really rely on a work order process whereby the occupants themselves have to enter the work orders, enter the work order information before they can get a maintenance personnel tasked to that job.
[00:16:51.800]
And so another issue that predictive maintenance helps alleviate I’m not going to say solves, but these occupants, they’re not going to be able to tell you very specifically, the information that your maintenance guide needs to know what part to grab before he leaves the shop. So he may have to go out there and then based on this work order himself, determine what is wrong and then go back to the shop, potentially to grab the parts and then go back again. So then you have this waste in time as well, but also in the trips.
[00:17:32.420]
If we’re talking, the maintenance shop is in a different building or perhaps geographically located several miles away or if it’s not in the same, even if it’s the same building, that’s just a lot of waste of time going back and forth for your maintenance personnel to diagnose an issue-based off of a vaguely worded work order by a lay-person occupant. And that was just another thing that occurred to me that were some of the pain points that we heard during a lot of our interviews that this type of program helps alleviate.
[00:18:08.880] – Scott Holstein
Great point. And I just feel like people’s time. When you’re looking at these facility managers, their time is valuable and it’s not only the time that they’re spending doing whatever that task is, it’s time that they’re not spending doing other things. So if they’re out there running around trying to find the right part, then it’s time that they’re not spending evaluating a new vendor for something or painting the lobby or whatever it might be. So I feel like that the man-hours oftentimes get kind of shuffled into it because it’s not a hard cost per se.
[00:18:59.410]
circling back, I wanted to ask you guys, I’ve read that article that mentioned earlier on predicted maintenance, and in that article, they refer to fault detection. And I wonder if you could just clarify for our listeners, fault detection, predictive maintenance. Are we talking about the same thing?
[00:19:16.380] – Rick Bennet
Yeah, I know. For me it’s not. Fault detection is something has already occurred. Predictive maintenance is, you know, notifying us that there is a potential for a fault, therefore, given us the ability to mobilize and catch that before it is a fault.
[00:19:38.140] – Scott Holstein
So I know facility managers out there have concerns about predictive maintenance and these kinds of programs being a replacement for their services. Where if you know, you know, when these things are going to break and how to fix them or what have you, you don’t need me. So what would you tell a facility, Mr with that concern?
[00:20:01.250] – Rick Bennet
This is a really good question. And facility managers are so important to their organizations and what they do. We touched on earlier. They’re very busy. And a lot of times, you know, facilities are under stat on maintenance staff as well. So the way I like to look at it is it’s really just another tool in the tool belt, you know, help them along. And we highlighted some of the benefits earlier. where it comes to reduce cost of repairs, mitigation, you know, the standards and regulatory compliance, the parts ordering, so on and so forth.
[00:20:42.900]
So, you know, I really think it frees them up to do their job again, it’s just another tool for them to be more effective at their job. I almost feel like too, as well is I don’t know, specifically, is across the board, but I believe it is there’s a shortage of facility staff. I really believe that this is another tool that can help them many their facility in a more efficient way. And the challenges with holidays, weekends and overall, just running a building again, it’s just another tool for them to use to be more efficient at their job.
[00:21:28.600] – Tim Darrah
Yeah, I’d agree with that. You know, at the end of the day, someone’s got to steer the ship and there’s not going to be any piece of technology or automation tool that is going to replace that job. And secondly, all ships need a crew, automated tools are not going to replace the job of a person who has to physically go down to where this equipment is located and repair it or service it in any manner. As Rick mentioned, this augments their job so that they can be more effective and thinking of one organization that I talked to, they had one guy whose job was to triage work orders.
[00:22:17.040]
That was his full-time job was to go through and triage work orders, and he’s not actually performing any maintenance task there. And so when you have tools like this in place, you’re going to start to see less work orders because the system is going to tell you, hey, you probably need to change the filter on their handling unit number two within the next couple of weeks, for example, or you’ll get alerts that could tell you the blower motor on this air handling unit is becoming unstable and might go bad in the next couple of weeks.
[00:23:00.650]
For example, you’ll have these preemptive notifications, and so you’re going to be servicing the equipment before any of the occupants ever noticed the problem ever happened. And you’re going to start having reduced work orders come through. And so that guy that whose sole job is to do work orders, he’s now freed up to do something more productive for the organization.
[00:23:22.300] – Scott Holstein
I think the majority of people are asked to do more with less LAN their jobs today. And facility managers are certainly different. As Rick mentioned, there’s already a shortage of facility managers out there, this is for you guys listening. This is another tool in your belt. It’s not a way to replace you. is a way to make your job easier. So when we’re talking about this solution, we talked a little bit about before how it’s gathering all of this data processing and then ultimately giving you some kind of meaningful recommendations.
[00:24:02.030]
So what does it look like to implement a solution like this?
[00:24:06.220] – Tim Darrah
So going back to kind of the beginning or how we differentiate predictive versus preventive the key differentiator there is data. So not only do you need the historical data, you need real-time data and you need to know be able to have continued access to that data stream. So, for one, let’s go back to the air filter change for example, If, you’re not measuring the pressure differential and airflow across your filters. You’re not going to be able to implement a predictive maintenance solution. For, your air filter changes.
[00:24:45.880]
And that’s just the nature of the Beast. Predictive maintenance relies on this data. each specific maintenance task that you want to perform or that you want to add to a predictive maintenance program. You need to make sure that you’re able to have access to the data points for that maintenance task. And Rick might be able to speak in a little more detail in the realm of HVAC and facility issues here, again, because, my expertise is more along the lines of the data. And so that leads into the domain knowledge then needs to be encoded.
[00:25:29.280]
Stat relates these data points, which for someone like, they’re just numbers. But the domain experts going to translate these numbers into a human-readable language that can inform the end-user the stat of the system. Because essentially what you want to know is you just want to track how the system degrades over time, because all systems, they undergo degradation from various sources. There’s no perfectly operating system out there that stat never degrades. And also, I’m using the term degradation and not the term fault, because typically degradation over time leads to a fault, if you contract the degradation, then you could prevent the fault.
[00:26:19.110]
And, this is something that there’s a lot of companies out there that say they have. We’re tracking all the data points, we get alerts and we get alarms. But really, when you peel behind the layers and look under the hood, as Rick mentioned earlier, with fault detection, fault detection is reactive. And so a lot of these companies that are out there that are saying that they have these alerts and these alarms and they give us the probable cause for failure.
[00:26:55.360]
That’s what they are, meaning your system is already reached the point where it’s failing or has failed. That’s sort of the crux of implementing a predictive maintenance solution, making sure that you have the data and making sure that you have a domain expert that can tell you if we want to prevent this fault. These are the data points we need to track. And it’s not just by tracking singular data points. , in some cases, like I mentioned with the air filters, it’s not just tracking the pressure differential, it’s not just tracking the airflow.
[00:27:39.020]
You want to track the joint interaction, of the pressure differential and airflow. And so that’s the second layer to the predictive maintenance aspect is. You know, you have the right data. Now you need to make sure that you combine the right data points in the right way to give you the proper information that you need. And so this is where simple trending you begin to not be as effective because, with simple trending, you’re just trending singular data points. But then you start to get into more advanced learning, more advanced machine learning techniques that can actually look at the data and understand relationships among the data points.
[00:28:27.600]
Stat as humans, we’re not able to look at data and understand that because of the non-linearities inherent. And basically, that means that not everything is going to be linear with using the advanced machine learning techniques, and you could start to get more creative and get deeper in to understand where and when these faults might occur, so that way you can prevent them.
[00:28:58.080] – Scott Holstein
Alright, Rick, anything to add to that?
[00:29:02.760] – Rick Bennet
I mean, really. I mean, Tim says it very well. It’s run through my mind, typically, we look at chill water pumps, and it’s a great example of they’re just they’re working, right. With this technology, we can actually monitor those pumps and detect a change in vibration, which is telling us, you know, there’s potential for bearing failure. And that can be huge in certain facilities. Right? Versus, you know, a fault detection saying, hey, you know, the pump is failed, even without the fault detection. At some point in time, you’re going to know that pump failed.
[00:29:46.960]
So the predictive model is a more inclusive model. You know, key from a critical piece of equipment, such as a chill water pump from, you know, failing at an inopportune time.
[00:30:01.840] – Scott Holstein
Tim, I’m curious, with your machine learning and data analysis background, just kind of touching on some things that you had said before. You know, one of the reasons why we need machines to find this information, because we, as humans, can only look at so much information at once. And identifying abnormalities over a period of time makes it even more difficult. I mean, even as humans if we had all of these, we have access to all the same information in our building automation systems. But we, as humans, don’t recognize stat, slow degradation, or that blip, the pattern that may actually indicate something meaningful.
[00:30:46.140]
Would that be an accurate way to describe it?
[00:30:48.440] – Tim Darrah
Yeah. That’s a very accurate depiction of kind of what I was trying to convey there about our limits of our ability to see these trends inherent within the machines and the data that they produce. That say “Hey, I’m starting to get a little sicker.” And like you said, we just can’t detect that, but algorithms can.
[00:31:17.200] – Scott Holstein
When we’re looking at our facility managers that are listening, this podcast like, okay, this sounds like something maybe I need to learn more about. And now they want to go and evaluate these types of solutions. Obviously, to facility coach going to i-performance, talk to National Facility Solutions. That’s a great place to start. But can you guys give us some kind of objective means of evaluating a solution and a service partner in this predictive maintenance arena?
[00:31:49.280] – Rick Bennet
I really like this question. And, you know, one thing that I always try to share or, make sure whoever I’m working for understand is, you know, we should understand your mission. Each building or each business has a mission. Hospital is going to have a different mission than, say, a factory. And so having someone who really understands what the mission of facility is very important, in my view. And just taking that to the next level. Okay. When all these points being monitored are important. But, again, you know, some could be more critical than others.
[00:32:40.700]
Right. So just having someone, again, a service partner that is in line with your mission, it is a very important thing to have. And again, to totally understand the difference between fault detection and predictive maintenance. Again, because, we don’t want to be, youknow, reactive. We’re trying to find the problems before they are a problem, right? Understanding of those two things is critical, in my view.
[00:33:14.800] – Tim Darrah
And that’s something that I didn’t even think about. But knowing the mission of the building, definitely, as Rick said, is the first thing you want to know, because, you know, a factory might not be as concerned as indoor air quality as a hospital. And so the solutions that an organization might want to implement. In that case, are going to be different. And then now, if we’re talking about a predictive maintenance program, or I would rather call it a data-informed maintenance, just because these terms preventive and predictive are now thrown around so much that it’s easy for people to lose sight of the specific differences between what they mean in a data-informed maintenance program.
[00:34:10.920]
You also want to know what kind of time horizon you have. What kind of time horizon, does this offer you over other approaches? And what I mean by that is if I’m doing predictive maintenance on my air filters, how soon am I going to get noticed that I should change them, how soon am I going to get a little “blurp” that says, hey, you should change your filters pretty soon. What does that pretty soon mean? Does that mean 24 hours? Does that mean a week or two weeks?
[00:34:41.810]
In all of these cases, it’s even 24 hours notice is still better than, hey, we got a bunch of people complaints of a headache and some stuffy noses. Maybe we need to go change the air filter. And now you got to find a guy who’s not already doing something else to go replace it. Any time horizon is going to be better than a reactive approach. But knowing that time horizon is very important in terms of implementing the predictive maintenance program beyond just the predictive maintenance aspect.
[00:35:17.700]
So if you know that if you’re looking at air filters and you’ve got a one-week horizon, but on your air handling units, when you’re looking at the blower motors that you have a two-week horizon or a three-day horizon, that your lead times on these different problems are going to be different. And knowing them is going to enable you to do all those other things that we talked about earlier, such as resource allocation, staff scheduling, and preemptive parts ordering. Once you know that the solution that you’re being offered fits within the mission of your building, and then you know that the time horizon on the different aspects of the solution that’s being provided fits within the program, the type of program that you want to implement, then you can start to say that, you know, it looks like this solution is going to fit our needs and then start to move forward with it.
[00:36:24.040] – Scott Holstein
So regarding the time horizon. find that one really interesting because we talk about this stuff in generalities all the time, but that’s extremely important. , now that you bring it up, I don’t want to know two days before I want to do two weeks before, especially if you’re running a critical facility where I don’t want to wait until I’m getting close to where I need to change my filters, I want to change them before. Is that something within your programming that you can adjust based on the customer’s needs?
[00:36:55.950] – Tim Darrah
That’s something that really depends on the Fidelity and level of complexity of the algorithms that are behind the scenes performing this analysis of the data. And in some cases, like a Rick mentioned vibration analysis earlier. I’ve implemented some vibration fault monitoring for bearings, for example, and some of my work with NASA, and that was an approach that used neural networks. So there’s some artificial intelligence in there, and I was able to detect failure a week before actual failure. But in another case, it might be such that it might not be possible for the algorithm to detect failure for a certain, I’ll say a specific type of failure for a specific type of component.
[00:37:56.360]
So part of that is within the limitations of the programming or within the limitations of the skill set of the engineers and the data scientists that you have behind the scenes generating these algorithms.
[00:38:14.110] – Scott Holstein
Yeah, it absolutely does. More or less my takeaway is it depends on the piece of equipment, depends on how tight the algorithm is around those recognizing those upcoming faults potentially and the short answer is it depends. But there are cases where it is probably something you can do, there are cases where it may not be as simple as just saying, oh, I want to know two weeks before instead of one week before. Is that fair?
[00:38:49.420] – Tim Darrah
Yeah, that’s very fair. It might just be the case that, hey, man, if you want to know two weeks before, I need to put another 500 man-hours of my data scientists on this problem so they can improve the models or it might just be that, hey, maybe we can get your two weeks notice next year when technology is a little better.
[00:39:11.420] – Scott Holstein
Lan, that’s all the questions I have for you. I’ll open up the floor is there anything that we didn’t talk about here today that you think your facility manager out there might want to know about predicted maintenance or anything around that topic.
[00:39:24.740] – Tim Darrah
I do have one thing to add on that topic of evaluating a solution. You also want to know the false positive rate because and I say false positive rate specifically over the other types of errors. You have false positive, false negative. You type one and type two errors because if my program tells me that I need to replace the compressor and one of my rooftop units, do I really need to do it? How much can I trust what this program is telling me? Because at the end of the day, it’s an algorithm that someone wrote, you know, based on data.
[00:40:04.100]
And if you got a company coming to you and they say that they have this very new tech, this is great stuff, but we have a false positive rate. You probably might want a second guess going forward with that, because, the money that you would be saving and the man-hours you’d be saving through this allocation and resource scheduling is going to go to waste if half of the alerts that you get turn out to be false alarms. And that also goes into trust.
[00:40:42.010]
If you don’t trust the system or if your maintenance staff doesn’t trust the system, then it’s simply not going to be used, so any of the time and money savings that program might offer just won’t be captured because it won’t be used. And this is something that I’ve experienced a lot in, specifically on this Navy project that I’m working on doing something for some of their patrol boats. The program managers over there and some of the other points of contact, they tell us all the time and they hammer it in us all the time. Oh, hey, if these guys don’t trust it, they’re just going to turn it off. And if they turn it off, they’re not going to use it. And so those are some other things to keep in mind when looking at this type of solution.
[00:41:34.600] – Scott Holstein
That’s a great point. I’m glad you brought that up Tim I think that’s true of any system. If you don’t trust it, you’re not going to use it. Your example of going out and change it the compressor, it’s going to work. The new compressor is going to work, and you’re probably going to get a little bit more efficient. But did you take that other one to the end of its useful life? the question so asking about those false positives is a great recommendation. Anything else? Guys, like I said, I want to open the floor to you, if there’s anything else you think that might be helpful here, let us know.
[00:42:08.530] – Rick Bennet
Yeah, I think. I don’t know if we really touched on it, but just for the audience out there to, think about, you know, as this type of predictive maintenance program would be put in place. I mean, the starting point is a baseline, right? So there would be baseline readings that would be obviously put in place. And then from that point forward is when for predictive maintenance would be launched and be looking at the data for all of these data points, for the predictive maintenance or repairs, so I’m not quite sure if we covered the baseline portion of it, but that would be part of it as well.
[00:42:56.480] – Scott Holstein
I’m glad you brought that up, Rick. We talked about the implementation of this product and we didn’t really touch on that. We talked about the gathering of data. What does that look like in real life? My experience has been a… Your customer would go to their building automation company and say, hey, we need you to expose these points via BACnet IP to our data analysis, to our predictive maintenance company that we want to start working with here. By the way, we also – Tim mentioned, you don’t have a sensor, you don’t have a sensor in the right place.
[00:43:40.450]
You don’t have the sensor getting the correct information that you need to make these decisions. Then, it’s not something that you can do. So you do you mind going back to that question real quick? Just talk about a little bit in real life. Is that what you’re asking your customer do is, hey, go talk to your building automation provider. All we need is BACnet IP. Can you go into that a little bit?
[00:44:04.100] – Rick Bennet
Yeah, I know for me I’m just going to kind of go back to again the mission of the building. That’s part of it, right? I mean, we need to understand what the mission of the facility is. And then, you know, if there are certain things that we want to have in the predictive maintenance model, is like Tim said earlier, is the hardware there? So that’s part of it and understanding the controls, what you have and is there anything that you may need for that to be implemented? And I will defer to Tim on the more the implementation of the predictive maintenance model on his end, for the rest of it.
[00:44:55.920] – Tim Darrah
It certainly is a matter of asking the customer, can we get these data points off of your BAS system? Because National Facility Solutions has the hardware to tie into the building automation system regardless of the vendor. So secondly, once we know that we can tie into the system, then it’s a matter of all right, what points are we able to collect? And that goes back to what I talked about earlier. Are we collecting the right data? And then if the customer asks for some type of predictive algorithm for a component or for some operation that, you know, when we work with domain expert and realize, well, we’re actually not capturing that data point, then it’s a matter of, well can we? If we don’t have the hardware sensors in place to capture that data point, what does it look like to get that sensor place and start capturing that data point?
[00:46:01.460]
So, it is a matter of figuring out what the customer has in terms of data and then figuring out if they don’t have something that is needed for specific tasks. Can we get it?
[00:46:15.920] – Scott Holstein
Alright, gentlemen, I greatly appreciate you guys taking the time to walk us through this. I think that our audience probably get a lot out of this topic, and I have a much better understanding what predictive maintenance is. What the benefits of it. How do you evaluate a vendor? What does it look like when you actually go to implement the solution?
[00:46:45.560]
Before we sign off here any last words?
[00:46:48.750] – Rick Bennet
I just appreciate the opportunity to come on and share with everyone.
[00:46:53.440] – Tim Darrah
Thanks for the opportunity. And I hope that there’s listeners out there who have a better understanding about predictive maintenance or if they were already well versed, then I hope that they were hearing the choir preach.
[00:47:06.660] – Scott Holstein
All right. Thanks again, to Tim Dara of Intelligent Systems and Rick Bennett of National Facilities Solutions, and thank you all for listening. Today. We look forward to seeing you at our next podcast. This is Scott Holstein signing off.
