114. Idea – Tackling Machine Learning and AI Head On Transcript


Brandon: 00:00 

It’s not a matter of if it’s a matter of when and if you’re not starting to invest now and starting to figure out how you can build that into your organization to drive value, then you’re behind the eight ball. The whole concept of machine learning is the speed to iterate, right? It’s helping your teams make better decisions and make those decisions faster. So the people that adopt quickly and drive value, they’re only going to iterate and drive improvements at an even faster pace than before.

Chris: 00:27 

Welcome to EECO Ask Why. A podcast that dives into industrial manufacturing topics, spotlights the heroes that keep America running. I’m your host, Chris Grainger, and on this podcast we do not cover the latest features of benefits based on products that come to market. Instead, we focused on advice and insight from the top minds of industry because people and ideas will be how America remains number one in manufacturing in the world. 

Welcome to EECO Ask Why. Today we have an idea episode and we’ll be talking about tackling machine learning and artificial intelligence head on. And I’m excited about this is going to be a fun topic to help you walk through we’ll have Mr. Brandon Mendoza, who is the director of sales at Odin technology.

So welcome Brandon. 

Brandon: 01:15 

Thanks, Chris, excited to be here. 

Chris: 01:18 

This is the fun one. This is a topic people think different things about it is definitely cutting edge. I’m excited. We haven’t really talked about this a lot on eco S why machine learning artificial intelligence. That could be a gray area for some of our listeners.

How would you explain this to someone who’s new to manufacturing or maybe new to these topics? 

Brandon: 01:37

Yeah. For anyone in manufacturing, hopefully you realize the abundance of data and manufacturing there are millions, if not trillions of sensors and data being generated every millisecond across manufacturing.

And I think that’s why, machine learning and AI are keen areas of interest for manufacturers right now, because with big data becomes a lot of effort to really extract insights from the data around where are there opportunities to improve production and go drive the action or prescribe the step it takes to achieve those results.

And so when you think about machine learning and AI, it’s really about creating a digital model to represent physical processes. So that you can drive optimization, start to predict things before they occur and prescribe the best action to deliver the best results. 

Chris: 02:35 

Wow. that was a great explanation, Brandon.

One of the best I’ve ever heard, because it really can go and everybody has a different perception of what machine learning and artificial intelligence is. For the people out there, there may be some myths floating around or, so far as machine learning and AI, you know, you get a chance to debunk them here, you know, what, what would they be?

What would you like to just say, Hey, you may, you may think of this, but this is not really reality. It’s more like this. 

Brandon: 03:03 

I think first off it’s there’s definitely some myths around time to value, right? I think. When think when people think about machine learning and AI, they often probably think, this is a significant endeavor that’s going to take me know six, 12, 24 months to produce a model that’s going to be predictive and prescriptive. And I need a massive amount of data. When in reality, depending on the use case and where you’re starting, we’ve been able to generate predictive and prescriptive models as fast as 30 to 60 days.

There’s some precursors to that around having the right data and the infrastructure to make that happen, but if the data is there you can actually drive value pretty fast in, in several different use cases and really start reaping the benefits quickly. In addition, I think, when people think about machine learning and AI, they think, wow, this is, that sounds complex.

I’m worried about my readiness. Do I have. The data infrastructure in place? Do I have the data I need? Do I have the skilled resources? Do I have data scientists on my team? And I think to think about this differently yes, there are some foundational elements, as far as, smart systems, sensors and instrumentation need to have some basic infrastructure as far as, centralized networks or, historians, OPC servers, things to centralize the data.

But when it comes to being able to then drive or extract action from the data, you don’t necessarily need to have data scientists on hand or somebody who’s, very adept in statistical analysis. The point of the solutions is to simplify that and to really focus on extracting the insight or the action from the data so that your teams are more focused on using that insight to drive operational improvements versus becoming a data scientist. And so I think it can be a lot simpler than people think and their readiness. They don’t need to have that infrastructure, consistent and established across the entire enterprise. You look at, Nestle, for example, they have 450 factories.

I’m sure not every factory has the same infrastructure, the same nomenclature, all of that. And so there’s ways to tackle it piecemeal and to scale start small and scale. And lastly, the complexity of the machine learning that’s being applied. Some of it’s very basic analytics that can drive significant value.

For example, what if the machine learning can just tell you where you’re running best and where you’re running worse, like that, sounds pretty simple, but when you think about the value behind that, there’s quite a bit there, right? If you can start to understand, “Hey, this machine is consistently running better or the shift or this product and start to extract out why that is, and then replicate and drive that best practice throughout the organization.

It’s a way to take something that’s pretty simple. I just tell me where I’m performing best and worse. That’s a use case for machine learning. 

Chris: 05:55 

Wow. That was a great explanation, Brandon. Lot of myths you covered there. The 30 to 60 days to time to value really jumped out to me, Brandon. That was awesome. And I love the advice on don’t overcomplicate it, it can really be a simple word, where are we running good versus where are we running bad and focus on that. If you look at a industry like automotive, could you give us a comparison from an automotive manufacturer as an example on how they’re approaching machine learning and artificial intelligence differently?

Brandon: 06:28 

Yeah, I think every industry has different use cases that they’re focused on. I think at the end of the day across manufacturing, the main areas that people are focused on are quality, performance, utilization, workforce productivity, and overall operational agility. And when it comes to automotive, for example, versus, food and beverage or pharma I think they’re very focused on cycle time, right? Most of the operations in automotive are more discrete in nature and therefore, utilization and performance are definitely, high areas of focus. They’re trying to reduce the amount of unplanned downtime and when machines are scheduled to run. You know, Running those as often, or as at a highest utilization as possible and at closest point to target throughput. So often they’re using machine learning to do things like predictive maintenance and figure out how to replace, tools or change over processes and time, those perfectly based on the data or, using machine learning to.

I understand again, where they have the best of worst performance across, throughput on different parts of their process. But also, even in automotive, quality’s clearly a key component. So how do they leverage maybe offline quality systems with their process data to create a more predictive quality models so they can, allow their operations teams to be more proactive in how they’re running their facilities.

Chris: 07:59 

Very cool. Thank you for that. I mean, So where are you seeing some of these adoptions early taking place that maybe you didn’t anticipate? 

Brandon: 08:10 

Yeah, you know, I think COVID, for me, at least has surprised me in some ways, as far as what industries have been positively or negatively impacted to give you a couple of examples the building materials industry, as well as like data infrastructure, so like wire and cable and things like that, have surprisingly gone up significantly. And when you think about it, it makes sense, right? Because people are spending more time at home and therefore, people potentially want to buy or build new homes as well as people are using the internet more than ever at their homes and therefore the need for bandwidth and things like that has gone up.

And so I think that dramatic shift in demand or increasing demand has caused those two industries to look to adopt machine learning and AI to increase capacity. And I think the other reason why you’re seeing that as COVID has definitely inserted some market uncertainty, right? Whether your demand is up or down whether that’s going to be a long-term or short-term impact is kind of uncertain.

And therefore, for example, on the building materials, if your demand’s up right now, then you’re not, you’re worried. That’s not going to be there for, six months, 12 months from now. Are you going to go out and buy a bunch of new equipment to, to increase your demand? Or are you going to look to adopt technologies that can allow you to take advantage of your current equipment that would drive higher throughput, better utilization, better quality and that’s a more sustainable action that we’re seeing manufacturers take right now because of that market uncertainty. 

Chris: 09:40 

Well, That was two great examples of you’re right. It’s through the roof with the pandemic, the building materials, man. You try to build anything lately on my gracious, it’s just, everybody’s doing projects.

 So what about entry points? If you were to give some advice to that manufacturer out there right now, are there any good entry points for that you should start looking at from a machine learning or AI standpoint? 

Brandon: 10:04 

Yeah. Yeah. I somewhat use a bad example earlier and it’s not necessarily a bad example, but I think when you think about machine learning, the top use case that I think most people think of is predictive maintenance. And I’ll be honest, that is not the area that we traditionally recommend starting with customers because when you think about machine learning, it all comes down to data and building out an accurate model on order to build out an accurate model, to predict certain behavior or prescribed or optimized, you need enough data to make the model accurate.

And when you think about predictive maintenance, the number of failures that you have specific to a specific problem might be far and few between. And so your data set might be limited, right? It might take you six months before you have a hundred of the same failure. And therefore it slows the ability of the machine learning technology to build out an accurate model that can predict that in advance.

We’ve actually seen much better success and much faster success in performance and quality. So those are our top two use cases for machine learning on how you can drive value to quickest. And I’ll just unpack those a little bit. Performance for example, it’s often a goal of manufacturers to say, how do I maximize throughput while maintaining or improving quality?

And we have an algorithm for example, called golden run. That looks at their processes and it looks at their production runs and automatically extracts out where they maximize throughput while maintaining or improving quality and actually prescribes back the process set points needed to achieve that.

Another example is, and I mentioned this earlier, but I’ll unpack it a bit further, quality. I think quality gets pretty complex because. You know depending on the process of the in industry, the amount of inline quality that they have might be limited. For example, most manufacturers don’t have inline instrumentation that actually determines whether it’s final, good or bad product, meaning if it’s, good or scrapped product, it’s often, an end of line or uh, an offline test that’s performed that determines whether it passes final quality.

And the ability to combine both offline quality with inline quality and production data is something that machine learning can help you sift through faster and be able to build a predictive model so you can predict quality issues before they occur or predict scrap for example. If you’re running a production run and your average scrap is 5%, but you range from, two to 10%.

If you don’t know until, after you’ve completed the order on how much scrapping is produced, how does that operator know whether they’re going to overshoot or undershoot their current production. So being able to take that offline data and build a predictive scrap model would allow that operator to understand, am I producing more or less scrap than I normally do? And therefore, do I need to produce for a longer or less period of time? 

Chris: 13:14 

Wow. Okay. That really helps. So cause you’re right. I think initially people go straight to that predictive model, but with that you’re all over it. That’s a limited data set and you may not see that return that you’re needing the performance and quality.

The sounds like that. What’d you call it the golden run? Was that what you called that? 

Brandon: 13:35

I did. And you will hear terms like golden batch golden run, but it’s really about finding that ideal recipe and using the machine learning to help you identify that. And sometimes it’s as simple as looking at your best of the best and your worst of worst when it comes to analyzing the production lines and allowing. The machine learning to just extract out where you perform best. 

Chris: 13:55 

Exactly. Exactly. And then from a quality standpoint, the way you tied it together with that, you call it the inline process, getting that. So I make sure I’m on the same page, the inline process? Data. Getting the data as the process is happening, versus just looking in the rear view when it’s done and making decisions and making more in the moment decisions for optimization is am I on the right path or help me if I’m off.

Brandon: 14:21 

Yeah. And I’ll just clarify a bit further. So again, some manufacturers have some inline quality, but I would say most don’t have inline quality that actually determines final, good or bad quality. Most of the time, it’s an offline test that happens after the fact that determines good or bad quality. And so what you can do with machine learning is leverage the offline quality combined with the inline quality and other process data to build a predictive model so that you’re not looking in the rear view mirror and finding out after the production run, I created 5% more scrap than I normally do.

Like when you have that predictive model, you can instream that in real time and alert or identifies to various personnel, whether things are operating below or above performance. You know, If there’s a certain quality KPI that you have to keep an upper and lower limits. Instead of looking in the rear view mirror, you can use that predictive quality to start to be proactive on controlling the process and making sure you’re not producing more scrap than you want to.

Chris: 15:25 

That tied it all together. Perfect example, what thank you for bringing me out of the weeds. Sorry about that. But I mean, that was a great explanation and it painted the complete picture for me now, for someone who wants to learn more. And they’re really interested. They want to sharpen their skills in these areas, where would you point them? 

Brandon: 15:48 

Yeah. I mean, There’s several companies that, have machine learning technologies whether it’s, the cloud providers and GCP, Microsoft, Amazon all the way down to the, the niche players like Oden that are very specialized in manufacturing and driving improvement there. All of these folks are continuously hope hosting webinars. So I would look to take advantage of those with our COVID environment. There’s not as many trade shows, right? So online webinars and virtual panels where people are discussing AI and machine learning and how it’s applied to manufacturing will be the first place that I’d recommend.

I’m sure like this podcast, there’s probably some podcasts out there that are solely focused on AI. I don’t know any off the top of my head, but I’m sure with a quick Google search we could find some top ones. And I think it’s interesting to look at ones that are maybe generic outside of manufacturing because there’s probably some parallels, but also look at ones that are maybe more focused on manufacturing. 

And then lastly, I actually just got done. I took a course at MIT called Smart Manufacturing Shifting from Dynamic Manufacturing. And it’s really around, smart manufacturing principles and leveraging advanced analytics and machine learning.

And MIT is a great institution, but I’m sure there’s several others out there that are offering similar courses. So that, that’s definitely a more in-depth commitment, but if you’re looking to learn a lot in an area to consider.

Chris: 17:14 

Right. No doubt. And we can even link in the show notes for our listeners if they want to check that out.

So thank you for those, for that insight on where to go, how to learn. Let’s talk about the who for a minute, who would typically lead the evaluation, integrating things like that from machine learning and AI in the manufacturing plant? 

Brandon: 17:38 

Yeah. It definitely depends. I think it’s similar to, when you think about who’s leading digital transformation and smart manufacturing as a whole, which we know is a cross-functional team, typically across engineering, IT operations, but even into, finance and HR, but as far as who’s the key lead for AI and machine learning in particular, it often comes down to the engineering team.

And I would say like continuous improvement or process engineering teams, because these are the folks that are already looking to optimize their process to the fullest level and take advantage of data and SPC and other things. And so these are people that are used to this to a certain extent and can more quickly understand and apply the technology.

Some manufacturers have data science teams, but not everyone has, that luxury. If you don’t have data science teams look to leverage solutions that are more turnkey where they’re doing the data science for you, smart manufacturing as a service, for example, but yeah, those are the, typically the teams that we’re seeing lead it, it often requires help from your it teams, right? Data is a big part of this. So whether it’s to data, infrastructure, or even I saw a customer recently that had a data governance team, which I thought was pretty interesting where they’re starting to define standards around, who owns data, who has access to data, what data do they need? What systems are connected things like that. Um, So that would be another thing though. Encourage folks to think about is you know you should you have data governance team within your organization. 

Chris: 19:09 

I guess also you have to, for our listeners out there to maybe in the moment in the facilities working got any headwinds that you’d like to point out that could exist when you start implementing a lot of these solutions? And just with your experience, Brandon, I’m sure you, you could offer a lot of guidance in that area. So any advice there? 

Brandon: 19:33 

Yeah, I think in general, a trend I’m noticing within IIOT platforms is a lot of them are great platform, but they’re often a toolbox meaning, “Hey, this is a platform that you can then build upon.”

And a lot of them are becoming, drag and drop zero code, low code environments, but the reality is they require a significant development. And in my experience, you end up paying two to five X to the staff costs, for example, two, to start generating ROI and delivering value. And sometimes it takes six to 12 months before you’re delivering that value and you end up with something that’s much more custom and difficult to scale.

I would look to leverage systems that are more turnkey, meaning, out of the box, the system is providing insights. It has built-in data taxonomies built in visualizations, built in analytics and machine learning so that your teams are more focused on using the insights from the tool versus, building out that custom experience.

That way your time to value can be greatly reduced where, in 60 to 90 days, you’ll have a clear idea of whether it makes sense to continue to invest. I think in general it also requires, cross-functional teams from an execution standpoint, as we talked to on the last one.

Uh, but then lastly, I would say focused on driving adoption and whenever you’re trying to implement a new technology, nobody likes change. And so it’s important to again, get buy-in on the why behind the project. And make it very clear folks, how it’s going to change their role and how it’s going to improve it.

If they see the technology as a burden to their job, they’re probably not going to welcome it, but if they see it as something that you know, is integral to making them more productive in their role, then you know, I’m guessing they would embrace it with open arms. 

Chris: 21:26 

We’ve got a win those advocates, right? Absolutely. This is, this has been a fun conversation. Brandon a lot of insights you brought for sure. And we call it EECO Asks Why we wrap up with the why every time. So why is embracing machine learning and artificial intelligence important for any industry out there to consider through their evolution?

Brandon: 21:48 

Yeah, I like analogies. I’m going to answer this one with an analogy. When you think about autonomous vehicles, I don’t think there’s anybody out there that’s saying is autonomous vehicles, are they going to happen? I think everyone’s convinced they are going to happen. And it’s not a matter of if, but a matter of when.

And therefore, if you look at automotive companies, they’ve been investing in autonomous technology for several years now and they’re continuing to do so even though, you can’t have an autonomous vehicle, operating on the road today, so same thing for manufacturing machine learning it’s not a matter of if it’s a matter of when and if you’re not starting to invest now and starting to figure out how you can build that into your organization to drive value, then you’re behind the eight ball. 

I think McKinsey just released a report the other day that the early adopters will be reaping ROI at 122% those that are lagging because again, the whole concept of machine learning is the speed to iterate, right? It’s helping your teams make better decisions and make those decisions faster. So the people that adopt quickly and drive value, they’re only going to iterate and drive improvements at even faster pace than before.

Chris: 22:59 

Right. Well, Brandon, this has been a wonderful episode. Thank you so much for taking the time with us on EECO Asks Why a lot of insight, a lot of good nuggets for our listeners that may be new to machine learning or artificial intelligence. So thank you for taking the time with us today. 

Brandon: 23:14 

Appreciate it. Thanks for having me on.