The Power of AI and Machine Learning in Manufacturing with Prateek Joshi

Connect with Prateek Joshi: prateek@plutoshift.com

Website: www.PrateekJ.com

LinkedIn: https://www.linkedin.com/in/prateek-joshi-91047b19/

Lisa Ryan: Hey, it's Lisa Ryan. Welcome to the Manufacturers' Network Podcast. Our guest today is Prateek Joshi. Prateek is the founder of Pluto shift and a published author of 13 books. He's been featured in publications such as Forbes, CNBC, TechCrunch, and Bloomberg. You can also visit his website at Prateekj.com to learn more about him. Prateek, welcome to the show.

Prateek Joshi: Thanks Lisa for having me. It's great to be here.

Lisa Ryan: Alright, please tell us about your background and what got you both interested and involved in Ai.

Prateek Joshi: I grew up in a small town in the southern part of India. When I was growing up, water was a beautiful luxury. In college, it stuck with me as I began my professional career. I studied machine learning Ai across law, a natural inclination. But there's a big gap in Ai in the physical world. It's not nearly as ubiquitous as it could be. We all know how to use it every day, as in search engines, but it was about machine learning.

But there's a big gap when it comes to the physical world. That was the core motivation behind doing this, to bring Ai to the physical world. It started with water, meaning any physical infrastructure that touches the water and gathers data. So, we use Ai to solve a fundamental problem like water. How can we use it efficiently? How can we distribute it? How can we make sure it's not wasted? So, that's how it all started.

Lisa Ryan: What are some of the applications of water that Ai would be used for? It's something that I would have never even thought of.

Prateek Joshi: I ended up with a simple example. As the average consumers here in America, we get our water to a network of pipes. An essential part within that setup is called the cleaning process, meaning you'll get raw water from somewhere, and there are these very large treatment facilities that convert raw water into the water that we can consume.

This is a very energy-intensive process, meaning you need to use a lot of electricity and chemicals to make sure the water is clean. If you're not careful, the water can leak and waste electricity, which is again a huge problem. A simple application collects data for pressure-temperature flow rates and then uses a tool like Ai to make sure you're not wasting water. You're not letting it leak, and you're not wasting electricity to clean water. This is one application of how you can use Ai efficiently. There are 100 other use cases as well.

Lisa Ryan: When it comes to manufacturing, of course, water plays a significant role with many factories and plants. But as far as taking this into the manufacturing sector, how do you make Ai practical for manufacturers?

Prateek Joshi: Let's look at a manufacturer. A simple example would be a company that produces food or beverage, and part of the process is to get raw material. Let's say you're making ketchup. You've got to get your raw material in, and then you need at least sources such as electricity, chemicals, and water. That goes into it, so raw material sources go into the facility when the ketchup comes.

A manufacturing operation can be broken down into several steps if you look at a manufacturing operation. Each step has a certain efficiency level, meaning how much are you consuming to produce a unit of output now if we don't use any new technology? A human would have to do it. Imagine a facility where you have 300 membranes, like a filter. What stops this from going through, and you got to make sure that it's all functioning. You can't keep an eye on such extensive infrastructure by planning around as a human. The goal is how do we make sure that if let's say, Monday number 49 is acting up. How will we know if the pressure is going way up or it's going way down? You need to know.

In the corporate world, you'll be sitting inside your office, and you want to know what's happening. That's where technology like remote monitoring can be beneficial. For example, a tool can automatically detect memory numbers 40 minutes. Of course, it would help if you did something about it, so what this does is it makes sure that they get produced regularly, and you don't need to run around to make sure the system is working perfectly. That's just one example.

The production supply chain is another vast area of many use cases. Operations-specific use cases can be deployed inefficiency here.

Lisa Ryan: Let's go back to artificial intelligence. Since you've written 13 books on it, you're the expert. So what exactly is artificial intelligence? What does it do, and how does it work?

Prateek Joshi: Artificial intelligence is a State. It's the goal, meaning you can build a system. To build an Ai system that is intelligent enough to take actions independently. Ai is a state of being. Now machine learning is a vehicle to achieve that goal. Data is the field for the vehicle. That's how we relate these terms like Ai, machine learning, and data.

That's how they put it together. Until then, we use the umbrella term artificial intelligence to describe any system hardware, software, or combo that can do things on its own. We use it every day. The simplest form of intelligence is like a calculator. That is simple, but we don't think of it as Ai, but technically it is doing the little thing. I have two massive numbers, and it shows you the result.

More complex Ai systems can drive cars. They can detect danger when flying in the air, which ranges how much intelligence they have. We can put in a system, but that's how we look at Ai. Today, as you've seen, there is a very successful implementation of Ai in the form of automation, meaning if you're in a factory, sometimes it's hazardous for a human to approach a hot furnace. So, a machine does the specific task of taking this and putting it there. It sounds like a simple example of how Ai manifests itself in the video.

Lisa Ryan: So, Ai then would be a part of machine learning? Because if I hear you correctly, machine learning is the machine is doing something, and, over time, it learns from itself and gets better.

Prateek Joshi: Yes, machine learning is as more data comes in, the machine learns how to behave in various scenarios. It learns more and more it approaches. As of today, we are not fully there yet. We don't have an Ai system that's indistinguishable from humans. We're not there yet, but plans are becoming more intelligent, and machine learning is all the algorithms. The umbrella term is machine learning. All the algorithms, tools, and frameworks you use to make a system intelligent.

Lisa Ryan: What are some of how you would deploy some of these Ai technologies in the world of physical infrastructure?

Prateek Joshi: An excellent way to look at it is we work backward from the use case. Let's say the goal is to reduce energy consumption. Let's say you're a food processing company and want to reduce the energy. You can do per unit of output produced. That's a problem, and once you've identified that goal, you work backward to figure out what tool or system I should use to attack this problem. Because Ai is pretty vast, no single model can solve everything. Working backward on the issue, and when you do that.

Once you define that, you also assess what data we have available. We want to reduce energy consumption, but do we even know how much we can do today? Do we know the initial primary lowers that make the energy consumption go up and down? I'm collecting temperature data, so these are basic questions. Then, we have some pressure-temperature data. We know what you want to do; then, you build a tool. It could be software. It could be hardware. It may be both. The whole time it is achieved, it starts driving that. Today, you can do X. Maybe three months from now. You will consume point eight of that. Meaning 80% of that and then eventually 50. The goal is to drive towards that goal in fashion, and that's how they appear. We're continually improving with more data.

Lisa Ryan: If somebody's thinking about incorporating Ai or machine learning into their plant, what would be some ways to drive the behavior change needed for that implementation? Because people are going to be afraid they're going to lose their jobs, or they may not like the robots, or they may be afraid of them, or they have all these misconceptions as far as what I can do. They start with that conversation and change the employees' behavior to get the buy-in you need.

Prateek Joshi: We commissioned a study to understand that. In March of 2020 and then a few months later, we wanted to understand people in manufacturing. It could be operations managers, operators directors, and people running facilities. So we surveyed those professionals to know how you use it. Because you can go into the facility every single day, and yet you need to know everything that's happening at all times.

So what are you doing so? We conducted a survey and found very interesting results. It's not that people or companies do digital transformation. If you're doing pen and paper, how do you digitize that work if you don't lose it in a fire? The goal is, how do you digitize the operations? And 94 of the participants said that the company's primary way of doing it is boiling the ocean. Meaning somebody comes up with a big initiative, their entire company of 40,000 people. They try to do everything all at once, and in many instances, it's not feasible to do because it's a massive company. So they are introducing a drastic change in our stakes. That's why. But boiling the ocean shouldn't be the plan, yet people do this. Within that, 78% of the participants said that they were supported by the Department heads when they took an op-specific approach. I mean that you choose a piece of work that could be monitoring a membrane or detecting pump failure. A small piece of work, and then you transform that; you digitize that in an already focused manner and a bite-sized way. What that does is it creates a success template meaning oh this facility in Los Angeles, or this was somebody in in in Miami did it. That template can now be implemented in Chicago, Boston, and Seattle.

A groundswell builds up, and that's how you transform a particular piece of work in and digitize that. So what we call out specific digital transformation. I would say start with the bite-sized approach and a clear success template that can be used to transform work. That's what we have seen as a third party did quite a bit here.

Lisa Ryan: Is there a best practice for deciding where to start? Are you looking for the most mundane tasks that human effort is wasted on, or are you looking at the most dangerous tasks? Are you looking for those detailed tasks that maybe humans miss the mistakes? Is there some best practice you recommend?

Prateek Joshi: That's a good question. What happens is the initial choice determines the entire approach, meaning people are either completely turned off by it, or they become very enthusiastic about it. So, the initial choice matters a lot. What we have seen is pick a workflow that is low-hanging fruit with high impact, meaning something could be easy. But if the company doesn't care, the initiative will be killed. So, the goal is to figure out what is low-hanging fruit. I can have the highest impact on the business, which is a great first try, so let's automate it. Because, when it comes to bringing a new piece of technology, it has to make sense for both the user, meaning, it has to be easy to use. Because the company has to invest capital, Roi needs to see the return. The sooner you can make that happen, the better it is.

Because of that, I think this combo that using a simple example would be monitoring as a simple example or other examples could be if you are scanning the seats by hand, digitize that. Nobody wants to sit and do that. People have other work to do, so the goal is to identify those tasks that nobody will mess with. So it's going to have a significant impact.

Lisa Ryan: So, with the clients you've worked with within the study you've done in this field, do you have some examples of success stories that you can share of where the company was before and what happened after they started using Ai.

Prateek Joshi: Yeah, we have a perfect example for company, where before we started working with them they know they produce the beverage company that makes a product and what was happening was water as you can imagine, is a crucial part of the business; meaning how this how they treated it and how they use it. How they discard all of that matters because every incremental benefit, every incremental ISM efficiency, will translate to the old dollars because it's a pretty big company. Because of that, within the operation, water treatment with a huge part of it. We started working with them on that and what happened was over a year, we ended up saving more than 15 million gallons of water. That was one side, and you can imagine 15 million gallons can feed a family for years. It's a lot of water on an industrial scale, it's large, so the goal is to identify these high-impact areas were before doing this. After using a product, they started out doing that, saving this water. What this also does is, in addition to the Auto I benefit, that is, a positive impact on the ESC metrics carbon footprint. So that is another very key benefit that you can achieve by solving problems and energy, water, and chemicals that can significantly impact your PSG footprint.

Lisa Ryan: Are there particular types of industries that you've mentioned beverage a couple of times, and you said your connection with water. What are some of the other big industries that you've seen that have successfully taken off taken off with us?

Prateek Joshi: The trend we have seen as any industry, where the energy consumption is high, is that those industries are active. They actively pursue new technologies because it's costly. So the source has a significant impact on carbon footprint. So are the water, beverage, food processing, chemical, manufacturing, and data centers.

We've seen a lot of activity in these industries. Food and beverage have been very active, so say that's one of the most active verticals. All of these industries have been actively pursuing new technologies here.

Lisa Ryan: Do you have any other ideas or tips if somebody is starting to think about this? What are some resources they could turn to? What would be the best way for somebody to begin exploring the topic?

Prateek Joshi: I think many of the resources are available online and on our website. We have several White Papers and case studies that were just published, mostly talking about how to think about it. In addition to that, many of the publications, depending on the industry or smart water magazine publishers, water-related technologies, food, and beverage, have something similar. The chemical industry has something similar, but the goal is to try things out and create a framework to quickly try things without disrupting the business and see exactly what works for you. It would be best to look at simple things like automating. But an easy starting point because this is only something we can automate, something as simple as scheduling. We automate that. It makes our life easier. With the manufacturing, you can look at it. You automate the work of monitoring. You can automate the work of ticketing.

What happens after taking what happens when you do save energy? How does that work? Determining what can be automated is step one—after that, working with a partner who can do a small, well-defined pilot. I'm going to go from there, so I think that's an excellent way to get started here.

Lisa Ryan: And is there a way to test that it's working, or is this something that once you commit Ai, you're all in or not. Is there a way to try it, see if it works, and then decide yes, I'm going to stay with this or no, I'm going back to the old way?

Prateek Joshi: 100%. That matters a lot, and that's precisely how we should think about it, so as a new technology, the goal is within them. Let's see what our director of operations says. The goal is, can you create a framework where, if you want to test out. Let's say you saw a new Ai cloud tool that can reduce chemical consumption, so the goal is, how can we quickly pass it out. It usually takes about 90 days. No matter what piece of tech it is, within 90 days, you should be able to see the impact.

I know. I think it should take like two years, that's too long. But I've seen 90 days has a good time at the break where you deploy it. You use it honestly for 90 days. Then you measure before it was a surprise, meaning you have to be diligent about the status quo before you start doing it to measure the impact, and then you can do it. So it's not a one-and-done thing. It's like Ai is almost becoming a ubiquitous cloud or Internet. It's already here. The goal is to make it work for you, and it's all it's ubiquitous. Almost all of us use it every day in some shape or form. It's just going to figure out how to make it work for you.

Lisa Ryan: And what about tying it in with your existing system systems? You think that some manufacturers have equipment that is decades old. So they're trying to incorporate some new equipment, or is it better if you're starting fresh I mean, how difficult is it to tie in all the different systems that you're going to find in a manufacturing plant.

Prateek Joshi: That's a legitimate concern, but potential customers are when we talk to noncustomers. We do see that quite a bit. These companies have been around for decades; some companies have been around for more than 100 years, so they have a long history of decisions and associated hardware. The goal is to ensure that they stay up to date on simple things like data collection, so if you're not collecting data, step one is to get the sensors in place.

Start collecting the data, but yeah, I think if you're not on the journey, some people are on all parts of the spectrum. If you have nothing going on the step, one would be talking to a hardware company, which can deploy sensors and send the data to a database. Once you have that, you go to the following type of data analysis, if you don't do data analysis without any data in place, I think. I think assessing where you are on that spectrum and then choosing the right initiatives matters a lot, and I think many, many companies have successfully moved along the spectrum. They're moving fast, and many vendors and companies are showing up with amazing technologies.

Lisa Ryan: And then the last thing that comes to mind is just from a security standpoint from a cyber security standpoint, how are you making sure that all of this is being kept safe that nobody can. Access that or hack into your system. What are some of the things you would recommend as far as that goes, or is that not an issue?

Prateek Joshi: No, it's an issue. Any data company has to have access to your customers' data. Make sure that data and your system are secure. A...