Leveraging Data Science and AI to Drive Innovation in Manufacturing

In this podcast episode, Dr. Beverly Wright, Vice President of Data Science & AI at Wavicle Data Solutions, and Prateek Shrivastava, Principal Data Scientist at Cummins, explore how manufacturing companies can leverage AI for quality control, supply chain optimization, and predictive maintenance. Prateek also discusses focusing on people and processes to build effective data systems. Tune in or keep reading to learn more about gen AI for new product development and energy analytics.  

Speaker details:   

• Dr. Beverly Wright, Vice President – Data Science & AI at Wavicle Data Solutions    

• Prateek Shrivastava – Principal Data Scientist at Cummins  

Watch the full podcast here or keep scrolling to read a transcript of the discussion between Beverly and Prateek: 

Beverly: Hello, I’m Dr Beverly Wright, and welcome TAG Data Talk. With us today, we have Prateek Shrivastava, Principal Data Scientist with Cummins, and he is so dedicated that he has arrived on crutches. Thank you for being here.  

Prateek: Thank you. Beverly, thanks for hosting me here. Yeah, the crutches are a different story. Let’s not get into that today.  

Beverly: I’m trying to remember if I’ve ever had a TAG Data Talk guest that arrived on crutches. So, I’m very pleased to have you on TAG Data Talk, talking about analytics challenges specific to the manufacturing sector. Let’s start off with a little bit of background. Why are you so cool? Besides running around podcast interviews with a crutch.  

Prateek: I think I’m cool because what I do, I didn’t go to school for that. I did not study statistics in college, but I started working and then realized this is a really cool field. This is just for all the people who want to get into this field: don’t let anything stop you. I had a background in computer science and information systems but not specifically any analytics. But I took a lot of trainings to enhance my career, and that’s how I got where I am right now.  

Beverly: I love that. Sometimes people come to me for mentorship, and they’ll say they have a scientific background. In some cases, that’s a good fit, because you can apply the scientific method to some of the data science work we do. So, very good. Thank you for that encouragement.  

Today we’re talking about some of the analytics methods that are in manufacturing, specifically. When we talk about data science and AI in manufacturing, what would you consider to be the maturity level? Are they the first ones out of the gate? Are they slower to adopt? Where do they stand in your mind? 

Prateek: I think most of the manufacturing companies that are there, they’ve been there for a very long time, if not hundreds of years, then at least 50 to 60 years. So, because of that, there is some rigidity in these organizations. They are not the first movers when you talk about analytics, because that’s not their core area. It takes them a while to get there, but once they get there, they can really do wonders. There are many use cases specific to manufacturing and analytics, and we need people who can solve those challenges.  

Beverly: Isn’t it funny that the people at younger companies sometimes feel like their companies are not mature enough, old enough, or don’t have enough resources yet. And then people at older companies are saying that their company is too mature. Is there a happy middle ground? In manufacturing, in particular, it seems like they’re a little bit later in the adoption curve.  

Prateek: Yeah, I agree with that. But at the same time, whenever there is a new technology that comes in, like gen AI right now, I feel like every company is trying to adopt it as fast as possible. That’s where you get to the sweet point where both things are happening together, where the old companies are adopting, and newer companies are developing. It’s a good merging point, where we are right now.  

Beverly: Tell us about some of the use cases that are in manufacturing. Like, are you all doing some cool things?  

Prateek: Yeah, definitely. There are some use cases which are very specific to manufacturing. One of the things that comes to mind is product quality. There are a lot of production processes that go on when we are building any part. And if any one thing that happens inside it changes, that changes the whole dynamic, and then we might end up with a faulty product. So, we want to ensure that we have the highest quality available. I compare it with the with the use case of credit card fraud. It would happen once in 1,000, but then it could happen. Once it happens, you really need to figure out why it happened.  

The same thing happens with manufacturing, when there are several parts that we are building, we want to make sure that every part has some good sense that has gone into that, and that’s where this particular use case comes into manufacturing. And then there are other things, like supply chain optimization, that’s everywhere. We are getting parts from all over the place.  

Beverly: So, supply chain optimization and then quality. But let me double click on quality a little bit, because that can mean a lot of things to a lot of people. There’s quality process, quality of your suppliers, there’s the quality you’re talking about, for the actual tangible widget or things, whatever asset that somebody’s building, right?  

Prateek: Right. At the company where I work, we have different types of quality, like product quality, supplier quality, and warranty quality to ensure our products are of the highest quality. Then there are lots of analytics that go into that process as well, just to make sure that the product that we are delivering is of the highest quality.  

Back to the point of analytics, we look at those warranty claims and try to make sure that those don’t occur over and over again. Another use case that comes into this is predictive maintenance because all these engines that we are delivering are computers now. All of these are IoT devices that are in the wild. We get all the data back from them, and then by using that data, we can try to predict when scheduled maintenance should happen. 

Beverly: I think I told you when we were at the conference recently about a client that I had, and they had this giant asset. It had all these IoT devices on it, and they weren’t quite sure what to do with the data. Are you seeing trends like that in manufacturing, where you need to get some IoT devices and AI but are sort of stuck on collecting the data? 

Prateek: I think that has happened. Even here, we have been collecting IoT data for a very long time. Initially, when the data comes in, nobody knows how it is structured and how to even make it readable. So, a lot of effort goes into the data engineering part of it. Then, already, we have several years of data accumulating until we get something out of that data. So yes, I agree completely with you, while it is a lot, it takes a while to use that data.  

Beverly: Is the manufacturing sector doing other things that everybody else is doing, like people analytics, marketing analytics, and operational improvement? 

Prateek: Definitely. There are some specific use cases and a lot of generic use cases that are happening everywhere. One of the newer ones that I recently found out about is energy analytics. So, we are using tons of energy, then we as a company also want to be carbon neutral and want to make sure that whatever energy that we are using goes into the plants, and then that is more sustainable than the previous methods that we’ve been using. There we are using some analytics to figure out what would be the best way of getting that energy. 

Beverly: Nice. TAG Data Science and AI did a joint event with TAG sustainability, and that topic came up of energy and how hard it is to have enough energy to maintain some sort of sustainability environment. So, I thought that was an interesting discussion, and one that scares me a little bit. But of all the different things that happen in manufacturing with data science today, it sounds like IoT is the thing that makes it unique, or where you see the opportunity.  

Prateek: That is true and another one with the current AI boom has to do with how, a lot of times, what happened was the trucks would go to service stations, but the technicians would write handwritten notes. There are a lot of manual parts involved in that process still. I mean, it will always be there, because there are people who fix those trucks whenever something goes wrong with them, but we were not able to use that data, since the advent of time. Now with these new technologies, we can build better summarizations and use that data to build our future products, and I’m excited about that. 

Beverly: So, the data before; why was it collected? Was it collected because you have to? 

Prateek: There are several reasons you would collect data, and there are several audits that go through. The data is collected, not for analytics as such, but for manual help. So, if somebody gets a similar claim, they can go to the past claims and see what somebody has written about that. Before, humans were able to read it, but machines could not read it, and it was not structured. It would write a lot of coding that would go inside that text.  

Beverly: Okay, so manual processing of text, notes, and comments that were originally taken because of operations or regulatory needs, now can be used as well.  

Prateek: Yes, it would help us significantly when we start to use it. It would go into both new product development as well as using that data to make sure that the products we have in existence are of the highest quality. 

Beverly: In manufacturing, you mentioned a couple things that are specific challenges that you all have with data science and AI. Things like how a lot of times manufacturing companies are older. You didn’t say this specifically, but I think sometimes, and this is cumulative with some of the manufacturing clients I know, the product kind of runs the house. So, if the product’s doing well, it kind of just takes care of everything, and there’s not a drive for data science and AI. The vertical, in and of itself, can be a little slower to adopt newer technologies, and so the age of companies and all these things get factored in together. Are there other special challenges? 

Prateek: Yeah, there are a lot of regulations and paperwork inside every one of those aspects. So even if we want to use that, there are multiple contracts that that have gone through, and the process itself is super complicated. We are translating a mechanical engineering problem into analytics, and without knowing all the rules of how those things work inside, it’s a very hard thing to do. Throughout my experience, I found what helps our analytics process is to have somebody who has business experience in that domain to sit alongside you. While I can know the technical aspects well, I need someone who knows the business as well as I do in the technical sense. So, that has been a big help with building any of the products.  

Beverly: That’s a good hack. So if you’re trying to figure out, knowing that maybe this vertical can be a little “not jumping on it first” when it comes to things like AI and new technologies, and knowing that we’ve got older products and that it’s difficult because the problems themselves are difficult. So, one thing you mentioned as a way to solve this is you got to have multidisciplinary. I know manufacturing companies, from my experience, have been very engineer-happy. They’re big on engineers, mechanical engineers, chemical engineers, depending on whatever the industry is, for doing everything. There’s one company I know that is a chemical company, and they have PhDs in chemical engineering that are in their marketing department, just because that’s what they do, they hire chemical engineers. So having a multidisciplinary is a great suggestion. What other advice would you give to these manufacturers that are trying to get through this?  

Prateek: So, what I felt is at least in our company, what we have put a little bit more focus on is to do more trainings. For instance, there are mechanical engineers who are doing lots of different things, like product engineers. Because they come from a mechanical background, they can break out a bigger, complex problem into smaller things. So, they are the project managers as well. But at the same time, if they get some exposure to more cross-disciplinary stuff, like studying some analytics, I’m trying to say that if they can get some training, they can even help the data scientists build those models. That has helped our organization.  

Beverly: Okay. Are there things that you can do with senior leadership. Or do you think they are the ones that are faster to jump on this? Or is it more ground up? What do you see?  

Prateek: From my perspective, I work in a technical team, so my director and my people come from technical backgrounds. So, in my experience, they are the ones who are already jumping into these things and coming up with ideas which have helped us. But I’m not sure about how it would work out in any other organization.  

Beverly: It can be tricky to advance and bring more AI technology into an organization when you don’t have senior leadership. So, I didn’t know if you had any ways of looking at how manufacturing has solved this, because it seems like if you make one little change in manufacturing, it can make giant dividends or giant differences. Maybe that’s the solution.  

Prateek: Yeah, true, that that could be the solution. If the top-down leadership focuses on something, then it could be achieved. There are business incentives for everybody else to achieve the same goals. So, the leadership should think about these solutions.

Beverly: A lot of our listeners are technical; how can you get in their ear? How can you help influence the leaders to try something new? Or do you have to just show this is how much we can save, or here’s how much we can make? 

Prateek: If you can provide a value, even if it is a rough ballpark number, that helps move the needle a bit. But then, of course, there are cases where you have to work through the political systems around the organization. If you can have good relations with people and talk them through, then it takes the whole direction.  

Beverly: That’s tricky. We’re saying just go do these things that are very, very hard. What do you think the future might hold for data science and AI in manufacturing? Is it going to be more of the same? Or is it going to be some dramatic shift? Or is there a gen AI play in here?  

Prateek: There is definitely a gen AI play in here. That’s what I have realized from the last several of my meetings where there is a lot of buzz about that. While we would not directly go into it headfirst, we would be thinking about those things for a very long time now. As I said, gen AI helps us with new product development. There are several white papers that we have written over several years which point out what goes into a new engine. Over several years, we have found out what the problems are with those things.  

If you combine those two together, you have a new draft produced for a newer set of engines. That gives us a very good starting phase that has already seen through the past deficiencies. So, I definitely feel that gen AI would come into picture. But at the same time, there are a lot of use cases which have not been explored yet. Even smaller things like supply chains, with the newer set of technologies that we have available and the new company computation we have available, we are currently improving our model due to that.  

I feel like both things are going to stay here for a while. Some gen AI will keep coming in, and it will grow bigger as time goes by. But there are many traditional analytics methods that would also be used for now.  

Beverly: Do you think that gen AI helps increase the number of hands that are going to touch it like compared to data science. I feel like it’s more user friendly and it’s more for people.  

Prateek: It would democratize the whole data science area. I don’t see that a lot of businesspeople that will directly go in and start coding tomorrow, using gen AI. But it would help them come up with new use cases. They can think of something and ask gen AI to develop something out of there, and then the things would go to a data scientist to implement, and it will help with conceptualization.  

Beverly: What final piece of advice would you give to people that are trying to better understand analytics challenges specific to manufacturing? 

Prateek: It’s all about people. When you build a system, by thinking about the people who will be using it, you will be able to build a better system. So, the reason I’m saying is because I see the data that comes in, and when the data comes in, the front-end structures are also not there for them to use it properly. There are not enough guidelines for them to build better data. So, for the people who will be using it, who are in manufacturing, and then who want to use it, my suggestion is to just build your processes around people, and then you’ll succeed.  

Beverly: Build your processes around people and add technology. Wow, that’s good advice. Thank you so much to Prateek Shrivastava, Principal Data Scientist at Cummins, for joining us today on TAG Data Talk.  

Prateek: Thank you so much for inviting me, Beverly.  

Explore the full catalog of TAG Data Talk conversations here: TAG Data Talk with Dr. Beverly Wright – TAG Online.