Predictive Maintenance in the Semiconductor Industry- with Semi.org

Trend Detection Podcast ·

Show notes

Welcome to the Trend Detection podcast, brought to you by Senseye Predictive Maintenance – the platform which enables predictive maintenance at scale across all of your assets, across all of your plants.

In this episode, we are joined by Tim Hunt and Todd Martin to discuss PdM in the Semiconductor Industry- with Semi.org.

You can find out more about how Senseye Predictive Maintenance can reduce unplanned downtime and contribute towards improved sustainability within your manufacturing plants, by visiting: www.siemens.com/senseye-predictive-maintenance

Transcript

Welcome podcast powered by Siemens, and now your host. Good morning. Good afternoon. My name is Eric Rood, director with SIMMI, and it is my honor to welcome you to Siemens' webinar from Reactive to Proactive Mastering Predictive Maintenance with AI. SEMI's committed to providing a platform to bring technologies and efficiencies to our members, and I hope you will find Siemens' presentation valuable and actionable for your own operations. To begin, it is my pleasure to introduce Ryan Mishin, global account manager for Semiconductor at Siemens, and our moderator today. Ryan, could you please take it from here? Appreciate the introduction, Eric. Uh, as you mentioned, my name is Ryan, Ms. Jan. I work at Siemens, uh, in the semiconductor space, and I'm happy to host this conversation today with some of our, uh, some of Siemens experts in the predictive maintenance and AI, uh, technologies and just to host this conversation about what we've been seeing, ah, around semiconductor with these predictive maintenance and AI type strategies. So, As the title suggests, this is really a, it's a hybridization of predictive maintenance, which is Not such a new topic, ah, it's a little bit more mature. Um, a lot of companies have a predictive maintenance strategy or are leaning to leaning that way, and we'll look at what that looks like. But then also with the, ah, emergence of AI, ah, especially in the industrial field, ah, how to couple that to really make a more robust and impactful predictive maintenance strategy. So, Uh, we're happy to discuss this with our experts and, and hopefully, ah, provide some, some additional insights to, to people out there in the semi industry. So just a couple of table setting statements um that we've heard from, number one, there's a an an automotive manufacturer that we've been working with for quite some time. Um, and I'm just gonna go ahead and read it. Ah, so we saved $45 million in labor and unplanned downtime avoidance at a single site since 2019. So, cumulatively $45 million in, in this labor and unplanned downtime. Ah, just by implementing a predictive maintenance strategy. So, ah, by, by working with their historical data and sensing and, and tending to issues before they break, can really have some, some huge upsides in, in, ah, for your bottom line. Ah, the next statement is really more of an industry neutral statement, so it's from a, uh, consultant that's, that we've worked with called McKinsey Company. I think a lot of people may be familiar with them. And again, this is more industry neutral, um, but digital tools have dramatically reduced unplanned outages. Higher availability and a more efficient workforce have increased profitability by 4 to 10% in some organizations. So a lot of these digital tools are available. Um, it's really how you implement them to best suit your application, could yield that 4 to 10% um. Ah, impact on your business. So, again, that's more of the industry neutral one, where that top one is, is really pertinent to to automotive. But you know, really there's, there's not an industry out there that that couldn't benefit from a predictive maintenance strategy, most notably the semiconductor industry which we're talking about today. So just a quick introduction to the speakers, panelists. So I'll be moderating, ah, sharing some kind of high-level information at the beginning. So, as Eric mentioned, my name's Ryan Mischan. I'm a global account manager at Siemens, ah, servicing the semiconductor market. I've been with Siemens almost 18 years, so grew out of the automation drives, um, a business, and now I support our whole portfolio through our Through our software portfolio, um, so yeah, uh, that's me. I'll introduce Tim next. Hello, I'm Tim Hunt. Uh, I've been with Siemens, uh, Siemens for 2.5 years now. Before that, I was with a company called Logan Aluminum. It's an aluminum rolling mill out of Russville, Kentucky. I was, uh, uh, I was there for 25 years. I retired from there. And uh I had the opportunity to really sink my teeth into predictive maintenance for that. I was able to do it in one of the departments, the finishing department, I was able to, I was there for 5 years and had so much success by using all the predictive maintenance technologies that I was got promoted to being able to do that for the plant and that's where I really grew a passion for uh predictive maintenance. And that's why I was able to build programs and, and, in the departments and train PDM coordinators and, and I just love this journey of using all the uh PDM tools and then bringing on AI as being another part of this PDM tool. So uh and I'm the uh it's a pleasure to be here with y'all. Um, good morning, everybody, and thank you, Ryan. Um, Todd Martin. Uh, just recently joined Siemens myself, but before Siemens, I, um, worked at Nissan at their, um, automotive production facility, and it was there that I began my predictive maintenance, uh, journey. Before that, I was heavily into automation and machine building, um, industries. So it was there at Nissan that I began, um, Designing and implementing uh condition-based monitoring systems and then taking those and bringing on uh predictive maintenance technologies. Um, and we were quite successful at it. So that's brought me to Siemens, um, to continue my career. Awesome. Thanks, Todd. So, today, today's agenda, what does it look like? Um, yeah, just, um, maybe just a, a more of a, on a personal note, um, uh, a little maintenance thing I've got going on at, at my own property. Uh, I have a gate latch out there that I had to reinstall this morning. Uh, we have some rather inclement weather in Portland, Oregon, and, um, yeah, I'm all of a sudden kind of wishing I had more of a Proactive maintenance strategy for it. Uh, so, yeah, just on a personal note, um, but yeah, what are we looking for? What are we looking at in terms of the agenda today? So, we're gonna look at some of the overall, so overarching challenges that the semiconductor industry, uh, are facing from, from our perspective. Uh, we, we talk or we work regularly with all sorts of, uh, end users out there. And yeah, to no surprise, a lot of their challenges ring true from customer to customer, so we'll kind of share what those look like. Um, how can predictive maintenance address those challenges? Um, yeah, a lot of those challenges are easily addressable or, or have some, some big impacts if, if you implement a good, ah, solid PDM strategy. How is maintenance typically performed today? You know, we'll talk about, you know, the classic, you know, the old school break fix strategy. Through a proactive maintenance strategy all the way up through a predictive maintenance, um, maintenance strategy. So what is predictive maintenance and what are some of the benefits that you could, you could gain from implementing a strategy like this. And then I think a lot of what we're interested in today is, is how can this emerging AI artificial intelligence. Um, technology really enhance those benefits, um, that, that you may already be using in a predictive maintenance strategy, or how can you really accelerate that predictive maintenance, um, that you already have going. Ah, implementation of PDM and then we'll wrap it up with a little summary. Uh, we'll get some time for Q&A with, with Tim and Todd, and, um, and we'll keep on moving. So, we have a little bit of an interactive poll that we'd like to, to ask the audience members. And so Hole number one. So what aspects of predictive maintenance are you most curious about? Let's say #1, is it the implementation strategy? Is it 2, cost savings that you can realize by predictive maintenance, 3, the use of AI, you know, the, the emerging technology, or something else. OK, I'm not, uh, not sure where the poll questions are. I have a chat here. Does everyone see the results? Or Ryan, please check the post section. The results are coming in. OK. Um, let's see. Imagine there's some real juicy statistics out there for what we have coming in, but uh. I'm not sure if I can see the controls here for poll showing. Audience view. OK. At audio web sharing. One second. Let me share the results with you. Oh, excellent. Thank you. Doing some live sharing of the results. OK. Scared attendance, timer. OK. Questions. Um, If you could go to post, then you should be able to see. OK, here we go, there we go. So, uh, to, to share this with the audience, about half of, half of everyone said the use of AI, that's the number one thing that you're that they're, that you're all here to, to learn about. Second, ah, by a close margin was, uh, the implementation strategies on how you implement PDM and then, you know, roughly 11% said cost savings that you can realize with with the predictive maintenance strategy. So, ah, we'll definitely hit heavy on the, on the use of AI as we, as we talk through some of these high points. OK. Thanks, Elle. OK, so like we talked about, um, our teams work with semiconductor manufacturers across the board. And to no surprise, a lot of these topics should not be, um, should not be foreign, I would, I would guess to a lot of the people that are on this webinar. So if you're on this webinar, you're probably working directly with or around some of these topics anyway, so, um, yeah, let's see how How these might impact your business. So sustainability is, is one area that by far, it's, it's one of the key pillars behind all of our semiconductor manufacturers, um, kind of their, their company ethos. So, that, that doesn't necessarily only apply to environmental impacts, um, of sustainability, but that also it means sustainability of machines, um, the sustainability of the business itself. Um, but certainly you can implement a predictive maintenance strategy to, to optimize your energy bill, um, by, by learning from historical data to, to, ah, interact with machines in a certain way. You can really, ah, lower your, your energy bill. So that's one use case of predictive maintenance. You know, of course, you know, by, by properly taking care of machines and processes. You ultimately lower your waste of raw goods and and lost products, so that's, you know, inherently a sustainability topic that we see all the time. I think, you know, this, the next one goes without without saying is is maximizing your overall machine lifespan. So that's probably the most understood application of PDM. And then of course, you know, we'd like to help reduce the potential overall environmental impacts by driving down the waste and driving down energy consumption. This as a, as a whole, ah, has, has a big, has a big impact on sustainability. Um, data usage and AI modeling is probably another one that we're, like we talked about, we're all here for, um. So, your companies out there are sitting on what I'm guessing is a huge stockpile of data, OK. So, you have historical data that's usable in these predictive maintenance strategies. You can use the data to help model your predictive maintenance by by inputting them in a clever way. Um, so, yeah, leveraging your data to to better enhance your predictive maintenance strategies. I think a lot of people on the on the webinar are probably well down the line of uh being able to connect and talk to uh different machines and different processes in their in their fab or back end facility. um, so that's kind of inherently inherently critical for managing sorry, I have a dog that's barking a lot. It shouldn't last, um. Yeah, being able to talk to these machines is critical for um for for being able to to manage them and send send schedules out to them. And, and then of course, implementing an AI strategy based on that historical data, that's sort of how we would enable AI is, is a lot of that historical data and being able to bring in ah other data sources and data streams to make your processes more intelligent. Digitalization is, is probably another more of like a company-wide ah strategy that a lot of companies are using to transform their business. Ah, and we see this a lot in, in the way that people are now starting to, ah, Use digital tools to simulate different parts of their processes or machines or figure out how different parts of their company work together. Um, we're seeing a lot of this stuff, a lot of digital tools implemented to, to, ah, to start to address that stuff. And Digital tools are also uh helping us learn more about our labor optimism or labor um and how we, how we implement um uh or how we optimize labor in the workforce. So, before, ah, digital tools and and predictive maintenance around those digital tools were really just seen as the ability to simulate and maybe um maybe optimize throughput, but lately, what we're seeing is, is using these digital tools, predictive maintenance included, in ways where we can optimize our labor. So, um, One application would be ah the scheduling of labor to perform maintenance, um, you know, not just sending people out on a, on a scheduled time, but actually learning about machines or processes in an intelligent enough way to then say, hey, at this date, I need to schedule this this resource out to make X, Y, Z changes, um, really being much more intelligent about how, how you deploy your labor. And then also you can use that as as a training, um, ah, ah, a training platform as well, um, using the predictive maintenance platform to to gain information, to train personnel, and of course, uh, inherently making the um the operations more, more safe and secure and, and easy to operate for, for your employees. So, at this point, we have a question that I'd like to pose to Tim and Todd. So, yeah, question for the for the fellows. How is maintenance typically performed today? And what are the challenges of doing it this way? Well, you see a lot today and, and I was surprised because the world I come from 30 years ago we done reactive maintenance, but we moved on to that from that to predictive maintenance and I thought that's what everybody in the world was doing until I came to work for Siemens and uh 2.5 years into this uh being the customer success manager, there are so many folks doing. Active maintenance. They're running into failure. They're run into this thing just melts down, then they go out there and take a crane and pull it out and put something else in and, and there's a better way to do it. We don't have to do that no more. That was good back in the 80s, but we don't have to run like that no more. We can predict now uh uh and. You can predict 30 days, 60 days, 90 days. I mean you can depend on the line speed and, and the pressures and torques and all, but I mean we can see way out in time to give you plenty of time to order the parts, get the manpower and get everything together. You don't have to tear all this. Equipment up anymore because secondary failures is our biggest cost anyway when it fails. Um, this slide right here, I love talking to this slide because it, it brings you a big picture because there might be some folks here on the call uh that, that don't understand really how PM works, what the, what the technologies are. Uh, you'll see this is the PDF curve, the prevention to failure, prevention being on your left. Uh, this is when your machine is starting to fail that goes into a failure mode like you got a bearing that creates a small on the, uh, say the outer race and now it's going into a failure mode. That's your top left corner. Well, we can use, uh, ultrasound or we can use, uh, uh, lubrication, uh, oil analysis to pick that up real early. But as it goes on, now, then we can use vibration to pick up that failure and we're actually at that point, we're picking it up with AI and as it trends on down and then you'll start getting the audible noise. At this point, uh, before vibration, you can't even see these faults with the human eye. I've split these bearings open before. Hey, there, there is no fault. Well, it's there in the oil analysis because your iron counts is spiking. But as it continues to fail, then the operators and maintenance will say, hey, I hear something, something don't sound right on this line. Well, that's that bearing, the spa is getting bigger and bigger and as the rolling elements roll over, it's creating the impact with a frequency. Uh, then you'll bring in your thermography, well, you'll see it then the temperature is rising. Well, then the next point is failure, ultimate failure. This is where you melt the bearings. This is when you score the shaft, probably destroy your fan housing. Uh, so there's a better way to do it. There, there's a system here and that we can really make a difference and I get to work with customers every week and see how we're making a difference with our, with our equipment. So, so Tim, so then what's the, what's the alternative then? Well, I mean, Ryan, the alternative is, is, is, and we've been doing this for a long time, you know, reactive went from reactive to condition monitoring and condition-based maintenance. We've all added technologies to our equipment to measure key factors on that equipment, temperature, motor current, torque, um, You know, vibration, time, you know, time to activation, things like that, and those technologies were a big, huge improvement in, you know, predicting failures ahead of time, but not scalable because you would have to know what to set the limits that you were going to react to those conditions. Um, that means If you've got a few assets or a few pieces of equipment, you can keep up with it, but as I said, not scalable to thousands of assets on the floor. So, That led to what do we, you know, how do we scale and be able to monitor all our pieces of equipment in the facility and leading into predictive maintenance and, you know, data analytics techniques and all that. And this predictive maintenance technologies, very mature, um, you know, I myself have used them for, you know, coming up close on 7 years and with great success. Um, So, of course, a lot of people here are going to be familiar that yes, we have these predictive maintenance technologies, um, but what's the new and up and coming, what are we, and that's, it's, that's AI. Now I imagine a lot of people think, OK, we're going to actually use the AI to look at those data signatures um to predict the failures. In some cases, yes, but one of the biggest exciting things about using AI in this is the ability to AI to ingest not just those um data signals and measurement values, but also your contextual data that you've been entering into your maintenance monitoring systems, you know, work histories, um, oh, we've seen this on another one and we input this because Tim said the, the bearing blew up. um. So now with, with the AI integrated with PDM it is able to tap that reservoir of knowledge that us as human beings have put in and spent a lot of time, you know, accumulating. Another portion of this AI is now conversational PDM. And what is that is that rather than just say, you know, we saw this X thing happen on this piece of equipment and this, this and this, it now starts speaking to us in conversational tones that we've become familiar with in other, um, you know, conversational AI tools. So it's, it really Expands the um the group of people that can use predictive maintenance to maintain their equipment. Even the advent for a long time, but you still had to have your, um, your, your subject matter experts, you know, people who were familiar with the equipment that could say, oh yes, this predictive is pointed at something's going on here, but then it was still. Almost 100% the human being had to look at what it was telling you, take that data. Investigate, interpret, to get really to the root issue. Now with this conversational and generational AI included with it. You don't have to have that specific subject matter expert as much. I'm not say anymore, but as much. …

Full transcript (WebVTT)