How CT Scans Can Improve EV Safety: with Efrat Avnet Steinberg

[00:00:22] Hello everyone and welcome to Kilowatt, a podcast about electric vehicles, renewable energy, autonomous driving and much, much more. My name is Bodhi and I am your host and on today's episode we are going to have a chat with Efrat Avnet Steinberg. So Efrat is the co-founder and CEO of Inner. Inner is a startup rethinking how we test and trust EV batteries, but that's not the only thing that Efrat has done. Her background spans law, product

[00:00:51] development and business leadership. She has experiences ranging from mobile finance to smart city projects and now battery tech. At Inner, she's leading a team developing a system that scans the entire EV battery pack to find hidden problems like cracks, leaks or swelling. These are problems that regular electrical tests can't detect. And this is all done by using x-ray machines in tandem with

[00:01:21] artificial intelligence. I'm going to be honest with you. Whenever I interview the CEO of any company, I'm a little bit apprehensive because some CEOs have a message that they want to get out there and no matter what you ask them, they're going to give you that message. And sometimes it could be quite a frustrating interview. I'm happy to report that Efrat was not that kind of CEO. She was so lovely to talk to her.

[00:01:51] Really got along well with her. Really got along well with her. Enjoyed the interview. She's full of good information and I just thought she was a delightful human. So without further ado, Efrat, welcome to the show. Well, thank you. Thank you for having me.

[00:02:09] All right. So I think that I think what we're going to talk about today is super interesting. But before we get to that, and I teased a little bit in the intro, but before we get to that, I want to talk a little bit about you because I think you've had a pretty interesting career, which, you know, according to your bio, it goes law, product management to business development. Kind of give us a little bit of your career and how it's developed.

[00:02:39] Yeah, thank you. Yes, I think I'm the farther from being the linear career person. I enjoy pivoting every few years. And I think it just builds an extra layer of what I've done before. So as you said, I started my career about 25 years ago as a lawyer. And I've been a commercial lawyer for a few years.

[00:03:03] But I think from the very beginning, it was clear that I want to take the decisions and not enjoy long term sitting in the advisor seats. So and then I switched to product and business development in different industries, always non-core. So I always enjoyed the new ideas and the concept of bringing an idea and getting it from a corridor conversation all the way to product.

[00:03:33] Always technology, always deep tech. I just love that. And I've done that in telecom and then in smart cities in the last 10 years working closely with startups.

[00:03:49] So I took this non-core entrepreneurial spirit that always excited me and started working with startups at the beginning, advising and working in a large startup accelerator. And then the last two and a half years actually leading my own startup together with my co-founders, which is Inner. You say like you like the new thing, you like the challenge. Inner Tech seems different than what you've done in the past.

[00:04:19] So let's talk a little bit about Inner Tech and then what led you to... It's Inner. I think a lot of people confuse. Yeah, it's Inner. Oh, because the website's Inner Tech. That's just to confuse you. Aha. Okay. Inner. And that's I-N-N-E-R. So what led you to founding Inner? So we came across a technology that actually was used in the medical world.

[00:04:49] This was very innovative and allowed building extremely large X-ray scanners in the medical world to bypass the constraints of the silicon wafer. Size and costs and challenging development processes.

[00:05:09] And this technology enabled building an X-ray detector from CMOS sensors, the ones that you have in your smartphone, in a modular way. And my co-founders come from medical innovation expertise, specifically from imaging, Philips, and from IT.

[00:05:35] And we started investigating the industrial world because we came to the understanding that actually this kind of approach of building extremely large CT scanners in a modular way actually could solve a huge problem in the automotive and energy world. So this was our starting point.

[00:06:26] So based on what I've read, your technology can, because there's sensors in the battery, there's a battery management system. You know, modules have them if they have modules nowadays. Nowadays, we're getting rid of the modules as a general rule. How does your CT scanner detect issues faster than the sensors do? Yeah, it's very interesting what you're asking.

[00:06:55] Because, of course, we acknowledge, we know that diagnosis is getting improved and better all the time. And there are plenty of BMS companies and platforms and smart sensors. Some of them even cling to be predictive. We get that. And we even acknowledge that they probably catch more than 90% of the faults that are out there. But there's a whole area of physical faults that those sensors fail to catch on time.

[00:07:23] We're talking about those cracks, dendrites, even torn wire. Let's talk about torn wire for a second. Because this story we keep hearing from the automakers across the world. We have been busy talking to automakers in the U.S., in Europe, in Asia. And the story keeps getting told.

[00:07:42] Once the pack is sealed, torn wires is an issue that existing sensors cannot test. And they acknowledge that close to 5% of the packs are sold to the public faulty. And this is outrageous, right? We know that the wires get loose and the sensors cannot detect it. Because you can't test it with voltage.

[00:08:10] You can't test it that nothing in the performance of the pack from day one will be affected. But if you had visuals, if the, in theory, if the box, the casing of the battery was transparent, you could actually look inside and see, oh, I have a torn wire there. I have a loose welding of the wire. And today, the existing measures cannot catch it. So I 100% agree with you.

[00:08:39] Diagnostics are really good. You can compare it even to, if we go back to our favorite analogy of the medical world. When you go to the hospital, because you're not feeling well, then the doctor will probably run a thousand blood tests on you and test your temperature. And they would get a pretty good idea about your health.

[00:09:01] But if your arm is broken, or if your appendicitis is really swollen, if you have pneumonia, then they actually need the visuals, the visual access to really diagnose what's wrong with you. So it's a bit similar in the battery pack. And there's quite a wide list known in literature for a long time, nothing new about this list, of physical faults.

[00:09:26] Again, we're talking about the torn wires, the swelling, cracks, dendrites, anode cutted overhang. There's really a long list of things that would not be impacting the performance of the pack at all at the beginning. You would need to have visuals to identify them. And as long as they're 50 micron or up, our scanner will identify them.

[00:09:52] Identify, measure, classify, alert, and give you very quickly a report so you can active on it. So you do this by using machine learning and AI. I would imagine if you're going to be throwing a battery pack or even a car in this machine, it's got to be quite large. But when somebody that's outside the company, somebody that buys your technology,

[00:10:20] do they need special training on being able to recognize these things? Because a radiologist is always going to look over any sort of imaging done in the hospital. Yeah, it's a very good question. Thank you for that. Not at all. The answer is very short. Although behind the scenes there's a lot of imaging and, of course, thousands of slices, because this is how you do CT or, in this case, laminography is the term,

[00:10:50] then the user does not need to see a single image. The user gets a recommendation, gets a diagnostic. So we would say something like, we found a crack. It's a module number ABC. It's the size of, I don't know, 75 micron. And therefore, our accommodation is this and this and this. Replace the module, fix the module, or whatever you can do about it. It really depends on the fault, of course.

[00:11:20] But it's our software is going to do it for you. So there's no need for you to go. Gotcha. And what would be the margin of error on that? Because obviously, we want to know if there's a problem. And then we also want to know if maybe there's a false positive that may end up us taking apart a whole battery pack unnecessarily. Yeah.

[00:11:43] The good thing about imaging, when the patient is a battery, then if you're suspecting something, there's no harm in doing it again. Yeah. If you're really concerned and you want to make sure, then you just take another scan for five minutes to run through it. But within, we're pretty confident it's not going to happen too often.

[00:12:07] And this is why the expertise in our team, specifically around imaging and around pattern recognition, is really the strong point. Our CTO, Hans Berman, spent 30 years in Philips developing imaging solutions.

[00:12:26] And actually handling and avoiding the case of false positives and false negatives, of course, is such an important issue in imaging. And it has solutions. There are solutions. There are ways to minimize that. And we don't think it's going to be a major problem. Okay.

[00:12:49] And I think we should maybe step back for just a second because I've been around CT scanners quite a bit in the last 20 years of my career. So how does a CT scanner work? Because you mentioned layers. And I think that's important because an x-ray machine just puts through radiation, takes an image. CT scanner is a little bit more surgical, I guess you could say. It's a little more deliberate in the scans that it takes. Can you explain how a CT scanner works? Yeah.

[00:13:18] So when it comes to human beings, if you would take a CT now, you would lie there still. And then the detector and the tube would rotate around you because they need to take the images from multiple directions, right? Across your body. And you're quite symmetrical. So they would take it from 360 around you.

[00:13:43] When you're trying to do imaging for a battery pack, then there are good directions and very bad directions. Because from here, because the radiation needs to penetrate from one side to the other. So the narrow side of the battery is pretty easy to penetrate if you use enough radiation, enough energy. If you're trying to do it from the side, you'll just see darkness because it's very long and very deep. This is why you're going to do it from the good sides of the battery.

[00:14:12] And this is why it's called luminography. And how exactly we do it, of course, is part of the IP. So it's hard for me to get into the specifics of that. But we're covering from the good sides of the battery. And the detector, the dimensions of our detector, which are very large, so they can cover the full dimensions of the pack,

[00:14:37] which is, I think, the main or not the main, one of the main differentiators, actually allows us to do it very, very quickly. Because some of the CT scanners that are out there that scan a battery pack are more research-oriented or sampling-oriented. They use a small detector. Therefore, in order to cover the two and a half meters dimensions of the battery pack, you need to do multiple iterations.

[00:15:06] This is why it takes them a long time, sometimes even hours. Well, our approach is really, we want to fit the workflow. We want to match the high throughput of a pack assembly line, for instance. So we need to be very, very quick. So this is why we designed the whole machine to fit the workflow and not to never slow you down, but rather to help you be quicker and better. It sounds like this is something that could be used by battery manufacturers.

[00:15:36] It sounds like it's something that could also be used by recyclers. And I read somewhere where you have either a version of this or the same thing where you can use, you can put the whole car in there and run the test. So who is your customer? Yeah. Yeah.

[00:15:57] So we start from the automakers because we, after interviewing and talking to, I think, across the board, we've talked to everybody. We understood that their pain is really the largest and the most urgent one because they're suffering from high warranty costs. And we really want to help them. There are too many recalls.

[00:16:22] And I think it's the, we keep track and it's more than a hundred thousand battery recalls a year, which is really crazy because we're talking about new batteries. And this is, we need to help them. So then this is the first use case that we're trying to help with in production. Then, of course, you can imagine that helping with recalls.

[00:16:47] Once there's a recall already, then boosting repairability is a major cause of ours as well. Because repairability today is extremely limited and is very manual, very unstandardized, extremely lengthy, relying on manual labor. So this is also where we could add huge value.

[00:17:10] You mentioned scanning the battery still as part of the car, which is exactly also part of our vision or part of our mission because it's exactly the same technology. There's not much in a car besides the battery. There's the battery holds the chassis. There are a bunch of seats there. There's not really much you need to ignore in your analysis.

[00:17:37] And we think it could boost secondhand sales. Today, when you go with a normal car, you just go and they check the brakes and the steering and the pollution of the engine for you. So tomorrow you can go to the same place and test instead of the engine, you test the battery to really know what's the history of the car, how the previous owner treated it.

[00:18:03] And then if we look ahead, then Second Life. Today, we see that this is extremely slowly progressing. And again, the testing processes are extremely unstandardized. I may even say sometimes unprofessional. The businesses of Second Life many times look like a couple of guys with a screwdriver.

[00:18:27] And if you want to scale, if you want to really build a Second Life for the dozens of millions of batteries that are going to be out there, you need to be extremely quick and extremely standardized. So we aim to be the industry 4.0 solution of the Second Life companies. And then if I'm continuing along with this life cycle of the battery, then of course recyclers.

[00:18:57] Because recyclers many times face also safety issues. They process batteries where they often have no idea what's the history of the pack. And maybe it's unsafe. So maybe some of the processes that the battery need to go through need to happen outside the facility to not endanger the facility from getting burnt.

[00:19:22] So there's a whole life cycle of the battery packs where we could add value. And don't forget that underneath all of that, there's a lot of data. We are going to look inside batteries along their life. Well, today the cell producers, they isolate cells. They age them.

[00:19:46] They do abuse tests in order to simulate how cells would look like after two years, five years, seven years on the road. With us, they don't need to simulate. We actually can see what goes wrong after years of use. We can identify unknown patterns to help the cell producers improve the design. So underneath all of that, there's extremely valuable data.

[00:20:11] So if I owned a mechanic shop and I wanted to buy one of these, is this something that's practical for Bodie's single guy mechanic shop? Or is this something that you're going to maybe deal with dealers first until the prices come down? Because I would imagine this is not, I know what a CT scanner roughly costs for a hospital. I can't imagine this is something that the average person will be able to afford for their business.

[00:20:41] Yeah, I think it's more medium-sized companies. You need to have a volume. As long as you have the volume, if Bodie now opens a business where Bodie scans cars, you know, all day, then 100% I would tell you buy the scanner. Because the ROI on that is going to be very, very clear. It's not for, if you do it as a hobby, probably not.

[00:21:09] But once there's a volume, once this is your business, the return of investment is going to be extremely clear. Extremely clear. It pays itself back. Because the report is so actionable and it helps you take decisions, then it's a no-brainer.

[00:21:29] For automakers, for the garages, if their business is that, helping people buy a second-hand EV or if they need to repair batteries, 100%. Do you have any of these devices deployed in the field? Not yet. Not yet. But we do already provide a service because our machines are not ready yet. We're a young company. We're still building them.

[00:21:59] But in the meantime, because there's demand, we already started offering a service with the help of our very prestigious partner, Fraunhofer, EZRT, in Germany. I don't know if you came across them. They're one of the most known applied research institutions based in Germany. And what they specialize in is doing research projects for the German automotive industry.

[00:22:27] And together with them, partners already today can send us suspected packs. We scan it there over in their research-oriented scanner. And then our team does the analysis. So analyzing those thousands of slices that we talked about before, identifying abnormalities, measuring them, flagging them, and giving the customer a report. So this is something we are already doing.

[00:22:56] And while, of course, building our scanner. During this testing, is there anything that has stood out to you as like, oh, wow, we didn't really account for this. But obviously, it does this very well. I think what I love hearing the most is what the customers are actually saying. Because they feel this is their bread and butter.

[00:23:24] And they live and breathe 24-7 batteries. And to hear from them how surprising it was to learn about, for instance, nonconformity between the different cells is really exactly the kind of impact that you want to make. Because you want to make sure that your customer has full knowledge about their own product.

[00:23:46] And to really empower them and make sure that they can deliver to their own customers the best product, highest standards of quality, highest standards of safety. So I think that's the best outcome of the work that we're doing. Now, I would imagine that the actual physical technology of the CT scanner remains largely the same.

[00:24:13] But when we look at new battery chemistries or even solid state batteries, I would imagine that's just a simple, like, we got to train the AI on these new systems. Exactly. Exactly. We're completely agnostic. We don't care about cell formation. We don't care about chemistry. It could be solid state. We're looking for abnormalities. And abnormalities are going to be there, whatever the chemistry is, whatever the cell formation is.

[00:24:42] So this is exactly why we're going to maintain the software to all the time perform the diagnosis on the existing and future battery types. This might be a hard question to answer, but is there anything out there that you see right now as a challenge the way that battery technology is going? For us, challenge for InR. Yes, for InR, yes.

[00:25:10] I think it's if they start building the batteries as a cube, which I've never seen anywhere. Yeah. But given the dimensions importance that I mentioned before, if the battery starts being a perfect cube where there's no angle that X-ray can penetrate, this makes our life very hard.

[00:25:38] Luckily, I've never seen a single automaker going this way. We have seen when they build this, but even when they build stationary applications, then sometimes, right, a big stationary device could be a cube. But even then, they're built from modules that are very easily scanned.

[00:26:04] So you can just take the one level of the big box, you scan it, you put it back easy. But again, if they start building it really as a perfect cube, that would be harder to perform imaging on.

[00:26:19] Other than that, in all of the scenarios and simulations and testing that we're doing and conversations with industry players, we don't see a challenge that is not solvable. But of course, we keep track all the time. And this is such a hectic and crazy market. So you can never know what's coming.

[00:26:49] Is there a form factor for the cells that works better than others, like a pouch versus prismatic versus cylinder? I think what we've noticed so far is that it's the easiest with pouch, I think. It's the easiest to see things. But it's not something, there's no drama there.

[00:27:14] I mean, you just need to do a little bit more work on the software. But I think, yeah, that's what I can say about that. Let me think, what else have we seen that makes it easy or hard? Yeah, it's... No, I think that's one thing that I can say about the form factor.

[00:27:42] The prismatic cells don't give you any trouble, just the way that they're stacked in there? No, it's... That's relatively more similar to the pouch. No, I think it's... But maybe it would be an interesting topic for a follow-up with my CTO once we can maybe release one of our white papers later on, so we can actually maybe make it around the images. Oh, yeah, that'd be awesome. Cool, yeah.

[00:28:12] Then we can actually show findings. That would be interesting for you. We can look into... Oh, yeah. I would love that. We're getting close on time here, so I don't want to keep you for too long, because you are in the Netherlands, and I'm just starting my day here in Arizona, but you are ready to end your day. In terms of safety and standards, if you're going to be selling these all over the world, is there anything specific that you have to...

[00:28:39] Is there a specific standard that you're building them to that kind of matches what the safety standards in different countries? Like in the EU, it's a little easier because they all kind of match roughly, but if you're going to the Middle East, to Asia, to the US, maybe there's some different standards that you have to overcome. We're just talking about the X-ray. Yeah, just the machine itself.

[00:29:02] Yeah, so I mean, although the architecture and the software are completely new, the radiation part is very standardized. There is nothing really unique about that. So I think in every different area in the world, they're quite similar in the sense of what kind of shielding do you need to protect people that are standing outside of the machine?

[00:29:31] It's quite standard in this sense. So there's nothing really unique about that. And apart from that, we're just going to... Every market we will go, we'll make sure, for instance, when we build a production line, then it just fits the criteria and the demands of the customers and their existing factories. So I don't think it's going to be a major challenge. We're quite prepared to handle that.

[00:30:00] So that's fine. Are there two different versions of the CT scanner? Is there one that you can drive a car in and one that you could just use for the battery pack? Yeah, exactly. It's the scanner that's intended for standalone packs is, of course, a little bit smaller. It's still a big machine, of course, because the battery pack is big.

[00:30:26] But the other one where the whole car goes inside, it looks a bit like a car wash. Main difference is, of course, you need to step out of it, stand outside, close the door, because we don't want to fry you. And then within a few minutes, you get a report. So those are the main two versions. Excellent. Excellent. And so I always ask this of every single person who's building something, especially something new.

[00:30:55] What were some of the unforeseen challenges that you came across when you started building this company? Obviously, you have a lot of expertise in the medical world with CT scans and stuff like that, but those are scanning flesh bags. They're not scanning metal objects. So what are some of the unforeseen challenges and how did you overcome them?

[00:31:26] I think there's probably additional challenges that we're still going to discover along the way. But I think one of the most interesting things that we saw is that a lot of the existing solutions out there position the pack on its side, because you want, of course, we said, yeah, if you're a human being and you're lying in a CT machine,

[00:31:56] then you just lie there and the machine rotates around you. Many of the CT scanners that are designed for battery packs position the pack standing on its side and then it rotates. What is interesting is that... The pack rotates, not the machine. The pack rotates. It's a huge machine. So the machine stands there and the pack rotates. Gotcha.

[00:32:23] But then we played a bit with it just to see what happens. But then, of course, the battery pack was never designed to stand on its side. And there are fluids there also. There's fluids. There could be air bubbles. There could be so many things that are actually manifest when you do this. So it really interferes with the scan. And then when you do the scan, when the battery is, again, in its natural position,

[00:32:54] that gives you a more accurate diagnosis because the battery never stands, never... You don't need to change the layout. You don't need to change its positioning. So I think playing with that and testing what impacts the battery when you position it and how does it look and what you can actually find, I think it was a very interesting test or a very interesting game to play.

[00:33:25] Yeah, I think that's a very interesting one. I know that one of the things that we're extremely excited about is the quick processing part because you're playing with terabytes of data. This is a huge object, very extremely heavy. And we're trying to... We're not trying. We're aiming to show you a fault that's 50 microns its size. You're talking about terabytes of data.

[00:33:53] So I think one of the challenges that we're most excited about is doing that so quickly. Yeah, but that's extremely exciting for us. Is all of that AI happening locally or is that shooting information up into the cloud and then being sorted out there and shot back down? Both.

[00:34:22] There are some elements of it that are happening on the cloud, 100%. But some of them are going to happen locally. There are also demands or needs of the customers to do things sometimes locally. So it's going to be a combined processing work. About how long does it take to do a scan of a standard battery pack? A few minutes. Okay.

[00:34:50] Yeah, that seems like it's pretty fast for terabytes of data. So, Everett, I want to give you an opportunity. If there's anything that you want to let everybody know about that I did not ask you about, I want to give you an opportunity to kind of be the last word, wrap things up. Perfect. Thank you. I think what we aim to achieve is a world of zero faults.

[00:35:19] The current situation where there's so many recalls, so many warranty costs, and lack of repairability need to change. And we aim to bring diagnostics closer to perfection, closer to zero faults for a better, safer, EV world for all of us. Excellent. And I think that's great.

[00:35:48] And I actually, I lied. I have one more question. You hear all the time, I'm sorry. You hear all the time somebody goes into the doctor for one reason, and then they find out they have another medical condition they didn't know about. When you are scanning cars, have you, to this point, ever found anything that maybe was involved with a high voltage system?

[00:36:11] Not necessarily the battery itself, but the high voltage system that is attached to the battery or something else in the car that you could be like, this is something you might want to pay attention to. 100%. But is it a bad thing? I think it's a good thing. No, it's a good thing. You're American. I would expect this kind of, I mean, if the question was about over the diagnostics, it usually would be a European concept or, yeah, just kidding. It's, of course, we are looking for any kind of abnormality.

[00:36:41] It could be in the BMS, it could be the cooling system, it could be any part of the battery that actually goes wrong. And this is also something to be defined together with a customer, because a customer could be worried about the specific fault. So we can focus on the specific fault that's interesting for the customer and then make the diagnostics even quicker. So it's, but 100%, we're looking inside and we can see everything. Okay, excellent.

[00:37:11] Efrat, thank you so much for joining me today. I had a lovely chat. So, where can people find out more about Inner? So, first of all, thank you. I've had a great time as well. Thank you for having me. We have a website. That's innertech.ai. With a lot of information. We're happy to, if people want to know more and talk to us, feel free to reach out.

[00:37:38] We have also a LinkedIn page, so people can also follow us there and know what's new. We're trying to share any kind of announcements or events or news there as well. And feel free to reach out to me as well. So, yeah, Efrat at innertech.ai. Yeah, we love talking to the industry. And every conversation teaches you more.

[00:38:08] And there's a lot of collaboration going on. So, extremely happy to explore it. Excellent. Excellent. Well, thank you so much. I appreciate you coming on. Thank you. All right, everybody. I want to thank Efrat Avnit Steinberg for coming on and being such a good guest. It's just a lot of fun. I really enjoyed chatting with her. And hopefully we'll get to do more with Inner in the future.

[00:38:36] If you want to check out what they are up to, links to their website and social media in the show notes. But you can also go to innertech.nl for more information on what they're up to. Again, I'll put links in the show notes so you don't have to necessarily remember their website. But it's pretty easy.

[00:39:05] Innertech.nl. Oh, before I let you go, I do also need to thank Martina for setting this up. Again, I can't express the amount of moving parts that have to take place in order to get an interview done. Martina was a delight to work with. So thank you, Martina, for being so easy to work with and helpful. It makes my job easier. All right, everybody.

[00:39:32] On Friday's episode, we get back to our regularly scheduled EV news. So I'm looking forward to getting back into the news. It's really hard for me right now being on break and not being able to just throw a couple things here and there about deliveries or, you know, some little tidbit of information. But I'm trying to keep each episode specific to our guest so as not to take anything away. So thank you, everybody, for listening.

[00:40:00] And I will talk to you on Friday.