How Machine Learning is Building Trust in 3D Printing with Howco Group

While additive manufacturing, also known as 3D printing, offers transformative potential for the marine and offshore industries, broad adoption will require trust and buy-in from key stakeholders.

In this episode of Setting Course, an ABS Podcast, Conrad Kao, Director of Additive Manufacturing for Howco Group, and Michael Kei, Vice President of Technology for ABS, joined host Brad Cox to discuss the role of data analysis and machine learning in building trust in and validating components produced with AM.

Share this episode on social media, leave a review on your favorite podcast platform or send feedback to podcast@eagle.org. Learn more about how ABS is supporting the maritime industry at www.eagle.org.

Key Points

• Machine learning can help optimize additive manufacturing processes.
• Trust in equipment is critical for marine applications.
• Data-driven certification can reduce lead times and costs.
• Human oversight is essential for safety and compliance.
• Standards are crucial for the adoption of new technologies.
• The future of manufacturing is increasingly digital.
• Collaboration is needed to develop universal standards.

Guests

Conrad Kao, PE, is a seasoned engineering leader and Director of Additive Manufacturing at Howco Group, where he drives strategic integration of advanced metal additive technologies to solve complex supply chain and production challenges, particularly in the oil and gas, aerospace, and high-performance industrial sectors. With a strong foundation in mechanical engineering and a Professional Engineering (PE) license, Conrad brings a unique blend of technical expertise, business insight, and practical leadership to the rapidly evolving world of additive manufacturing (AM). He is passionate about advancing additive solutions that improve part performance, reduce lead times, and enable new levels of customization and efficiency in critical applications. At Howco, Conrad oversees the company’s additive operations, championing the adoption of cutting-edge AM processes such as laser powder bed fusion and hybrid manufacturing. Under his leadership, Howco Additive has expanded its capabilities to serve demanding markets with complex geometries and high-value metal components — driving innovation from concept through qualification and production.

Michael Kei is Vice President of Technology for ABS. In his role, Michael is responsible for leading and approving technical development and maintenance of products & services to meet organizational and business needs. He has 20 years of experience in the marine and offshore industry where he has driven innovation and led high-performing teams. Michael holds a bachelor’s degree in civil engineering and a master’s in offshore engineering and has a proven track record of implementing cutting-edge solutions that enhance operational efficiency. 

Chapters

• 0:00 Introduction
• 0:58 Additive Manufacturing Overview
• 2:45 Leveraging Data to Validate Components
• 9:08 Machine Learning vs. Artificial Intelligence
• 14:08 Human Oversight for Data-Driven Certification
• 15:59 Evolving Standards for AM-Produced Components

Transcript

Brad Cox (00:07)
Welcome to Setting Course, an ABS Podcast, where we're charting the future of the marine and offshore industries. I'm your host, Brad Cox, and today we're diving into a couple of cutting-edge technologies that I find really interesting. We've talked about additive manufacturing on the show before, highlighting some of the big benefits and the role of digital warehousing. Today, we're going to be exploring how data analysis and machine learning could help build trust in those 3D-printed components and what that could mean for AM's role in the maritime industry.

To provide insight from the manufacturer's perspective, we have Conrad Kao, Director of Additive Manufacturing at Howco Group. Howco manufactures and distributes alloys and components for use in some of the world's most challenging applications, including oil, gas, chemical processing and the commercial space industry. Thanks for joining us in the podcast studio, Conrad.

Conrad Kao (00:49)
Thanks for having me.

Brad Cox (00:50)
It's also great to welcome back Michael Kei, ABS Vice President of Technology for the Americas. Thanks for joining us, Michael.

Michael Kei (00:55)
Thanks, Brad. Always.

Brad Cox (00:57)
So, to get us started, I do want to look at the bigger picture. So, Conrad, why is additive manufacturing getting so much attention these days and what can it mean for industries that really embrace it?

Conrad Kao (01:07)
Yeah, absolutely. So, I got my start in additive manufacturing about 10 years ago, working for one of the service companies, and we were trying to get engineers to adopt this new metal printing technology. They approached it from strictly a supply chain perspective of just trying to get existing components and seeing if it could print them. Turns out we can print them, but it's not necessarily cost effective, right? 

So, what we're finding now is there's so much more experience on the design engineering side of things. Engineers actually know how to design for additive and take advantage of the complex geometries that are made available to them. That paired with better understanding of post-processing and material properties, those two aspects combined, we're seeing a lot more adoption by the oil and gas companies.

Brad Cox (01:47)
So, what's been the experience with those parts so far? What's been the performance? Is it on par with traditional manufacturing?

Conrad Kao (01:54)
Yeah, absolutely. We have parts in the field right now. Almost every single major player in the service industry has downhole components. And in most cases, you can either meet or exceed traditionally wrought material properties with specific heat treats, hip cycles, especially of stainless steel alloys and nickel alloys.

Brad Cox (02:10)
Michael, what can AM mean for marine and offshore in particular, and what's that potential?

Michael Kei (02:16)
To reinforce what Conrad has actually mentioned, AM allows for design freedom. So basically you can design parts that are traditionally not manufacturable through the traditional methods. In addition, you can look at new materials. There are lot of exotic materials that are being studied right now that are not part of the traditional supply chain at the moment. So you can just imagine if you are able to print something that is of a different shape, different component but serves a better function in terms of strength and endurance. That is what AM can actually bring.

Brad Cox (02:45)
So, focusing in on our core topic, Michael, I expect you'll have lot of insight to this. Trusting your equipment when you're out in the ocean is critical. How are components validated today and how are technologies like machine learning changing that approach?

Michael Kei (02:58)
Additive manufacturing takes what used to be a series of manufacturing tasks and rolls them into a single and digital build. That's powerful. It also changes how we help make sure that the parts are safe and reliable. 

In traditional manufacturing, quality checks happen at every stage of the process. It's like guardrails along the road. Whereas with AM, many of those checkpoints disappeared. So, we need smarter, data-driven ways to verify the outcomes and keep the same level of trust and certification. So, certification for additive manufacturing isn't just a tick in the box. It's about meeting industry standards, proving safety, controlling the process, and building confidence for shipowners and OEMs. 

Traditional manufacturing leans on decades of data and mature processes like casting, forging and machining. So, qualification is lighter, variability is lower and testing is minimal. Whereas AM is newer and more complex, so, qualification is more rigorous at the moment, with heavier testing and detailed process records. Today, that means extensive physical testing, material certification, process qualification to show repeatability. It's a lot more rigorous than traditional manufacturing, which relies on decades of proven methods. 

But here's the exciting part for AM is that it is digital in nature. So, it's designed to production integration, rich process data and full traceability. So it opens the door to faster, smarter qualification. The industry is moving towards model-based approaches, physics simulations to predict performance, machine learning to reduce testing and in situ monitoring with digital twins to validate builds in real time. So, it's basically an end-to-end digital record of the full manufacturing cycle.

Brad Cox (04:38)
So, you mentioned trust, and I know that's going to be the buzzword today, because I think a lot of this is trust in the data, trust in data collection, trust in data analysis, and even trust in additive manufacturing. Conrad, can you provide some more insight into Howco’s approach to validation, and what's your perspective on using machine learning to support this?

Conrad Kao (04:55)
Yeah, so Howco, we operate as a contract manufacturer for additive manufacturing. And so our approach is dictated by our customers, which is still based on traditional methods, right? So, we still do mechanical testing, NDE where it's required. But I guess to emphasize what Michael is saying, there's a tremendous amount of data being captured currently with additive processes. And it only makes sense to utilize that data and eventually build profiles or models to pre-qualify, if you will, end-use components. The cost and lead time involved with traditional validation can be thousands to tens of thousands of dollars per part, often adding several weeks to months of lead time as well. If we approach this from a digital perspective, the opportunity is there to save a lot of money for everybody involved.

Brad Cox (05:37)
So Conrad, can you go into a little more detail as to what traditional testing involves?

Conrad Kao (05:42)
Sure, absolutely. I should take a step back even further and say for the oil and gas space, you know, we have standards like the API 20S guideline that basically separates components into different criticality groups. Depending on the criticality, maybe we would do some density testing, mechanical properties such as yield strength and tensile strength, all the way to the most critical where you do non-destructive evaluation as far as CT scan and other similar types of testing. 

Currently for the vast majority of parts, we're doing yield strength, tensile strength, charpy, hardness testing. And then in certain instances, if it's required by customer, we'll do the final CT scan.

Brad Cox (06:17)
And so, in the scope of what we're talking about here is you're capturing that data upfront and that tells you that it meets standards before you have to do any of that stuff.

Conrad Kao (06:26)
Sort of. So, with the process, you're capturing, you're utilizing the in situ data to build a profile for what a good part would look like. At least this is how I understand the potential. So, you would run enough tests using a known component, build it, do the physical testing, build that profile. And then based on that, you can say, this is the profile of what we've input and assuming there's no anomalies, we're going to get a good quality part at the end.

Brad Cox (06:52)
So, it's like the training people may have heard about when referring to things like AI and ML?

Conrad Kao (06:56)
Absolutely.

Brad Cox (06:57)
So, I'd like to talk a little bit more about that machine learning aspect. We had a recent episode about robotics, and we discussed how the growing availability of processing power and the AI/ML boom is really supporting some big advances. So, Michael, how are these advances overlapping the additive manufacturing world?

Michael Kei (07:14)
One of the biggest differences between additive manufacturing and traditional methods is that AM is fundamentally digital. That means that we are not just printing parts, it's about enabling digital inputs, outputs, real-time data capture, feedback loops and advanced controls that traditional manufacturing simply doesn't offer. 

Where it gets exciting is with the advancement of sensors and data collection technology, that has improved dramatically, giving us incredible visibility into the build process. So, you combine that with machine learning, the impact is huge. AI-driven insights are helping optimize processes, detect defects in real-time, and even predict part performance before printing. So, this is critical for industry like energy and maritime, where certification and reliability is non-negotiable.

The way ABS sees it is that machine learning can actually contribute to additive manufacturing in maybe four ways. One, process monitoring and control. So, it detects defects on the fly, actually while you're producing the parts. Predicting quality assurance. So, it forecasts strengths without the need for extensive physical testing. Validation and certification using digital twins and automated inspection to speed up approval. And of course, design optimization where the machine learning and potentially even generative design that can give you more optimized designs.

Brad Cox (08:31)
And Conrad, how do you see all this data processing capability impacting additive manufacturing? I know we've talked a little bit specifically about the validation aspect, but what about the big picture?

Conrad Kao (08:41)
I think the bigger picture is that it's going to help increase adoption. We're already seeing an increase in adoption across multiple industries for additive just due to the education of designing for the manufacturing method. One of the restrictions now is trusting that we're going to have a quality part. The more advances we can make in reducing lead time and cost of validating components, the more likely engineers are going to use this process. I see it as a key way to mass adoption for additive manufacturing.

Brad Cox (09:08)
For the sake of our listeners and just kind of setting some, I guess, guardrails in a sense, where a lot of people will hear AI, machine learning, LLM, all this stuff, and just think it's all just one thing, and it's really not. Can one of you provide some, to use the word again, guardrails as to what we're talking about here when we are talking about using this data-driven approach for additive manufacturing?

Conrad Kao (09:29)
I don't claim to be an expert on either AI nor ML, but my simple understanding of it is machine learning is much more large data set focused for a specific use case. And artificial intelligence is also a large data set focus, but using all of that data to predict future outcomes. Whereas, the ML approach is taking that data set and finding the discrepancies or anomalies within that data set. Correct me if I'm wrong, Michael.

Michael Kei (09:54)
Yeah, I think machine learning is a subset of artificial intelligence. What machine learning does is to get a large set of data and to figure out what the norm is. So the predictive part comes from what has happened in the past. Whereas, from a larger scale, AI can actually be predictive. It can generate or create something out of, I would say nothing.

From a certification point of view, AI would be slightly more challenging to adopt at the moment, where machine learning is basically collecting what historically we have and the experience we have, and then have the machine go through whatever that we have picked up during production and make a judgment based on that. So that would be a lot easier in terms of building confidence rather than going right into AI where it starts to make estimation or prediction on its own without having actually the guardrails that you mentioned.

Conrad Kao (10:45)
For me, it's taking all this existing data. We're literally talking about gigabytes and gigabytes of data per build, per part, of information regarding that entire process. I think using that as kind of its big data center or big data hub and using the ML approach can kind of characterize that specific process, right? And additive to me is, I guess, much more of a predictive thing like if we have a new material, what are the process parameters going to look like and then will that actually predict a good quality component at the end of the day?

Michael Kei (11:16)
I think from a design point of view, we can look into AI, which is basically generative design. From the AI part, they can definitely create new designs. Also start to understand, help to rapid prototype new materials with new designs. But when we are looking at certification, I think AI is a little bit more into the future of being predictive. What we want to rely on right now is the huge data set that Conrad mentioned and to actually use ML to figure out what the norm is and to use machine learning to be able to pick up outliers very, very quickly without human interference. So, that will create a much faster certification process. It also allows the industry to build confidence as we go along because confidence needs time. So, if you start throwing in too much advanced technology, it's not going to work. But from the design part, yes, we can start talking about AI, not on the certification part.

Conrad Kao (12:07)
Yeah, I think just to follow up, I think it's less of a leap to go from current validation methods to an ML method where we're looking at all this data and finding outliers. That's not as drastic of a leap of faith as it is to go to some sort of AI predictive validation. To Michael's point, that first small step is as much easier to gain trust in the general public. And then we build from there. You don't want to jump off the cliff right off the bat.

Michael Kei (12:33)
And maybe one of, one of the things that I think we have not touched on, which we want to use in situ certification is that a lot of test coupons actually end up in the wrong position. So, you're not picking up real critical points. You're just trying to do a checkbox. I used the words checking the box. I think sample testing could actually fall into checking the box because class says 15 coupons. You give me 15 coupons, but are they at the critical locations? We don't know. Are they picking up potential defects? We don't know. But in situ printing will allow you to do that.

Conrad Kao (13:05)
Yeah, I mean, that's a great point. And I don't know if this is too far into the weeds, but the witness coupons we print along with every component obviously don't occupy the same volume that your printed component does. So really this is just a representative sample of what that build job is seeing. But to Michael's point, the actual part is located in one XY coordinate. The samples are located in another XY coordinate. Yes, we're testing that because it's being built with the same process parameters. But at the end of the day, are these two actually identical? That's, you know, you could make an argument that they're not. But in situ monitoring does allow you to say, okay, what is going on and my specific part during that process. So I think having that data profile would be key to ML qualifying parts in the future.

Michael Kei (13:46)
It actually enhances safety rather than actually reduce it. Right now, it's just about building confidence on the database that we have.

Brad Cox (13:53)
It's the difference between taking one out of a batch or knowing what every single unit looks like.

Conrad Kao (13:59)
Yeah, exactly what Michael's saying, this in situ data per part, you would literally be inspecting every single component as opposed to taking 10 out of 100.

Brad Cox (14:08)
When we talk about robots and autonomous systems, people always want to know what kind of human oversight there is. Michael, what sort of oversight is needed for this data-driven certification? And, so, I know I mentioned this is the buzzword, but how important is adding that extra layer of trust?

Michael Kei (14:23)
Yeah, I think when everybody talks about data-driven manufacturing, it's always how much human oversight. Is it no human or there will be human-in-the-loop and to what extent? So actually, the answer to question right now is it's still going to be quite a lot. So what is happening is that in industry like marine and offshore, a single failure can be catastrophic. So, humans will provide an additional layer of judgment, accountability beyond what algorithms can actually deliver.

In fact, most regulatory bodies still require documented human sign-off for certification. And why is it so? Because machine learning models aren't perfect, at least not at the moment. So, they can misinterpret sensor data or fail under unusual conditions. So human oversights help ensure decisions are made and aligned with safety and compliance standards. It also builds confidence. As companies adopt new technology like additive manufacturing, oversights will help to mitigate risk and data biasness, sensor errors and even cyber threats. 

So, what does oversight look like? Maybe just process validation, quality assurance certification and continuous improvement as well because models need to continuously improve and human-in-the-loop will definitely assist.

Brad Cox (15:32)
And Conrad, how do you see that human-in-the-loop role as data-driven validation gets to be more common?

Conrad Kao (15:37)
Yeah, absolutely. I think, especially in the short-term, near mid-term future, there's no substitute for human oversight. We want the community to trust in this process of validation. And I don't think anybody trusts a robot to make those decisions quite yet. But I think that long-term, it could be a reality someday. At the end of the day, right now, an algorithm is not going to make a decision on whether something's safe or not.

Brad Cox (15:58)
In a lot of our episodes we talk about regulations, whether that's safety or environmental, and how they impact technology in the industry. I think additive manufacturing lives in a little different space. So, Michael, what sort of standards or requirements are there for AM and what's needed to demonstrate that reliability?

Michael Kei (16:14)
So right now, ABS has several key publications. We have requirements for AM technology, which provides formal requirements for standard qualifications. We are also working on new guidance notes that is basically working on model-based qualification. These aren't just theory, they are actually backed by real projects that ABS has been involved with. For example, ABS just completed a project with NIST to speed up qualification for metal-based AM.

Instead of relying solely on extensive physical testing, we introduced model-based approaches like expert systems, in situ monitoring with machine learning and physics-based simulation. The goal? Faster, cost-effective certification without compromising safety. Building on that, ABS is also leading a maritime-focused project to create a qualification framework for data-driven defect detection.

Another initiative under NSRP is developing best practices for what we call rapid qualification, delivering guidance, case studies and methods to reduce testing and low certification costs. And right now, IACS Recommendation 186 is the first global framework for AM in marine and offshore settings. It applies to metallic additive manufactured parts used in marine and offshore equipment, focused on powder bed fusion, direct energy deposition and binder jetting technology. 

One of the big challenge right now is a lot of all these standards, and as what Conrad has mentioned, a lot of certification focus on strength. Moving ahead, ABS is also looking for partnership and collaboration with industry to work on endurance and fatigue life. That will complete the loop of certification of AM parts.

Brad Cox (17:48)
And Conrad, what about your thoughts on the role of standards in driving that adoption of not just AM, but also the data-driven certification?

Conrad Kao (17:55)
Yeah, absolutely. Right now, we kind of have a landscape where every industry has their own body of standards or certifications that they abide by. You know, API 20 is for the oil and gas, aerospace has their framework, dental and health have a separate set of standards. It makes me excited to hear all the opportunities that Michael spoke about that's already going on and opportunities for collaboration in the future on additional updates to the standards as well. 

I think, moving forward, if there was some sort of universal cross-industry specification that everyone could reference, that would help drive adoption even further. I'm much more of pragmatic person. I don't think that's going to happen anytime soon, but it's exciting to hear the activity going on behind the scenes.

Brad Cox (18:33)
Closing things out, I wanted to give each of you an opportunity to provide some closing thoughts for our listeners. Conrad, you want to take that one first?

Conrad Kao (18:39)
Yeah, absolutely. Like I said, I started in additive about 10 years ago. It's come a tremendous way since then. First generation of machines were much more manual. Now there's so much data being collected. The adoption of the design for additive practices has really taken off. We’re seeing so many good applications in all industries really. And I think the next step is really trying to converge on a data-driven validation model. Once we get to that point, that really reduces lead time, cost of testing, physical destructive testing that's required currently, and a lot more industries I think will take notice and start to implement more of the additive parts in their assemblies.

Brad Cox (19:15)
And Michael, any closing thoughts?

Michael Kei (19:17)
Yeah, for us additive manufacturing definitely seems to be the game changer for the future of manufacturing. It brings huge benefits like rapid prototyping, cost-effective designs and even using less materials. But here's the challenge, which is certification. So, building trust and bringing trust into the final product is critical. And right now, that's the biggest hurdle to widespread adoption. 

The exciting part? Data-driven certification powered by machine learning could be the key to unlocking AM as a mainstream process. So, at ABS, we are leading the charge, partnering with industry, investing in R&D to prepare the marine and offshore sectors for the future where additive manufacturing is not just an option, it becomes the norm.

Brad Cox (19:57)
All right, great. So, machine learning, AI, all those things are certainly buzzy topics, and people in different spheres are exploring how they can impact what they're doing. And that includes additive manufacturing. It really sounds like there's a lot of exciting stuff happening in that space right now. So, Conrad, Michael, thank you both for a really insightful discussion.

Conrad Kao (20:14)
Thanks having me.

Michael Kei (20:15)
My pleasure.

Brad Cox (20:16)
And for everybody listening, thank you for joining us for another episode of Setting Course. Be sure to subscribe, leave a review, and share this episode. To learn more about how ABS is supporting additive manufacturing, visit us at www.eagle.org. Thank you for listening.