Diego Braido's Wooden AI: Transforming the Lumber Industry

‍

Diego is a forest engineer and holds a master’s degree in wood sciences. He has spent the last ten years working at BID Group, where he currently holds the position of Director of the Applied Artificial Intelligence team.

Diego and his team are dedicated to improving existing optimization systems through the implementation of artificial intelligence, as well as developing new and innovative products that can expand the applications of AI in the lumber industry.

‍

Host:

Hey, how are you?

Diego Braido:
Hey, how’s it going?

Host:
Pretty good. How’s your day going so far?

Diego Braido:
Pretty awesome.

Host:
You think it’s a little early for a podcast?

Diego Braido:
No, it’s just right. I usually wake up at 7 AM.

Host:
7 AM? Yeah, sorry. It’s not bad. Well, I’m going to introduce you here. This is Diego Braido, engineer, PMP, and he's currently the director of Applied Artificial Intelligence at the ID Group. And I think, at least today, I think this is going to be a really interesting conversation, because when he said he applies AI, you’ll be very shocked about what that means, and that’s one of the reasons I wanted to name him today. Thank you for joining us.

Diego Braido:
Yeah, thanks for inviting me.

Host:
So, one quick question: how did you get into AI in the first place? Like, why did you choose this career path?

Diego Braido:
My background, I'm a forest engineer. I graduated 12 years ago. Then I started to go into the academic path. I did my master’s degree, and I switched to go into industrial processes and transformation. I started working for Beat Group Technologies, and we’re the leading suppliers of integrated solutions for sawmillers—people that buy logs in the forest and transform them to cut them and get the boards that are used for construction. And ever since I started working for Beat, we were already using some pretty advanced technology to grade the lumber, analyze some visual and geometric aspects, but we weren’t using AI deep learning at the beginning. So, I started doing this field work for this company or working on machine learning, which is a subset of AI. Then, since 2018, we switched to AI deep learning, and that’s where magical things started happening.

Host:
Before you even get into that... why AI? Why study that at a high level? I’m assuming even before you got into it, everyone was... And that may not be true, but everyone... What are you doing exactly?

Diego Braido:
My initial choice to solve academic problems was a little bit random. I just liked forests and would like to understand how to plant and grow trees. For three years, I was concentrating on plantations—how to protect the environment, how to protect someone. So, it was a natural initiative.

Host:
You were very interested in protecting the environment and how that works, so that’s how you got into the doctorate?

Diego Braido:
Yeah, okay. All right. Then I saw there’s a lot of potential to modernize this industry, because it’s one of the oldest industries in the world, right? We’ve been using wood for construction for a long time. And what really attracts me is this potential and opportunities to integrate new technology and optimize it. Not too long ago, people were still just cutting logs as they wanted—humans were taking decisions based on their knowledge, but they’d always go for the safe side. So, there was a lot of waste, and the clients started realizing, “Hey, this commodity is increasing in price, so we need to implement new technologies. We need to go beyond just human-based solutions.” So, we need to use technology to help us out to extract the maximum output, minimizing the waste, to be competitive. People were popping up a lot of mills, creating new sawmills processes, and they were competing with each other, so they had to come up with strategies to get the highest quality product and dominate the market. So, then I saw this potential. I saw this company that was doing these pretty amazing scanners with cutting-edge technology, and I always liked technology itself. Then I discovered this whole world of programming optimization, and then eventually machine learning and AI, and it’s just so amazing. So one thing led to another and this is where I am here right now.

Host: 

He was saying it was supernatural.

Diego Braido: 

Supernatural, yeah.

Host: 

Okay, so um, optimization of the logging industry was a big push for you once you got it. You found that there was an opportunity there.

Diego Braido: 

Yeah, so then what were some of the inefficiencies that were very obvious to you initially that you could speak about when you first got in? It's like, this clearly could be improved.

Host: 

Hopefully that gives the audience a map of the industry a little bit, you know what I mean?

Diego Braido: 

Yeah, exactly. Just like the automobile industry, everything was done by humans. It's a production chain; you put a labor in every step of the process. And to get from log to boards, there are about 20 processes. People kind of don't know how many processes are involved in breaking out the logs, removing the bark, cutting to smaller logs, removing the boards from the logs, cutting them, grading, then drying, and then packing and selling. All this process was all done by humans. No advanced computer or machinery was available until we started—and the other competitors as well.

And then, even then, we were putting very modest scanners in the beginning. As the demand was growing, the clients were like, "Okay, we need to do better, we need to do better, we need to do better." This constant demand to extract the highest quality and minimize the waste.

And then when you look at the potential, it's still a cheap material. Not today, for instance, the wood prices dropped a lot, but during COVID, it hit a high that people had never seen before. So it is a material that will always be used, and we're having some shortage, especially in Canada, with wood fires and all the protected areas. We need to pay attention to it and the climate. The trees, they don't grow as fast as in South America, for instance, where you plant and in 20 years they're already 20 feet high and 15 inches in diameter. Here, you take care of the forest and it takes a long time to grow and be ready to be extracted and exploited.

So we need to take care of this resource, and as Canadians, we need to be proud of the forest. And this is the part that kind of saddens me the most, is I feel like this industry still needs more love. And now people are finally realizing, it's time to invest, it's time to make sure we maintain the sustainable business, and it's not too late, but people need to take actions. A lot of companies that I've visited, I've seen some people, they're like, they just want to push the numbers, get the most they can get and that's it. Some started looking, okay, can we do better, spending less money, less resources, and can we be competitive?

That's where the computer-based solutions came up to be very handy because you replace a human that sometimes is tired on a Monday morning or a Friday night. And this human is essentially taking the decisions, it's the human who's deciding each single piece in every process, what it should be done by the machine. So you push buttons to control the mechanics. So say a human's not feeling well or tired or something, it's not going to take the best out of it. So there's a lot of potential to not replace these humans by machines, but install machines so the human can better work with machines—just the system.

And this whole change process is fascinating because you need to convince people with their old-school mentality. They've been doing this work for 20, 25 years, some 41 years until retirement. And now you're coming and saying, "Hey, it's not you who's going to take the season anymore, it's a scanner using computers and artificial intelligence. You're just going to assist it and make sure it's doing its proper job." So people are like, "What? I don't trust it. It's never going to be as good as me."

Yeah, and it ended up being much better than the humans because, like I say, it can work 24 hours every day, right? It doesn't think about its girlfriend or going out on a Friday night. It's constant, with accuracy.

Host: 

Nice, so you spoke about the scanners as being sort of a huge advantage in the process. So the assumption is the scanner is doing something a person would have been doing—scanning for something a person would have been scanning for. So can we talk about what that is and how it's able to do a better job than a human?

Diego Braido: 

Yeah, so when we say the word optimize, we're trying to find the best solution.

Diego Braido: 

So for every single piece that will be scanned in our scanners and we need to take a decision, so start with a log. A log is never perfect, like a conic, there's always some defects, it goes like deformation and stuff. So you need to scan, find the best solution, close to 1/8 of an inch to cut it out, remove the defects, find the optimal solution. A human cannot make all those calculations in a matter of 50 milliseconds, but it's scanner scanning because we have all the data. We map the whole geometry of the piece, then we have recent cameras that will scan and we can detect where all the defects are that we can see on the image. So we size them perfectly, and we measure, and we can take all the trials to find the best solution possible. A human could do it, but it would take a really long time to do it. And, you know, in production teams, you don’t have much time. You need to get this here, and there’s a few items coming. You need to take a decision so you barely have like three seconds. That was the time a human would grade, say, boards like two by fours would take. They have five grades, and each grade has some characteristics that must respect the amount of defects, how pretty the lumber is, etc. So the human would need to remember all these rules: What makes a number one, oh is this, this, and that. What makes a number two, and you have to calculate every single piece, and in one shift, it could grade up to 50,000 boards. So do you think someone can be concentrated and take decisions 50,000 times with the same accuracy? For sure, sometimes they move tired, but they'd be like, "Oh yeah, I can't do this job anymore, just do whatever," because there’s so many pieces coming back. Yeah, it’s fine. It’s just one board, I’m gonna reject or not do the best job, right? But I scan it.
Host: 

Yeah, yeah, we'll do a constant, reliable job and it's always better than a human decision, right?
Diego Braido: 

So, and this is... is this integrated into the bigger machinery, or is there someone who is standing over the boards as they go through scanning them?
Host: 

No, it's static scanners, and you have chains or conveyors that will push logs or the boards or other parts of the logs that we cut down, and we'll be scanned. It will be controlled in a controlled environment, so there's all the PLC and controls and make sure the pieces are aligned, touching the line bar. We call it the reference point, so we always scan the same way in a controlled environment.
Diego Braido: 

So how much improvement did this sort of bring into the process? Like how much, um, how much of a removal of errors did the scanners introduce?
Host: 

Well, you know, it all depends on the client’s situation before they buy or he buys the scanners, but just say like we can improve just 60 percent of the productivity. If they go from just labor, no machines at all, you put all scanners, you're gonna see a tremendous increase in modules. Uh, yeah, you guys are throwing the errors because you can make errors, you can hurt people down the way, because these numbers, if you're not grading them properly, you go to Home Depot or Rona and you buy these numbers and go build your house with. But there's a breakage in the lumber, for instance, or something, and your house gets damaged, the sawmiller where you bought the lumber from. His job is to make sure he delivers a product that is safe for people to buy, build houses, or do other stuff with. So this job must be controlled.

So we have agencies that go visit the clients to do inspections to control the quality. So it's a very controlled environment in the wood transformation, and this is very important to be grading properly and not losing too much value. With the implementation of automation in wood processing, we ensure the accuracy and consistency of the grading process. People always play on the safe side because they're scared of making mistakes and getting people suing, like, "Ah, about your lumber and it wasn’t the proper grade, you did a bad job, so I’m gonna sue you." This can be mitigated by ensuring proper efficiency in the lumber industry through automation.

By using AI-driven inventory management, companies can track lumber stocks, helping prevent mistakes and optimize storage and distribution. Additionally, focusing on AI and sustainable forestry ensures that the logging and wood processing processes align with environmentally friendly practices. Finally, adopting predictive maintenance in the lumber industry will prevent costly machine breakdowns, reducing downtime and improving overall productivity.

Diego Braido: 

There’s this legal aspect as well, but when you play on the safe side, you tend to overestimate. So that’s where the scanners come to give an excellent ROI, because you make a lot more money, but you're not making mistakes.
Host: 

Yeah, I start being safe and you're being accurate.
Diego Braido: 

Nice. 

Host: 

With all this being said, how do you, you know, sell this to an industry of this age? Because you talked about introducing new technology, especially AI, which a lot of people are getting a grasp of, even with newer industries. So how do you introduce that kind of technology to an industry that is truly one of the oldest industries that you could have industrialized and has been under tons, like hundreds of years of industrialization?

Diego Braido: 

Exactly, so there's a lot of trepidation when it comes to what machines are going to represent in that industry. So how do you have that conversation? Obviously, there's an improvement, but like anything else, before you get to the person who's gonna make that decision, there are quite a few conversations you have to have before that. Yeah, it just sounds tough. The numbers sound self-evident, but it's almost like it needs to be that in order to even get any sort of headway in that industry.

Host: 

Yeah, they have no choice. Even the clients that were small, small sawmillers, they didn't have much cash flow. They got to a point where either they act and buy scanners or they die and shut down, because the competition was so strong around them. They were killing them because they were not making enough profits to pay the whole process and continue to buy logs. Because what is the commodity, right? So it keeps the commodity. Yeah, it varies like the stock market. So yeah, high, you gotta buy the logs high, but you're gonna sell the product high, so your profit margin will shrink, and then the labor salaries, etc.

Diego Braido: 

So you need to add. I know it. So people ended up saying, "Yeah, you know what? I have to buy these scanners, even though I don't trust them. I buy it." And once we install, they're like, "Oh, damn, I have been passing by this opportunity until now because I had this preconception that a machine would never be as good as a person." But they ended up discovering it's way better than like, "Wow!"

Host: 

Is the competition coming? Is it because competition is coming internationally?

Diego Braido: 

Yeah, we do have some good competitors within Canada and the United States, but in Europe, they have their own market. It's not the same thing. We're also having clients all across North America and a few in Europe, but that market is different. It's mostly within our North America. There's a growing market right now. There are a lot of sawmillers all spread across the globe, and during the COVID pandemic, a lot of new people decided to buy new mills and keep growing and increasing the number of sites. So yeah, it's a good opportunity, a good market.

Host: 

Okay, let's talk a little bit about the technology, the actual technology that's powering all this. So I guess a simple question is, you know, what are you using? Your source technology? Do you guys have something very specific within your ecosystem that enables you to actually do this? So where do you land on that?

Diego Braido: 

So yeah, it was an evolution process, but we have really good programmers, people who are highly skilled in the programming field. So we kind of took something that was in the open library and we completely adopted this ecosystem of AI neural networks to do our applications, because we don't do like ChatGPT. That's not the kind of AI we do. Our AI is to detect defects on the image or to detect some issues that a human can see by looking at an image. So this is the AI that we deploy. We analyze vision images to identify where the things that we're concerned about are, so we can properly flag them. And then, with our computer algorithm, we go calculate how big it is, how widely spread, etc. So then we can get that input and optimize the solution for each single log or board, depending on the part of the process.

Host: 

Alright, so is that TensorFlow? Is that PyTorch? What's your stack?

Diego Braido: 

Well, more... I cannot disclose exactly what we use, but it's kind of mixed from what you can download and use as open source. But we had to adapt because, like I said, this is a very niche application of AI, right? There's not many people out there doing this, so it's really hard to find this kind of stuff available to download and quality and use it right. So we actually kind of base ourselves from something that exists to learn how it works and then we put a lot of efforts to how we can do this on our applications and specialize this technology to do what we need to do right.

Yeah, it wasn't an evolution process, but we have really good programmers, people that are highly skilled in the programming field, so we kind of took something that was in the open library and we completely adopted this ecosystem of AI neural network to do our applications because we don't do like ChatGPT; that's not the kind of AI we do. Our AI is used for automation in wood processing to detect defects on the image or to detect some issues that a human can see by looking at an image. So this is the AI that we deploy. We analyze vision images to identify where the things that we're concerned with are, so we can properly flag them and then with our computer algorithm, we go calculate how big it is, how widely spread, etc.

This allows us to ensure the efficiency in the lumber industry. With this input, we can optimize the solution for each single log or board where the grainy pattern is, depending on the part of the process. And through AI-driven inventory management, we make sure that every log is processed with maximum precision and minimal waste. By leveraging AI and sustainable forestry, we are not only improving productivity but also aligning with environmental goals. Finally, predictive maintenance in the lumber industry ensures that our machinery runs efficiently and downtime is minimized, which is critical for long-term sustainability.

‍

Host: 

All right, so is that uh tensorflow is that high torch interior stack?

Diego Braido: 

Well, more I cannot disclose exactly what we use uh but it's kind of it's kind of mixed from what you can download and you open resource but we have to adapt because like I said this is a very niche application of AI right there's not many people out there doing this so it's really hard to find this kind of stuff available to download and use it correctly.

Host: 

And how are you finding this is affecting the industry, this kind of move towards automation, especially with the robots coming into play?

Diego Braido: 

Yeah, that’s another part of the evolution we're seeing in this industry. Now people, they accepted they need to buy scanners so pretty much there's only a few handful clients without any scanners at all. I don’t think they're going to survive much longer unless they're in a specialty product where they're the only ones who make a kind of product. So now the next challenge is the labor shortage.

Host: 

Yeah, is there a reason they don’t want to buy the scanners if the scanners are well done, also the price right? It’s not like a thousand dollar scanner you can go to a meeting to meet him, depending on.

Diego Braido: 

Yeah, it's quite expensive because the ROI is so fast and you're gonna cut maybe three or four salaries when you buy those scanners per year so you still sum them up with all the social benefits, you get to pay back less than a year depending on the application. But then now in Canada, we're seeing this labor shortage issue where we need labor. We're hiring people from overseas, we're trying to get more employees, but it's really hard to find people that are willing to move out from the big cities to go to the isolated remote areas where the sawmills are located. They're not in downtown, right? So you have the challenge of there's not many people in Canada so our labor is short. Then the field labor that is existing, they never want to go live in up north or in a decent area unless they were born in those specific regions. So now some clients they used to run three shifts a day, so 20, 23 hours or 24 hours with one hour break probably. Now some they have to completely stop one shift because they didn’t have enough operators to run the mill. So this is really, really hurting them so bad because you don’t make anything during the whole shift because you don’t have enough labor. So now they're like wow we cannot find anyone to come, and the people they hire, they stay there for like two weeks and they just quit because they don’t like the job. It's monotonous, there's some risks of injuries because you're constantly manipulating heavy stuff close to machinery so you could get hurt, right? And young people they have different mentality, right? They want to work with computers, they want to be high tech, that's awesome. No one wants to go, they have to be up north in Quebec or in the remote areas. Extreme West Coast BC, it’s pretty places, but people don’t want to work in those areas sometimes.

Host: 

And now the sawmillers, they're like we need to start looking for robots because if we cannot fulfill that position with the human, we're not going to be able to run.

Diego Braido: 

So they need to pay salaries, we have a business to keep alive, so they have to now buy robots. That's why we started these initiatives offering robotics optimizers within the scene. So we use AI in the lumber industry to detect when there's a problem, say you're running out of boards, they're supposed to be straight, there's one board that is skewed, so it's going to cause a jam because it's going to block somewhere down the road. So AI in the lumber industry will detect there is a problem and the robot, okay got it, I’m gonna remove it and go pick it up, remove any container, run instead of just asking for a guy to climb, remove the board, there’s a risk of being very safety aspects, etc.

This is where automation in wood processing comes into play. It allows us to minimize human error and speed up the process, all while maintaining safety and precision. The adoption of robotics enhances efficiency in the lumber industry, reducing delays caused by manual intervention and increasing throughput. Moreover, AI-driven inventory management helps track the inventory, ensuring that we can predict shortages or excess stock, making the process even more seamless.

As we continue integrating more AI and sustainable forestry practices, we ensure that the technology used not only improves the operational side of the business but also aligns with environmentally friendly practices. Finally, the implementation of predictive maintenance in the lumber industry allows us to anticipate equipment issues before they arise, reducing downtime and preventing costly repairs.

‍

Host: 

I know Tesla is for you know manufacturing engineering, Tesla is doing or using quite a few trainable robotic machines to not necessarily do what you talked about but definitely do quite a few things.

Diego Braido: 

Yeah, in their factory.

Host: 

I don’t know whether Bid you become who you're in is in that space as well and in terms of making those or at all or it just is something that you're in you know that is necessary and it's something that you're looking into so which is that, is that something you guys are gonna look into making robots objectively? So we're purchasing the robots, their design and manufactured by a third party, and we're just developing the code that will do the actions and then the communications with our scanners.

Diego Braido: 

Yeah, right.

Host: 

So can we talk—I mean, I know you can talk about the specific technology—but sort of talk about, uh, maybe architecture or something, so that the audience can learn about, you know, if they ever got into this industry, what it would mean to write for or code for some of these solutions? Because I think a lot of people want to hear Robotics and AI—it's, you know, so there's a certain kind of person that's very excited to be able to play with both those things. So it'd be great to hear sort of one of the challenges and some of the, um, the ways in which architectures and things that you're building solutions, uh, or problems that you have, because you're trying to sort of merge these two things together.

Diego Braido: 

Yeah, exactly. So first things first, you need some highly skilled programmers because they're the core of this technology, right? You need to get someone or a crew that knows how it works—machine learning, AI, deep neural networks. It's not that complicated because, like I said, you can download some open-source stuff, and you and me, we can start doing some AI on our computer just by downloading very light software. You just need to understand the basics of programming. Some people do is fighting the one of the symbols programming language, other, and it's the best one for AI. And I know Google, Microsoft Azure, they just launched their machine learning portal, so you kind of, if you don't have time to do that or don't have the resources, you can also go see people that have this structure, and you rent their machine and you just upload your data, and it's going to analyze for you for at least testing purposes.

Host: 

Right.

Diego Braido: 

And then you need to have this whole server structure with all the computers, the GPU graphic processing units—cards—that's like the fuel for AI. This is what's going to give you the power to crunch the data, build their models, and then to run live predictions because you need first to train, and that's the biggest challenge is the data.

Host: 

So did you, for this particular situation, were you at a factory capturing information? Yeah, and then using that to filter the real maps, is that setup?

Diego Braido: 

Yeah, exactly. Because another thing is, because of sort of the nature of this, you can—the continuous improvement cycle can be sort of infinite. Because as it's training, it can—as it's working that training, it can continue. You can have a camera there that consistently just takes pictures and you can error check against the AI itself, and it can just sort of train itself and infinitely improve.

Host: 

So which is not just sort of more of an idea, but for how you started it initially, was that how that happened? You just took pictures, you went to Amelia, asked them, “Hey, we're going to take pictures of all your boards,” and just like…

Diego Braido: 

So the animation over, because our scanners that we have been selling since 2008, this is one of our biggest products that we sell. There is the auto grader, which is the machine that will grade the boards and gives the perfect optimal solution for every single two by four, two by six, etc., that is scanned. So we already have all these images because we were using conventional programming to detect all the wood defects without AI. It was really hard to achieve excellent accuracy. We were doing pretty good, yeah, but there was some room for improvement.

Host: 

Sure.

Diego Braido: 

So that would leverage our successes. We already have this whole ecosystem with all the optimization scanners, the vision acquisition image. So we had all this huge data set available for us. We just had to say, "Okay, how can we take this image? We need to build an in-house program to train the AI, show those. Okay, in this image, I need you to identify this knot or this opening in the wood." So we would identify and then need to build this architecture to train the model with these images that were tagged by him.

Host: 

No, absolutely. So what is—I'm actually now interested in—what was the solution before AI? Like, oh man, if you agree with it, it was really complex, thousands of lines of code, thousands.

Diego Braido: 

Because, yeah, no, yeah. Because we have a line. So, it's not like a snap camera. You collect every line, there's RGB colors, and you merge them, and you have the full image. So, it's really good resolution that we have. You can clearly see the piece of lumber, you can see all the natural defects that are present, and you, as a human grader, if you've gone through training, you would be able to grade that lumber most likely just by looking at the image. Sure, you need to get your measuring tape and measure stuff, but you could look and say, "Okay, that board, it's a number one or a number two because that defect seems to be too big," etc. So, you have a pretty good image of what you need to do.

This is where automation in wood processing comes into play. By using advanced technology, we can automate the grading process, reducing human error and increasing speed. It directly contributes to efficiency in the lumber industry, making the whole operation more streamlined and less prone to mistakes. Thanks to AI-driven inventory management, we can now track lumber batches in real-time, ensuring better control over stock levels and preventing any shortages or excess inventory.

As we continue to innovate, AI and sustainable forestry become increasingly important. The technology can help us assess not just the quality of the wood, but also its source, ensuring that we're using materials that align with environmentally friendly practices. Additionally, the application of predictive maintenance in the lumber industry enables us to foresee potential issues with equipment, allowing us to perform maintenance proactively and minimize downtime.

‍

Host: 

And what is the number one and number two defect, is that?

Diego Braido: 

Oh, number one is a number one grade, so there's a set of criteria that says you cannot have oversized more than half an inch. I'm just seeing numbers or, uh, yeah, yeah, it cannot be more than a third of the length. There's a bunch of rules because, right, the people that came up with those rules, they ran a lot of testing to test, "Okay, amongst all these wood natural defects, how far can we go until the board is not safe anymore?" And if you put a weight on it, you put load on it, it's kind of stressed. You're gonna stress rip it off with the fibers and break it down. So, they had to come up with rules saying, "Okay, until you can go up to here to be acceptable in that grade of lumber." Then you go to number two, and if you exceed that parameter or that kind of allowance, you go to number three. As you go down, the quality deteriorates, and then it's just for fancy or building trails, railways. You cannot build houses anymore. It's just number one, number two, and what they call the MSRS is the machine structure. You kind of measure the MoE, the modulus of elasticity of the lumber, how much you can bend until it comes back up. AI in the lumber industry has started to make a significant impact on how these measurements and testing are conducted. With the help of AI in the lumber industry, automation tools can now predict defects and the overall quality of wood, making it easier to classify the lumber and speed up the process.

Host: 

Right. So, there's a bunch of calculations you need to do. But you could tell that just by looking at it, you can tell the MoE by looking at the limits?

Diego Braido: 

No, we have a machine, a little hammer that will hit the lumber, and by resonance, we're gonna measure how fast it goes. So, you can kind of deduce what is the MoE, the modulus of elasticity.

Host: 

Okay, so it's a coefficient though that you measure to see if this number is strong enough to build a house or it's not strong enough.

Diego Braido: 

Okay, you going back, can your scanner detect that somehow?

Host: 

Uh, no. You need a sensor that will measure.

Diego Braido: 

You definitely need the sensor.

Host: 

Okay, got it. So, going back to the original way you guys were working on it, you would take these RGB colors where you could actually, to some degree, see what you need to actually grade it. There's other things in terms of like physical properties that are tough, but in terms of defects specifically, yeah, at least you can say it's at the upper-lower limit based on the images that you are collecting.

Diego Braido: 

Right, so how would the machine be able to tell, "Hey, this is..." Is it testing against what you consider to be perfect, and then it's just all these categories of things? You're just checking for if they exist or not computationally, just by checking where the image, can you find something?

Host: 

Yeah, so just marching down, marching down color, marching down the image and looking for defects, basically?

Diego Braido: 

Exactly. For every pixel in the image, we would have its intensity for red colors, green and blue, and the average intensity. So, each pixel had a value. So, we were trying by a bunch of parameters to say, "Okay, a knot, every single knot on the board is usually between 70 and 80 intensity, and the rest is sound wood." So, that's how we would flag these pixels to say, "Okay, that piece is most likely a knot." So, we would have a cluster. Then we'd do a lot of processing to kind of box it out without AI, and it was working good.

Host: 

The only issue was when some wood species—because there's a lot of different wood species—the sawmillers, they cut some do a lodgeable pine, all others, awesome fur, or hardwood for flooring. So, it depends on the appearance of the lumber.

Diego Braido: 

The color intensity, the challenge would increase because it wouldn't be that easy to separate perfectly, okay, these pixels are a defect, and these are not a defect. And that's where we realized we need something else than just trying constantly to change parameters.

Host: Right, because customers would call, "Hey, my lumber I got from a different forest, it has a different appearance. We need to adjust the scanners." So, it would be a lot of resources, a lot of time So, according to how we can adapt our code for this kind of image right now, we would change it and then with machine learning, we started seeing some really amazing gains. Like, okay, now we can kind of cluster the image into specific groups, and the system will try to do it on its own. But it wasn't perfect because it hadn't had the deep neural network yet. But when we put this last subset of AI, that's when the magic happened. Because all you had to do was train the system with a few images, and it would learn and reproduce what it learned, and it would be so accurate, it's just amazing.

So moving back to how this actually ties into the robotic system, how does that overlay? How would that work? Right? So, you even talked about where it was before, what it is now. Now, you're going past this deep neural network part to making it into something that is actionable by itself and actually, you know, not just reporting results. So how hard is it to get the robot to behave based on those images?

Diego Braido: 

So our strategy right now is the robot doesn't have any sensors. Our robots are only the arms or the equipment that will place something or pick something up or remove something. So we use cameras, fusion cameras, or a scanner that will give the robot the locations, the coordinates, X and Y, right? And then it will say, "Look, take that piece right there." So the robot is calibrated. He knows his intervention area, so he will get, "You told me to go to this location right here. Okay, I'll take it." So the robot doesn't do AI; the AI is done by something else that feeds the robot with the output.

Host: 

So you're just sending it location information?

Diego Braido: 

Yeah, that's pretty much it. It's just an arm. You're not giving it any sort of brain. It's just acting as an arm.

Host: 

Exactly. But they need each other to do the job, right?

Diego Braido: 

Yeah, absolutely.

Host: 

So, it sounds easy in the way you presented it, but are you having challenges in making that happen? You've done some truly amazing things so far, and it's very interesting to see how you've been able to really accelerate this solution to a point where the identification is no longer an issue. But I'm sure there's still more similar challenges, right? What are the challenges in getting that arm to do the right thing? Because you don't want it to pick up the wrong piece, and the pieces that are defective—that you sliced off, like you spoke about initially—those are very hard pieces to pick up because they're small. That was the goal, to make them smaller and smaller. So it has to get better as both systems improve. Or, I suppose, as the AI improves, the arm has to get better at picking up smaller pieces. And that kind of dexterity for a machine is difficult.

Diego Braido: 

Yeah, you know, adversarial behavior there in a way where one is bounded by the other. Absolutely, we have challenges all the way. And every time we encounter some new issue, we kind of like it because it pushes us to say, "Okay, we have now another issue. How can we fix it?" So now it forces us to understand the technology even better. Because when everything is easy, you're like, "Okay, I mastered this technology. I don't need to do anything else." But that's not always the case. There's always a challenge that shows up. "Oh, is the AI in the lumber industry going to be able to distinguish the situation from everything else?" We don't know. Let's try. So then we try, and wow, it managed to do it. It's cool.

And the robots, we just started. We have two systems installed, and we are still working on it. It's working good. They're still like you said, when we test stuff, we think, "Okay, we place it in our test bench. We have a little production line in our office." So we train the robots, and they'll pick up the skewed boards and stuff. It was easy because we were making them. But when you go into real production, all kinds of scenarios can pop up. It's never predictable. You can find one piece in a specific position you did not think would happen, and it did happen. So now you need, "Okay, now we need to train the AI in the lumber industry to see that and train the robot to go, like, how can you reach out to that piece and remove it without disturbing the others and making more mess?" It's just supposed to help it out. So there's always this challenge.

Host:

Like you said, with more data it's gonna get better with the more applications, the more testing, it will just improve, improve, improve. So it's always a growth, right?

Diego Braido: 

Exactly.

Host: 

And it also sounds like if it picks it up or moves something the wrong way, you can shift the board to the point where it can get stuck later down the assembly line. So it's not just being able to pick it up; like you said, shifting things around, you have to make sure that doesn't happen either. So it's sort of like, there's another angle to it where it can stop the production line, so that could make it really tough.

Diego Braido: 

Exactly, yeah. So you need to maximize the productivity time of the client and not create more downtime.

Host: 

Yeah, exactly. And the two issues: to make sure you detect the issues and you don't create something that doesn't exist.

Diego Braido: 

Exactly.

Host: 

It also sounds like your initial way that you were building, which was tweak for recognition, you kind of have to now do that for the robot.

Diego Braido: 

Yeah, it's gonna be all over again because it's a new technology. We need now to give not only the locations in two dimensions but also three dimensions because if there's one piece over each other, you need to be able to tell the robot, "Hey buddy, this is the piece that's above you, not the one below." So that's gonna push us to improve our capabilities. Now we need to use other sensors to help out, not only 2D cameras, right? We need to have the depth data. So that's where we're gonna have to, again, do some research. We're already doing it, moving forward with it, so it's going to push us to improve again because it's a new challenge.

Host: 

What do you say this industry is going to be in the next few years, given that you're sort of the one pushing it forward?

Diego Braido: 

Well, I see the future will be machinists connected to each other, sharing a kind of cloud system where one part of the process will say, "I'm making more than usual errors, or there are too many ways there might be something wrong with machine number three." So it will send a message saying, "Hey buddy, I think your saws are offset. Do a tweak." And automatically, it will readjust based on the output it's getting down the road. All of this, they'll all be using AI in the lumber industry to analyze and do their job, plus cameras, monitoring the whole area to identify production anomalies. The machinists will be communicating with each other, and there will probably be an extra operator required to confirm. Because you don't want to leave a meal fully automated, taking all decisions based on data. What if it's a bad decision? You could really screw up your production. However, AI in the lumber industry could significantly optimize processes by analyzing trends, predicting maintenance needs, and reducing downtime. With AI models continuously learning and improving, these technologies will enable more precise decisions in the future.

Host: 

Yeah, that's right. People are often scared that AI and robots are going to replace humans. In some industries, they will do it, and some are already doing it—there's no humans at all, it's all robots and computers. But we kind of need to look at the pros and cons, right? If your industry is not that well controlled, there's always stuff that can happen. You're better off having an operator that will assist the machinist. It's like, "Yeah, your decision is right," because sometimes you need to interpret what's happening. You need to take your discretion, analyze, and say, "Hey, this situation is not good, but I know if I change it now, it's going to cause even worse issues down the road." And that AI could learn too, but what if you make some mistakes and you're relying your whole production on machines that communicate, exchange information, and change parameters? If it's not all right, you're gonna screw up your production rate.

Diego Braido: 

Exactly. And especially in these remote areas that nobody wants to go to, so yeah, you really need to make sure it's as close to foolproof as you can.

Host: 

Yeah, absolutely.

Diego Braido: 

Sorry.

Host: But you know what, thank you so much for joining. I really do appreciate it. I learned quite a lot. Who would have thought that this was the newest put together? It's, you know... this is not something anybody would have imagined, right?

Diego Braido: 

I know, and it's so interesting to see that juxtaposition. And also, I can't even imagine the kind of conversations you're having with people, like the cultural clashes. Even though it's something that obviously improves things and pushes things forward, that sort of cultural clash and conversations and the outcomes you're always part of, and the kind of person you need to be to have those conversations must be very interesting in this industry. So thanks for sharing some of that with us.

Host: 

Yeah, we'll find you if they want to learn more or be part of this.

Diego Braido: 

Yeah, so they can always reach us on LinkedIn. If there are some potential people that see if we could help them with our technology, they can reach us out. We're always open to new projects. We have this niche in the industry, but now we have all this expertise and experience. We're looking forward to how we can leverage this technology and expertise to sell new applications outside of the laundry industry. So this is probably where we'll be going in the next five years.

Host: 

Well, thank you so much and I hope you enjoyed it.

Diego Braido: 

Yeah, thank you.

Host: 

It's done. It's done. How'd you like it?

Diego Braido: 

Yeah, it's really fun. It's fun.

Host: 

Yeah.

‍