[00:00:00] Speaker A: What is AI for me is a way to solve a problem that cannot be easily described with rules, simple logics. So anything that require complexity, AI would be a good option to consider.
[00:00:22] Speaker B: Hello everyone and welcome to the Robot Industry podcast. I'm Jim Beretta and I'm your host and I am excited to introduce you today to Young Yao from Soralink. Young Yao holds a PhD in electrical engineering from McGill University. She's a dynamic professional known for her creativity and ambition. In 2010 she co founded Tactile Labs, a nonprofit organization that advances haptics research. Her leadership has fostered innovation and collaboration.
After completing her PhD, Yun became the senior engineer designing wireless communication systems, excelling both technically and in leading innovation initiatives. In 2021 she she co founded Soralink to integrate AI and IoT into the food manufacturing and processing industry, aiming to apply cutting edge technologies to real world challenges. Yun's career reflects her commitment to pushing technological boundaries, fostering collaboration and driving transformational change across industries.
So welcome to the podcast Young. Thanks for joining me.
[00:01:28] Speaker A: Thank you very much for the invitation. Very glad to be here.
[00:01:31] Speaker B: Well, and we're kind of getting together a little bit at the ADM show in Toronto. So this is a bit of a preclude to that. I wanted to just ask you, why did you start the company Soralink 2021?
[00:01:42] Speaker A: I was thinking about career change. You know, it's the time to do something different and at the same time I saw the convergence of different technolog.
At the same time processing has become very efficient. Microprocessors and their processing very efficient in the energy wise as well as the wireless communication protocols. Now it's a lot more robust, it requires less energy as well as the AI. The boom of AI is now attracting many smart people to dedicate their time and effort in this. So all this together I see such a great moment to advance AI and smart sensors in manufacturing sector which I was part of for the past almost 20 years.
[00:02:27] Speaker B: Oh great. And can you tell us about Soralink? And I kind of guess you're an integrator. Is that what you call yourself sometimes like an AI integrator?
[00:02:34] Speaker A: I wouldn't call myself integrator because we do have a solution, a product.
So we provide a product that with sensors and the connectivity as well as the AI part analysis, part of the whole solution to our customers we are capable of integrating into the rest of the ecosystem such as ERPs, CMMs and other management system.
So that is also part of what we do. But primarily we have a Product and a solution that we sell.
[00:03:07] Speaker B: So we all have our own understanding of AI. But what's AI to you?
[00:03:13] Speaker A: AI for me is I took a course in AI when I was in university, which was a while ago and back then we were talking about neural network. And this was maybe 15, 20 years ago we talked about neural network. And back then we think the general feeling is that AI is like always the second best solution to anything you try to do. It wasn't good enough and it's kind of interesting, so kind of just keep it in the amount of options.
But today it has completely changed AI. It's by far in many cases the best solution for many problems, complicated problems, problem we're trying to solve. So just go back to your question. What is AI for me is a way to solve a problem that cannot be easily described with rules, a simple logics. So anything that require complexity, AI would be a good option to consider.
[00:04:10] Speaker B: So some people think that AI is just like ChatGPT. And can you explain the difference?
[00:04:16] Speaker A: ChatGPT is one of the best tool that pick can use regarding AI. It is interface. You can have conversation with can talk to the interface about many things. It's going to act like a very knowledgeable person with answer back.
But behind the interface there is a large language model which was just one part of AI. In the thing that we do, we deal, we don't necessary always need large language model. The LLMs we are working with telemetry data, meaning that the data comes in every second or every minute of every hour.
It's not exactly the same thing as what we are used to as ChatGPT. So there are many, many, many different ways we can build AI models and machine learning models which is not in the in the same area as ChatGPT.
[00:05:16] Speaker B: So what is, what is a model and why does it drift? Maybe you should explain drift first, I guess.
[00:05:22] Speaker A: Okay, well, I'll start with the model. What is a model? AI model is a routine, a function that could take complex input and produce output in many different ways.
Once you a model needs to be trained with data so it can behave the way you want it to be. For example, very simple example, you input pictures of cats and muffins. I don't know, maybe if you see that some cats look at the muffin, you tell them this is a cat, this is a dog, this is a muffin and that with enough pictures the model will be able to clearly to precisely predict what is a cat, what is a muffin and what is a dog.
Once trained, it's possible that model drift meaning that it doesn't perform as well. For example, I don't know you or some somebody come up with a muffin that's completely looks exactly like a dog. It's purpose made to be the dog. Then if you want to differentiate that really the muffin, the dog, my muffins or the muffins or dogs that you have to retrain. Again, drift means your model no longer behave the way it's supposed to be. So we had to retrain with new data.
[00:06:31] Speaker B: Jan Montreal is considered to be a hotspot in North America for artificial intelligence. Why is that?
[00:06:39] Speaker A: I think it goes back to this brilliant scientist, Yoshua Benjo who has decided to establish his career in Montreal in about I think 20 years ago or 30 years ago. And he had made, he has participated in the turning point of AI when he was at the University of Montreal and many people being inspired by his work come to follow him and they kind of create an environment with which he could interact and build more research on it from there. Of course from research you have applications and you have startups.
That's when the ecosystem was born. I would say Dr. Bengo contribute a lot to this.
[00:07:36] Speaker B: That's great. And we'll put his information a little bit on the show. Notes.
So tell us about AI in Canadian manufacturing.
[00:07:45] Speaker A: Canadian manufacturing compared to the rest of world has a lot to catch up.
I think many of this is the historical reason.
For example, I'm thinking about Europe. So Europe has went through a two world world where a lot of the infrastructure were being destroyed. Canada, we're lucky to not have that. So many of our, a lot of our infrastructure are still dated back like 50 or 60 or like 80 years back. So we didn't have to rebuild after the war.
Which makes us lagging in the speed of adaptation for new technology. AI being the latest technology that could help us become more efficient.
I don't think we are the best we can be in terms of taking this technology for us to benefit from it.
[00:08:42] Speaker B: So that's like our labor and aging population and that transfer of knowledge, is that correct?
[00:08:49] Speaker A: Of course.
As everybody know, we are having a shift in the age of, in the labor market.
Many people are retiring and with the retirement comes a lot of knowledge.
With years of experience they accumulate in the, in the industry.
I talk to many, many company owners, people own the company. They say I have like this 60, 70 year old person I rely on for the past 50 years. Now they are returning. What should I do? And that is in Itself is a big problem because then you have to train a new person. A new person is as may as well intentioned as the person wants to be. It takes time to become as good as experienced person. I wouldn't say, I would say experienced.
And that I think that's where technologies such as AI could come into play to help us with this knowledge transfer.
[00:09:54] Speaker B: So what are some of the challenges? Like you've got a big job, right? What are the challenges you face when collecting information from machines?
[00:10:02] Speaker A: We all know AI requires data. Data is like the raw material in this AI pipeline to produce the outcome that we wish for.
If your data is not consistent, lacking or small in quantity, the performance of your model will not be as good as it could be. With large language models like ChatGPT, this is less a problem because it's got the sheer quantity of the data they have access to. But in manufacturing we don't have a lot of data.
So if the little data we have are not well collected, then you will not, you will not get the result you, you hope for.
[00:10:50] Speaker B: When you're talking to Canadian manufacturers or manufacturers anywhere, who are you having the conversations with? Are they like the CEOs or Chief Technical officers? Are they plant managers?
Who's really interested in knowing more about AI?
[00:11:06] Speaker A: I try not to put AI up front and first talk to the maintenance director or maintenance supervisor.
What I try to do is provide them immediate quick wins.
So what I try to do is provide them quick wins so they can very fast be able to justify the return of investment of this solution.
But ultimately it would be the production director or even the factory director who could really see the benefits based on the money saves, how efficient we can become now with AI aided tools.
[00:11:53] Speaker B: Yes, and I would imagine the quality people would be interested in that as well. Right? Especially if you're focusing on the food industry.
[00:11:59] Speaker A: Yes, of course.
First of all, some food, for example in my sector would be milk or meat.
They must be preserved within a certain temperature range.
Otherwise you are going to put your product produce product that could endanger the population.
Only that part we are already not, we're not at the place we could be given how affordable and how easy the new tool are at our disposal today. So I would say we should start from the basics. Let's just try to follow the regulations in a way that it doesn't require extra person to do this. You can use tools, sensors and control boxes, hardware to make it easier so that you can and employ your people in the place who is the best for them to Be So what keeps.
[00:12:59] Speaker B: Food manufacturers awake at night these days?
[00:13:02] Speaker A: I would say the first thing is, is would be something very important would be cash flow. Yeah, right. Cash flow is king. We need to sell product and we need to be efficient. So how AI can help in this worries that people have that keep them from sleeping at night would be using AI you can be more efficient. Meaning that with this you can produce the same output using less resources. You can use the same resources, same money and you can produce more product such they can sell increase your revenue.
[00:13:41] Speaker B: So how do you start that conversation? I mean, I know you're not talking about AI all the time, but how do you start that conversation with your prospects and your customers?
[00:13:50] Speaker A: We are in the area of predictive maintenance.
So one thing I ask people is that do you recently have machine breakdown that you think could have been prevented if there was something data collected with the machine all the time. For example, we know that most mechanical components will show signs before it start to fail. For example, your car, there's like if some of the problem you can power here feel something, right? It's the same thing for all mechanical machines. And the problem with them is that sometimes they don't have time to check all the machine all the time. Sometimes it's a place where difficulty accessible. They have to climb like three floors or ladder before they can reach a certain place. So they don't go check there every time. And when machine fails because something that could have been prevented if somebody just go check. We can use technology to do that part. So to give you a better coverage for the machine that has to make sure the machine that are important keep running all the time.
[00:14:59] Speaker B: And is that what you mean by like Fitbit for machines?
[00:15:02] Speaker A: Fitbit machines is like. It's interesting analogy that everybody understand, right?
As I said, to capture the sign of failure before it fails, you need a sensor on the machines a lot. A lot. Like when you have the Fitbit or Smartwatch in your wrist, you know when you are, you're having, you know you're doing exercise, it knows that you're doing exercise. So yes, that's why, that's why we call it fit for your machines. Because it's analogy that everybody understands.
[00:15:32] Speaker B: Can Yan, can you kind of maybe share some stories that you might be able to share? Like we call them use cases, right.
In the industry.
[00:15:41] Speaker A: One of the first good catch that we did was in the milk manufacturers.
It's about a bearing on the boiler fan, right? Boiler in the milk manufacturer they have to run all the time in this case. So to continuously produce, to process milk, if the fan is not working, boiler overheats, it breaks. So in this case, there was a bearing, it was lacking lubricant, or maybe the wrong kind of lubricant was put in place like many months ago.
It was producing a high amount of vibration. Suddenly. In this case, we're able to alert the customers and they were able to have a plan shut down the week following our alerts. In this case, saved, if you account for the cost of downtime, is saved for about $250,000. Wow. Just one time.
[00:16:47] Speaker B: Yeah, yeah. The cost of unplanned downtime is just horrific for a food manufacturer.
[00:16:53] Speaker A: Yeah. Especially for a manufacturer that's running 24 7. And these are kind of costs you don't want to have to deal with as little as possible.
[00:17:03] Speaker B: Yes. Because it goes to reputation, it goes to. You can't ship product. Like there's just so many fallouts from something like that.
Yun, when is AI a bad fit for. For a prospector, for a manufacturer?
[00:17:16] Speaker A: I would say before you start using AI, you need to have an idea what you want to achieve. Otherwise you will probably be disappointed because nothing is magical, especially in manufacturing. Right. Every factory is different, the team is different, what you produce is different. And they will not be a magical solution that fix all your problems. So I would say when it is not a good fit is when you have this magical thinking. Thinking that, oh well, you will just magically do the work without you having to do anything. So that would be one thing.
Another case I can think of could be when your team takes.
You don't have the buying of the team to do this.
Unfortunately, one of the biggest challenge we have in adapting the technology is not the technology, it is people resisting the changes.
We are forcing them to go outside of their comfort zone. And that I would say I am optimist, could almost always be mitigated by good change management, by good change management tools.
[00:18:29] Speaker B: Why is data important?
I know you said it was. It's like the heart of AI, right?
[00:18:34] Speaker A: Data is what is fed into your models. AI models.
[00:18:39] Speaker B: Yes.
[00:18:40] Speaker A: If the data is faulty or it is lacking in quantity or is not collected properly, then your machine learning models would not behave as you wish it to be. I give you an example. In one case, we had a lot of data measuring the vibration of a particular machine, but the sensor was so sensitive that it was picking up the machines next to it first.
Second, the lifts that go around it and the trucks that goes by.
So let's say Every Monday morning at 8am Something vibrates and it was a truck on the street that passes by. You cannot associate that with what you have in the factory. So good data, not too much noise is essential, especially in manufacturing sector where the quantity of the data is limited.
[00:19:44] Speaker B: Jan, is it when a company maybe is thinking about doing an AI project, is it better to start with a simple project or a big project?
[00:19:54] Speaker A: That's like a rhetorical question, right?
[00:19:56] Speaker B: Yeah. How long is a piece of string? Right? Yeah.
[00:19:58] Speaker A: Yes, of course.
I would say the best way to go forward is to start small, have quick wins so that you can have the buying of people of the people working on a project first and has with milestones. That's realistic, right? Of course to know if what you want is realistic, you have to work with specialists. And that's why they come in, they say this is possible and this. You better wait for these five steps to recomplete before you can do the other step. So small steps, quick wins and frequent milestones is what I recommend.
[00:20:34] Speaker B: So if I'm going to start a small project, let's say in a factory, in a food environment, how long does the first one take?
[00:20:40] Speaker A: The first project, I would say start with something that's between three or six months that's totally feasible.
You can use technology that are made to be ready to deploy that doesn't require a lengthy amount of integration into your system.
I was thinking about something like document processing that's a quick win for automation or you know, ready to deploy sensor on machines that you can quickly know what is the vibration level of your critical machines. And that kind of solution would be easy to work with and you can demonstrate the return of investment easily.
[00:21:24] Speaker B: Yung, what are some of the deliverables on an AI project?
[00:21:28] Speaker A: The first deliverables could be feasibility study. Do you have the data that you as required to produce what you need? Right. Just first off, if you don't have the data, then you have to go collect the data that will be the first one. Otherwise one of the deliverables could be, let's produce a very simple model trying to produce something that a smart person, not necessarily trained, will be able to do. And from there it can, you can build upon that model so it become more sophisticated and to meet your need.
[00:22:03] Speaker B: And so what outcomes then can a company expect?
[00:22:08] Speaker A: The outcome would first be you with the same amount of people, they can be more efficient. For example, if you use AI to automate some of the manual steps, then these people will have more time to do other things that require their real people skills or things they are more trained to. So the first thing want to replace are repetitive steps that does not require the high level of human intelligence to do and that we can have now your smart employees to do other things. That's more interesting.
[00:22:45] Speaker B: Young, thanks for coming on the podcast today. Did we forget to talk about anything?
[00:22:50] Speaker A: I think we cover everything.
There was a very assessive list of questions. Thank you.
[00:22:56] Speaker B: Hey, what's your booth number at the ADM Toronto show?
[00:22:59] Speaker A: Yes, we're going to be at ADM and the booth number is 602 and I'm going to also give a presentation on Wednesday.
I think the time is around 11:30. So yes, please come by to rebooth. I will be delighted to have a conversation with anybody interested in AI and manufacturing.
[00:23:21] Speaker B: Yeah, and I think you're in the tech theater, so I think that. So people can come to the ADM Toronto show. And also how can people find more about you and get in touch?
[00:23:31] Speaker A: Well, there's our website, S O R A L I N K CO Sora co. Yeah. And from there you can write me a message or you can go to our LinkedIn page again, sorority code and you'll be able to find me easily.
[00:23:51] Speaker B: And you're young Yao and that's spelled Y U N Y A O. And your website again, soralink co. So thanks again for coming on today.
[00:24:00] Speaker A: Thank you so much.
[00:24:01] Speaker B: We'll see you in Toronto.
[00:24:03] Speaker A: Yes, see you in Toronto.
[00:24:04] Speaker B: And thanks to our sponsor, Mecademic Industrial Robot. Mecademic builds the most compact and precise robots used in industries such as photonics, medical devices, optics and electronics. Mechademic continues to set new standards in precision footprint flexibility and accelerating small component automation for manufacturers, robot integrators and innovators alike. And you can find
[email protected] I'd like to acknowledge a three the association for Advancing Automation. They are the leading automation trade association for robotics, vision and imaging, motion control and motors and the industrial artificial intelligence technologies. Visit automate.org to learn more. And if you'd like to get in touch with us at the Robot Industry podcast, you can find me Jim Beretta on LinkedIn. Today's podcast was produced by Customer Attraction Industrial Marketing. And I'd like to thank my team, Chris Gray for the music, Jeffrey Bremner for audio production, my business partner and our sponsor, Mecademic Industrial Robots.
