Q-Cast

Podcast: If You Don’t Have Enough Defects for Training, AI Can Step In

Nicholas Blake of Advex AI explains how synthetic examples can be used to help improve training models, what machine vision offers, and how AI inspection can cut training time down from years to hours.

Listen to the Full Podcast ▼

Michelle: To start off with, for our listeners, can you tell me a little bit about your work and what you do?

Nicholas: I am in charge of sales and marketing here at Advex, and Advex AI is in the industrial automation space. We provide a platform that allows users create synthetic data for training vision systems to detect defects or do robotic pick and place applications.

Michelle: Very timely technology these days, it seems. So for manufacturers who are kind of considering these technologies, can you tell me what are the benefits of this type of machine vision and AI for manufacturers and how can it help them?

Nicholas: So, I want to make one thing kind of clear, we're not replacing vision systems, right? So ideally you already have a vision system in your factory or you're thinking of integrating one and you kind of have an idea of what hardware you want to use, how you want to connect it to your PLC. So we're not replacing all of that.

Really what we're helping with is making that process go faster and making it better. And so what we can do is, whether you're using some of the off shelf guys like Cognex and Keyence, maybe you're working with a startup AI company. There's lots of them out there, or you're building your own. There's always that process where you have the camera up and now you're collecting data.

And when you're collecting that data, if it's taking you too long, if you don't have enough examples of the defects, or you find that the examples you do have are not getting that vision system, to the level of performance that you want it to, that's when a software like us would come in handy.