Artificial intelligence (Al) is often promoted as the enabler for the fourth industrial revolution, and why wouldn’t it be? We’ve connected numerous machines and we’re now harvesting more data than ever before, but turning that data into actionable insights at line speed is a big challenge. And without actionable insights, data is just noise!

Another critical issue with any form of digital transformation is the so called ‘digital dividend’. Can the money invested in any digital transformation program drive real added value in terms of efficiency or quality to provide a tangible return on investment? If it can’t deliver a dividend, there really isn’t much point in making the investment and going to all the trouble.

Finally. I think it is worth asking if the application of Al, and any digital transformation for that, is solving a real problem. Are we trying to improve a quality issue that is real and costly, or are we simply tweaking a process that already has exceptional performance? As one EMS CEO recently reminded me, going from failure rate of 2PPM (parts per million) to 1PPM might not be worth too much investment.

Where does Al really come into its own?

Operators have many amazing talents and a few limitations, understanding both is essential to selecting the right application for Al. Humans are great at multitasking and at absorbing, consciously and subconsciously, data from multiple sources in complex environments. What they are not great at is studying detail in extremely high volume. Show us a cat. any cat, and we know it’s a cat. Trying teaching Al to recognize a cat and you’ll understand the challenge, it will need many learning examples before it can really tell what is a cat and what is a dog with short snout.

But if you show us a thousand images of something complex like a semiconductor and ask us to spot the one that is slightly different, we’d have little to no chance of spotting the outlier. What’s more we’d be much too slow to process the images at a speed that makes it viable.

Al at Cybord

Take the Cybord SaaS Al-based component analysis system as an example. We set out to
solve a real problem, to use Al for something that was impossible with humans, and to deliver real incremental value to the user. Let’s break that down:

  • The problem was simply one of surgical traceability for component in a zero-trust supply chain. Surgical traceability means being able to trace every device to every PCBA (printed circuits board assembly), and hence to every product. Zero-trust supply chains are supply chain that do not assume the veracity of the parts just because of where they come from. In a zero-trust supply chain, parts are verified once inspected. This is important all the time, but comes into sharper focus when devices are in short supply and the grey market is employed.
  • So far, we have inspected more than five billion devices at goods in and on the SMT line, ail in real-time. This is something that would be unimaginable for the human eye or brain to process. Hence, it is a task ideally suited to Al.
    Big data and Al really do make good bedfellows and this is very big data!
  • When it comes to the value proposition that’s also straightforward. Firstly, the systems ensure bad parts do not make it into products that reach the user. And by assuring every part assembled is a good one, the cost of assembly is reduced as waste is cut out and yields improve. What is more, this level of surgical traceability, which has been redefined in I PC standards, offers that ability to reduce any recall by identifying the products affected by faulty or unfit components. Recently half a million cars were recalled in South Korea due to faulty components, we suspect this number could have been as small as 5,000 if our system had been deployed throughout the supply chain.

Al systems just keep getting smarter

What is amazing about these learning systems is that they never forget, they just keep getting smarter. That means that today each image is compared to a database of around six billion other images. In the future that could be sixty billion images.

The system is generally being deployed by brands that want to reduce the risk and increase traceability in their supply chain as it applies to components. But as we deploy it for those brands we see their contract manufacturers gaining from the insight they add into their own systems and processes. This means that, while the brands (or OEMs) might be insisting that their suppliers use the system, those suppliers are driving additional value themselves. This win-win is important for the value proposition and reduces any friction during the deployment phase. What many love about the system is that it largely uses hardware that already exists. Cybord sources images from the placement head and from AOI on the line. We also use images from reel-to-reel inspection in goods inwards to catch any faulty parts early.

Our experience with Al is that it can deliver incremental and substantial value when used in the right way, but each case should be explored on its own merits using the criteria set out above. It should solve a problem, it should do something that humans are not suited to, and it should add real value. Digital transformation should have a digital dividend!