Artificial Intelligence seems to be everywhere at the moment. As engineers, however, we’re rather more interested in what Embedded AI can actually do than what the latest hype says it might do.
At Arrow Technical, our experience with neural networks goes back much further than the current AI boom. We’ve seen first-hand how machine learning-based approaches can sometimes solve problems that become extremely difficult using conventional maths-based algorithms.
What’s changing now is the practicality of putting that intelligence directly into everyday electronic products without massively inflating the unit price.
The right tool for the job
Take a conventional control problem. A well-designed PID controller can be an elegant, reliable and inexpensive solution — and in many applications it remains exactly what we would recommend.
But real-world systems aren’t always so accommodating.
Changing loads, temperature, component ageing and multiple interacting real-world variables can make conventional control increasingly difficult. Engineers can find themselves adding ever more complicated rules, exceptions and algorithms to accommodate situations that are difficult to predict in advance.
This is where embedded AI and machine learning become interesting.
Rather than attempting to describe every possible situation mathematically, a system can learn characteristics from data, recognise patterns and potentially adapt its response to changing conditions.
It doesn’t necessarily replace conventional control either. In many applications, the best solution may combine proven deterministic control techniques with machine learning providing an additional layer of intelligence.
What’s changed?
The principles behind neural networks certainly aren’t new. What has changed dramatically is the processing power, development tools and specialist AI hardware now available at an embedded-system level.
Capabilities that once required expensive computing hardware can increasingly be incorporated directly into PCB-based products — often without requiring a permanent connection to a cloud service.
That creates opportunities for smarter control, predictive maintenance, anomaly detection, sensor interpretation and products that can respond more intelligently to the real world around them.
But there is still an important engineering question to answer:
Will AI actually make this product better?
At Arrow Technical, we believe emerging technologies are most valuable when they’re treated as engineering tools rather than fashionable marketing features.
Sometimes the right answer will still be a simple microcontroller running a conventional algorithm. In more challenging applications, embedded AI may now offer capabilities that simply weren’t commercially practical before.
Knowing the difference is where experience counts.
