Manufacturing’s black art is retiring. AI may be the best chance to keep it

EVERY factory has them. Veteran engineers who can walk the floor, hear a machine, smell something off, and know what the defect is and what caused it. Decades of experience have given them a kind of intuition that manuals don’t capture.

Some call it the “black art” of manufacturing where operators diagnose with such remarkable efficiency. And if you ask an engineer “how do you know?” – the answer is “I just do”.

Manufacturing makes the modern world. But some of its own tools are a generation behind. Artificial intelligence has promised productivity gains everywhere else. Can it do the same for manufacturing, or will the industry’s own constraints hold it back?

Retiring experts, thin pipeline

In my younger days, I had never considered manufacturing for a job. Privileged to be born in Singapore, the “obvious” ideal aspirational job often can be described as a white-collar job; in a comfortable air-con environment and avoiding the hard sweaty work. This is perhaps where the challenge begins for the manufacturing industry – where talents view the sector as “not sexy” or unappealing.

As several key manufacturing hubs including Singapore continue to age, many of these key talent engineers are on the brink of retirement and new talents are not entering manufacturing at the same rate. That tacit knowledge lives in people’s heads, and when they retire, it leaves with them.

Some manufacturers shared first-hand that it can easily take two to three engineers to replace a seasoned veteran, and it doesn’t guarantee a job well done. Demand for manufactured goods is undoubtedly rising, and this expertise gap is probably one of the biggest under-discussed constraints.

Data-rich, insight-poor

Over the past decades, factories have digitalised rapidly. Sensors track nearly every stage of production, and manual logging has given way to continuous data capture.

However, capturing data is only half the equation. Lots of data does not mean lots of insights. Dashboards multiply, yet the decisions still depend on the same few experienced people who know what these numbers really mean.

Some even concede that such sensor data tracking has become a box-ticking exercise; it helps factories hit their “tech” KPIs, but does little to help workers with the task at hand.

SEE ALSO

Who gets ahead in the enterprise consulting business with AI now in the game?

Beyond the backflipping humanoid: where will the robotics returns actually be earned?

AI for life science: Queen’s Gambit or fool’s wager?

It seems only timely that AI can step in to help. Today’s AI systems, with appropriate relevant training, can absorb vast amounts of production data, find patterns humans can’t see at scale, predict challenges and support real-time decisions on the floor.

They act as a capable assistant to the veterans and may even help new engineers to improve their productivity and performance.

Take a familiar scene on the factory floor where the yield on a line suddenly dips. Traditionally, the factory would wait for its senior engineer to walk the floor, check a few machines, and trace the problem back to its roots. This process can take hours.

Now imagine an AI system that has been trained on the plant’s engineering drawings and years of production data. Within minutes, it flags that the dip began shortly after a slight temperature drift at one upstream station, traces how that drift affects the steps after it, and recommends a specific adjustment, along with how to verify the fix on the actual equipment.

In essence, it captures judgment in a form that can be checked, explained and passed on to the next generation of engineers.

Budget-ready, change-averse

Interestingly in manufacturing, budget constraint for productivity is rarely a concern. Most manufacturers already systematically budget monetary resources to trial advanced technology, new materials and new manufacturing methods. But the hardest part to get right is change management.

In manufacturing, changing a small process may have consequences all the way down the line, especially in process manufacturing, where production flows in a linear sequence and an adjustment at one step ripples through every step after it.

Companies run a tight schedule and cannot bear the risk of operation downtime. This makes managers and workers reluctant to experiment, even with promising changes.

Then there is the question of data. Manufacturers treat their production data as sacred, and for good reason.

Process parameters and yield histories are, in many ways, what keeps them ahead of their competitors. Any tool sending this data outside the factory would not even make it to a trial.

This means viable AI for manufacturing has to be built with security at its core, running in-house, and focused on the intelligence rather than becoming yet another data aggregator or dashboard.

One line at a time

So what kind of AI solution will actually work in a factory? First, it has to fit into the existing workflow rather than disrupt it.

Factories have little appetite for new hardware or extra sensors, so the strongest solutions run on the data and equipment already in place, with minimal add-ons.

Second, it has to work alongside floor engineers, not around them. They own the decisions on the factory floor and carry the responsibility when something goes wrong.

A good system recommends and explains, but the engineer ultimately decides. Tools that bypass that judgment may not survive on the line.

Deployment is where many companies win or lose and define the trajectory of the company. This is where forward-deployed engineers play a crucial role.

Rather than promising to transform the whole plant at once, they work on-site with the customer to narrow the first deployment to a single line or problem with a clear, measurable outcome, such as a yield improvement, reduced scrap or less downtime.

Demonstrating ROI (return on investment) on the factory floor, in numbers the plant manager already tracks, will earn the trust of the people on the floor.

Skilled veterans won’t be on the floor forever, and the next generation isn’t arriving fast enough to learn from them. AI won’t replace that intuition, but it may finally give manufacturers a way to capture it, explain it and pass it on before it retires.

Manufacturing may not be the most glamorous sector, but it could be where AI proves its real worth.

The writer is associate director, investment at Vertex Ventures SEA & India