Quality assurance is a critical part of manufacturing; it’s just humans are not very good at it. Their faculties are not trained to pick out anomalies in rapid carousels of industrial assemblages. More than this, they just can’t do it.
Modern manufacturing, particularly of discrete electronics, is too complex and too various for human eyes. But machine learning, at the edge, is turning fallible old production checks into fail-safe new productivity gains.
Case in point: HPE and Foxconn have developed a machine version of Where’s Wally? to discern the production-line equivalent striped hats in grey boxes, and raise the alarm. It is a symbiotic arrangement, using HPE edge components to automate quality assurance on a Foxconn production line carrying HPE servers.
The new system, developed with video analytics firm Realmetrics, is being positioned as a support line for the standing workforce, to relieve it of duties it was never qualified for in the first place. “It’s
always there, and it’s always checking,” says Steve Fearn, global chief technologist at HPE.
Machines bring precision and control, and they are eminently biddable – the Foxconn system won’t rush the work to clock-off early, “to get home to the family”. We encounter him at an HPE showcase in Geneva; he shows a picture of a slide from a Space X product. “It’s important that part is correct –
otherwise there’ll be a big bang and a squeal from someone inside the vehicle.”
He shows another image, this time of a carbon sample from a Formula One car; it’s somehow critical, but it could be anything. The stakes are high, and humans are fallible, he explains. “Things are getting smaller, things are getting more varied, and yet we need more precision.”
The Foxconn customer-and-supplier circle has been put to work at Foxconn’s European manufacturing facility in Kutna Hora, in the Czech Republic. John Gallagher, operations manager of Foxconn, is in charge. “Quality underpins everything that we do,” he says.
But, as with discrete manufacturing, its products are complex, its batch sizes are small, its production is swift, and its schedule is packed. The Kutna Hora line for HPE servers jumps to a new configuration every three or four units; 50-60 per cent of its units are shipped within 24 hours of joining the
assembly line.
“We’ve got people inspecting, and they miss a lot of things,” says Gallagher. The new edge system catches the faults before they cause delays, and makes good on Foxconn’s quality promise. “If we say it will get there on Tuesday, it will get there on Tuesday.”
On the conveyor in Kutna Hora, every unit enters a section for video review; mobile cameras move into inspect every aspect of the unit, from every angle. The cameras are 2D; Foxconn has no requirement for 3D machine vision at the site, as yet.
Its machine learning is trained in the cloud and executed on site. Anything that does not pass muster – “damaged components, missing components, foreign objects, incorrect configurations” – goes to a human referee. Any question, and it is seized for a second opinion, and either righted and re-entered, or else held back; the production lines keeps chugging along all the while.
Nothing goes to waste. The facility has moved away from a “he-said, she-said” development path, says Fearn, towards iterative production based on hard data. “Variation is the killer; we can now capture every configuration and every fault and improve the manufacturing process.”
Foxconn is working to run the same video checks on individual assembly stages, to identify at an earlier stage in the production process. “We would basically break down the analytics one by one, and then eliminate the need for a final inspection,” comments Gallagher
A new report and webinar on edge computing in industrial IoT setups, called AI and IoT at the cutting edge – when to move intelligence closer to the action, is available; go here for the report, go here for the webinar.
