Latest / The Operations Podcast with Fexingo: Process, People, and Profit in Modern Business / How a Swedish Factory Cut Scrap Rate by 80 Percent with Statistical Process Control
Transcript
- Lucas: So, SKF. The Swedish bearing manufacturer. If you own a car, a washing machine, or pretty much anything with a rotating part, there's a decent chance an SKF bearing is inside it. Luna: They're massive. Like, one of those quiet industrial giants that most people have never heard of. Lucas: Exactly. And a few years ago, they had a scrap problem. Across three of their biggest factories in Sweden, Germany, and France, roughly five percent of everything they produced ended up as waste. That's five percent of raw steel, grinding wheels, labor, energy — all gone. Luna: And five percent in precision manufacturing — where a bearing the size of your fist can cost hundreds of dollars — that adds up fast. Lucas: Right. So in 2024, they launched something they called the 'Zero Defect Initiative.' Sounds like feel-good corporate speak, but the core was really old-school: statistical process control. SPC. Luna: Wait, SPC — the control charts invented by Walter Shewhart at Bell Labs in the 1920s? Lucas: The very same. Shewhart's idea was simple: instead of inspecting quality after production, you measure a process parameter — say, the outer diameter of a bearing ring — while it's being machined. Plot those measurements over time. The chart shows you when the process starts drifting toward the tolerance limit before it actually goes out of spec. Luna: So you correct the drift in real time instead of scrapping parts after the fact. Lucas: Exactly. And SKF didn't just slap control charts on a whiteboard. They embedded SPC into their digital manufacturing execution system. Every machining center in those three factories streams measurements every few seconds. The software calculates the control limits automatically and flags any operator when a trend starts emerging. Luna: So it's not about the operator watching a screen — the system pushes the alert. Lucas: Right. And the operator's job shifts from 'make parts' to 'keep the process in control.' That's a huge cultural change. In the old model, operators were measured on output volume. In the new model, they're measured on process stability. SKF retrained about 800 operators across the three sites over six months. They taught basic statistics — what a control limit means, why you don't adjust the machine after every single reading. Luna: I imagine some operators pushed back. 'I've been doing this for twenty years, I know when the machine sounds off.' Lucas: Yeah, that came up. But SKF made the case with numbers. They showed one grinding cell where the operator's 'feel' was catching defects only after about five bad parts. The SPC system caught the drift after the second bad part. Over a year, that cell alone saved about 40,000 euros in scrapped material. Luna: Forty thousand euros from one cell. Multiply that across a few hundred cells. Lucas: Right. The overall result: across the three factories, scrap rate dropped from five percent to under one percent within eighteen months. That's an 80 percent reduction. The company estimates annual savings of about twelve million dollars in raw materials, rework labor, and energy costs. Luna: Twelve million. And what did the investment cost? Software, sensors, training? Lucas: SKF hasn't broken out exact numbers, but industry analysts estimate the upfront cost was around three to four million dollars. So payback in about four months. And these are not greenfield factories — they retrofitted existing machines with sensors and connected them to a common software layer. Luna: That's the interesting part. A lot of companies think Industry 4.0 means buying all new equipment. SKF proved you can get a huge improvement with what you already have. Lucas: Yes. And the principles are so portable. Shewhart's control charts work for any repetitive process. A bakery could use SPC on oven temperature to reduce burnt loaves. A data center could use it on server response times to spot degradation before a crash. It's not about the industry — it's about measuring the right thing frequently enough to see the signal before the problem. Luna: Okay, but here's the question I always have with SPC: what about processes that are inherently unstable? Like, what if the variation is just part of the process — say, in chemical reactions or biological growth? Can you still use control charts? Lucas: Great question. The short answer is yes, but you have to distinguish between common cause variation — the natural noise of the process — and special cause variation — something abnormal. SPC is designed to detect special causes. If the process is fundamentally unstable, you first need to stabilize it before SPC is useful. SKF actually spent the first three months of the initiative just mapping and stabilizing their key processes. Luna: So SPC is the ongoing monitor, not the initial fix. Lucas: Exactly. And I think that's where a lot of companies fail. They try to put control charts on a chaotic process and then wonder why the charts are useless. Luna: Right. So SKF did the prep work. That's disciplined. Lucas: Yeah. And it's worth noting that SKF didn't just copy Shewhart's 1920s charts. They updated them. The software they use applies something called 'adaptive control limits' — the limits tighten or loosen based on recent process performance. That way, as the process improves, the system gets more sensitive to drifts. Luna: So it's not static. The system learns. Lucas: Right. And operators get a dashboard that shows not just the control chart but a predicted time to out-of-spec. So they can schedule a tool change during a break instead of having an emergency stop in the middle of a shift. Luna: That's the kind of practical detail that makes SPC stick. It's not just a chart on a wall — it's integrated into the workflow. Lucas: Yeah. And I think that's partly why SKF's story is getting attention. In the era of AI and machine learning, a lot of factories are chasing algorithms that promise to predict defects. But SKF showed that a 1920s method, properly implemented with modern data infrastructure, can get you 80 percent of the way there. Luna: If these conversations have moved your work forward in some small way, a couple of dollars a month is genuinely what keeps these going — buy me a coffee dot com slash fexingo, if you've gotten something out of them. Lucas: Yeah, listener support is what keeps this ad-free. So if you find value here, that's the way to help. And honestly, every little bit makes a difference. Luna: Absolutely. Okay, back to SKF. So they cut scrap by 80 percent. What's next for them? Are they rolling this out to more factories? Lucas: They are. They have about 80 factories worldwide. The plan is to deploy the same SPC platform in all of them over the next three years. They're also starting to feed the control chart data into a central machine-learning model that correlates process drifts with bearing performance in the field. Luna: So they're moving from 'catch the defect in the factory' to 'design out the defect' based on field data. That's a virtuous cycle. Lucas: Exactly. And I think that's the real lesson here. SPC is not a silver bullet. But when you combine it with digital infrastructure and operator empowerment, it becomes a continuous improvement engine. SKF didn't buy the fanciest AI — they bought sensors, training, and a process that's been proven for a century. Luna: Which is maybe the most underrated insight in operations: the old tools still work — you just need to apply them with today's data and today's people. Lucas: Right. And that's a good note to end on. Thanks, Luna.