In an ancient brick building on the outskirts of Copenhagen, translucent green bottles of a pilsener that is “probably the best beer in the world” clink along a conveyor track.
At least, that’s how most of the world recognizes Carlsberg beer, from the deep, sonorous tones of Orson Welles intoning the phrase in commercials since 1975.<!- mfunc feat_school ->
It’s hard to beat Carlsberg for tradition. The company has been brewing beer since 1847. But today, it’s also hard to beat them for innovation. Carlsberg is the fourth largest brewing conglomerate in the world, with operations on every continent except Antarctica.
And those bottles of beer, all of them, in all the various factories all over the world? From the moment the empties come into the factory to the moment they hit the racks of a distribution truck, they are tracked and managed by a Manufacturing Execution System (MES) powered by Big Data and managed by data scientists.
Carlsberg and other manufacturers use the wave of information unleashed by modern manufacturing methods to:
- Identify inefficiencies and tune the production process
- Scale production to immediately respond to customer demand
- Tie factory inputs and outputs to external supply chains, cascading requirements and availability in real-time to both vendors and customers
Lean Manufacturing is Data-Driven Manufacturing
Factory floors, on breweries and elsewhere, do not immediately leap to mind when the concept of Big Data comes up. Yet the manufacturing industry was at the cusp of the lean revolution in business management.
It all started at Toyota.
In the 1950’s, Toyota plant manager Taiichi Ohno began to implement a series of processes that would come to be known as the Toyota Production System (TPS). All micro-processes stressed operational excellence and waste reduction. But, critically, many of the procedures relied on increased access to data about the manufacturing operation in progress.
An inventory system called Kanban was devised to send signals about stock levels to allow for on-demand parts-replenishment. Improving sightlines in factories to make the work-in-progress easy to see and providing a visual representation of workflow in the form of “Kanban boards” proved critical.
Ohno didn’t have sophisticated computer systems to help with any of this, but his emphasis on data as the ultimate driver behind every decision made on the manufacturing floor was clear.
TPS has been co-opted and has come to be known simply as lean manufacturing– and it has taken the business world by storm. Companies that don’t run lean often rely on Six Sigma, a related process-improvement approach that is almost entirely data-driven.
Today, Kanban boards that provide a visual representation of work flow and Kanban flow-based systems are found in factories from Taiwan to Rio de Janeiro. And with the low cost of computerized automation, many of them are running on volumes of data that Taiichi Ohno could never have imagined.
Tying The Corporation Together with Manufacturing Execution Systems
It was relatively easy to digitally acquire and integrate data from the sales, marketing, accounting, and other backend business functions. They were among the first to acquire computers and begin managing information electronically. Although integration wasn’t easy, at least the data was entirely digital to begin with.
Things were different on the manufacturing floor. Even at Toyota, the Kanban boards and pull-inventory systems were written out by hand and managed manually.
But electronics came to the factory floor eventually.
- Robots began to take over laborious, repetitive tasks
- Automated inventory systems started to take care of monitoring stock levels
- Production levels were tracked by production line sensors
- Audits and quality control were taken over by automated testing suites
MESs coordinate all those various discrete tasks and functions so that robots work at the pace that inventory is available, and additional parts are ordered when the robots are running below optimal efficiency. Faults in the line or in individual stations are quickly detected and either routed around, or the product flow adjusted to avoid wasteful pileups.
The data from this is raised up from the shop floor, as well. Executives half a world away trying to plan an expansion into new markets can instantly consult a dashboard and see the state of the plant and stock available.
The Devil is in the Seemingly Unrelated Details
And unexpected insights can arise from additional data. One unnamed European chemical manufacturer, described in this 2014 McKinsey research article, was widely renowned for running a model factory. Yet after implementing detailed data collection into their systems, they found that apparently minor and previously unconnected factors were having significant impacts on production efficiency. The company was able to reduce production waste by 20% by stabilizing variables that no one had even realized were critical.
At Carlsberg, the data goes both ways. Having tied the MES in the brewery to Enterprise Resource Planning (ERP) systems that predict demand and schedule product deliveries, the brewery operators now can automatically adjust their production based on customer orders. In effect, the breweries are responding directly to customer demand, without the friction of intermediate processes.
The company estimates that the system has bumped both sales and gross margins up by more than one percent, and reduced factory downtime ratio by fifteen percent.
From Planning the Obsolescence of Goods to Predicting the Failure of Machines at the Point of Production
Automating and instrumenting the production process may only have been the first step in moving manufacturing into the future. As the products themselves increasingly become Internet-enabled and have their own sensors and logic embedded, a phenomena being called the “Industrial Internet of Things” is coming into being.
The additional sensors available in the production line and the data they generate are all contributing to more in-depth and reliable predictive modeling for production. Predictive analytics are now being used to calculate machine maintenance windows and product lifespan.
These aren’t entirely new tasks, of course; managers have always understood that parts would wear out and products would fail. The difference that data makes is in the reliability of the predictions. Wear is a complex process that involves many interacting parts and pieces. When it is poorly understood, failure can only be predicted in a relatively wide window.
But when it is modeled with enough fidelity, it can be predicted with accuracy. This allows managers to get every last bit of work out of machinery before replacing it, while also reducing the risk of unexpected failures.
This is all too complex for manual calculation, which is where data scientists come into the equation. Machine-learning is now used to produce heuristics to analyze the masses of incoming data and perform the predictive analytics that was once thought to be impossible in the days when it relied on human calculations.