Sandvik: Pioneering new software to revolutionize metal machining and manufacturing

In the 1860s, the engineer Göran Fredrik Göransson pioneered the Bessemer process, creating the first mass-produced steel at an industrial scale. The innovation transformed engineering, putting inexpensive, high-quality steel into the hands of workshops worldwide, where it was used to build railways, ships, factories, and all the infrastructure of the industrial world. The company Göransson founded, Sandvik, is now a global engineering group with 46,000 employees. Sandvik is now working on software as well as steel: devising 21st-century ways to transform industry yet again, through its Center of Digital Excellence, or CODE.

"Engineering typically focuses on hardware, but there are so many areas where software is ready to make a real difference in the way people work," says Joakim Ahrlin, Director of Engineering at Sandvik CODE. "Together with our world-class manufacturing experts and in partnership with Google Cloud, we're already reshaping a traditional industry."

A game-changing app for manufacturing businesses worldwide

There are well over 2 million CNC machines worldwide, a type of automated machining tool, used in machine shops globally for the precision cutting of metals. They're invaluable for standardizing production in workshops where more expensive alternatives aren't available, but programming them is a long, inefficient process that has hardly changed since the 1960s.

"The user interface on a CNC machine looks like something from a flight deck," says Joakim. "Engineers need to stand at the machine for hours to write the code, working out how it will look in 3D based on their own experience, piles of paper manuals, and complicated instruction sets. If they get it wrong, the machine might break or the item being produced is rejected, so it's careful work. A simple piece can easily take eight hours to code, during which time the machine cannot run. Workshops only make money from machines when they're running, so this drawn-out process is a real issue for them."

Sandvik CODE set to work creating Prism, a new app for tablets that uses 3D visualizations and a simplified touch interface to cut CNC coding times from around four hours to less than ten minutes.

To do that, Sandvik CODE developed a solution on Google Cloud, assisted by developers at Avalon Solutions, a Google Cloud Partner. "As far as we know, this was a brand-new concept," says Joakim. "Nobody had really managed to package so much machining expertise into a simplified interface, so we weren't sure what our data models would look like. With Cloud Datastore and Cloud Functions we could update and redesign our backend and data models at speed. That made us very agile and flexible every time we ran into new challenges that had to be solved."

The team developed the new app in only seven months, making the most of rapid prototyping. "Cloud Functions is so quick to deploy that you can get an internet-exposed endpoint in a heartbeat," says Joakim. "Because they integrate so well with Cloud Pub/Sub and Cloud Storage, we could build a prototype of a part of our backend in just a couple of minutes. We would usually expect that to take at least a couple of hours, with a lot more hammering around trying to figure out how to deploy it."

PRISM is built with React Native, using only Google Cloud products at the backend. Cloud Pub/Sub communicates between 3D files in Cloud Storage, monitoring on Stackdriver and analytics on BigQuery in a microservices architecture based on Google Kubernetes Engine (GKE). "We love Kubernetes but we don't want to manage the clusters ourselves, so GKE is a perfect product for us," says Joakim. "It's a fully managed solution that still gives us the flexibility we need to deploy things exactly as we want them."

"The developer experience on Google Cloud is world-class," adds Joakim. "As engineers, we appreciate that Google Cloud is designed by people with a deep understanding of the developer experience and what is needed to make it focused, so that we can concentrate on the things that really matter."

Connecting a community through data science

As well as creating external products, Sandvik CODE has reinvented Sandvik's internal approach to data science. "We wanted to help people do more with data," says Lukasz Tracewski, Lead Data Scientist at Sandvik CODE. "At Sandvik, you get a laptop when you join the company. It's good for office work, but if you want to do any sort of data analytics, it can be a real headache: you need to get admin rights, install software, and confront the computational limitations of the laptop itself."

Sandvik CODE looked to democratize data with a web-based data science platform. And when commercially available solutions proved too expensive, the team decided to build their own, using open source tools and Google Cloud products.

With GKE as a foundation, the team pulled together select open source tools such as Jupyter Notebook, a leading option for presenting data analytics in a web browser, and Discourse as a forum for data scientists, analysts, and amateurs to discuss approaches and share advice. "One of our main aims with data analytics is to form a community, so that colleagues with different expertise can connect and ask questions," says Lukasz. "In the past, people would work on data science problems more or less in isolation. We wanted to change that."

To help build that community, the platform also contains material to teach absolute beginners how to code, so they have all the materials they need to participate. It even hosts interactive books, written by experts at Sandvik to share machining or engineering knowledge across the company. "The book is completely live," says Lukasz. "As you play with formulae, you can see the results plotted live on charts. That makes it easier to communicate complex ideas between international teams, but it also makes data science more accessible. One of our colleagues in marketing uses it to work on his own data."

GKE helps make the solution much more cost-effective than commercial alternatives by sharing cluster resources between jobs. So when resources are not being used in one area of analysis, they can be used for other jobs on the same cluster.

"GKE can easily scale to however many people are on the platform," says Lukasz. "Costs don't jump that much when people join, because we can share out the computational power of the cluster. And when they leave, we can redistribute resources or power down instances. That makes it much less expensive than any commercial solution currently available."

Building solutions to go from scratch to scale

PRISM is already being used by customers in machine shops, with spectacular results in Sandvik's early testing. According to those tests, with PRISM, the process of coding a CNC machine is 94 percent faster than before, and error rates are down by almost as much. In most major markets, including North America and Europe, manufacturing shops work toward a standard 3 percent increase in productivity per year. With PRISM, workshops have delivered increases between 10 percent and 20 percent, which is more than triple the industry standard.

Sandvik's data science platform is no less impressive. The platform cost 50 times less than leading commercial alternatives to develop and deploy. "I firmly believe that this platform sets the stage for future savings well into the millions," says Lukasz. "It could be the basis for a recommendation engine, for example. At present, we don't have one in Sandvik Group workshops where customers buy parts, but a 2015 McKinsey study estimates that a simple recommendation engine can deliver revenue growth of over 30 percent."

"Everything we've done is built to scale," says Joakim. "We're fully prepared to take the current configuration on PRISM from a few hundred people to several hundred thousand users. It's really thanks to Google Cloud that we can do that, because the products scale so well, both individually and together."

"Our technology partnership with Google Cloud will help us to achieve our aim of digitally enabling the end-to-end fabrication life cycle, from CAD design and prototyping, through to production," adds Joakim.