Follow Datanami:
August 13, 2013

Manufacturing Real-Time Analytics on the Shop Floor

Isaac Lopez

With large production machines pumping out endless streams of data, and production processes that are in constant need of improvement, manufacturing and big data technologies seem like a match made in heaven.

Unfortunately, this hasn’t been the case, as the manufacturing seems to have misplaced their ticket on the big data train. In a report released earlier this month by Cap Gemini, they note that while historically, manufacturing companies have traditionally been early adopters of such technologies as ERP and Production Planning systems, the sector has been slow to adopt more recent digital technologies.

I spoke with Will Sobel, CEO of System Insights, a company trying to turn this trend around, and bring manufacturing into the age of big analytics, big insights, and ultimately, big productivity. He says the industry is lagging right now and suffering from outdated processes and metrics.

“If you look at the way a shop is often run, it’s still using a clipboard type technology to be able to collect information from the shop floor – somebody walking around looking at what’s going on, and there is very little connectivity with the machine tools,” he explained.  Sobel commented that the standard approach to assessing efficiencies has been through a standard called Overall Equipment Effectiveness (OEE) – an aggregate standard developed in the 1960’s that is still being used as a key performance indicator today.

“We’ve found that OEE is really insufficient,” says Sobel. “It’s a decent, high-level metric – an aggregate metric – but to actually action OEE is really hard. What we’ve found is you really have to start digging in and looking at the data to understand why something is happening on a given machine.”

Sobel’s company, System Insights, has developed such a system. Using an array of big data oriented technologies (including MongoDB, Hadoop, Amazon, MTConnect and some unique data collectors), they’ve developed an analytics package called “vimana,” to collect and analyze machine tool productivity, and then spit out insights on what can be done to improve equipment utilization and improve production line capacity.

The process starts with the production tool – Sobel says they typically connect directly to the machine controller, giving them access to all of the machine’s data. “If the machine is recent – within the last ten years, it’s fairly easy for us to connect to it,” he said. Once the machine is set up and ready to deliver the data, it’s wrapped up using MTConnect standard, which XMLizes data for delivery into a local concentrator, where it gets streamed into the vimana application in the Amazon cloud.

Once in the cloud, the data is loaded into MongoDB using an open source, complex event processing (CEP) engine called Esper (note: we covered EsperTech and the work they are doing in the CEP space last year). As a real-time, in-memory, CEP engine, Esper is designed for high-volume event correlation. Sobel says they’ve built rules inside of Esper that help understand the data coming off of the machines in the production facility.

Sobel adds that they use a few other tools to help contextualize the data, including the business rule management system (BRMS), Drools, as well as their own Scala-based statistical processing engine. Additional context is through a user interface, where machine operators (or other people on the shop floor) can add classifications to various intervals of time (or during particular events), giving meta-tags that add context.

“Esper is basically our pattern matching engine that we use – similar to like a financial data system – to understand what is happening at any given time on each machine,” explains Sobel. “We know whether it’s producing, whether it’s not producing, what kind of downtime it is, etc.” There is an array of different types of machine statuses that they’ve classified.

As the information is contextualized and put into MongoDB, certain alerts come out of the rules engine that will tell the end user important information about the machines they are operating. “That’s the predictive side of it,” says Sobel. “These alerts will tell the user that they need to go pay attention to a particular machine because it’s heading for a problem. Or it might see that a machine has been down too long, and somebody should check if help is needed.”

With the data contextualized and staged in MongoDB, it’s ready for analysis, says Sobel. “In the next stage, we have Hadoop and R analytics running against MongoDB – we slice, dice, analyze, look at different patterns and correlations that allow us to better understand some of the historical reporting, what’s been happening, and why it’s been happening.” The idea, says Sobel, is to get to the why of certain events so that they can achieve their purpose of improving equipment utilization and thus improve production line capacity.

Sobel says they typically experience productivity improvements between 10-40%. One of System Insight’s clients is Curtiss Wright Controls in Shelby, North Carolina, who deployed vimana on a production cell with four high-end CNC-based horizontal milling machine tools.  According to their case study, Curtis Wright Controls was able to monitor production efficiency and machine utilization without requiring any manual intervention of data collection. The result, claims the company, was an improvement of over 20% in equipment utilization in a period of 10 weeks.

Aside from controller data straight from the machines, Sobel says that the company also does power data correlations using their own high-speed power meter.  “If a machine tool can’t get you information directly – let’s say the controller is 25-30 years old – how do you understand what the machine is doing just by the external signals,” he rhetorically asks. “You get it either by energy consumption or vibration, or other types of sensors.”

To address this, they created a power meter custom made for the big data age. Where a standard power meter will only give around two samples a second, the System Insights meter samples at one-hundred per second. “Using that power meter, we can find patterns that indicate how the tool is operating,” he says, explaining that they are able to see events on the machine tool at any given time just by looking at the energy consumption of the machine. Sobel says that using this meter, they can layer the data on top of modern machines to get a deeper understanding of its operations in vimana.

With so much opportunity in the manufacturing arena, there are sure to be other entrants into the field.  Sobel says that anyone entering the market will have plenty of catching up to do. “There is tons of noise in the data. Somebody entering into this is going to have to put in quite a bit of time to bake that noise into any sort of usable context.”

In the meantime, we’ll be watching as big data and other digital manufacturing trends start to take root.

Datanami