The slightest variable can lead to batch yield issues in factories and R&D labs. Elemental Machines is using a combination of Internet of Things and machine learning to ensure consistent outcomes.
Serial entrepreneur Sridhar Iyengar’s first exposure to quality control problems in manufacturing came not long after founding his first company, AgaMatrix. The startup, which counted the French drug-maker Sanofi among its customers, specialized in disposable test strips to calculate the blood glucose levels of people with diabetes or hypoglycemia. It contracted with a factory to make several million strips a day but quickly ran into batch yield issues. “I spent 53 nights in 52 weeks in the same hotel in South Korea trying to fix the problem,” remembers Iyengar. He and his co- founder gathered all the data from the contractor, augmented that with sensors in the factory, and built their own dashboard and prediction algorithms. The yield went up to virtually 100%, he says.
“We were able to predict the quality of output two to three months in advance and if we saw a high risk we were able to use our own tech tools to pinpoint where the issues were,” says Iyengar. That helped the company snag CVS, a large U.S. drugstore chain, as a customer. AgaMatrix went on to create the first medical device (a glucose meter) that could be plugged into an iPhone, and was bought out by Sanofi. Next, Iyengar founded Misfit Wearables, together with John Sculley, the ex- CEO of Apple. The company’s health trackers created a data-rich, globally distributed sensor network. For Iyengar it was an “aha” moment. Why not create a Fitbit for machines and use that data to improve the yield of R&D trials in biotech and life sciences as well as manufacturing anything requiring precision and repeatable results?
No Waste, High Yield
So, Iyengar founded Elemental Machines, which uses the cloud, Internet of Things and machine learning technology to model and understand how processes in a laboratory or factory work and then use that knowledge to help obtain consistent, repeatable results, saving companies time and money. The Boston-based startup pulls data out of machines that aren’t currently connected to anything, collates it with data from equipment that is already monitored, analyzes what is happening in real time and displays it on dashboards. The company, which has so far raised $11.5 million, has 100 customers ranging from small startups to big pharmaceutical, petrochemical and food manufacturers. “We are building the hardware and software tools that allow companies to model and understand how complex processes work and very rapidly zero in on what went wrong, or right,” says Iyengar. “If things are done correctly, there is no waste and high yield, and it can also help prevent drugs and other products from being recalled.”
When valves or other parts fail machines don’t always automatically report the error. Many times errors are only discovered after a beta trial, which can cost hundreds of thousands of dollars, says Iyengar. Take the case of a synthetic biology company, one of Elemental Machine’s customers. When it grows cells it warms them to 37 degrees celsius and the flasks are gently shaken to both blend and aerate the cell mixture.
“The company was consistently getting bad results and couldn’t figure out why,” says Iyengar. Finally, they discovered that a tiny screw was loose in one of the machines that shakes the mixtures. The shaking vibrations and the rate of aeration were a little bit different so the cells grew at a different rate, expressed themselves differently, impacting the purity rate. “It took them months to figure this out,” says Iyengar. “Now they are using our technology to measure all the variables that may affect their yields and outcomes.” Earlier this month Elemental Machines announced it is partnering with PerkinElmer to allow scientists, as well as lab and facilities managers, to access information at any time from a computer or mobile device about what is happening inside their laboratories and be notified of potential problems when conditions change. While Elemental Machines started by focusing on helping R&D laboratories control variables it has since branched out into manufacturing. One of its manufacturing customers, a materials science company, was not measuring temperature, humidity or light in the areas where it was mixing chemicals.
«Turns out the microclimate made a huge difference,» says Iyengar. Now the manufacturer is using the startup’s technology to measure those elements and analyze the data.
Iyengar sees many more applications for the technology. “Now that we have a way to collect and analyze all of this data it allows us to generate a numerical representation of quality and use that to train our models on the way the process is actually run,” he says. While it is still early days, Iyengar says he believes it will be possible to develop a system for R&D trials and manufacturing that would operate much the same as Waze, a GPS navigation software that works on smartphones and tablets to provide drivers with turn-by-turn navigation information.
Elemental Machines’ “Waze for factories” could prove crucial not just to traditional R&D labs and manufacturers but to a variety of emerging fields, such as personalized medicine and next generation materials like lab-grown leather and meat, where precise and repeatable results are crucial, he says.