The ancient practice of augury involves the reading of omens, often through the flight patterns of birds. The birds, in this case, are in the machine, and by listening to the sounds a machine makes, Saar Yoskovitz can predict when it will break down. There are no birds involved, but the process can tell the future of your machines. Saar Yoskovitz, Founder and CEO, says the goal at Augury is to bring predictive maintenance to the Internet of Things (IoT) and new markets:
“We listen to machines, and based on the noise that they make, we can tell you if they’re working properly or if they have a malfunction,” he said. “Even what type of malfunction they have.”
Augury accomplishes this by connecting ultrasonic and vibration sensors to machines, and then to the Cloud, they can apply Machine Learning algorithms to an acoustic signal. “We’re kind of like Shazam, but for machines.”
About Augury
At present, Augury has two primary products. The first provides an instant analysis of a machine using a portable diagnostic tool. A vibration and ultrasonic sensor connected to a smartphone is attached to the machine by a magnet and, “Within seconds, we tell them exactly what they need to do in order to repair this machine. So not only is something wrong, but you need to replace this bearing, or you need to add more oil.”
The second product is a continuous diagnostic process, also called Diagnostics-as-a-Service (DaaS). “We take the same types of sensors, we mount them on the machine, connect that to the internet and then we have ongoing alerts if anything changes, or anything goes wrong,” he said.
Augury’s customers are enterprise companies, Fortune 500 companies, HVAC maintenance companies, facility management companies, as well as the facilities themselves:
“Initially, we’re focusing on commercial and industrial facilities, mostly on the infrastructure side – pumps, fans, compressors,” he said. “We’re very much agnostic to where this machine is and what it’s doing, so a pump could be pumping cold water for air conditioning in an office building, or it could be pumping chemicals in a pharmaceutical factory on the production line. We could still tell you that you have bearing wear or that you need to align the machine.”
Crowd Sourced and In-House Metadata
Augury uses a combination of on-staff diagnostic professionals and crowd sourcing to read the language of machines. How can they tell bearing wear, for example, from a worn belt?
“This whole market of predictive maintenance has existed for 30 years now and it’s been done using human experts. You have machine doctors that go through all this raw vibration data, and they know how to analyze the machine based on the vibrations,” he said. “We have three [experts] on staff and twenty external vibration analysts that go over our backlog and our existing data in order to make sure we have a lot of voices in the mix when we train the algorithms.”
The marketplace at Amazon Mechanical Turk (MTurk) uses crowd sourcing to do Human Intelligence tasks (HITs) like assigning tags to images, for example, and Saar uses a similar concept to tag machine data. Unlike MTurk, however, jobs are not farmed out to the general public:
“Because not every person can look at a vibration signal and tell you what’s wrong with that machine, so we’ve built our own Mechanical Turk network basically, in order to tag all of this data,” he said. “When you look at Machine Learning, it’s not only about how much data you have. It’s more about how much tagged data you have or labeled data you have.”
This is not just a recording of machine sounds, he said, “this is a recording of a machine that has bearing wear.”
Machine Driven Maintenance
Historically, data about machine maintenance has been tracked by technicians who use a paper log to note when the machine was serviced, and routine maintenance is performed on a set schedule. By connecting to the internet, data about the machine can be used to get a more comprehensive look at its performance.
“Some companies use this in order to streamline operations, [or to] make sure that the operating conditions of machines are always optimized,” he said. “But by far, the leading use of this new found connectivity is predictive maintenance.”
The US Department of Energy found that predictive maintenance can cut on-going maintenance costs by up to 30%, eliminate 75% of breakdowns, and reduce energy consumption by up to 20%, he said. Companies have achieved these results by “switching to a more advanced type of maintenance [regime], which is driven by the condition of the machine and not by a scheduled approach.”
The implications for manufacturing could be far-reaching, but before the potential of this technology can be maximized, Saar says a shift in focus is needed. “One of the largest challenges is not just slapping on sensors and throwing more technology at the problem. There’s also a mindset change that needs to happen within the existing maintenance staff, within the existing management.” Rather than having a regular schedule of oil changes every three months, or belt replacements every year, for example, the needs of each machine at any given moment should determine the workflow.
He used a power plant as an example, where the steam turbine is the most expensive piece of equipment. “It comes with a full-risk service solution from GE or Siemens,” and the company spends millions of dollars to monitor it.
“But this turbine is supported by 200 fans and pumps, and if any one of those pumps goes down, the steam cycle is broken and the turbine can’t work, so in effect, the infrastructure is as critical as the turbine, in this case,” he said. “But they can’t rationalize putting a $40,000 solution on a $15,000 pump.”
Edge Computing Meets the Internet of Things
Saar would like to see more discussion about edge computing. “One of the underlying tensions there is that in order to save bandwidth and to be operating at scale, you need to move as much of the computing as possible to the edge so you can remove all of the waste and only send the data you actually need,” he said. The flipside of that is, “To train a Machine Learning algorithm, you need access to as much raw data as you can physically get,” which isn’t as efficient when decentralized. “We still haven’t bridged that gap. How do you do research for advanced algorithms and how does that work with edge computing?”
As of today, Augury has records for 35,000 machines, and that number is growing daily. “On the one hand, we need to store it and be able to manage it. On the other hand, we need to be able to run algorithms on it, learn from it, and create insights from it, which is a different problem altogether.” The process of analyzing machine performance entails collecting data from each machine – approximately 15GB of data per machine per month. Multiply that by 50 or 100 machines in a building, and a decision needs to be made about how and where to store that data.
“You want to transfer all of that to the Cloud via 3G or 4G – that’s kind of a no go, right? So, a big part of our future and the whole Internet of Things future is edge computing. How do I analyze all this data as close as possible to the sensors in order to reserve that bandwidth, to reserve that battery power where things are battery-powered?”
Augury represents “a new wave of internet age technology” coming into a traditional market, Saar said.
“As the price of technology goes down, we enable them to move to lower end pieces of equipment, so from gas turbines to pumps and fans, and if you can do pumps and fans, why not offer it in a hospital because it’s pretty critical over there, and data centers, and office buildings?”
Augury’s business model and technology enables them to go down market, “to smaller and smaller pieces of equipment.” Similar to the Software-as-a-Service (SaaS) model, the Diagnostics-as-a-Service (DaaS) business model moves the investment “from capital to operational expense, which enables these facilities to do predictive maintenance without the upfront investment of a million dollars to wire up the whole facility.” He sees creating value for his customers as his company’s purpose.
“That is the big engineering challenge: how do you take this massive amount of data and understand what is needed and when? As well as how to crunch all this data and give them actionable insights that they can understand and put to use.”
A shift to condition-based maintenance has the potential to provide actionable insights across any markets where machines and the Internet of Things are used. “The smart home or even the more modern cars – everything is already connected,” he said. “You have an on-board processing unit, the CPU, the processing power, why not add a couple of sensors and analyze the data in a different way to also track the health of this equipment?”
“This can be done across the board, from industrial facilities where we are today, where we have the big expensive machinery, and eventually down the road, even in your home, where you have the washing machine, dishwasher, refrigerator or car?” Augury may not be in the business of reading omens, but with machine driven, predictive maintenance, perhaps in the future, your dishwasher may be able to tell you when it needs a new bearing.
Photo Credit: chombosan/Shutterstock.com