How Predictive Maintenance Can Use Machine Learning
Buying a car in the U.S. is a rite of passage for most American youth, and their first car purchase is oftentimes a hooptie from a used car lot that will ferry them around until they’re able to afford something nicer. Anyone who has driven a car like that knows the importance of listening for strange noises. It could be the exhaust pipe starting to drag a bit, it could be a brake pad that’s worn down to the metal, or it could be a wheezing sound coming from the radiator as the car overheats. All of these sounds will telegraph a problem with your 1982 Chevette that you’ll then need to spend the weekend trying to fix yourself to save money. If you don’t fix those problems and the car blows up, you just go and buy another hooptie. If we’re talking about strange sounds coming from industrial machinery, then the cost of failure becomes much more costly.
The ability to use the sounds that are coming from a machine to predict when it will need maintenance – something aptly named “predictive maintenance” – is now feasible because we have machine learning algorithms that can help us “listen” to the sounds machines make in a way that even humans can’t. The word “augury” refers to “a sign of what will happen in the future; an omen” according to Google, and it’s also the name of a company that wants to use machine learning for predictive maintenance.
Founded in 2011, New Yawk startup Augury has taken in $51 million in funding so far to develop machine learning algorithms that make machines more reliable. We first came across Augury a few years back when we wrote about 7 Industrial IoT Startups Using AI to Monitor Machines, and since then the startup has raised another $25 million which closed just days ago in the form of a Series C round. They are now keeping a watchful eye on more than 60,000 machines across the United States with 7,000+ failures detected and 3,000+ machines saved from being total write-offs. If your job involves machine maintenance in an industrial setting, one of your Key Performance Indicators (KPIs) is probably “machine uptime.” Think about how good a solution like this would make you look when it comes time to tell a compelling story for your end-of-the-year bonus.
Types of Predictive Maintenance
Another AI startup called MapR does a good job of defining how to predict failure in machines, and they presented the following list of the most common methods:
- Infrared Thermography (IRT): for seeing variations in temperature
- Ultrasonic Analysis: for measuring changes in frequency that could indicate an issue
- Current Analysis: tracking the voltage and current of electricity, usually as it’s supplied to a motor
- Vibration Analysis: an indicator of misalignment, wear, or imbalance
- Oil Analysis: a way to measure the overall systemic lubrication
While there are existing ways to monitor these things, they may not be as sensitive or cost-effective. For example, when it comes to vibration analysis, there are sensors already being used to detect such problems, but they’re not as sensitive as the solutions that Augury has developed. Some companies also use visual inspections for preventative maintenance which take time and therefore money, and are subject to human error.
The Augury Platform
Today’s product offerings from Augury consist of wireless sensors that are embedded among the machines to detect faulty equipment (Halo) and handheld sensors that allow you to analyze equipment out in the field (Scope) along with a platform that allows you to manage it all. The sensors detect vibration, temperature changes, and magnetic data, all of which are analyzed in milliseconds by intelligent machine learning algorithms.
The system only gets better over time, as the machine learning algorithms compare recording data from your assets to past recordings of similar machines around the world. With continuous diagnostics, you can be alerted to the first sign of any developing issue which may represent a looming malfunction. You don’t need to make any upfront investment because the company has cleverly built their offering around a Diagnostics as a Service (DaaS) pricing model, which means you don’t have to engage in hand-to-hand combat with your CTO in order to fight for the budget you need to adopt a solution that saves the company money.
Augury Success Stories
One compelling success story involves an appliance manufacturer with a one million square foot production facility that churns out refrigerators. Here’s a picture which gives you an idea of just how big a one million square foot facility is:
Inside their production facility, the client has four manufacturing lines, each with its own critical air compressor. If any of these 400-700 horsepower compressors fail, a rental compressor needs to be used (added cost) not to mention the cost of the actual downtime (even bigger cost). Historically, the company used “light preventive maintenance” such as visual checks to ensure the compressors stayed online. Then they adopted Augury’s solution – and as all those tools over at BuzzFeed like to say – you won’t believe what happens next:
At the onset of the installation of Augury’s Halo solution, a severe malfunction was detected, and was immediately diagnosed as worn thrust bearing on the main drive shaft. Basic vibration sensors would not have caught this malfunction, but as Augury’s sensors boast a detection range up to 10 kHz, this costly malfunction was detected at its earliest stages.
The diagnosis provided by Augury allowed the facility staff to perform a $7,000 repair on the machine only three days after the on-site installation of Halo. The fault was caught early enough for the customer to take action, avoiding a catastrophic failure that would have cost the customer upwards of $240,000 in addition to the downtime related costs. That’s the sort of performance improvement that will make you look like a rock star during the next year-end performance review.
If you’re an investor in today’s market, you probably feel a great deal of uncertainty as the market whipsaws back and forth. It certainly feels like the market is not just due for a correction, but a major bear market. Then, we think about the many companies out there like Augury that are creating unprecedented efficiencies and savings by using the newly-found powers of machine learning algorithms. All these savings will impact the bottom line, which translates into increased earnings, which then translates into stock prices rising as earnings surprises become more frequent. That’s the bull case.
Then there’s the bear case, which says that no matter how great things look, you can always be blindsided by some black swan event that suddenly destroys immense amounts of wealth. Consequently, the best thing we can do is just adopt the same stance as financial pundits do. Give absolutely no useful guidance, but instead, suggest that investors “remain cautiously optimistic” about the market going forward in the face of amazing technological advancements like those on offer from Augury.
A tech company founded by a self-made billionaire has an acceptance rate of 0.41%, making Harvard’s acceptance rate of 5% look like a cakewalk. They've deployed more IoT sensors than anyone else, and they're processing 1.5 million transactions a second. It's one of a few IoT stocks we're invested in right now. Become a Nanalyze Premium annual subscriber and immediately access our entire portfolio of more than 30 tech stocks.