The manufacturing sector faces enormous challenges, as do the supply chains linked to it. A series of man-made and other crises have contributed to a perfect storm of risk: skyrocketing energy, fuel and raw material costs; impending recession; war in Europe; transport disruption; skills shortages; Brexit (for UK companies); environmental and sustainability needs; and the long tail of the COVID-19 pandemic, which has left many businesses struggling.
Thus, factories cannot afford unplanned downtime and costly machine breakdowns, as this can snowball the industry, damaging customers and their own dependent customers. Ultimately, failures could impact us all in terms of shortages and empty shelves, driving prices even higher.
One company thinks artificial intelligence and machine learning (AI and ML) are the answers, but not like other companies in the hype cycle are doing – we’re surrounded by vendors making similar claims. This one uses an unusual approach: sensors that detect vibration, sound, and electrical faults up front, then it deploys AI and ML behind the scenes.
With complex machinery running 24/7 on production lines, sometimes in safety cages or shut down factories, failures are often not discovered until too late: when a critical component fails. But unless you can see inside every machine and monitor every widget – which is expensive and inconvenient – how can you tell if a hidden problem could lead to a catastrophic failure?
The answer is by detecting minute changes in the way machines sound or vibrate, says Israeli entrepreneur Saar Yoskovitz, co-founder and CEO of Augury, a New York-based machine health unicorn (“Machines speak, we listen”).
Founded in 2011, with a full launch and first funding in 2014, the company today employs over 400 people worldwide and works in a wide range of manufacturing sectors, such as food, beverage, paper, forest products, chemicals and pharmaceuticals. Blue chip customers include Pepsico, Nestlé, Hersheys, Colgate-Palmolive, Heineken, Danone, Canfor and Bausch Health, among others.
The first spark of inspiration came from an unlikely source: national service in Israel. Yoskovitz says:
My co-founder [CPTO Gal Shaul] I was in the navy and I was in the artillery. For both of us, our lives depended on very big machines. And in those scenarios, you get very “intimate” with the machines, in that with every little creak or creak, you can really understand what’s going on. So it was intuitive for us to say, “If I hear something wrong, why can’t we make machines and computers understand it too?”
In college, my studies focused on speech recognition using AI and machine learning algorithms. What we are doing today is very similar. We listen to the machine, we take an audio wave, and we try to find meaning in it. But instead of looking for words in speech recognition, we look for patterns of malfunctions.
We pick up vibration signals, but we also have magnetic sensors that can detect electrical faults in motors and drives, temperature sensors, etc. But the driving force was, if we can hear something is wrong, then we can train a computer model to do that too.
Unique insights from unique data
So, Augury is all about data, but of an unusual, sloppy kind. By gathering large amounts of sound data from sound machines and components, any minor deviations can be flagged as a potential problem. Yoskovitz adds:
It’s exactly that. If you have a pump in your factory, we don’t need to build a new model for your specific model, because we’ve already seen over 20,000 pumps. We know exactly what cavitation or cracked bars look like. We have over 200 million machine hours that we have monitored, and all of that data is in our cloud.
We can use it to further refine our algorithms to make them more accurate and create new levels of insights, and we have predefined templates for all of these different types of machines. From the first moment the sensor is installed on your machine, we can tell you what’s wrong. So, within three months, all of our clients pay off our annual program in full and they start to grow.
Bold claims, and the company website claims 7x ROI or more in certain industries.
But a number of recent reports have revealed that in general, smart factory initiatives often fail, due to difficulties in scaling up a successful pilot project, upgrading old facilities and finding the right skills, as well as gaps in change management. For example, pre-COVID research from Capgemini suggested that only 14% of large-scale smart manufacturing projects succeed. Why do companies struggle to make their factories more efficient?
Acknowledging the problem, Yoskovitz says:
There’s this term “pilot purgatory,” where pilots go to die. Our market today is not good at proving the value of a small pilot project and then scaling it up, that’s true. But we were able to crack the proven value code very, very quickly.
But then, how are you evolving, on a global scale, with large companies? Turns out the key to unlocking it is time to value: the sooner you can deliver value to the customer, the sooner they can go to their manager and get more investment and scale.
We have the fastest time to value possible, and the reason for that is that we have a complete solution: we have our own hardware, we manage the connectivity, we run proprietary AI diagnostics on the dataset and we provide professional services. around change management, reliability, etc.
The value of machine health
Which brings us to the company name. An augur is a sign or omen of the future (something that does or does not bode well). To what extent does Augury deal with this definitive added value of Industry 4.0: predictive rather than reactive maintenance? Yoskovitz says:
That’s exactly what we do, that’s where we started. And over time, as we work with large customers like Pepsico, we understand that machine health is not a maintenance issue. It’s much more than that: it’s a supply chain risk management issue, it’s a workforce empowerment issue.
With [snack manufacturer] we helped them make a million pounds of their product that they couldn’t have made, because we detected and fixed a problem that increased their availability, and therefore increased their capacity.
And we have the largest cement manufacturer in North America. They predicted that we will increase their capacity by two full factories. So machine health isn’t just “Hey, you need to replace that bearing”, it’s “Hey, you can avoid spending $800 million building two new factories”, not to mention the environmental impact of this.
So we renamed this category from predictive maintenance to machine health. But that’s only one side of the coin. The other aspect is the health of the process, because we know that the mechanical health has an impact on the quality of the product produced by the machine, and therefore on the energy consumption of the motor, or on the throughput and the yield.
In May, Augury acquired process health AI specialist Seebo in a cash and stock deal worth an estimated $140 million. The target is the alleged $1 trillion untapped manufacturing capacity from inefficient factories and processes.
If I detect a malfunction in a gearbox, and we know that the lead time for a spare part is eight weeks, we can tell the operator: “If you slow down the line, you can maintain throughput and quality, while increasing component life. But if you don’t, it will fail and you will be closed for a month.
We want to build a world where everyone can rely on the machines that matter, where the power is always on and the water keeps flowing. But as we can see, that’s not the world we live in yet. Every week you hear about supply chain disruptions, chip shortages from China, power shortages via Russia and Ukraine…
What we do is at the most basic level: we work with the biggest manufacturers, and we help them to make their production lines more reliable, more productive and more sustainable, so that they can go and restock the shelves of your supermarket. local.
An innovative and exciting adventure. But can Augury’s data help solve the broader issues we face: soaring energy bills and global warming, which can push many businesses to the brink? Yoskovitz says yes:
Basically. Predictive maintenance can reduce energy consumption by up to 20%. And on the process health side, from Seebo we can take data from existing datasets and help customers optimize their production lines to reduce their energy consumption and carbon footprint.