THE NEW INDUSTRIAL REVOLUTION – FLOWSERVE IS DEPLOYING ARTIFICIAL INTELLIGENCE TO TRANSFORM OUR CRITICAL INDUSTRIAL SYSTEMS

Marla RosnerWritten by | Artificial Intelligence, Featured

George Washington gave the very first State of the Union address in 1790. That same year, as a nation was being born, Thomas Simpson, a civil engineer living an ocean away in London, founded a pump and steam engine workshop and named it Simpson & Thompson. While most of what we know from 1790 is relegated to history books, museum exhibits, and plaster replicas, Simpson & Thompson can today be found in its latest incarnation as Flowserve.

Flowserve has certainly had a long life, and it’s done so by a tradition of constant innovation and reinvention. Throughout their exceptionally long past, Flowserve’s products have always been “things of the future.” We often associate innovation with tools such as the printing press, the factory line, the Internet, mobile devices, and wearable technology. But pumps? What could possibly be so exciting and innovative about pumps? The truth is, pumps are what keeps the lights on and the water flowing. The type of machinery that Flowserve manufactures is the central organ—the heart if you will—of society’s most critical civil services. From energy production to water utilities, Flowserve’s equipment keeps our most vital societal functions operating. And the manufacturing of this critical machinery started over 220 years ago.

Just a few years before founding his company, Simpson invented the first bell and spigot joint, a new method for making the relatively new iron water pipes watertight at socket joints. This may not seem like a big deal, but it was this type of technology that enabled modern plumbing. The bell and spigot joint would remain the primary pipe joint used across the world for 170 years, all the way until 1956. Simpson & Thompson also pioneered the use of air vessels in water towers, marking a shift from pumping water from cisterns at the top of water towers to the more efficient models we use today. In 1828, they built the first slow sand filter bed, a method of water purification that gave rise to the first treated public water supply in the world. In the 1840s, they invented the double-acting beam rotative compound engine, a more efficient and powerful steam engine that helped drive further modernization in the Industrial Age.

Most of these technologies are no longer in use today, and mean little to nothing to the modern reader. But they were critical innovations that formed the building blocks of modern society and all of its trappings we rely on, from clean water to powerful motors. They and many other such inventions allowed the company Simpson founded to grow and flourish, spreading its influence further with international mergers that eventually resulted in the Texas-based Flowserve, almost 230 years later. That tradition of pioneering at the technological cutting edge has shown no sign of stopping.

In addition to building the industry’s top hardware, today Flowserve’s innovation is found in the intelligence of the software running their equipment—predictive maintenance powered by artificial intelligence.

THE PROBLEM OF PREDICTIVE MAINTENANCE

Flowserve’s bread and butter is designing, developing, manufacturing, and repairing precision-engineered flow control equipment for their customers’ critical processes. This includes pumps, valves, seals, and support systems, mostly intended for use by some of the world’s largest companies in oil and gas, power, chemical, water, and other critical industries. Even now, they’re continuing to explore ways to push the industry and its technology forward.

No one is more aware of this than Eric van Gemeren. As Vice President of Flowserve’s Product Management and R&D, van Gemeren is responsible for generating new growth opportunities and driving innovative practices across the product portfolio, new product introduction, product cost reduction, and has overall accountability for managing Flowserve’s global portfolio of investments in emerging technologies.

van Gemeren is no stranger to any of these roles. A registered professional engineer with a Bachelor’s degree in Mechanical Engineering and a Master’s in Systems Engineering, van Gemeren has been developing and leading high-performance teams in high-tech organizations for more than 25 years. He has held positions ranging from military service and management consulting, to executive leadership in a multi-national organization. Throughout all this, he says, his passion has always been in driving growth through product and service innovation, helping to transform old-economy business models. In this respect, Flowserve is a standout.

“What’s amazing to me is the creativity with which our customers and engineers are thinking about new ways to apply technology to better understand and continuously improve their operations,” he says. “I’ve seen instances where some of our latest solutions can combine vibration, pressure, fluid, and thermal signatures to understand more of what’s happening inside the pump or the flow system—and provide that data to a user in almost real time so they can take action quickly.”

The latest frontier for van Gemeren and Flowserve is to expand even further on these aftermarket services— specifically, by selling assets that are packaged with intelligent software, such as machine learning solutions, to determine asset health and predict impending failures.

Predictive maintenance is something Flowserve has been working on for a very long time. The ability to predict asset failures before they occur is invaluable in the industries Flowserve serves, as it can massively reduce operating and maintenance costs as well as unscheduled asset downtime. This translates to more operational efficiency.

van Gemeren sees this as a crucial part of Flowserve’s evolving business model. “Flowserve is unique in that we not only are a long-standing OEM and service provider of leading edge flow control technology, but we also now have the ability to detect, diagnose, and then of course repair or even upgrade our customers’ flow equipment,” he says. “We’ve been providing solutions for remotely monitoring and diagnosing pumps, valves, seals, and actuators in our industry for many years—and we are continuing to invest in this space.” Having a broad services capability—Flowserve maintains over 200 service centers globally—is a critical element in the equation.

Generally speaking, predictive maintenance is approached by using models. A model is best thought of as a mathematical function, such as f(x)=y. The input for the model, or the x in the function, is sensor data from various assets, such as temperature and pressure. The output of the model—the y value—then creates the desired insights on how the assets are operating or if they’re close to failure.

In the past, Flowserve and other similar companies have approached this challenge with physics-based modeling. This kind of pump monitoring involved measuring predetermined thresholds of vibration and temperature and sending alerts whenever these measurements exceeded a specified level. This methodology has worked well for years, and provides operators with immediate condition-based information. However, it stops somewhat short in that the information is generally minimal or incomplete. This pump is running hotter than expected, or this asset is vibrating beyond its specified limits. But it couldn’t tell the operator why this might be happening.

“Individuating” a root cause is what Flowserve is working on, and is what will tell the plant engineer the root cause of the pending failure. Perhaps the shaft is no longer aligned, or the bearings are wearing out, or there is blockage in the line somewhere upstream of the pump. This is extremely valuable to know, and will tell the operations team where to look, and when they might expect an asset to fail in these conditions.

It was clear that a better solution would be needed. According to van Gemeren, Flowserve’s customers wanted to augment better information with what their engineering, operations, and maintenance teams already do, so they could optimize operations. The goal is always to lower operating costs, reduce unplanned downtime, and improve reliability and safety—and Flowserve’s customer base sees technology as the way to achieve this.

As for what van Gemeren himself wants? “I tend to think about this more from the standpoint of expanding the envelope of information readily available to the customer to improve performance and to make better decisions. Today, that window is pretty small for some equipment or systems, and it may not be all that clear. We are working to push the boundaries of that envelope, and make it more transparent, so that the customer can truly see what is happening in nearreal time to his equipment—and then act upon it. First is knowing, then taking action with that data.”

The Industrial Revolution – Flowserve is deploying Artificial Intelligence to transform our critical industrial systems

The best way to accomplish all this has been years in the making. Building on known methods, using increased data collection and cloud computing, Flowserve is now adding more advanced modeling and machine learning to the mix.

With the same nose for cutting-edge innovation that has carried them through for centuries, Flowserve had been keeping an eye on machine learning as a possible solution for almost two decades. In the 1990s, machine learning was experiencing a new renaissance, as scientists shifted their focus from previous rules-and knowledge-based approaches to a more flexible data-driven approach, feeding programs large amounts of data to permit those programs to learn from the data in an organic fashion. This meant that machine learning could feasibly be applied to a model for predictive maintenance. It could be fed sensor data as the input, analyze that data, and produce maintenance information as the output—in theory. In practice, the growing technology just wasn’t ready for such an application, since it was too expensive and difficult to implement.

This has all finally changed in the past few years, however, as advances both in data science and in raw computing power have finally allowed machine learning to start to reach its full potential. van Gemeren also cites the drop in hardware costs and computing costs as having allowed businesses to deploy broader and deeper computing solutions than ever before at a lower cost profile to their customers. “No longer is this a thing of the future,” he says. “The tools are now becoming much more usable by engineers and practitioners who don’t necessarily need a PhD in science to fully interpret the results. We are making it easier to know more about what is happening with the asset, act upon that information, optimize the system, and even predict what might happen next.” In response, the technology landscape has been undergoing massive shifts, as the newly augmented machine learning takes its place exactly where businesses like Flowserve long believed it would eventually be—poised at the forefront of a technological revolution.

Flowserve wanted to take better advantage of these growing capabilities, and found part of their solution in SparkCognition’s artificial intelligence technologies. In this sense, Flowserve did not have to hire their own data scientists or build a large support team, but could move quickly into this space with confidence.

INTEGRATING AI

It was in 2014 that Flowserve first partnered with SparkCognition, an Austin-based AI company.

Providing deep analytics and actionable insight via machine learning was the logical next step for Flowserve. “But we cannot do it alone,” van Gemeren says in explaining how the partnership with SparkCognition came to be. “Great partners are fast, flexible, and bring a unique capability to the final solution. SparkCognition is great at drawing conclusions using disparate data sources, and is ideal for layering on top of other more traditional analytics capabilities that we have and continue to develop. Partners can also bring a different perspective, and SparkCognition has been great at bringing new ideas and technology, like natural language processing, to the forefront of what is possible. This capability, added to Flowserve expertise in pumps, flow equipment, and service looks to be a potent combination of talent.

Unlike the rules-based models Flowserve had previously employed, the machine learning models that SparkCognition developed for Flowserve rely on hypothesis generation based on near-real time data streams and trained on historical sensor data. By leveraging data in this way, machine learning can combine human insight and experience with a machine’s speed, scalability, and lack of bias, to create models of unprecedented accuracy and sophistication.

The Industrial Revolution4

Eric van Gemeren, Rob Miller, and Kumar Ramasundaram of Flowserve

The results are perhaps the best way to illustrate this point: SparkCognition’s machine learning model allowed Flowserve to give their clients significant forewarning before a failure, an increase of as much twenty times in early testing. Models are also showing false positive rates of less than two percent in early testing, meaning that the models accurately tell the user exactly what will happen 98% of the time, helping the customers individuate true root causes, beyond traditional sensor capabilities. The next step is testing this methodology with several pilots now being deployed in the field.

“And partnerships work both ways, of course. We think partners mutually support each other, and Flowserve can bring a long history of over 200 years in an industry that brings both credibility and deep intellectual property to the relationship.”

SparkCognition has also worked with Flowserve on projects that can detect the current operating state of a pump based on a relatively small number of sensor inputs, such as pressure or vibration. Using this cognitive solution, Flowserve is able to effectively identify whether a pump is operating in the right state, if there’s a problem with the pump, what kind of problem there may be with the pump, and if it is a problem that has been encountered before. This is a wide range of highly valuable data, and being able to output all of it based on such a constrained data set is not only a technological achievement, but also is part of what makes this model both efficient and scalable. The scalability of this machine learning solution means these analyses can be performed not just at the pump unit level, but at the fleet level as well. This allows Flowserve to monitor large numbers of pumps as well as helping their customers gauge the efficiency of an entire operation.

All in all, Flowserve now has the ability to create dynamic models of pumps under a wide variety of operating conditions, while using multi-dimensional analysis to provide early asset failure warnings. Not only that, but Flowserve is now able to identify the factors responsible for a failure, making it easier to repair assets and avoid future failures.

van Gemeren is excited about these results, and what they mean for the future of Flowserve. He believes that this is a significant expansion in both the depth and breadth of diagnostics Flowserve can perform on running machinery, and one that also improves the predictive power of those analytics. “Together, these advances increase the return on investment our customers get from these solutions,” he says. He doesn’t plan on stopping there, either. His next goal is working to make internal use of SparkCognition’s natural language processing (NLP) capabilities. NLP is software that can learn written and spoken language in much the same way as a human. With this capability, SparkCognition has developed solutions to understand, analyze, and use what is known as “unstructured data”—data in a format such as natural language that would normally be impossible for a machine to understand. Flowserve’s next project with SparkCognition is to feed this NLP technology a corpus of technical documentation—data sheets, troubleshooting guides, electrical diagrams, and so on—in order to create a database in which their technicians can quickly and easily find relevant documentation for whatever they’re working on.

Even this, van Gemeren believes, is only the beginning. “Today, we are working more closely than ever with our customers to continuously evaluate and evolve our technology to remain at the forefront of what’s happening in this space,” he says. “In the longer term, we see a more connected, more predictive operating environment for our customers where the combination of rapidly advancing AI technology, coupled with the reduced cost of gathering and analyzing data, allows plant owners and operators to instrument not only the ‘critical few,’ but also the ‘important many’ pieces of equipment that are key to their day-to-day operations.

“There will always be new challenges and new technologies to apply to help improve reliability and performance,” he adds. It’s a perfect summation, it seems, of both van Gemeren and Flowserve’s approach to just about anything they do.

Flowserve may not look or sound much like the workshop of Simpson & Thompson from 1790, but the world at large doesn’t look much like it did in 1790, either. The ability to change, grow, and invent along with the world around them is what has allowed Flowserve to survive for almost two and a half centuries, and it’s what will keep them going in the years to come. No matter how much technology changes, there will always be a need for companies that innovate. The only way not to become a thing of the past is to always keep an eye on the future.

Last modified: November 22, 2017

Leave a Reply

Your email address will not be published. Required fields are marked *