By Cassandra Balentine
Predictive maintenance solutions are gaining popularity, with increased usage across several major verticals, including government, aerospace, and defense, as well as manufacturing, energy and utilities, healthcare, and transportation and logistics.
A recent report by MarketsandMarkets, Predictive Maintenance Market by Component (Cloud and On Premise), Organization Size (SMES and Enterprises), Vertical, and Region — Global Forecast to 2021, revealed that the predictive maintenance market size is estimated to grow from 1,404.3 million USD in 2016 to 4,904.0 million USD by 2021—a compound annual growth rate of 28.5 percent during the forecast period.
Predictive maintenance systems (PdM) software utilizes business intelligence (BI)-driven artificial intelligence (AI) to help determine when maintenance should be performed on a piece of tracked equipment. With this information, organizations can reduce down time, related costs, and lost revenue.
Deployment Options
There are several ways to look at deployment for PdM. The simplest comparison is whether it is cloud or on premise based. However, the definitions by provider vary as highlighted below.
Richard Howells, VP of solutions management, Digital Supply Chain, SAP, addresses the argument for on premise versus cloud deployments. He points out that on premise solutions require every organization to have its own IT staff to manage the backend work on a dedicated server. “Conversely, having all data live in the cloud means that the company relies on a vendor to host their information as an additional service. A cloud solution requires much less IT work on the client’s end, meaning the data is accessible through the internet for any user at any time and via any location.”
Pedro Alves Nogueira, PhD, director of engineering, Toptal, says cloud solutions are easier to spin up, scale, and configure. “However, they offer less in the way of customization or performance tuning or support for some very specific scenarios. A custom-built solution offers more freedom of choice and can easily be tuned into specific needs, but will require a much larger budget footprint in terms of infrastructure.”
Ruban Phukan, VP product, Progress/co-founder, DataRPM, believes that at the highest level there are two types of PdM offerings, full-stack platform specialized for PdM and PdM solutions available embedded in an existing industrial solution like EAM, ERP, MES, SCADA, and other systems. For each of these, deployments can be edge, where the PdM solution is offered on the edge device; self-managed data centers that collect, store, and process data from various systems at different locations together; cloud, which is similar to the central data center just that the infrastructure is managed by a cloud provider; and hybrid cloud, which uses a mix of the above.
Mark S. Benak, VP business ventures, Uptake, sees three options in terms of deployment options. These include build from scratch, assemble and integrate disparate tools, and deploy and innovate. “Building your own means you’re the master of your digital universe. The tradeoff, however, includes a significant delay in time to value, the creation of a less robust platform due to a talent gap, and execution risk.” He details pros and cons below.
Build from scratch solutions are built on top of a pubic cloud infrastructure as a service with a team of in house data scientists, engineers, developers, and consultants to create a custom internet of things (IoT) platform from the ground up. “Piecing together disparate tools is an option to consider if your business has a reduced budget or a smaller team of engineers and data scientists. With this option, you purchase tools that do the legwork of creating an infrastructure for your teams upon which to build. While this approach means faster speed to path than building your own, this process can be complex and difficult to execute successfully.”
For assembling and integrating disparate tools, you patch together a platform from several best of breed technologies, either through self-aggregation or with a professional services team. “An IoT platform that is fully equipped to handle industrial data, while still allowing customers to extend, customize, and innovate on top of the platform’s built in functionality is the best approach in my opinion,” shares Benak.
Eitan Vesely, co-founder/CEO, Presenso, points to four basic options, the first is for the industrial plant to build its own machine learning solutions using a combination of commercial and open source solutions. “For instance, Amazon Machine Learning provides visualization tools and wizards to generate machine learning algorithms without having to learn complex ML algorithms and technology.”
The second option starting to emerge is for OEMs to bundle PdM with the hardware that they provide. Instead of selling equipment, the OEM can lease it and assume responsibility for analytics. “The so-called hardware as a service is not a new concept—the Rolls Royce introduced Power by the Hour in the early 1960s,” he says.
The third option Vesely lists is to use a third-party application that performs machine learning to the big data that is generated by the sensors within industrial equipment. There are cloud, edge, and on premise options for these applications.
And finally, Vesely says the fourth option is the digital twin. The digital twin is a virtual clone of industrial machinery that provides real time performance information to the industrial plant including early warning of potential failure.
“Historically, deployment of predictive maintenance still required a lot of manual work. Manufacturers would hire a team of data scientists, run a machine to failure, and record the data that is produced during this period. This would be used as a benchmark against which to spot similar patterns occurring in the future,” explains Dr. Simon Kampa, CEO, Senseye.
He says the data produced by individual machines, even the same models from the same manufacturers is as unique as a human fingerprint; however, identifying these crucial indicators and developing specific algorithms to spot them can be a long and laborious process. “We’ve made deployment of AI-driven predictive maintenance analysis a lot easier with a series of fairly generic algorithms that can be applied to any machine from any manufacturer, and looks at the data that is already being produced. Crucially, the AI then teaches itself to become smarter and more predictive, fine tuning its own performance for each monitored asset as it goes. This approach pretty much eliminates the need for human intervention and allows rapid development and rapid return on investment (ROI) to be achieved factory wide,” says Dr. Kampa.
The key consideration for predictive maintenance is usability. “Can you get information to your employees so that they act on it quickly? The predictive part of predictive maintenance is actually often fairly straightforward. Often, it revolves around focusing on one or two data points,” offers Michael Kanellos, IoT analyst, OSlsoft.
He offers two examples. Syncrude, a tar sands mining company in Canada was experiencing a rash of what they call ventilator incidents, for example engines on their trucks were suddenly exploding. They started tracking 44 parameters on around 136 trucks—which meant around 1,716 data signals per second. They had the answer fairly quickly, the sudden drop in oil pressure was the signal. Syncrude started serving up the signals, employees knew that meant a failure in a few days and took action. Syncrude cut maintenance by $20 million a year. Similarly, TasWater, a water utility in Tasmania, figured out that it could predict sewage overflows—which potentially could damage the local oyster harvest, a large business—13 hours in advance by tracking the rate of fill on certain sewage cisterns. A straightforward alarm system is solving the problem. “In both cases, the predictive part of analytics were accomplished by giving people access to information that was difficult to find or not entirely readily available,” shares Kanellos.
Target Markets
Predictive maintenance products are thought to be particularly useful in manufacturing and transportation industries, although its use can extend to other verticals as well.
“With manufacturing specifically, we see digital twins having a huge impact when it comes to keeping plant and supply chain operations running by predicting, sensing, and fixing potential issues before they cause major problems—usually at a much lower cost and without a negative impact on customer service,” offers Howells.
Benak says PdM is important for any asset intensive industry given the high cost of downtime. This can range from oil and gas to energy, manufacturing, transportation, and beyond.
Due to advancements in AI automation and analytics, predictive maintenance techniques that were traditionally developed to transform aircraft maintenance and keep man safely in the air can now be deployed affordably and at scale in more down to earth settings. “Rather than only being used in regulated markets or on a few critical assets, we are now seeing predictive maintenance products being rolled out at scale across a number of industries. We are seeing most traction in areas such as automotive, consumer goods manufacturing, and heavy industry, but there is also growing demand from the oil and gas, renewable energy, facilities equipment, fleet monitoring, and pharmaceuticals sectors.”
Nogueria believes PdM solutions are a generic product overall, and therefore don’t cater to any specific market. From an academic perspective, he says PdM products stem from a series of subfields within AI/machine learning, operational research, simulation, and in many cases—mostly manufacturing—robotics. “They are applicable in a wide range of verticals, from predicting the failure of Formula 1 engines, heavy-duty machinery on oil rigs, power plant equipment, self-driving cars, or even sensor enabled crops. Anything that has a high associated cost with a failure event is a candidate from PdM, so long as it can be properly instrumented and accurately measured.”
Kanellos also sees two categories, human centric and algorithm centric. With human-centric analysis, data is served up to employees that as soon as they see the trend line, can predict what will happen. “The only reason they couldn’t predict it before was that they didn’t have access to the data. In truth, this covers about 80 percent of the science,” explains Kanellos. Algorithm centric predictive maintenance occurs when there is a large amount of data streams and/or the symptoms go beyond the experience of your typical employee.
Phukan says predictive maintenance is the future of the maintenance industry. “It helps identify the optimal time to maintenance and provides information about which machine or component is likely to have an issue, much before the problem occurs. This dramatically helps in reduction of unplanned downtime, unforeseen quality issues, and unexpected drop in efficiency and performance. All of which dramatically reduce costs and improve productivity.”
IoT and PdM
The growing spotlight on the IoT and the industrial IoT affects the drive and demand for PdM systems.
Phukan says predictive maintenance is essential in delivering value from IoT to industrial context. “The availability of high resolution data from sensors enable algorithms to create a digital twin of machines and processes to model the normal operating behavior and identify early signals of problems much ahead of time. This helps deliver highly accurate predictions of faults and issues that can deliver significant cost savings, minimize risks, plan inventory, and manage field services better.”
IoT is the backbone of any predictive maintenance project and offers three key benefits when it comes to PdM systems, offers Howells. “It improves service profitability, reduce maintenance costs, and increases asset viability. Connected technology allows organizations to listen to their machines and structures, saving time and money when it comes to maintaining day to day operations.”
AI-driven predictive maintenance relies on data gathered via the industrial IoT, which makes it possible to analyze data from thousands of machines remotely, comments Dr. Kampa.
“Predictive maintenance is the gateway drug of industrial IoT,” suggests Kanellos. “Unplanned downtime costs U.S. manufacturers alone around $20 billion a year and probably 80 percent of it could be avoided. Utilities and oil and gas companies are investing because they have loads of assets and maintenance costs are high. These assets—like gas pipelines are often remote. Sending maintenance crews out on boats to check out offshore oil platforms isn’t fun.”
“I believe that PdM will pay a critical role in making IoT one of the most prevalent and relevant markets in the next two decades. For one, the cost of data storage has been shrinking steadily for the past 50-plus years,” says Nogueria. He says no one would even consider the idea of storing terabytes of data on the cloud 20 years ago simply due to the cost. Now, storing that same amount of data can cost roughly $25. “The same can be said about sensors, which have been steadily reducing in price and size. Coupled with the lower data storage and processing costs available today. I believe we have perfect conditions to start taking steps in order to analyze the massive amounts of data we are creating. In addition, the sensor networks that are now so easy to deploy will be instrumental in this effort. In short, PdM may very well give us such a high ROI that will be unthinkable not to join the IoT revolution.”
Kanellos says mining companies and water utilities are investing for the same reasons. “There are approximately 1.2 million miles of water mains in the U.S. and 1.5 million miles of sewer lines. We see water agencies losing 30 percent or more of their water to leaks. If you could predict breaks, leakage could be dramatically reduced. You can also reduce maintenance. San Francisco CA’s water agency expects to reduce employee time for repairs by 9,000 hours through predictive maintenance. It’s also big in manufacturing, but often these companies are tackling other problems. Pharmaceutical makers for instance really look to IoT to do these things like improving product consistency first.”
North America
PdM solutions are in great demand in North America. There are several reasons for this.
Benak says North America has traditionally led the development and maturation of reliability theory, staring with the publication of the RCM methodology in 1978. “The global opportunity for future adoption of predictive maintenance is very much there, however. Regions like the Middle East for example, where asset-intensive industries like oil and gas are densely populated will see widespread adoption of predictive maintenance.”
“There are a few reasons the U.S. will dominate the predictive maintenance market in the coming years. First, as MarketsandMarkets confirms, predictive maintenance is increasingly important in the U.S. given the large amount of solution and service vendors located there. Furthermore, as more manufacturers increase output in the U.S., we are seeing increasing demand for monitoring and predictive technology that allows plant operations to run without disruptions and save organizations money,” shares Howells.
“North America is a huge market for predictive maintenance. There are lots of potential customers there and they are keen to explore new technologies and approaches to predictive maintenance can make them smarter, more efficient, and more productive,” days Dr. Kampa. “That said, the demand for predictive maintenance solutions is growing worldwide. We are doing more and more business in pretty much every corner of the globe.”
Predicting Profit
Predictive maintenance solutions are gaining popularity. As the benefits of IoT and industrial IoT come to fruition, organizations equip their assets with the technology that allows them to monitor and predict when maintenance is needed. This allows businesses to harness technology to reduce downtime, improve efficiency, and increase the bottom line. SWM
Nov2018, Software Magazine
