Predictive Maintenance is critical in industry 4.0.
Executive Summary
Predictive maintenance is not a new concept, but as the cost of industrial IoT sensors and infrastructure continues to come down, cloud-based
predictive maintenance solutions are becoming an intriguing option for many companies in the manufacturing and industrial sectors.
In this report, we compare the costs and efficiency of different maintenance strategies, including reactive, preventive and pro-active strategies. Reactive maintenance is shown to be the least efficient, most costly option, followed by planned maintenance, then proactive maintenance, with predictive maintenance being the most efficient, rated by Deloitte to deliver over 90% OEE. Proactive maintenance, in which equipment is serviced based on its operating condition by highly knowledgeable maintenance personnel, is efficient but challenging due to its requirement of highly skilled workers. In contrast, predictive maintenance is user-friendly and does not place high demands on personnel knowledge or experience.
The options in the market for predictive maintenance solutions is shown to be a spectrum, ranging from simple condition-based anomaly detectors to advanced AI-enabled systems with digital twin and simulation functionality. Customers are truly able to find a solution that fits their needs and budget.
Customers seeking predictive maintenance solutions must choose between an on-premises solution or a cloud-based platform. Experts agree that cloud- based platforms are the superior choice, providing better computing power, usability, and scalability at a lower cost, especially for customers without a well-characterized IT infrastructure in-house.
Lastly, this report examines emerging business models enabled by IIoT and predictive maintenance, namely the example of Kaeser Compressors, which used predictive maintenance data to greatly improve service and maintenance, successfully reducing customer unscheduled downtime by 60%.
Customers interested in developing predictive maintenance capability are advised to focus on starting with simpler IoT applications such as condition monitoring. In addition, many vendors can run pilot projects to help prove the viability of these systems.
The industrial internet of things is shown to greatly enhance the functionality and opportunities afforded by predictive analytics technology as applied to equipment maintenance.
While IIoT-enabled predictive analytics and asset management has been in use for years by large, powerful companies like GE, ABB and Boeing, smaller enterprises have been slower to adopt the technology. However, the tools required to execute IIoT have become increasingly democratized in recent years, opening up realistic avenues to ROI for smaller companies.
“When it comes to predictive maintenance solutions, the difference that I see between large multinational companies and smaller ones is that I think the motivation to use these solutions is slightly different,” explained Christoph Inauen, vice-president of strategy at Siemens Mindsphere.
“When I think about the multinationals, for them it’s also about being able to compare how their assets perform, for example, across multiple locations or even regions to find out which ones are the, call it, the low performers and then, you know, act upon that.
For smaller companies, it’s more about making sure that the assets that they have bought and that they operate perform perfectly so that they get the most value out of those assets in terms of workload or utilization. So, they want to make sure that it’s always up and running, there’s no outages, and so on. So, predictive maintenance is equally important to very large companies as well as small.”
In the balance of this research report, we’ll examine how today’s IIoT market enables valuable predictive maintenance solutions, and how manufacturing customers can leverage the technology for themselves.
USING THE IIOT TO ENHANCE PREDICTIVE MAINTENANCE
“In this world nothing can be said to be certain except death and taxes.”
Benjamin Franklin wrote this in a 1789 letter. If Franklin had been an industrial engineer, not just a scientist, politician, author, freemason and inventor, he probably would have included “maintenance” in that set.
No piece of machinery can operate at 100% efficiency 100% of the time. Maintenance has always been part of all industrial processes, and probably always will be. What is changing, though, is the technology and methods we use to conduct maintenance.
Today, most maintenance is done on a set schedule, with emergency repairs executed in the event of breakdown. However, this system is inefficient, and results in unplanned downtime, high repair costs and lost productivity. Fortunately, the growing accessibility of low-cost sensors, software and advanced computing allows machines to predict failure by the data signatures that indicate it up to weeks in advance. This enables a radical advance in the efficiency and life expectancy of industrial assets and drastically lower cost maintenance.
Predictive maintenance is not a new concept, but the increasing use of the internet of things (IoT) has unlocked much greater capability for the technology to find applications across industry.
So, how does the IIoT add value to predictive maintenance solutions? And why should you get on board?
While IIoT-enabled predictive analytics and asset management has been in use for years by large, powerful companies like GE, ABB and Boeing, smaller enterprises have been slower to adopt the technology. However, the tools required to execute IIoT have become increasingly democratized in recent years, opening up realistic avenues to ROI for smaller companies.
“When it comes to predictive maintenance solutions, the difference that I see between large multinational companies and smaller ones is that I think the motivation to use these solutions is slightly different,” explained Christoph Inauen, vice-president of strategy at Siemens Mindsphere.
“When I think about the multinationals, for them it’s also about being able to compare how their assets perform, for example, across multiple locations or even regions to find out which ones are the, call it, the low performers and then, you know, act upon that.
For smaller companies, it’s more about making sure that the assets that they have bought and that they operate perform perfectly so that they get the most value out of those assets in terms of workload or utilization. So, they want to make sure that it’s always up and running, there’s no outages, and so on. So, predictive maintenance is equally important to very large companies as well as small.”
In the balance of this research report, we’ll examine how today’s IIoT market enables valuable predictive maintenance solutions, and how manufacturing customers can leverage the technology for themselves.
PREDICTIVE VS. OTHER MAINTENANCE STRATEGIES
Figure 1. Maintenance strategy continuum
In a perfect world, equipment wouldn’t wear and break down. Unfortunately, maintenance is a routine part of every equipment life cycle. There are three main approaches to maintenance, involving varying levels of planning, costs, and efficiency. Below, we’ll briefly describe three other common maintenance strategies that all industrial engineers and millwrights know well.
Matt , senior portfolio marketing manager, asset performance management at Industrial software provider AVEVA, described the transition from traditional maintenance strategies to predictive maintenance as part of a maturity process.
“We look at the maintenance strategy as an evolutionary process. The investment in technology to support the highest level of reliability and availability of an asset is based on the balancing of risk, cost and performance. If you have an asset that is very low cost to repair and that does not have a significant impact on business if it goes down, you may run that asset until
it breaks down and then just swap it out. On the other hand, you may have an asset that’s very critical to your process and to your business, and that’s where we deploy more advanced technologies like predictive analytics.” The
key is to partner with a technology provider that can engage across the entire maintenance maturity process and assist the business in operationalizing their maintenance strategy no matter where the organization is currently at in their maintenance maturation journey.
REACTIVE MAINTENANCE: RUN-TO-FAIL
In this strategy, maintenance staff prepare to fix breakdowns by keeping an inventory of common spare parts. When equipment does fail, it becomes a mad scramble to identify the problem, repair the equipment and get it back up and running. Maintenance staff must always be on-hand and ready to deal
with an emergency job. Besides stopped production, equipment failure can also result in damage to other machines, tools or parts. In addition, it’s impossible for maintenance personnel to effectively manage their time.
“In all of these predictive applications, none of them are ever going to prevent a machine from breaking down,” said Robert Golightly, marketing manager, asset performance management at AspenTech. “What they are going to be able to do is to shift downtime from an unplanned mode to a planned mode. It’s been shown that emergency maintenance can be five to ten times more expensive than maintenance activities done on a planned basis.”
According to Deloitte analysis, reactive maintenance results in less than 50% OEE. For comparison, Deloitte ranks predictive maintenance at >90% OEE. The same cost study places data-driven predictive maintenance at $9 per horsepower per year. However, this study evaluated non-IIoT-connected predictive maintenance, in which data readings are periodically collected at the machine.
PREVENTIVE MAINTENANCE: SCHEDULED ACTIVITIES
Regular, scheduled shutdowns are routine in manufacturing operations. No matter which primary strategy is in use, planned maintenance will always be a part of the equation. Under this strategy, equipment is serviced at predetermined time intervals, repairing or replacing damaged equipment before it breaks down. As every maintenance specialist knows, this strategy reduces breakdowns at the cost of some redundancy. Parts replaced before failure may seem to be perfectly fine. Per the Deloitte analysis mentioned above, preventive maintenance supports 50-75% OEE.
PROACTIVE MAINTENANCE: IDENTIFYING PROBLEMS
While reactive maintenance is focused on identifying the cause of a breakdown of equipment, proactive maintenance is focused on identifying those causes before the breakdown occurs. For example, if a machinist reports more chatter and vibration on a certain tool, maintenance personnel can start to work on the problem before it causes a machine failure or a broken tool.
Like preventive maintenance, this strategy works best in concert with other strategies, such as planned maintenance. The disadvantage of proactive maintenance as a primary maintenance staff is that it requires highly skilled personnel to correctly diagnose and address problems based on the warning signs. In addition, the maintenance department must be given sufficient power and agency to make the decisions required, such as shutting down a machine that appears to be running fine; ordering parts that may not be apparently needed, and other executive calls. Depending on the company culture, this may not be easy. Deloitte analysis shows this strategy delivers 75-90% OEE.
THE PREDICTIVE MAINTENANCE SOLUTION COMPLEXITY SPECTRUM
Predictive maintenance was possible before cloud computing and the IIoT. In fact, many of today’s predictive analytics solutions can be implemented as on- premises solutions not connected to the internet. However, the benefits of the cloud, including lower cost of ownership, high-power computing resources and IoT connectivity capabilities, have enabled more advanced AI-based systems that leverage machine learning to generate higher-value analytics than more basic, model-based systems.
Today’s predictive maintenance solutions market follows a curve or spectrum. At the low end are simple, rule-based systems which are computationally and monetarily inexpensive. At the high end are advanced AI-enabled systems which can crunch larger sets of data and deliver high-fidelity insights, often with automation features (such as placing work orders, for example.)
Patrick Crampton-Thomas is global head of digital products and asset management at SAP. He explained that while advanced technology is what’s exciting in today’s IIoT market, it isn’t needed in every plant.
“Obviously, the more intelligence you can apply, the better decisions you can make, resulting in a better outcome, lower costs, better service, et cetera. So, the more sophisticated you are, sure—it’s better. But does it make sense?
Should a company realistically invest a million dollars just to save that extra hundred bucks?”
According to Golightly, higher-end systems are more useful in cases where earlier warning is needed. In addition, these systems can save maintenance time by precisely pinpointing the issue, rather than simply flagging an anomaly state.
“A model-based system can basically be an anomaly detector. It creates an alert if something is different. Then it either has a rule base or an expert or a combination of those two that interprets the anomaly and tells what that anomaly means. The fidelity of that is quite poor,” said Golightly.
“With a model-based system, resolving one anomaly alarm is pretty straightforward. But when you have ten or twenty of them going off at once, which isn’t that uncommon in production, now you have a different kind of problem. Now you have a prioritization problem,” Golightly explained.
Aspen Mtell is an asset performance management system that uses temporal multivariate analysis to deliver prescriptive analytics. “In comparison, Mtell signatures are tied to a particular failure. So, let’s use a compressor as an example. The Mtell signature will tell you which valve, which piston, which cylinder—the component level of what it identifies as the looming problem.
That’s radically different than an anomaly detector. Once an anomaly alert has been raised, you then have to have the experts figure out what that means.
So that not only has resource implications, bill implications, but it has time implications to complete that workflow so that you can react to the alert.”
In addition, accurate models of normal operation of an asset can be costly and time-consuming to create. Depending on the level of sophistication of the model and the process being modeled, the development of an application can take months. In contrast, a machine learning algorithm can intake historical data and begin developing its analysis in a matter of days or weeks, without requiring human expertise.
On the other hand, there are advantages choosing these low-end systems for some customers. For example, a small or medium-size enterprise that can’t justify a million-dollar outlay for a high-end predictive analytics system, but they can still realize an ROI if a simpler system can help keep a machine from overheating. So, to them, predictive maintenance may simply mean a system that sends a mobile device notification to the maintenance technician when a temperature sensor reading exceeds a certain threshold.
Low-end systems are less expensive, both in terms of cost and in computational requirements. The lower computational and data bandwidth requirements can save money on cloud-hosted systems, and are easier to build on-premises, for companies that choose to do so.
AVEVA offers both a model-based system and a more comprehensive solution to accommodate different customer needs on the curve.
“We have two kinds of analytics solutions on the portfolio: one is our condition management software, which is essentially an advanced set of rules that customers can put together, and the other is called Predictive Asset Analytics,” explained Newton.
” Our condition management software will look at a variety of data sources and enables the customer to easily put together a rules engine that performs condition-based logic. This is a solution for entry-level predictive analytics and works well when the faults and subsequent actions are well known.
When we need to look at very large volumes of data and utilize digital twin technology, modelling how physical equipment and assets should operate, that’s where we deploy Predictive Asset Analytics. We leverage existing customer data to build out a model that describes what good behavior of that asset looks like. From there, the software monitors the asset in real time using those models and issues alerts on any type of deviation from the model.”
Siemens Mindsphere also offers multiple solutions along the spectrum. “We have on one hand something we call an analyze predict package. That is a pre-combined package that has multiple, let’s say, capabilities from a
solution point of view,” said Inauen. “On one hand, we have in here a predictive learning solution environment that allows data scientists, for example, to be able to build algorithms, employ algorithms and build solutions. On the other hand, we also have a dedicated data science workbench where we actually use the Zeppelin notebook environment that is quite common in the market for data scientists to use. This allows users to build the model, to train the model, visualize the model, support the model. This is the high-level offering in our portfolio when it comes to data holdings and to predictive analytics in particular.”
“Condition-based monitoring is a complete spectrum, and on the left-hand side it starts with basic sensor-based condition monitoring: ‘hey, just tell me when this equipment hits 99 degrees, then I’ll go service it.’ There’s nothing rocket science about that. It’s so simple and quick to adopt,” said Crampton- Thomas. “In some assets, that’s all you need.”
As we move along the spectrum, as Crampton-Thomas described, we start to add in technology such as predictive algorithms, advanced data science, and machine learning.
“To go to the extreme on the right-hand side of the spectrum, where it isn’t just predictive and machine learning things like algorithms; you really get into material science,” explained Crampton-Thomas. “An example of this could include an extremely high-fidelity digital twin of the asset, simulating the behavior. So, if it’s a pump, for example, then you could start to simulate corrosion and things like that, which could inform maintenance decisions. This could be used to prolong the life of an asset, especially if it’s three miles underground or in the sea, for example. One opportunity could be to do fluid dynamic analysis to extend the life of the asset by reducing the flow, and by less corrosion or abrasion.
The way I see it is that you have to understand the asset. Pinpoint what you really need and apply the right level of data science to it. Often, simple is fine. You don’t have to overcook it. For some high-value critical assets, you move up the curve. You may start doing predictive analytics, you may start using machine learning. And then, of course, for the really critical assets, difficult ones to get to, et cetera, you start using more and more simulation and machine and data and material science approaches.”
ON-PREMISES OR CLOUD-BASED?
Internet connectivity is part of the Industry 4.0 megatrend, but some customers still have reservations about the security, cost, utility and flexibility of cloud platforms from companies like ABB, GE, Siemens Mindsphere and Microsoft Azure. In a nutshell, our research shows that on-premises solutions can be viable where there is a developed IT infrastructure, but cloud-based solutions are more cost effective and scalable in most cases.
In many cases, asset management solutions providers (Including AVEVA and Aspen Tech) design their solutions to be as flexible as possible, allowing the customer to use the product on a cloud platform or on-premises, as needed.
LOWER COST
Offloading the cost of the infrastructure, hardware, IT and OT support, and maintenance of an on-premise server lowers the total cost of ownership of an IIoT system at almost every scale. In addition, cloud-based systems are instantly scalable, eliminating the need to build or buy hardware to adapt to changing needs.
According to Inauen, cloud platforms effectively spread the costs of an IoT solution across all customers, which is more efficient than one customer building the network and computing infrastructure in-house.
“What we’re seeing in the IIoT space is that, at the end of the day, the cloud options are more cost-effective,” said Inauen. “In most cases, this typically has to do with the fact that if you want to do something on-prem, you have to buy infrastructure and hardware such as servers, et cetera. Someone has to maintain those servers. Then, you load the software, and then operational
aspects kick in as well. So, all of that is basically, let’s call it, hidden or additional cost.”
The usability and computational power of cloud services also contribute to cost savings on the user side.
USABILITY
One of the effects of the growing trend toward connectivity as a standard feature in industrial equipment is that data analytics is becoming more user- friendly. Historically, interpreting data and generating actionable insights required data scientists and production process experts. Today, analytics applications are designed to be usable by non-experts.
“You don’t have to be a data scientist. Our interface is all graphical, there’s no programming required, you don’t need to be an engineer to get started with it,” said Newton. “There is a little bit of getting up to speed on obviously how to work the software, how to train models, but it’s incredibly user friendly. We typically have people up and running building models within a few hours.”
Crampton-Thomas used the analogy of popular social networking apps to explain the value of cloud-based IIoT platforms. “When we want to use LinkedIn and Facebook, we just use it. We don’t worry about installing it, we don’t worry about having to configure it and set it up. And in fact, the
configuration that we do in those applications is really user driven. We add our own photos and so on,” he explained.
“And it’s exactly the same in the context of business applications. The idea of a cloud application, it’s not that you can just host an off-premise app. This isn’t just hosting. That’s been around for years. I think the true value of cloud
is when you combine it with how the apps are built, because the apps have to be built differently, right? You have to be able to onboard yourself. The apps have to be simpler to use, more intuitive. I can’t bring in IT to help to configure them, so then they have to come with much easier to use configuration utility so that I can self-configure the apps,” he said.
In short, because part of the value proposition of cloud platforms is that it lessens the need for dedicated IT personnel, it would be counterproductive for a platform to require IT personnel to use it effectively. Therefore, the majority of IIoT cloud platforms on the market have a strong incentive to make their services as user-friendly as possible.
COMPUTATION POWER
The algorithms that process machine data to perform predictive analytics can be highly CPU-intensive, especially when processing large volumes of data or computing complex machine learning algorithms. With on-premises solutions, the IT department is responsible for scaling and maintaining the computing and networking hardware, which is costly and requires in-house expertise.
“In the cloud, you open up almost infinite computing capability,” said Newton. “You’ve got access to processors, storage, and RAM resources available basically at the flick of a switch. This allows you to expand your solution set and the number of assets pretty quickly. you’ve got quite a bit of power on tap.”
Inauen agreed. “The cloud can easily scale out as more capacity is needed, versus an on-prem system that must be designed to deal with peaks, and that’s certainly going to become very expensive. On the other hand, the cloud world is a more elegant way of dealing with that problem, because you simply sign up for capacity on a permanent basis. You can deal with peaks easily and then scale back again on the capacity if you don’t need it.”
SECURITY
Some customers think that on-premises solutions, especially those that are not connected to the internet, are more secure than cloud-based solutions because no control of sensitive data is handed over to a third party. However, this is not the case. Cyber-physical attacks can occur on any system. The weakest link in most industrial cybersecurity systems is the ‘cyber hygiene’ of employees. Can you be sure that no employee in your facility will not click a suspicious link in an email? Will not plug in a USB stick without knowing where it came from? In addition, does your in-house IT staff have the resources and capability to protect against attacks?
In contrast, while a cloud solution may seem more vulnerable because sensitive data is hosted on a server somewhere in out in the world, the companies hosting your data have much stronger cybersecurity and much greater resources to protect assets.
To put it bluntly, who would you rather try to hack: a company such as Cisco Systems? Or the “Podunk Plastics” company network?
THE COMPATIBILITY QUESTION: MACHINES, PLATFORMS AND APPLICATIONS
There are several IIoT platforms available on the market, such as Siemens Mindsphere, PTC Thingworx and GE Predix. Some platforms include applications, and some offer compatibility with third-party applications, allowing users to run, for example, a predictive analytics application from AVEVA on a platform made by Siemens. However, it’s important for users to research these compatibilities before committing to a platform provider or a software ecosystem.
“That’s part of what we try to do with the entire asset performance management portfolio,” said Newton. “Customers may have, for example, a historian software in place today, and that’s where we can deploy Predictive Asset Analytics and analyze the data that’s in that historian. They may have a variety of software that they’re already invested in, and our solutions can plug right in and help them get started with predictive analytics pretty quickly.”
I asked Inauen about application compatibility on Siemens Mindsphere. Mindsphere has open APIs, allowing third party application developers and their customers to build compatibility with the platform more effectively.
“MindSphere basically exposes all of its capability through well-defined APIs, those APIs are documented on a webpage,” said Inauen. “Anyone can access them and can easily view what those API’s can do. So, some of these API’s for example, can push data out into another environment and then third-party solution environments can consume that data, and build their models there and run predictive analytics, and push feedback downwards into MindSphere. So, through APIs, because it’s all defined in an open way, this is certainly possible.”
“These days, there is a huge focus on interoperability between different systems within IIoT. There’s a lot of convergence. For example, I could work my system with OSIsoft PI, I could work with SAP platforms or I can use different platforms,” said Crampton-Thomas. “I think this whole world, in a way, has to get to interoperability. And I think it is moving that way. But there’s probably more journey to go.”
NEW OPPORTUNITIES IN THE FUTURE OF PREDICTIVE MAINTENANCE
So, we’ve shown how cloud platforms lower the cost barrier of access to computing resources by
replacing tangible products with a service. Interestingly, this new business model is finding its way into industrial equipment market, as well.
Kaeser Compressor is a popular example of
this paradigm. By leveraging IIoT-connected predictive maintenance, the company was able to cut customers’ downtime by 60%.
KAESER COMPRESSORS
Realizing the potential of predictive maintenance to support better servicing to customer equipment, Kaeser equipped its products with sensors to capture environmental and performance data, including temperatures, humidity and vibration.
This data is transmitted to a central system that conducts real-time predictive analytics, allowing Kaeser to identify which parts are prone to failure,
and replacing those parts during regular service calls, helping to prevent unplanned failures.
This application of IIoT technology has resulted in a 60% reduction in unscheduled downtime and emergency service visits, saving money for both the company itself and its customers.
In essence, Kaeser customers aren’t paying for a piece of equipment.
They’re paying for compressed air as a service. This simplifies the customer experience and adds significant value for customers who now do not have to think about compressor maintenance.
According to Inauen, this business model is likely to grow across the industrial equipment market.
“This new business model for Kaeser Compressors really has a lot of benefits, and I see companies across industries that are absolutely innovative in the space. This is the way they’re trying to differentiate, and something like predictive maintenance absolutely helps them,” Inauen said.
“I’m sure that these types of outcome-based business models will increase over time. This is definitely something that will come and it’s already happening in some places,” agreed Crampton-Thomas.
He continued: “So, the ability to collaborate around a given machine is enabled much easier through a cloud environment, enabling a shift of responsibility. The second aspect of it is not just the collaboration; it’s also about maintaining the service level, and the level of connectivity required between the service provider and the machine. Because if that machine stops or the service provider stops the production line, they will have an extremely unhappy customer. So, the customer experience becomes increasingly important, and these new service providers will be ultimately responsible
for that customer experience because they are offering a service instead of a product.”
