Using IIoT for predictive maintenance isn't a panacea -

Using IIoT for predictive maintenance isn’t a panacea


The ability to perform predictive maintenance is one of the most often mentioned advantages of industrial internet of things (IIoT) implementations. IIoT systems are not typically implemented primarily as an aid for predictive maintenance – rather, the initial motivation for the deployment of these systems is in order to achieve greater control over production, manufacturing, and power production processes.

Once firms have secured their IoT connectivity and have in place software services for IIoT projects, however, many spot the opportunity to use the data they are already collecting to move to a predictive maintenance model. This shift is usually underpinned by two key concerns. The first is the sheer cost of unplanned downtime, which amounts to at least $50 billion annually across the manufacturing and production industries. The second is that firms are recognising that the way in which they model the failure frequencies of key components is becoming obsolete.

In this article, we’ll look at the reasons why many companies in the aerospace, power generation, and manufacturing sectors are embracing predictive maintenance. Then, we’ll look at some of the challenges involved in doing so.

Failure Profiles and Predictive Maintenance

An increasing body of evidence is showing that traditional maintenance approaches are failing. Recent research by Watson IoT and IBM shows, for instance, that the paradigms of preventative maintenance that most firms have been working with since the 1960s do not reflect the actual failure profiles of their machinery.

This research points out that most preventative maintenance schedules are based on a “bathtub” curve, in which components are more likely to fail during the early and late stages of their deployment, but that in reality only 4% of components fit this failure profile. Instead, this research found that fully 89% of failures occur at random, as shown in the figure below.

Research shows that nearly 90% of failures can occur at random (Source: IBM).

By deploying internet of things systems that are able to alert operators to small-scale faults in key components, or to assess the quality of finished products, these seemingly “random” failure profiles can be analysed. This means that maintenance can take place when components are at risk of failure, rather than wasting resources on unneeded maintenance or risking unplanned downtime.

This kind of predictive maintenance has, according to its supporters, two main benefits. It has been shown to reduce maintenance costs by 10-40%, largely due to the more targeted resource deployment it affords. It also has a value in reducing unplanned downtime, since components are less likely to fail unpredictably.

The Predictive Maintenance Stack

It is certainly possible for some companies in some sectors, working with particular types of machinery, to achieve these benefits. However, predictive maintenance might not be the panacea it seems when it comes to reducing maintenance costs and avoiding unplanned downtime.

In order to see why this is the case, it’s instructive to take a specific example. Hitachi US, in their guide to implementing predictive maintenance, presents just such an example: a company using wind turbines to produce electricity. They point out that a “traditional” maintenance schedule for this type of machinery would be to shut down each turbine at regular intervals, and for engineers to check it for faults.

This type of preventative maintenance is likely to be wasteful – because engineers will end up working on turbines that are not going to fail anytime soon – but also have a limited role in preventing unplanned downtime – because, as we’ve seen, the majority of faults happen at random intervals.

The Hitachi IIoT system utilizes a stack of technologies for implementing preventative maintenance. (Source: Hitachi)

The IIoT solution envisaged by Hitachi as seen above uses a stack of technologies to implement a preventative maintenance system. These include:

  • Partnering with a software-as-a-service (SaaS) provider to conduct pilot programs on machinery,
  • Investing in a technology suite to collect, process, prepare and structure massive amounts of device data
  • Using bespoke (and in some cases AI-driven) algorithms to identify and monitor patterns in this data.
  • Using these patterns to rebuild maintenance workflows around predictive models, including integrating these with ERP and CRM systems to order replacement parts and even dispatch technicians.
  • Integrating the implementation of all of these steps within a company-wide change management approach.

The Challenges

Although an IIoT stack of this kind would likely be able to deliver the benefits that are claimed for predictive maintenance approaches, it’s also apparent that Hitachi’s chosen example – of a wind farm – is an unusually well-suited system to IioT-driven predictive maintenance. In looking at why this is the case, we can begin to see where and when this model is not suitable.

The first, and most obvious, issue with this kind of system is the up-front cost. The type of IIoT system described above relies on the integration of (at least) four complex systems: IoT sensors, cloud storage, analysis software, and reporting systems. For the average firm, putting all these systems in place will represent a large investment in hardware and software, not to mention the operational disruption that will inevitably result from the deployment process.

The wind turbines in Hitachi’s example, being relatively newly-designed machinery, are assumed to possess built-in sensory systems: firms working with legacy machinery will not have this luxury, further increasing deployment costs.

The second issue is that not all production or manufacturing systems can be monitored in this way. Hitachi’s windfarm is a particularly good candidate for preventative maintenance because it is made up of multiple identical components. The number of failure points within each turbine is relatively low, and the power of the system is a consequence of the fact that it can spot when one turbine will fail.

For many firms, this will not be the case, because their manufacturing machinery will consist of single instances of critical machinery. The investment involved in building bespoke algorithms for all of the likely failure modes for thousands of unique components will, in most cases, far outstrip the productivity gains made from them.

The third problem with IIoT systems of this type, and one that Hitachi does not appear to consider, is cybersecurity. Implementing the kind of IIoT system in our example will involve providing each turbine with HTTP connectivity, SSL certificates for IoT access, and other IIoT security tools.

This will greatly increase the cyber threat profile of production systems for the average company, particularly because the average company today has 31% of its assets in the field. That makes effective remote security both immensely important and immensely challenging.

The Bottom Line

In short: predictive maintenance systems based on IIoT connectivity can provide some firms with dramatic productivity and efficiency gains. However, companies looking to leverage these systems should be cautious about the purported advantages they offer.

The cost of implementing a fully-formed IIoT edge system with bespoke, legacy machinery is likely to dwarf the maintenance efficiency gains made. If, on the other hand, your production machinery is modern, repetitive, and predictable enough to make the implementation of such a system cost effective, predictive maintenance will be a natural choice.

Sam Bocetta is a former security analyst for the Department of Defense and current freelance journalist specializing in writing about cybersecurity, technology, and cryptography.


Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.