ISA Interchange

Welcome to the official blog of the International Society of Automation (ISA).

This blog covers numerous topics on industrial automation such as operations & management, continuous & batch processing, connectivity, manufacturing & machine control, and Industry 4.0.

All Posts

How to Implement Predictive Maintenance

Predictive Maintenance is a concept that has recently gained a lot of traction as part of digital transformation strategies and the application of Industry 4.0 technologies. Like many applications, executing them is easier said than done. So, what exactly is predictive maintenance? How is it implemented? What are the benefits and risks? 

Types of Maintenance 

To understand Predictive Maintenance, it is helpful to understand other types of maintenance and reliability strategies. A quick overview is provided below in order of maintenance maturity. As companies enhance their reliability, the journey often follows these steps. In all cases, a maintenance strategy should be technically valid and economically viable. The purpose of maintenance and reliability is to ensure and restore normal operating conditions. 

  • Reactive Maintenance (RM) – Other terms for this are run-to-failure or plan-to-repair. The strategy is to run an asset until it literally fails and must be replaced. While it is a valid maintenance type, it is commonly the result of a lack of any other maintenance strategy. This is best for low-cost, less-critical assets.
  • Preventative Maintenance (PM) – This is the most common strategy, whereby maintenance is performed based on a predetermined interval, like running hours or calendar days. This is similar to changing the oil in your car every 3,000 miles. While it is easy to implement, it often results in the over-maintenance of assets, which is costly and introduces risk.
  • Condition-Based Maintenance (CBM) – Within the maintenance and reliability community, condition-based maintenance and predictive maintenance are understood to mean the same thing. For the purposes of this article, CBM refers to a strategy where maintenance is performed based on a defined set of operating conditions or states. CBM is most powerful when several conditions are met at the same time. This strategy should be used for most assets.
  • Predictive Maintenance (PdM) – Although this differs from the generally accepted definition, PdM uses machine learning techniques to determine what the future state of an asset should be. If the actual machine state differs from what was predicted, maintenance may be necessary. As more data points are available and machine learning algorithms improve, maintenance becomes much more targeted, and assets can be operated more efficiently. As expected, due to cost, PdM should be limited to high-value, high-critical assets.
  • Prescriptive Maintenance (RxM) – This maintenance strategy continues to be an emerging field and not commonly used. It leverages PdM strategies to not only predict the future state of an asset, it also provides guidance on how to operate an asset until some period in the future is reached. For example, if a shutdown will occur in 4 weeks, but the asset will likely fail in 3, RxM offers how to run the asset until the shutdown occurs. 

New Technologies 

While this article is about predictive maintenance, it is important to understand that PdM is not a one-size-fits-all strategy. New technology allows PdM to be more economically viable (the cost is much less than in the past). It also reduces the cost of other maintenance strategies. How technology is implemented is much more important than the technology itself. Once it has been determined which assets will implement a predictive maintenance strategy, I suggest the following steps below. 

Leverage Your Existing Data 

The next step is to identify data that is currently being collected and inventory the current monitoring technology. This will determine what opportunities exist and what the so-called “low hanging fruit” may be. 

A good place to start is to simply add analytic capabilities to the existing data. There are several technology companies that have done this for many years. By simply using historical data to develop models, the ability to analyze and predict is well-established. In the past few years, the cost to deploy these systems has come down and the ability to analyze more data has increased. You are no longer limited to critical and high-value systems. 

Add Some Sensors 

One of the tenets of big data, machine learning, and artificial intelligence is more data. A great following step is adding wireless sensors. This is a popular option for enhancing your reliability. With the advancement of battery technology, many of the sensors will last up to 10 years. They are easy to install and require minimal configuration. Many will communicate via a local gateway that transmits the data to a cloud application. End users are alerted to potential problems via email or a mobile app. 

Wireless vibration sensors are the most common, due to cost and ease of installation. Most will provide triaxial vibration and temperature measurements. Along with alerts to potential problems, the software enables vibration technicians to further analyze the data, including Fast Fourier Transforms (FFTs). Ultrasonic, infrared, and even lube oil sensors are gaining popularity too. 

It should be noted that not all wireless sensors and platforms are created equal. There are many offerings in the market that just make measurements and transmit data. However, there is not any kind of diagnostic capability. While they are an enhancement over traditional route-base vibration measurement, they are not truly a predictive maintenance solution. 

Combine Process and Reliability Data 

The final step is to combine the disparate data in your plant. Vibration, ultrasonic, and process data often reside in different locations. To enhance the predictive models, combining these data sources will be required. Consuming both types of data in a common dataset will provide additional insights in the observations. 

How the data is collected, analyzed, and presented takes on many forms. Using your current automation and control company is one of them. There are several well-established companies that support both process and reliability technologies. If you already have a relationship with them, the integration is simple and straightforward. The downside to using your current platform is that these solutions might be proprietary, and you may not be able to use the data elsewhere in your organization. 

Akin to the wireless vibration sensor companies, there are many new machine learning and artificial intelligence companies that have developed platforms specifically for predictive maintenance. These become a second option. Their solutions provide connectors to the most common sensors and process historians. Some of them also offer the sensors which makes the implementations even easier. The benefit to this path is that technology companies have expertise within machine learning and artificial intelligence. Moreover, they are a bespoke solution which leads to a faster time to value. 

Another option is to utilize a Unified Namespace (UNS) for combining the process and reliability data. A UNS is a design methodology that becomes the single course of truth and the structure of a business. While the concept of a UNS could be an article on its own, in this case the underlying technology will connect to the data sources and provide an access point for other systems. The benefit is that many different systems can consume this data. This provides the option of using several different ML/AI companies for different applications, including predictive maintenance. The challenge is that architecting and building a UNS is not as well understood, so the time to value may be longer than other paths. 

Conclusion 

The first step to implement predictive maintenance is to determine if PdM is the right strategy for an asset. 

The second step is to add analytic capabilities to currently collected data. This could mean simply installing software and building asset and process models. 

To enhance those data models, the third step is to add sensors for additional data sources. This means adding more of the same kind of sensors as well as adding different types of sensors. Wireless vibration sensors make for a great choice. 

Finally, integrate the reliability data with process data. Using your existing automation technology, adding a technology platform that specializes in PdM, or developing a UNS, are all great options.

David Schultz
David Schultz
David Schultz is the owner of G5 Consulting and a Senior Consultant for Matrix Technologies. He has 25 years of automation and process control experience. He currently works with companies to help them develop and execute their digital transformation strategies. Along with ISA he is a member of Society of Maintenance & Reliability Professionals (SMRP) and Project Management Institute (PMI).

Related Posts

OEE and Profit Forecasting

Low-Latency Analysis One of the promises of Industry 4.0 (I4.0) is the ability to connect everything and ...
Ryan Kershaw May 24, 2022 5:30:00 AM

Interaction and Decouple Control

Nowadays, all process industry equipment is furnished with controllers that allow the equipment to operat...
Abhishek Sharma May 20, 2022 5:30:00 AM

Technology Localization and its Role in Economic Development

What is Technology Transfer? Technology transfer is a process in which technology or knowledge developed ...
Elena Rios May 17, 2022 5:30:00 AM