Everyone is eager to find ways that AI can improve revenue and profit, and that’s helped drive predictive maintenance to be one of the most popular concepts in the industry. This graph from Google Trends shows that queries about predictive maintenance have risen steadily over the past 5 years.

Deloitte recently explored the importance of predictive maintenance and the smart factory, saying “Traditionally, maintenance professionals have combined many techniques, both quantitative and qualitative, in an effort to predict impending failures and mitigate downtime in their manufacturing facilities. Predictive maintenance offers them the potential to optimize maintenance tasks in real time, maximizing the useful life of their equipment while still avoiding disruption to operations.”

But many people are still asking what is predictive maintenance, and what can it do for your process plant?

We share everything you need to know about predictive maintenance to help you make informed decisions for your company.

What is predictive maintenance?

Predictive maintenance, or PdM, is the most advanced approach to managing maintenance within process plants. It’s part of the rise of industry 4.0, big data, and the Internet of Things (IoT), because it uses the newest applications of artificial intelligence (AI), machine learning (ML), and IoT sensors.

You could think of PdM as a subset of predictive analytics. Predictive maintenance uses AI, ML, the Internet of Things, and big data to monitor equipment and check for part failure.

Predictive maintenance is sometimes called condition monitoring, or CM, because it uses IoT data to track the condition of your parts.

What types of maintenance are there?

Predictive maintenance differs from other types of maintenance in many ways. Let’s start by looking at various different types of maintenance, such as:

  • Reactive maintenance, or run-to-failure
  • Preventive maintenance
  • Prescriptive maintenance
  • Predictive maintenance (PdM), or condition monitoring

Reactive maintenance means that after a part has already failed or when an anomaly or incident is already detected, you react to replace or repair the part, or to investigate what caused the anomaly. That’s why it’s also called run-to-failure, because every piece of equipment is used until it fails, and then it’s replaced.

With reactive maintenance, there’s no risk that you’ll waste time maintaining parts that currently don’t need any attention. However, reactive maintenance keeps you constantly on the back foot, stressfully chasing fires. Reactive maintenance pushes up maintenance costs, because a small early repair could extend the lifecycle of a costly part.

Preventive maintenance, also called time-based maintenance, means that you regularly check the condition of every part and make whatever small repairs are needed before equipment failures occur. You create a strict, condition based, proactive maintenance program that ensures that you don’t overlook any corner of the plant. Preventive maintenance can extend the lifecycle of your equipment. However, with preventive maintenance, there is a risk that you might waste time and money on parts that don’t need attention yet, and that you could overlook parts that do need your attention.

Predictive maintenance uses Artificial Intelligence and Machine Learning to direct maintenance management to the parts that need it most. It analyzes big data from industry 4.0 in real time for condition monitoring, to spot the early signs of equipment failures and detect tiny anomalies before they develop into costly incidents. Predictive maintenance helps you save on maintenance costs by addressing only the parts that need attention at the time, instead of using preventive maintenance which involves checking every item whether it needs it or not. Predictive maintenance condition monitoring also guides you to make timely small repairs that extend the lifecycle of your equipment and help reduce downtime.

Prescriptive maintenance takes predictive maintenance a stage further. In addition to condition monitoring to identify the earliest signs of potential part failure, it also recommends what you should do next. Prescriptive maintenance suggests which actions to take to mitigate the anomaly or fix the parts that are showing signs of failure, and anticipates the results of your interventions. In the process manufacturing industry specifically, it’s almost impossible to successfully apply prescriptive maintenance because there are simply too many constantly-changing variables.

How is predictive maintenance related to predictive analytics?

Predictive analytics and predictive maintenance both rest on the application of machine learning to real time big data from IoT sensors and other monitoring systems, but predictive analytics is a broader term.

Predictive maintenance focuses on equipment failures. It uses condition monitoring to track each part individually, spot the earliest signs of failure, and alert you to them. A predictive maintenance program helps prevent you being taken by surprise by sudden part failure.

Predictive analytics encompasses a variety of statistical techniques from data mining, predictive modelling, and machine learning, that analyze current and historical facts to make predictions about future or otherwise unknown events. It’s a wider term which can be applied to different disciplines/verticals: ecommerce, finance, etc. Using analytical tools, predictive analytics can spot small anomalies in production quality, output, part availability, and other ongoing metrics to improve the entire process. It can be used to dive deeper into what is expected to happen in the business, for example predicting fraud, or expected customer demand. 

It’s important to note that although it’s called “predictive” maintenance, it’s almost never possible to predict when a piece of equipment is liable to fail. Process plants simply have too many ever-changing variables, so nothing fails the same way twice. Instead, predictive maintenance helps pick up on early anomalies that are the first signs of failure.

How does predictive maintenance work?

Predictive maintenance begins with IoT sensors on every piece of equipment. The Internet of Things gathers data in real time and sends it back to the central condition monitoring system.

Software like SAM GUARD takes the condition monitoring data showing the regular running of the plant, and automatically analyzes it to establish a baseline for “normal” plant behavior.

The predictive maintenance system then uses ML to quickly process and analyze the flood of new data points coming in all the time from IoT sensors, monitoring equipment condition based on anomalies, and generating an alert when needed.

SAM GUARD’s predictive maintenance solution goes a step further by adding human domain knowledge, which we call human intelligence (HI) to the machine learning system in order to monitor the entire plant. This human enhancement of the machine learning helps the PdM system gain a better understanding of which anomalies are serious and which fall within the range of expected fluctuations. In this way, SAM GUARD’s predictive maintenance solution generates fewer, but more targeted, alerts. When a system produces hundreds of alerts a day, it’s inevitable that engineers will learn to overlook them.

Which industries benefit from predictive maintenance?

Predictive maintenance brings important benefits to many industries, such as health care, energy / utilities, banking, cybersecurity and others. One of the biggest beneficiaries of a predictive maintenance program is the process manufacturing industry, since it has an enormous number of interconnected moving parts, with many vital pieces of equipment, and production can never slow down or pause.

Process industries that benefit from predictive maintenance include:

  • Chemical processing plants
  • Petrochemical plant
  • Oil and gas industries
  • Refineries
  • Cement plants
  • Paper and pulp plants
  • Beverages
  • Pharmaceutical industries

Why does my company need predictive maintenance?

There are many ways that it will help your manufacturing company to make your maintenance predictive instead of using reactive or preventive maintenance. Each plant contains many pieces of equipment, and you rely on each one to perform correctly in order for production to continue smoothly and to avoid downtime.

Replacing a part is expensive, demands a significant amount of time input from your maintenance team, and might require you to stop production while you carry out the replacement. When predictive maintenance issues an early alert about potential part failure, you can deal with the problem while it’s still small and easier and inexpensive to repair. Early condition based intervention stops small problems from snowballing into major issues without your knowledge.

When you get an early condition monitoring warning, you can move the relevant piece of equipment higher up your maintenance management schedule, so that your team investigates it sooner rather than later. In this way, making maintenance predictive often enables them to repair the part in a way that prevents it from imminent failure, saving you from having to replace costly parts on a more frequent basis. 

If you receive an early alert that a part is not performing correctly, and you investigate it and find out that it will need replacing in the very near future, you probably still have some time before it fails entirely. You can plan to replace it at a time that suits both your maintenance team’s schedule and your production schedule. Thanks to the early condition monitoring warning, the issue isn’t so urgent that it has to be dealt with immediately. This is especially true if you’d have to shut down part of your manufacturing plant to carry out the replacement, because that way you can choose the least inconvenient point and reduce overall downtime.

In contrast, if you’re taken by surprise by sudden partial or complete part failure, or signs of imminent failure, you might not have on hand the replacement parts that you need to deal with it. You could have to suspend manufacturing for a day or two, or more, until the parts arrive. Equally, your maintenance team always has a full schedule with many urgent tasks that need their attention. When a part suddenly fails, they’ll have to put all their other responsibilities on hold while they deal with it, which could lead to a crisis developing elsewhere in the plant.

What are the benefits of predictive maintenance for process plants?

The main way that predictive maintenance benefits process plants is by reducing maintenance costs. When you use predictive maintenance and condition monitoring to get ahead of part failure, you’ll be able to repair parts in a way that extends their lifecycle, avoids downtime, and enables you to replace them less frequently. 

By preventing unexpected part failure, predictive maintenance can also help maintain a positive public brand image for process plants. Leaks, explosions, and pollution incidents can harm your reputation, plus most plant owners are truly concerned to do all they can to protect the environment. Gaining an early warning about imminent part failure helps you replace parts before they’re at risk of causing injuries or ecological damage.

Predictive maintenance also saves you time, because you can optimize a condition based maintenance program for greater savings and efficiency than if you’re using preventive maintenance. When maintenance is predictive, it ensures that each part is checked and examined when it’s most likely to be necessary, instead of holding your team to an arbitrary schedule that allocates equal time to every single part.

With predictive maintenance, you can decrease frustration among your employees and raise satisfaction levels for ops and process engineering employees. Predictive maintenance helps you and your managers gain control over your plant. Predictive maintenance means you can stop reacting to sudden emergencies, slash downtime, and gain stability across the organization.

Ultimately, predictive maintenance will help you to increase your overall revenue by reducing maintenance costs, overall downtime, and the costly last-minute shutdowns that you need to implement to replace a part that failed unexpectedly.

When would I use predictive maintenance?

Predictive maintenance has a wide number of use cases. Here are just three examples:

Predictive maintenance for pumps

Pump motors need to run smoothly and on full power to keep your plant efficient. When you make maintenance predictive, you can spot slight changes to vibrations in the pump which could indicate imbalance, due to deposits on the impeller or other parts of the pump. Addressing the vibrations via predictive maintenance enables you to clean the pump on schedule to prevent deposits building up enough to damage the equipment and the concrete structure of the pump.

Predictive maintenance for heat exchangers

Various measurements such as temperature in the heat exchanger and other related parts of the plant can indicate partial blockages in the heat exchanger. Predictive maintenance means that the heat exchanger can be cleaned, and the partial blockage removed. Fixing this on time reduces energy costs and other related issues such as erosion that can take place while it’s blocked. Without predictive maintenance, if these issues are not fixed in time, they could eventually lead to a plant shutdown and lost production.

Predictive maintenance for furnace

Blockages in a furnace can lead to damaged product and are a challenge to the operation.  These can be identified through faulty temperature readings that are the root cause of the blockage, and replace the faulty sensor. This prevents the product from being wasted by being cooked at the wrong temperature, and also the considerable damage that would have been caused by the time required to clean the oven, when it cannot be in production. So by identifying the small problem early, predictive maintenance prevented a larger, more expensive breakdown.

What to look for in a predictive maintenance solution for process plants

Choosing a predictive maintenance solution for process manufacturing requires research and careful consideration. These are the main factors that you should bear in mind when you compare your predictive maintenance options to make your maintenance predictive.

Does the vendor have experience working with process plants?

Process manufacturing plants have needs and pain points that differ from other industries. Process plants have thousands of sensors but sparse historical data, which is very difficult for predictive maintenance solutions that aren’t familiar with the conditions. Make sure your predictive maintenance vendor is familiar with these unique requirements.

How long will it take to implement the solution?

Look for a predictive maintenance vendor that has a solution that can be up and running within a couple of weeks, not months.

How much skill is required to manage the solution?

The most effective predictive maintenance solutions are those that can be run by the plant’s operational team, rather than needing to bring in a data scientist. Find out whether you’ll need ongoing support to learn how to master the predictive maintenance system, how long you’ll need it to continue, and what level of internal resources you’ll need to devote to it on a long-term basis.

Will you need a data scientist to analyze the reports?

Data scientists are a valuable resource, and you have many tasks that you need them to carry out. It isn’t cost effective to have to divert your data science team’s time to handling data reports produced by your predictive maintenance solution.

How many alerts does the system generate?

Alert fatigue is already a serious problem in process plants. Process engineers are very busy people, and even if you have a team dedicated to analyzing the alerts, they should not be bombarded with tons of irrelevant alerts or false alarms, as it will reduce the overall effectiveness of the system. Look for a predictive maintenance solution that keeps down the number of alerts it produces so as not to add to the alert fatigue burden.

What is the false alarm rate?

Process manufacturing plants are plagued by “the boy who cried wolf” syndrome, where process engineers respond to so many false alarms that they learn to ignore the system producing them. Predictive maintenance can bring so much value to your plant that it would be a shame to miss it because of false alarms.

Will the solution assist you in identifying the cause of alerts?

Some predictive maintenance solutions generate structured alerts that draw on a number of relevant datapoints. The combination of these insights can help engineers discover the root cause of the problem and work out how to resolve far more swiftly than if they received a generic alert. For example, a rise in vibrations in a pump while power output remains stable is more illuminating than simply hearing about a rise in vibrations.

Predictive Maintenance Changes the Paradigm

One of the main challenges of adopting predictive maintenance solutions in process plants is that process engineers often expect their job to be reactive; management expects them to put out the proverbial — and sometimes real — fires. By switching to a predictive mode, process engineers need to think like detectives, examine the clues, and investigate the possible scenarios that the evidence suggests. For example, if Indicators show a valve’s position at 90%, together with the temperature of a reformer tube dropping, these early indicators enable predictive maintenance, allowing engineers to take measures to avoid equipment failure and major emergency shutdown.