Predictive Maintenance: The Domain Overview

Discover the importance of asset maintenance for organizations and learn the difference between Reactive Maintenance, Preventive Maintenance, and Predictive Maintenance. You will find out how Machine Learning might be applied to keep your assets in top shape

We live in a time when technology can solve more business issues than ever before. The latest advancements can offer companies a competitive edge and significant improvement to their processes. Today we will cover the maintenance challenge for organizations and the ways to protect valuable assets and save on resources with the help of modern approaches like Preventive and Predictive Maintenance, as well as powerful Machine Learning technology.

Why maintenance is important, and what damage downtime can cause the organization?

Downtimes are one of the key problems every organization needs to learn how to reduce or eliminate entirely. Industrial downtime, in particular, is a constant pain point for manufacturers that needs to be dealt with. The biggest problem with downtimes and outages for organizations is their massive cost. Unexpected failure can cost significantly more than scheduled maintenance for an organization and can take an unspecified amount of time to fix the problem. The haul in production can be a massive factor that reduces productivity or stops the manufacturing processes entirely. There are also hidden costs for downtimes that can damage the reputation of the company in the future. Ineffective maintenance strategies and approaches often become the main factor in industrial downtime. Especially, when organizations are trying to fix the problems only after their appearance. The following infographics will show you the numbers that ineffective maintenance costs organizations across the world.

According to the research by the company Vanson Bourne, which specializes in helping technology vendors understand and talk to enterprise organizations across the globe since 1999, the biggest impacts of unplanned downtimes for companies are:

There are plenty of reasons for unplanned downtime. Vanson Bourne also states that among companies that experienced an outage during the last three years, 45% was caused by hardware malfunction or failure, while 39% of respondents claim that the cause was a software failure.

The bigger the organization, the more damaging the consequences of unplanned downtime can be. On August 8, 2016, Delta Air Lines lost power at its operations center in Atlanta due to an electrical equipment failure. This shut down computers that were used for booking passengers for nearly 5 hours. The damage cost the company nearly $150 million, with almost 1,000 flights canceled on the day of the outage and 1,000 more in the following two days. Delta Air Lines spent additional money on refunds and offering vouchers to the clients for future travel.

It is hard to precisely calculate the downtime losses, because of their complex nature and lasting consequences. However, there is a basic formula that can provide you with a high-level understanding of the potential damage:

LOST REVENUE = (GR/TH) x I x H

Where:

• “GR” is the gross yearly revenue
• “TH” is the total yearly business hours
• “I” stands for percentage impact
• “H” is the number of outage hours

Instead of calculating downtime losses, it is better to prevent them. In the next section of this article, we will talk about the popular ways of reducing or eliminating downtime in the organization.

What to choose: Reactive Maintenance, Preventive Maintenance, or Predictive Maintenance?

Depending on the extent of a system failure, there are multiple ways to solve the situation. Let’s take a closer look at the different types of system maintenance.

Download the White Paper

Reactive Maintenance

This type of maintenance happens after the system or equipment breaks down. All the focus goes to restoring everything as close to “normal” condition as possible. Obviously, emergency fixes are far more expensive compared to planned maintenance. Another reason for the high cost of Reactive Maintenance is the fact that critical outages are most likely to happen during peak hours, disrupting processes when they matter the most. The necessity to fix everything as fast as possible may lead to additional expenses on hiring additional services and buying unplanned equipment.

Benefits:

• Saving on the initial cost of systems
• No effort for planning is needed since the action will be taken only when the system fails

Challenges:

• It is hard to predict the budget expenses since you will realize the full impact of a particular downtime only when it happens
• The system will deteriorate quicker, compared to regular and planned maintenance
• Pinpointing the issue and replacing broken parts, or even waiting for delivery of required parts, will take a lot of time
• The occurrence of the next problem will always be an unpleasant surprise
• The planned upgrades of the system may be pushed or canceled, which can damage your business vision and strategy. This happens because emergency repairs are always prioritized over planned work
• There are no guarantees that the same problems won’t happen again
• An improperly maintained system uses more energy

Best fit for:

• Non-crucial equipment
• Equipment with low repairment costs
• Easily fixed equipment

Not suitable for situations:

• When equipment failure may lead to safety or security issues
• When the repairment process is time-consuming
• When equipment must be available 24/7

Preventive Maintenance

This is a popular approach that companies use to plan regular specific maintenance tasks to keep equipment in a working condition. The key difference from Reactive Maintenance is the fact that you have scheduled repairments. The interval of the checkups may vary dramatically, from once in a few days to annual tune-ups.

Benefits:

• Saving on downtime-related financial losses
• Improved reliability of the system and equipment
• Less emergency maintenance cases

Challenges:

• More expenses on constant maintenance and the risk of overpaying for unnecessary services
• Still having a risk of certain components or equipment breaking down randomly
• The need for constant training of your technicians

Best fit for:

• Equipment with a probability of failure that increases over time or use
• Failure modes that can be prevented (and not increased) with regular maintenance

Not suitable for situations:

• When your equipment may have random failures that are unrelated to maintenance or maintenance frequency

Predictive Maintenance

One of the goals of this type of maintenance is to continuously monitor equipment state in order to predict the potential problems that may lead to costly failures and taking actions to prevent those failures ahead of time. In order to do that, in Predictive Maintenance, the system is being constantly monitored and analysis is being made based on data collected from various sensors. As a result, the functioning of your equipment will be optimized and the costs of repair will be reduced. While it does not completely replace the traditional maintenance approaches, it is a great addition to the overall maintenance process.

Benefits:

• Maintenance costs reduction, sometimes up to 50%
• Up to 90% reduction of critical failure cases
• Downtime related to the actual repair process will also be lower because you could identify certain components that might break down in the first place. The complexity and cost of maintenance decrease as you only replace parts that are causing the failure and not all the parts that might have been broken during the failure
• Compared to Preventive Maintenance, you don’t need to have a stock of spare parts
• For expensive machinery, Predictive Maintenance can save you money on extending the life of particular parts
• Potentially saving lives of employees caused by destructive failures
• Helping to schedule planned repairment shutdowns

Challenges:

• More complex system setup with a higher level of required IT expertise, compared to Preventive Maintenance
• Additional expenses on sensors, data collection, storage, and other related infrastructure

Best fit for:

• Companies that have an ability to collect information about equipment state via various sensors and need failures to be cost-effectively predicted

Not suitable for situations:

• When you don’t have any sensors that provide diagnostic information about your equipment — in that case, it’s better to start with an equipment upgrade

Key Point

Using the Predictive Maintenance approach a company could do equipment maintenance right in time, so you don’t lose money either on complex maintenance when failure already occurred (in the case of Reactive Maintenance) or unnecessary maintenance when no failure was actually going to occur, but you did equipment maintenance anyway (in the case of Preventive Maintenance).

In terms of frequency of maintenance, the three aforementioned methods could be compared as:

Here is a visual representation of what happens in these three methods when the maintenance is being done:

Key Point

While requiring more IT expertise and some additional expenses, Predictive Maintenance is the most effective method that leads to preventing disastrous failures and saving a lot of money. However, at the same time, it is the most sophisticated method and needs a more detailed explanation, compared to quite obvious Reactive and Preventive Maintenance.

How can you implement the Predictive Maintenance system in your organization?

The first thing you need to do is to set up a proper data collection process, that will allow you to get all the necessary information, including data from equipment and sensors, as well as overall system data over a certain period of time. Next, you need to build a system that will make conclusions and predictions based on collected data. The number of parameters that are usually being collected will be impossible to process by humans. There are two most popular ways to go with this system, creating a rule-based AI system or using Machine Learning models. Both approaches have their own advantages and disadvantages.

Rule-based Artificial Intelligence systems

These systems generate pre-defined outcomes, according to rules programmed by human experts. Rule-based systems use “a set of rules” and “a set of facts” to make statements. The best usage of this approach is when you need speedy outputs, have a very high risk of error (however, ML systems can work in those scenarios as well), and don’t plan to implement Machine Learning in the future.

On the drawback side of rule-based AI systems:

• They are very expensive and time-consuming in implementation
• They require complete knowledge of the behavior of equipment
• They don’t perform well with previously unknown faults

Machine Learning systems

The key difference is that ML-based systems define their own set of rules that are based on data inputs. Speaking of data, ML systems require the data collected in the right way and in the specific volume, which we described in one of our previous articles. So, the ML systems could take the correct data and work based on a probabilistic approach. Why is it such a powerful alternative to a rule-based system?

• Machine Learning systems are always “training” and evolve, compared to static rule-based systems with a deterministic approach
• Rule-based systems are limited to certain business logic. When the event occurs outside this business logic, the rule-based system will ignore this event. Machine Learning systems are always aware of a “normal” state of the system. When something abnormal will happen, the ML system will detect it
• Machine Learning systems can detect patterns in abnormal behavior, while rule-based systems will look only for previously set scenarios
• Machine Learning systems can be used in more complex cases, compared to straightforward rule-based systems. Additionally, when there is no simple and obvious way to solve a specific challenge, a Machine Learning system could be the only solution to the problem
• When the situation and the type of data are changing fast, it could be ineffective to constantly rewrite business logic for the rule-based system. The Machine Learning approach will be able to keep up with the changes
• There can be possible conflicts between rules when their number grows exponentially. At some point, you may lack the computing power to control all possible combinations of rules in your system or have the challenge of managing all the rules properly

Key Point

Nowadays, the Machine Learning approach to Predictive Maintenance allows us to build more reliable and agile solutions than Rule-based AI systems.

Conclusion

The maintenance of assets is a major challenge for organizations across the world. The right choice of maintenance strategy can save an enormous amount of money that could be spent on the future growth of an organization.

Each mentioned approach can be used in a certain scenario, however, Predictive Maintenance with the power of Machine Learning technology is probably the most efficient way to deal with this challenge. For sure, you can’t completely eliminate Reactive and Preventive Maintenance from your processes, because some incidents will still happen, and your equipment requires regular maintenance and check-ups. However, investing your time and resources into Predictive Maintenance will help to reduce the number of unfortunate incidents and maintain the equipment in top condition and with lower costs.

In our next article “Predictive Maintenance: Machine learning techniques to solve a PdM problem”, we look into more technical details on developing a Predictive Maintenance solution for your organization, highlighting the Machine Learning approaches, in particular. We also discuss deep learning techniques and an ML-powered predictive maintenance solution. If you are interested in building a modern system for your company, read our next articles for more insights based on our experience!

We at Intelliarts AI love to help companies to solve the challenges with data strategy design and implementation, so if you have any questions related to ML pipelines in particular or other areas of Data Science — feel free to reach out.