Manufacturing and utilities companies today usually have no shortage of data. Thanks to the proliferation of sensors and the internet of things, there is an almost unstoppable flood of readings. What remains challenging, though, is to turn data into useful information that can fuel decisions. In the case of a preventive maintenance program, data can tell you when it is the right time to replace a component or when an asset is going to break down.

Strategic importance of data

Data has become a strategic asset, which is why it is so important for companies to have a vision on how to organize the growing volume of data. Just like any machine or physical asset, data needs to be managed from the day it is created to the day when it is no longer in use. 

To make sense of data, maintenance professionals increasingly rely on AI and machine learning to unearth critical information about their assets. Based on machine learning techniques, they can find patterns and learn about the remaining useful life (RUL), flag irregular behavior and make recommendations for mitigation or maintenance after failure. As such, data can help to prevent unexpected asset downtime, lower maintenance costs (less frequent, only when needed) and extend equipment lifetime.

Failure data

Other than process and business data, asset managers also need to look at failure data: data about maintenance works, operating parameters, malfunctioning and incidents. Failure data is crucial for implementing a defect elimination process and for understanding and maintaining productivity, efficiency and profitability. After all, you need to learn from past failures and see when and how they occur.

However, many organizations find it difficult to consistently collect the required failure data or they find that failure data is just not available with their assets. Generally speaking, there are two situations where a lack of failure data may occur.

1. There are no or hardly any incidents or failures.

When your assets show no failures, there is nothing to retrieve failure data from. Scarcity of problems may seem like a good thing. At least, the asset maintenance team seems to be successful at their job or seems to have things under control, because no incidents are reported. 

However, it’s not because things are scarce, that they cannot cause trouble in the future. A lack of historical asset failure data makes it harder to understand and foresee future incidents. If your assets or devices hardly present any failures, it is difficult to fully understand and foresee the possible failures that you must avoid. To draw an analogy from the medical world, global pandemics do not happen so frequently. And as the recent COVID-19 crisis has shown, there is relatively little historical information on how to handle a pandemic well. And yet, the human and economic consequences of COVID-19 today are dramatic. 

At the same time, a strange contradiction is starting to appear when failures are scarce, in that the closer a maintenance worker comes to the ideal scenario of incident-free operation, the more his job is being questioned. Maintenance workers may be satisfied with faultless assets, but they also may be bored out, wondering whether their job is actually needed in keeping the assets up and running. Even worse, company management may have the same idea and decide to reduce maintenance efforts and manpower. Contradictorily, this may lead to a riskier situation where assets are inadequately monitored. 

2. Failures do occur, but there is no knowledge about the failure cause.

Lack of failure data can also mean that there are enough asset failures, but there is no knowledge about the cause and effect of the failure. Insufficient data can lead to trial and error, speculation, theorization, or assumptions on how to prevent incidents. That’s an unreliable way to take care of assets. 

A lack of understanding can be frustrating for the maintenance team. Also here, company management may start to question the effectiveness of the maintenance team.

Why do companies lack failure data?

With today’s massive generation of data, it may be hard to believe, but there are many reasons why asset-rich organizations lack failure data. 

Let’s look at a few.

• With new devices or assets, the chances of failures are much smaller. 
• The asset is operating in new conditions (temperature, pressure, dust, etc.) from which there is no data. For example, a production line may have been tested in northern Europe, but that data may be insufficient when the line is deployed in a hotter, African climate. 
• The asset is used differently. The asset may be configured to produce higher numbers or with a higher intensity, making previous data sets incomplete.
• Stricter regulations or safety standards may apply, making the old data irrelevant.

Even with older assets, failure data may be scarce:

• An asset may be so robust and reliable that incidents are infrequent. 
• The maintenance work is excellent, thorough and frequent. 
• The asset is infrequently used, providing less cause for failures.  
• The conditions in which the asset is used are not critical.  
• The asset is very rare, making it difficult to collect large amounts of data. Just think of space technology or custom-made machines for niche use cases.

Sometimes, the conditions in which data needs to be collected are difficult, making data collection very expensive. A good example is wind turbines, which are typically hard to access. Defects may also occur outside the production environment (at third parties or in the field), making access to the assets difficult.

What should you do when you have no failure data?

Failure data are crucial for your preventive maintenance program. However, if you have no failure data yet, there are a few strategies you can follow.

#1 Generate additional data

If you have no data yet, you may need to crank up your data collection efforts, for example by implementing more sensors or other means of detection. 

Another strategy is to proactively set up test cases with failures. With a so-called Failure Mode and Effects Analysis (FMEA) procedure, you follow a step-by-step approach to identify all possible failures in a design, a manufacturing or an assembly process. FMEA is a bottom-up data collection approach, where you start at a low level of your process, working your way up to the effects on the system or subsystems. For each component, you can record the failure modes and their resulting effects on the system, which allows you to identify potential failures and their root causes.

However, testing and simulating failures is not always obvious. Asset downtime is expensive and it may not always be possible to take your asset out of production. Simulating your exact asset conditions in a test setup is not always possible. 

This brings us to other strategies.

#2 Make predictions within and beyond your existing data set. 

Integrate and link data sets with each other to come to new insights. Combine your process data and the little failure data that you have through extrapolation or interpolation. 

• Extrapolation is a way to make guesses about the future or about some hypothetical situation based on data that you already have. In other words, extrapolation is the use of your data set to predict beyond the data set. 
• Interpolation is a way to make predictions within your dataset. For example, if you have data about what happened at point A, C and E on your maintenance timeline, you may also want to know more about the condition of your asset at point B or D. You can do that through interpolation.

#3 Look for non-obvious data sources.

The internet of things is not the exclusive data generator. There are other (sometimes old-school) strategies to collect valuable data. 

• Look for non-quantitative data about your assets in accounts or company archives. 
• Look up indirect information, such as: 
  • Open source data, public archives or climatological information
  • Data from assets at third parties (clients or contractors)
  • Registered experiences, such as interviews, questionnaires, service reports, data from subcontractors, customers or after-sales service
• Search for similar data: For example, a kick scooter sharing service might look for data on bike and motorized scooter sharing, or for data on non-shared kick scooters.

Collect data now to predict failures tomorrow

In an ideal scenario, predictive maintenance teams can analyze huge volumes of asset data, giving them a better understanding of the ongoing health of their assets, and allowing them to anticipate failures. However, scarcity of incidents or lack of insight into the causes of incidents or failures cannot be a reason for maintenance departments to rest on their laurels. Instead, they need to continuously collect data in a proactive and creative way, even when the nature of the failure is not clear. 

Collecting data will eventually enable you to: 

• Prevent unforeseen failures and defects
• Mitigate the impact of future problems 
• Be prepared for changing asset conditions
• Monitor the effectiveness of your maintenance efforts. Even if your maintenance work seems successful, it does not mean that it is (cost-)efficient. 

Do you have trouble collecting valuable failure data? Or do you find it difficult to pinpoint the cause of your asset failures? Kapernikov may be able to help. Drop us a line to get in touch with a Kapernikov expert.

Further reading

https://www.manmonthly.com.au/features/perplexities-predictive-maintenance-generating-leveraging-failure-data/

https://www.youtube.com/watch?v=84j4RH_ACJI

https://www.accenture.com/us-en/insights/artificial-intelligence/ai-investments

https://www.embedded-computing.com/guest-blogs/the-perplexities-of-predictive-maintenance-generating-and-leveraging-failure-data

https://medium.com/the-industry-4-0-blog/zero-downtime-how-to-start-out-with-predictive-maintenance-3538bf0fec46

https://accendoreliability.com/types-of-failure-data/

https://accendoreliability.com/field-industry-and-public-failure-data/