Imagine a world where assets tell you exactly when they need maintenance, preventing costly breakdowns and downtime. This is the promise of Predictive Maintenance powered by Artificial Intelligence. In the ever evolving engineering world, integrating new and exciting technologies, Predictive Maintenance has been labelled a game changer for businesses aiming to optimise their maintenance practices. However, the challenges of actually implementing Predictive Maintenance often go unspoken. This blog series will dive deep into these challenges, exploring what Predictive Maintenance actually is, the hurdles to overcome with the data, and how to evaluate the outcomes.

Traditional Asset Maintenance

Within engineering industries, asset maintenance typically follows one of two approaches: Reactive maintenance or preventative maintenance.

Reactive Maintenance: This approach is simple in so far as maintenance only takes place once the asset has failed, prompting the maintenance activity to take place. Whilst this is simple, this method is the costliest as the assets have to have their operations stopped in order to repair the asset.

Preventative Maintenance: This strategy relies on regular scheduled maintenance tasks based on the known lifespans of the assets components, aiming to keep the assets running smoothly throughout their life. However this can lead to unnecessary repairs and replacements, which again is costly.

So Why Optimise Maintenance?

Asset downtime can cost businesses significantly through various means such as loss of stock, trading interruptions, reputation and costly repairs to fix the underlying issues that have led to failure.

With technology at the forefront, the overarching aim of modern maintenance is to ensure unrivalled reliability to be able to service customers and remove the need for traditional asset maintenance. This requires that the maintenance activities are synergised with the assets through continually monitoring the assets operation through telemetry data received through the internet of things (IoT) sensors affixed to assets.

What is Predictive Maintenance?

Predictive Maintenance is the prediction ahead of time for whether or not a component will fail in the near future, therefore allowing maintenance activities to be scheduled in at the right time to fix issues before they occur and become costly. It can therefore, based on continuous monitoring of telemetry data, allow for the early detection of failures, reducing costs and risks.

The whole aim is to strike a balance between maintenance costs and failure rates, as illustrated in the graph below.

Figure 1: Visualisation of the overall cost for reactive, preventative, and predictive maintenance regimes.

As can be seen, preventative maintenance aims to minimise failures but is costly due to frequent replacements and prevention costs, whereas reactive maintenance incurs high repair costs due to frequent failures. Predictive Maintenance on the other hand allows the total cost to be minimised by striking a balance between the repair and prevention costs and total number of failures.

What It Is Not

In practice, there are a lot of claims of successfully implemented Predictive Maintenance activities, however these can actually be classified as either rule based logic, whereby rules are built to flag issues based on logical operators, or anomaly detection, which aims to identify telemetry observations that are outside of the norm.

The issue with these are that they do not predict, i.e. there is no time element associated with the failure, and they often pick up issues as they occur or just detect inherent noise within a data set.

Whilst they are not predictive, they do however serve a purpose, and they can be used as a precursor to Predictive Maintenance by allowing a good training dataset of failures to be built up.

Implementing Predictive Maintenance: Challenges

Predictive Maintenance holds the promise of predicting asset failures before they happen, allowing for timely maintenance that minimises risk and ensures reliability. However, transitioning away from traditional methods requires more than just technology – it requires a shift in mindset.

To truly optimise maintenance and transition from traditional methods, the cost of predictive interventions must be lower than the costs incurred from reactive and preventative maintenance. This means not just adopting Artificial Intelligence technologies, but also rethinking how maintenance is perceived and valued within businesses.

One of the biggest obstacles is changing the culture around maintenance. Its often seen as a necessary evil, especially if the process or asset needs to be shut down for the duration. Convincing stakeholders to invest in Predictive Maintenance can be challenging especially when there are no visible signs of impending failure, yet this mindset shift is essential to realise the full benefits of Predictive Maintenance.

Where can we use it?

Predictive Maintenance can be used anywhere where telemetry data is being recorded for assets and where assets undergo maintenance activities.

To be able to use Predictive Maintenance, the following components are needed:

• Time series telemetry data – Data from sensors collected at regular intervals.
• Failure data – Historical failures data including timestamps for model training.

Summary

Predictive Maintenance shows a promising advancement to optimising asset maintenance and reducing the overall costs associated with traditional maintenance methods. Through predicting failures before they occur, Predictive Maintenance allows for the proactive interventions to take place, minimising downtime and enhancing asset reliability. However, the journey to Predictive Maintenance faces more challenges once the data has been collected. Stay tuned for the next instalment where the data challenges will be explored in detail.