What Is Predictive Maintenance? Types, Uses Cases, and How It Works

Maintenance is a tricky subject. Scheduling planned maintenance efforts has an opportunity cost: the cost of not running your systems and operations while your business could otherwise generate revenue.

However, any failure to maintain can lead to unplanned downtime and costly outages. Even worse, it could deteriorate your production line and systems. Or fail unexpectedly during peak hours — precisely when your business could generate the highest profits.

Another approach is to follow fixed maintenance cycles, but every system, device, and machine may require its own maintenance timeline. The idea here is to find an optimal tradeoff: how can you schedule maintenance during off-peak hours and within the required time window, when it is least impactful to the business but also early enough to avoid any system failure?

How much does downtime cost?

Let’s first understand the true cost of downtime — that’s what we’re really trying to avoid by ensuring our systems are reliable. In a complex production line with many interdependent assets, unplanned downtime occurs frequently.

What is the impact of this unplanned downtime? The direct revenue opportunity cost of downtime is $49 million. Through another lens: a single downtime incident can trigger a 2.5% drop in stock value.

However, if you’re a large enterprise, beware: the cost of downtime is significantly higher. Global 2000 organizations incur $400 billion in losses annually due to downtime incidents. This averages $200 million per organization annually, or around 9% of their total profits.

The key to improving maintenance schedule, increasing operational efficiency of maintenance processes, and eliminating unplanned downtime is data, analytics and intelligence as part of your predictive maintenance program.

How to predict maintenance schedules

If you’re organization runs on equipment of any kind, that equipment likely needs some sort of maintenance.

Predictive maintenance is a data-driven strategy designed to minimize downtime by scheduling maintenance only when needed, ensuring it reduces disruptions and avoids significant downstream impacts. It relies on real-time monitoring of systems, which can track:

• Equipment health and performance
• Failure risk and maintenance requirements
• The cost of unplanned downtime to the business

A traditional predictive maintenance strategy relies on information from many sources, including data (system and infrastructure logs, for example), business performance metrics, market trends, and external business demand. It’s also common that sensors may be installed on the machines or equipment to generate real-time information on health parameters, including energy:

• Consumption
• Vibration
• Temperature
• Sound
• And other outputs

Alternatively, real-time monitoring systems may be configured to:

• Measure the overall system performance and health.
• Identify the root cause in event of performance anomalies and outages.

(Related reading: IoT monitoring explained.)

AI and machine learning for predictive maintenance

Since the goal of predictive maintenance is to schedule maintenance only when it is most needed, business organizations need to answer two important questions:

1. What is the true cost of downtime right now?
2. What is the risk of scheduling (or delaying) maintenance to a particular time in the future?

To answer these questions, a predictive maintenance strategy uses advanced AI algorithms to map the real-time information generated by machine sensors and operational demands of the business to the cost of downtime.

Here, an optimal tradeoff is considered, to schedule maintenance during the period when operational demands are lower and well within the useful operational life of the machines. Secondary measures such as redundancy can help reduce the risk of outages during both planned and unplanned outages.

Other types of maintenance strategies

Predictive maintenance schedule helps minimize the cost of planned downtime and extend the useful life of expensive systems. However, the process may be too expensive, complicated and not a business imperative in some cases.

Let’s take a look at other types of maintenance strategies. Keep in mind, though, that these three strategies have one big difference when compared to predictive maintenance, these types rely on known (predefined) and current-state information.

Reactive maintenance

The “run to failure” approach. In reactive maintenance, you’re basically just waiting for equipment to fail.

This is typically used for non-critical applications where the machines can be replaced easily, at low cost and fast enough to incur any significant downtime. In most cases, an equipment failure may not cause an outage — due to higher risk tolerance and redundancy strategies. This strategy is also suitable for machines with high RUL (Remaining Useful Life) specifications and low failure rates.

Preventive maintenance

A moderate risk-averse approach, preventive maintenance aims to reduce the failure rate at the cost of frequent maintenance routines. These routines are planned and scheduled according to vendor specifications and known failure rates, at a safe margin well before a machine can fail.

This approach is suitable when your systems are:

• Your systems are easy to replace and maintain.
• Your systems lack adequate redundancy — meaning that any failure could cause costly outages.

Condition-based maintenance

As a subset of predictive maintenance, the condition-based maintenance program follows a data-driven approach to maintenance. Machine sensors may continuously monitor for system health: a maintenance alert is triggered when the health parameters exceed predefined threshold.

This approach reduces the risk of failure, without having to extensively analyze real-time machine data for its long-term health performance. Note that an anomaly or an erroneous reading on the sensor can also trigger false negative alerts.

Enable predictive maintenance for your organization

Predive maintenance relies on machine learning-based monitoring and analytics tools to acquire contextual knowledge based on historical trends and predict future-state information based on the current real-time sensor data.