From Vibrometers to Sensors, Cloud, and AI: How Modern Predictive Maintenance Works

In the 1940s, two Royal Air Force (RAF) squadrons faced a challenge — out of 40 bombers, only 20 were flight-ready. Routine maintenance was performed every 50 flight hours, yet reliability remained an issue. Increasing preventive measures seemed like the solution, but scientist Conrad Hal Waddington, responsible for the coastal RAF squadron's aircraft, challenged this.

He discovered a paradox: the number of failures increased after scheduled maintenance when the opposite should have happened. Waddington suggested performing maintenance not on a set schedule but based on the actual condition of the equipment, which was now checked after each flight.

Flight hours increased by 60 percent, and this approach became the precursor to modern predictive maintenance based on equipment condition data.

For a long time, data was collected and recorded manually. For example, to check the condition of bearings, a specialist had to apply a vibrometer to each engine on the premises. Today, equipment performance is monitored by digital sensors, and AI analyzes the data. In this column, I will dive deeper into the technologies used in modern PdM and how this approach is evolving.

Rising Costs of Failures and the Role of Predictive Maintenance

The cost of equipment failure is increasing. Back in the 1980s, airlines lost around $400,000 in daily revenue due to the downtime of a Boeing 747. Today, for large companies with hundreds of machines, the cost of downtime can amount to hundreds of thousands of dollars per hour.

Manufacturing and industrial companies on the Fortune Global 500 list lose $172 million annually due to unplanned downtime. Often, these failures could have been predicted. This is the path that companies implementing PdM are taking.

The development of PdM accelerated in the 2000s with the increase in computing power. Sensors for monitoring equipment conditions and the first software solutions for data analysis began to emerge. This allowed for partial automation of data collection and processing. PdM improved maintenance efficiency, as decisions are now based on real-time equipment data rather than just scheduled dates for part replacements or oil changes.

According to Deloitte, companies implementing PdM have reduced downtime by 10-15 percent and costs by 5-20 percent. For our client, one of the largest global players in e-commerce, this translates into hundreds of millions of dollars saved on equipment replacement, repairs, and downtime.  

There are an estimated 18.8 billion IoT devices worldwide, with a third of them used for monitoring equipment conditions and automating analysis.

Predictive Maintenance Architecture: How It Works
 

The PdM process, from detection to fault resolution, can be broken down into four steps:

1. Data Collection from Equipment: This is done using wireless IIoT sensors. They continuously gather data on vibration, temperature, and over a dozen other parameters. For example, from half a million sensors installed at our clients' sites, we receive over 10 billion data points daily.
2. Data Transmission: Sensors transmit information to cloud storage via mobile internet connections or, in our case, custom-designed protocols. The data must be delivered without loss or delay to ensure real-time analytics.
3. Data Processing and Analysis in the Cloud: ML models filter out irrelevant fluctuations, normalize signals, and group data. They then analyze and compare current data with historical trends, identifying patterns and anomalies in equipment performance. For example, vibration data analysis can detect issues with lubrication or loose engine connections before they become noticeable to the human eye.
4. Identifying the Cause of Failure and Repair: Based on the analysis results, analysts interpret equipment signal graphs to determine the cause of the anomaly. They then create a detailed task for the team: schedule maintenance, replace a part, or change the oil. If the company uses contractor services, recommendations from analysts can be sent through a specialized service, such as a control panel or mobile app.

Implementing PdM systems is primarily about building a large-scale IT infrastructure. IoT engineers and network engineers integrate IIoT sensors with cloud or local servers and set up the network infrastructure for stable data transmission. 

Data architects design the database architecture to handle terabytes of interval data per day. Such a system requires constant maintenance to ensure fast query processing, quality integration with analytics, and scalability, as data volume continues to grow.

Because PdM requires a large IT team, many businesses opt for the "PdM as a service" model, subscribing to analytics and sensors without the need to build the infrastructure themselves.

Where Predictive Maintenance is Heading

It’s forecasted that by 2030, the global PdM services market will grow to $60.13 billion, up from $7.85 billion in 2022. Over the past 10 years, interest in automated equipment condition analysis has increased by 275 percent.

Modern PdM systems not only detect issues but also predict them. In some cases, prediction accuracy reaches up to 99 percent. The combination of IIoT, Big Data, and AI is transforming maintenance into an analytical discipline.

Shortly, the main development focus of PdM will be the automation of decision-making and system scalability. For example, integration with digital twins – virtual replicas of industrial assets that simulate equipment behavior and allow testing of various operating scenarios. 

In addition to industry, PdM applications are expanding into other sectors such as healthcare, transportation, logistics, and energy. At the same time, the importance of cybersecurity is growing, as cloud-connected equipment must be secured as reliably as a server containing sensitive information.

In the future, PdM will become even more effective. Modern algorithms can already detect and alert about issues earlier than experienced analysts might notice them on a graph. For example, the Condition Monitoring System Waites can analyze spectral signals from bearings that appear long before overall vibration levels increase. The system signals when a part starts to wear out.

Businesses implementing PdM have an edge over competitors, as their equipment operates 10-20 percent longer, and failures occur much less frequently.