What Is Predictive Maintenance?

Why Predictive Maintenance Matters

To understand what makes predictive maintenance so attractive, it is important to understand the two ways maintenance is usually performed: reactive maintenance and preventive maintenance.

Reactive Maintenance and Preventive Maintenance

In a reactive maintenance approach, you only perform maintenance once the machine has failed. This approach may be suitable for a lightbulb, but unplanned failures and downtime can be very expensive and dangerous for industrial machines.

Many operators therefore perform preventive maintenance: scheduling maintenance at regular intervals without considering the actual condition of the machine. While this approach mitigates the risk of failure compared to reactive maintenance, it results in higher maintenance costs, increased downtime, and an associated increase in inventory and spare parts. It also does not prevent unexpected failures as the condition of the machine is only measured periodically, rather than monitored and analyzed continuously in real time.

Acquiring Data

Acquiring data is the first step in developing any predictive maintenance algorithm. AI algorithms are only accurate if they have robust training data that represents the types of failures you want to predict. It is therefore important to collect data that represents the machine under both healthy and failing conditions.

However, failure data is often hard to access—after all, the goal of any maintenance program is to prevent failure! This makes it hard for engineers to get the right kind of data to start building an accurate algorithm.

One solution to this problem is to generate synthetic data from physics-based models, such as those built in Simulink^® and Simscape™. For example, an engineer can build a model of an oil pump and simulate failures due to a leaky valve and a blocked pipe. Then it’s possible to generate failure data in a safe, cost-effective way that does not impact the performance of the actual oil pump. These physics-based models can then be used in operation as digital twins to predict future performance.

Identifying Condition Indicators

Once you have the right data, the next step is to reduce it to a set of features that can be used as condition indicators to train a predictive algorithm. Condition indicators are features that indicate the difference between healthy and faulty operation. They are typically extracted using a combination of statistical, signal processing, and model-based techniques using analysis and design tools such as MATLAB. The engineering team’s expertise is key here—they have insights into how the machines work and can help identify the best features.

Identifying the right features is key to the success of a predictive maintenance algorithm. The right features can be used to train algorithms to detect trends that cannot be easily observed. Additionally, feature extraction reduces the size of the raw data set. For example, commercial airplanes generate close to a terabyte of data per flight. Transmitting, storing, and analyzing such vast amounts of data can be difficult, which is why feature extraction is becoming increasingly important.

Developing Algorithms

Once you’ve extracted the best features, the next step is to train your predictive algorithm. These algorithms can fall into three major categories: anomaly detection, fault identification (diagnostics), and remaining useful life estimation (prognostics). Ultimately, the goal of predictive maintenance algorithms is to turn sensor data into maintenance decisions.

If data is labeled with failure modes, engineers can use supervised learning methods to train predictive models to differentiate between these failure modes. These models can then be connected to operational systems in the field where they can help determine the root cause of degraded performance.

Unsupervised learning methods are best suited to applications such as anomaly detection where the goal is to classify incoming condition indicator values from equipment as either normal or anomalous. As unsupervised learning methods do not require labeled training data corresponding to different failure modes, they tend to be very popular for engineers trying to develop predictive maintenance algorithms for the first time.

A separate class of probability and time-series based methods can be used to calculate the remaining useful life (RUL) of a machine. These models accept the current value of a condition indicator and estimate within a defined confidence interval when the equipment will fail. Knowing when the machine may fail allows engineers to schedule maintenance, order spare parts, or limit the operation to extend its life.

Deploying Algorithms in Operation

A predictive maintenance solution is more than just algorithms. The algorithms must be deployed in operation to realize the benefits of reduced downtime, lower maintenance costs, and improved operational efficiency.

The operational environment must securely manage data and scale computing resources to ensure the algorithms can run effectively in embedded or IT/OT systems. It must also integrate with other IT systems for managing inventory, raising service tickets, and presenting dashboards with the results of the algorithm to the operations team.

In many operational applications, predictive maintenance algorithms are not just running in the cloud on-prem servers. Parts of the algorithm—often signal processing and feature extraction—can be deployed directly onto edge devices like industrial controllers that can quickly process high frequency sensor data in real time. This helps reduce data storage and transmission costs.

Predictive Maintenance with MATLAB and Simulink

Engineers use MATLAB, Simulink, and Predictive Maintenance Toolbox™ to design, test, and deploy custom condition monitoring and predictive maintenance algorithms.

With MATLAB and Simulink, you can:

• Access streaming and archived data from cloud storage, databases, data historians, and industrial protocols.
• Interactively explore, extract, and rank features with the Diagnostic Feature Designer.
• Develop predictive models to detect anomalies, identify faults, and predict remaining useful life (RUL).
• Build physics-based models to generate synthetic sensor data and deploy digital twins.
• Generate C/C++ code for real-time edge processing.
• Integrate with your choice of IT/OT systems without recoding: scale algorithms in the cloud as shared libraries, packages, web apps, Docker containers, and more.

Related Topics