Condition monitoring is the process of collecting and analyzing sensor data from equipment to evaluate its health state during operation. Accurately identifying the current health state of equipment is critical to the development of predictive maintenance and condition-based maintenance programs.
Benefits of Condition Monitoring
Condition monitoring enables equipment manufacturers and operators to:
- Reduce unplanned failures: Detect anomalies and faults before they become major problems.
- Optimize maintenance schedules: Avoid the costs of unnecessary maintenance by scheduling service only when necessary.
- Minimize downtime: Isolate the source of faults more quickly to reduce the time required for diagnostics and maintenance.
Condition monitoring is not just about collecting data but also about using that data to evaluate the condition of a machine. In practice, it could be anything from a control chart that ensures a single sensor value does not exceed a safety threshold to a machine learning algorithm trained on hundreds of sensors with months of historical data.
Condition Monitoring Algorithms
Condition Monitoring vs. Prognostics
A predictive maintenance program may include both condition monitoring and prognostics algorithms. The main difference between condition monitoring and prognostics is the time frame.
| Time Frame | MATLAB Code Example | |
|---|---|---|
| Condition Monitoring | Current state | Detecting faults in bearings |
| Prognostics | Future state | Estimating the remaining useful life of an aircraft engine |
Condition monitoring focuses on the current state of machinery. It identifies faults and anomalies using real-time data and enables immediate corrective actions to prevent failures. Condition monitoring approaches include anomaly detection algorithms, which detect when machine behavior deviates from normal, and diagnostics algorithms, which identify specific component faults.
Prognostics, on the other hand, looks into the future to estimate the remaining useful life of equipment by analyzing trends and patterns in the data. This predictive aspect helps in planning maintenance activities well in advance, optimizing resource allocation and minimizing downtime.
Together, condition monitoring and prognostics algorithms can help form a comprehensive prognostics and health management (PHM) strategy that enhances the reliability and longevity of machinery.
Condition-Based Maintenance vs. Predictive Maintenance
Condition-based maintenance (CBM) is a maintenance strategy based on the current health state of the equipment. Learn how CBM works and how it differs from predictive maintenance.
Developing Condition Monitoring Algorithms in MATLAB
A typical workflow for developing condition monitoring algorithms in MATLAB® includes acquiring and preprocessing data, identifying condition indicators, training the model, and deploying and integrating the algorithm.
Condition monitoring algorithm development workflow.
Acquire Data
Acquiring data is always the first step in developing condition monitoring algorithms. If you have streaming or archived sensor data from operational machines, there are many ways to access it for analysis. You can obtain the data directly through test hardware or retrieve it by connecting to tools such as OPC UA, RESTful web services, databases, AWS S3, and Azure Blob.
It can be a challenge to acquire and organize the right data for training condition monitoring algorithms—especially data representing faults and failures. You can replace or augment existing operational data by generating synthetic data using physics-based models built in Simscape™.
Condition Monitoring with MATLAB
Learn how you can develop condition monitoring algorithms with MATLAB. Develop condition monitoring algorithms for the early detection of faults and anomalies to reduce downtime and costs due to unplanned failures and unnecessary maintenance.
Explore and Preprocess Data
Achieving a deep understanding of the data is key to designing valuable condition monitoring algorithms. This may require preprocessing data to remove outliers, noise, trends, and other artifacts. Preprocessing ensures that condition monitoring algorithms are trained on the most representative information available.
In this data exploration phase, it may be useful to visualize parts of the data set. Domain experts who understand what the data represents, may be able to detect anomalies and faults easily by eye. In this case, simple condition monitoring algorithms such as findchangepts or a control chart may be sufficient.
If the data contains many sensors and complex patterns that are not easily identifiable, more advanced methods will be required to extract meaning from the data set. This is where AI-based approaches like machine learning and deep learning become valuable.
Sometimes anomalies are easy to spot in sensor readings, as shown in this MATLAB plot. In this case, a simple condition monitoring algorithm would suffice.
Design Condition Monitoring Algorithms
The first step in designing condition monitoring algorithms usually involves identifying condition indicators: features that indicate the difference between normal and faulty operation. Identifying condition indicators may be straightforward or may require an iterative process of extracting and analyzing derived quantities from many sensors to find meaningful patterns. This process is known as feature engineering.
Predictive Maintenance Toolbox™ includes the Diagnostic Feature Designer app, which lets you extract, analyze, and rank the most relevant features for detecting faults interactively. This app simplifies the task of converting raw data into useful derived features, making it easier to develop effective condition monitoring algorithms.
With the Diagnostic Feature Designer app, you can interactively extract and rank features to train condition monitoring algorithms.
After identifying the right condition indicators, the next step is to use them to create a condition monitoring algorithm. This involves training machine learning or deep learning algorithms, which learn from the selected features to detect faults and anomalies accurately. The Classification Learner app enables you to interactively try out different fault classification methods to find the most suitable condition monitoring algorithm via comparison of various machine learning models.
With the Classification Learner app, you train a variety of classification models for condition monitoring.
Deploy and Integrate Condition Monitoring Algorithms
To provide business value, condition monitoring algorithms need to be deployed and integrated in an operational environment such as an on-premises server or cloud. Condition monitoring algorithms can also be deployed to an embedded system, enabling faster response times and reducing the amount of data sent over the network.
Deployment involves transferring the algorithm from a development environment into a real-world operational setting where it can start monitoring equipment. This step requires careful planning to ensure that the algorithm interfaces seamlessly with the machinery’s sensors and data collection systems. Integration, on the other hand, involves embedding the algorithm within the operational workflows, ensuring that it communicates effectively with other maintenance and monitoring systems. This might include setting up notifications for maintenance teams when the algorithm detects a potential issue or integrating with a dashboard that displays real-time health indicators of the machinery.
Using MATLAB and Simulink, Coca-Cola developed embedded code for a machine learning–based virtual pressure sensor used in Freestyle beverage dispensers.
The deployment and integration phase marks the transition from theoretical design to practical application, where the algorithm’s effectiveness in predicting and preventing equipment failures is truly tested. Condition monitoring algorithms can become key tools for predictive maintenance, helping to extend equipment life and improve operational efficiency.
For tools and examples related to condition monitoring, explore Predictive Maintenance Toolbox.
Condition monitoring is the process of collecting and analyzing sensor data from equipment to evaluate its health state during operation, which is critical for developing predictive maintenance and condition-based maintenance programs.
Condition monitoring reduces unplanned failures by detecting anomalies early, optimizes maintenance schedules by avoiding unnecessary service, and minimizes downtime by isolating fault sources more quickly.
Condition monitoring focuses on the current state of machinery to identify faults and anomalies using real-time data, while prognostics looks into the future to estimate remaining useful life by analyzing trends and patterns in the data.
Condition monitoring is a component of predictive maintenance programs that evaluates the current health state of equipment, while predictive maintenance is a broader strategy that may include both condition monitoring and prognostics algorithms.
Condition indicators, sometimes called health indicators, are features that indicate the difference between normal and faulty operation, derived through feature engineering by extracting and analyzing quantities from sensor data to find meaningful patterns.
Condition monitoring algorithms can be deployed to on-premises servers, cloud environments, or embedded systems, enabling faster response times and integration with operational workflows and maintenance systems.
Condition monitoring uses anomaly detection algorithms to detect when machine behavior deviates from normal or fault detection (diagnostics) algorithms to identify specific component faults. Each of these algorithm types often leverages machine learning and deep learning.
Yes, synthetic data generated using physics-based models in Simulink and Simscape can replace or augment operational data, especially when acquiring data representing faults and failures is challenging.
Resources
Expand your knowledge through documentation, examples, videos, and more.
