One of the major challenges in enabling predictive maintenance is lack of the right kind of data required for making accurate prognoses and diagnoses. Equipment is often instrumented solely in support of control and process optimization. Hence, additional sensors are usually required to support condition monitoring and facilitate predictive maintenance. However, installing the right sensors to collect needed data is often made difficult when the equipment does not have provisions for mounting instrumentation in locations that are conducive to condition monitoring. This is especially true for older equipment, where retrofit costs and effort can be impractical or prohibitive.

A recent and significant advancement in asset health monitoring is the use of virtual sensors, which can estimate critical parameters without the need for direct physical sensors. This approach not only reduces installation costs and complexity, but also enhances the reliability and accuracy of maintenance predictions. In this blog post, we explain the benefits of using virtual sensors in predictive maintenance, illustrated by a real-world example involving reciprocating compressors.

Understanding Virtual Sensors

Virtual sensors are algorithms or models that infer the value of a physical quantity based on other measured parameters. Instead of relying solely on physical sensors, virtual sensors use existing data from other sources to calculate the desired information. This approach can be particularly advantageous in environments where installing physical sensors is challenging or cost-prohibitive.

In addition to reducing the overall sensor count, virtual sensors can provide deep insight into an asset’s operation and condition by generating time-series data of important physical parameters that may be difficult to obtain by other means. For example, internal temperatures or pressures can be estimated from available sensors to give operators a better sense of how an asset is performing, without actually installing such sensors.

Case Study: Predicting Compressor Valve Failures

A practical example of virtual sensors in action comes from one of Novity’s commercial installations at a natural gas processing plant in the Permian Basin. The plant in question uses several reciprocating compressors, which are crucial for maintaining pipeline pressure.

Valve leaks are a well-known problem with reciprocating compressors. Traditionally, monitoring these compressors for valve leaks involved installing pressure and rotational position sensors—a process that often requires shutting down the equipment and incurring significant and costly downtime. Installing rotational position sensors typically needs partial disassembly of the compressor to access the crankshaft, where a multi-event wheel or band is mounted, along with a position sensor installed on the frame to detect the rotating signal. Often times, older compressors don’t have the proper provisions for installing such components, thus requiring significant modifications to the equipment.

Instead, Novity’s TruPrognostics AI only uses high-bandwidth pressure transmitters installed in the indicator tap ports and is able to accurately predict the rotational position of the compressor’s crankshaft without needing a physical rotational position sensor. It does this by leveraging the fact that the pressure curve is synchronous with the crankshaft’s rotation, allowing for accurate rotational position estimation. Furthermore, suction and discharge pressures can be estimated from the cylinder pressure, adding additional insight into the operation of the compressor.

The graph above illustrates how accurately this technique allows us to estimate the crankshaft angle. If you look closely there are both a yellow and a black line that are nearly at the same position throughout the whole chart, they yellow line shows the virtual sensor estimate and the balck line shows the actual crank angle.

Advantages of Using Virtual Sensors

1. Reduced Installation Costs and Downtime: Installing traditional crank-angle measurement systems requires shutting down compressors, which is both time-consuming and costly. Virtual sensors eliminate this need, as they can be implemented using existing pressure transmitters. This results in significant cost savings and minimal disruption to operations.
2. Enhanced Predictive Maintenance Capabilities: The virtual sensor approach proved to be highly effective in predicting valve failures. By analyzing the pressure data and estimating the rotational position, Novity’s TruPrognostics AI detects valve leakage with nearly the same accuracy as systems using physical rotational sensors. Moreover, the prognostics method applied to the extracted features predicted gradual valve failures several weeks in advance, providing ample time for intervention. Using a model-based approach and leveraging the pressure estimates, compression ratios for each throw and cylinder can be monitored, providing additional equipment insights.
3. Increased Flexibility and Scalability: Virtual sensors offer greater flexibility in data acquisition hardware choices, as high-frequency signals do not need to be synchronized. This flexibility makes it easier to retrofit older equipment with predictive maintenance capabilities, extending the benefits of Predictive Maintenance to a broader range of assets.

Implementing Virtual Sensors: A Step-by-Step Approach

To successfully implement virtual sensors, operators should follow a structured approach:

1. Data Collection: Gather high-quality data from existing sensors. This involves identifying relevant data sources and ensuring accurate and consistent data collection.
2. Feature Engineering: Identify and extract meaningful features from the data that are indicative of the desired parameters. These features should be robust and reliable under various operating conditions.
3. Model Development: Develop algorithms to estimate the desired parameters based on the extracted features. Validate the model against actual measurements to ensure accuracy. Physics-based models in particular are very well suited to take advantage of virtual sensors, since the virtual sensor signal is often already included in the model formulation.
4. Prognostics Integration: Apply prognostics algorithms to the estimated data to predict potential failures and estimate the remaining useful life of components. With a flexible algorithm framework, virtual sensors can be incorporated seamlessly.
5. Continuous Monitoring and Feedback: Implement a continuous monitoring system that tracks the condition of equipment and provides real-time alerts to maintenance personnel. Use feedback from actual interventions to refine and improve the virtual sensor models.

At Novity, we use virtual sensors whenever practical, providing early warnings of equipment failures and enabling more effective maintenance scheduling without adding additional installation costs or complexity.

Do you want to learn more about how Novity’s TruPrognostics AI can help you with your maintenance operations and reduce unplanned downtime?

Get in touch today!