By clicking “Accept”, you agree to the storing of cookies on your device to enhance site navigation, analyze site usage, and assist in our marketing efforts. View our Cookie Policy and Privacy Policy for more information.
HomeArticles

Getting Started with Machine Learning for Compressors

Getting Started with Machine Learning for Compressors

Downtime is often a significant cost and source of revenue loss for operations requiring compression. Learn how to get started with machine learning to reduce downtime.

If you work with natural gas, chemicals, or other industrial processes, it’s likely you deal with compressor systems. These systems tend to be complex, fit-for-purpose, and expensive. However, due to their role in maintaining phase states, compressor malfunction or downtime can affect many parts of an operation. Downtime – whether planned for scheduled maintenance, equipment movement, upgrades, and so forth, or unplanned due to failures or disruptions – it is often a significant cost and source of revenue loss for operations requiring compression.

As industrial companies start to look at machine learning (ML) and artificial intelligence ( AI) applications for critical equipment, compressors are often a natural place to start. In this article, I’ll walk you through how to get started.

WHAT CAN MACHINE LEARNING DO FOR YOUR COMPRESSORS?

Machine learning refers to computer programs that are trained to identify relationships, patterns, and classifications based on data sets, rather than specific underlying knowledge of how a system works. By contrast, a compressor simulation program might use formulae based on gas mix and thermodynamic conditions to estimate compressor performance.

A machine learning program would instead be trained on data specific to a system, in order to understand how input conditions, e.g. pressure and temperature, relate to compressor performance measurements in a specific system. Often by combining a machine learning approach with parameters from traditional simulations, a better understanding of the overall system is possible. However, machine learning programs don’t necessarily need to know why a system should operate in a certain way. They simply analyze what has actually happened in the past to better understand what’s happening now or what may happen in the future.

For compressor reliability and maintenance, you are often most interested in understanding whether specific patterns in operating data, such as pressure, temperature, flow, and vibration, are indicative of undesired operation modes, especially critical failures. Ideally, you’d like to know about specific failure modes, such as valve failure, lubrication system failure, dry gas seal failure, corrosion, and so forth. Moreover, wherein the compression system such failures may occur.

One challenge you face, however, is that industrial compressors are highly engineered systems that rarely fail. A historical data set for a compression system is unlikely to show multiple episodes of failure. Especially across the universe of potential failure modes. Thus standard approaches focused on predicting future incidents based on past patterns aren’t often applicable to compressors due to your lack of historical examples to learn from.

Also read: Pump Analytics: Solving Pump Inefficiency and CO2 Emissions

ANOTHER WAY YOU CAN GET STARTED

Modern industrial monitoring systems allow operators real-time evaluation and alarming of critical parameters. More advanced systems may allow increasingly sophisticated monitoring options, for example, presenting derived values such as efficiency or identifying temporal trends in operational data. However, such monitoring typically relies on hard-coded alerting thresholds and may still require significant expertise and analysis time in relating alerts to real problems needing action. Adding a well designed machine learning layer to this workflow can further increase sophistication and decrease the time needed to identify mission-critical events when or even before they occur.

Such a layer may contain one or both of the following approaches based on learnings from historical observations combined with an operator’s expert domain knowledge:

  • Interpretable generic multi-sensor anomaly detection: Despite having few failure examples, a historian will contain plenty of sensor data from regular operations. An approach can therefore be developed to monitor all sensors simultaneously from a compressor and alert an operator when any of those measurements are outside of the regular operational envelope based on historical observations. Preference should be given to machine learning algorithms which can relate such an alert back to the primary sensors which caused the alert. Thus, guiding the operator to where in the compression system a problem might be occurring.
  • Failure heuristic-based anomaly detection: Compressors are physical systems and information on how a critical failure mode manifests itself in some subset of sensors is often known even if a given compressor hasn’t observed that failure mode in its own history. This allows a more focused approach with the regular operational envelope learned for that sensor subset rather than across all sensors. The resulting algorithm then allows a more direct relation to be drawn between an alert and potential cause of failure, significantly reducing the required diagnosis time.

CONCLUSION

Compression systems are a good place to start when building on existing monitoring solutions with applications that add a more contextually aware machine learning layer to significantly streamline the work of diagnosing problems. Lack of failure data in the lifetime of a given compressor isn’t a significant barrier as well designed anomaly detection approaches already provide significant value. Furthermore, as solutions are scaled across a fleet of comparable compression systems, information can be learned across that fleet as a whole further increasing the value machine learning approaches can provide.