Condition indicator for centrifugal pumps in data center cooling systems

Aingura insights

Electrointensive Industries

Condition Monitoring

ARTIFICIAL INTELLIGENCE (AI) TECHNOLOGIES DEVELOPED

Novelty Detection Techniques: Utilizing advanced novelty detection methods to identify anomalies and deviations in the performance of centrifugal pumps.
Online AI-Driven Baseline Generation: Employing AI algorithms to continuously generate and update baseline data for real-time comparison with current pump performance.
Probability-Based AI Technology: Implementing probability-based AI models to assess the likelihood of component failure or degradation.
Unique KPI Representing Overall Behavior: Creating a distinctive Key Performance Indicator (KPI) that offers a comprehensive view of the centrifugal pump's overall performance and behavior.
Online Feature Subset Selection: Employing online feature subset selection techniques to extract the most relevant and informative features for accurate condition assessment in real-time.

Develop a nonintrusive sensor system for data collection from centrifugal pumps.
Create a comprehensive dataset from limited available information about the analyzed system.
Implement advanced data preprocessing techniques to clean and format the collected data.
Explore novel feature engineering methods to extract relevant information.
Apply machine learning algorithms to develop a KPI that can effectively monitor the health of centrifugal pumps.
Ensure the KPI is capable of detecting anomalies or faults in real-time or near-real-time.
Establish a threshold or baseline for the KPI to trigger alerts when anomalies are detected.
Integrate the KPI into the existing monitoring system for seamless deployment and operation.
Validate the KPI's accuracy and reliability through testing and performance evaluation.
Provide actionable insights and recommendations for maintenance and corrective actions based on KPI alerts.

Generation of an explainable model that accurately represents the dynamic behaviour of the electrical and mechanical system of the pump.

Extraction of modal parameters that are highly correlated with the natural modes of the analyzed system, providing interpretable information to the end user for diagnosing abnormal operation.

Utilization of the estimated modal parameters to create a probabilistic machine learning baseline, which in turn produces a Key Performance Indicator (KPI) that signifies the healthy state of the pump's operation.