Innovative data techniques for centrifugal pump optimization with machine learning and AI model.

Area of Science:

• Data Science
• Mechanical Engineering
• Machine Learning

Background:

• Modern centrifugal pump machines (CPM) require robust data acquisition systems for performance monitoring.
• Data quality is critical for the effectiveness of machine learning (ML) and deep learning (DL) models in analyzing CPM data.
• Sensor fusion technology, exemplified by the Dewesoft FFT DAQ system, is key to extracting high-fidelity data from CPMs.

Purpose of the Study:

• To highlight the significance of data cleaning, pre-processing, and transformation for ML/AI model utilization.
• To detail methodologies like Exploratory Data Analysis (EDA), Data Visualization, and Feature Engineering (FE) for data enhancement.
• To demonstrate the application of validated data in training ML/DL models for optimizing CPM operations.

Main Methods:

• Utilizing the Dewesoft FFT DAQ system with sensor fusion for data acquisition from CPMs.
• Implementing data cleaning, pre-processing, Exploratory Data Analysis (EDA), Data Visualization, and Feature Engineering (FE).
• Applying hypothesis testing for data integrity validation and subsequently training ML classifiers and DL algorithms.

Main Results:

• Achieved a 27.25% enhancement in operational efficiency, measured by the F1 score.
• Reduced model training time by 180 seconds, enabling faster predictive maintenance.
• Demonstrated improved model performance using Precision, Recall, and F1 score metrics.

Conclusions:

• The integration of advanced data science techniques significantly enhances CPM operational efficiency and predictive maintenance capabilities.
• Thorough data pre-processing and validation are essential for reliable ML/AI model performance in industrial applications.
• This approach provides actionable insights for informed decision-making and optimization of centrifugal pump operations.