Use of the K-Nearest Neighbour Classifier in Wear Condition Classification of a Positive Displacement Pump

Area of Science:

• Mechanical Engineering
• Condition Monitoring
• Machine Learning

Background:

• Multi-piston positive displacement pumps are critical in various industries.
• Pump failures can lead to significant downtime and economic losses.
• Existing diagnostic methods have limitations in accurately assessing wear conditions.

Purpose of the Study:

• To develop and validate a learning system for classifying the wear condition of multi-piston positive displacement pumps.
• To investigate the effectiveness of K-nearest neighbour classification for pump diagnostics.
• To establish a reliable method for early detection of pump wear.

Main Methods:

• A diagnostic experiment was conducted to collect vibration signals from pump bodies.
• Time-frequency analysis was applied to the acquired vibration signals.
• K-nearest neighbour (KNN) classifier was developed and trained using signal features.

Main Results:

• Signal features were extracted and correlated with specific pump wear conditions.
• The KNN classifier demonstrated high accuracy in classifying pump wear states.
• The validated model successfully predicted wear conditions using new vibration data.

Conclusions:

• The K-nearest neighbour classifier is a viable and accurate tool for diagnosing the wear condition of multi-piston positive displacement pumps.
• Vibration signal analysis combined with machine learning offers a promising approach for predictive maintenance.
• This method can improve the reliability and operational efficiency of positive displacement pumps.