An automated health indicator construction methodology for prognostics based on multi-criteria optimization.

Area of Science:

• Engineering
• Computer Science
• Data Science

Background:

• Autonomous health management systems are crucial for improving industrial performance and minimizing downtime.
• Effective health indicators (HI) derived from intelligent sensors are essential for system monitoring and predictive maintenance.
• Current methods may lack automated feature extraction and selection capabilities for diverse sensor data.

Purpose of the Study:

• To develop an automated methodology for selecting pertinent measurements from multiple sources.
• To handle high-frequency sensor data for extracting low-level features.
• To combine extracted features into optimal health indicators (HI) based on defined evaluation criteria.

Main Methods:

• A novel methodology is proposed for automatic measurement selection and feature extraction from raw sensor data.
• Genetic programming is utilized for its flexibility in combining features and creating health indicators.
• The methodology does not require prior knowledge of system degradation trends but can integrate such information.

Main Results:

• The proposed methodology successfully automates the selection of relevant measurements and feature extraction.
• It effectively combines low-level features into appropriate health indicators (HI) according to multiple criteria.
• Performance was validated through two real-world application case studies.

Conclusions:

• The developed methodology offers an automated and flexible approach to creating robust health indicators for autonomous systems.
• It enhances system monitoring and supports informed maintenance decisions, contributing to reduced operational losses.
• The study provides a valuable overview of health indicator evaluation criteria for industrial applications.