Improved correlation analysis and visualization of industrial alarm data.

Area of Science:

• Industrial engineering
• Data science
• Control systems

Background:

• Multivariate alarm analysis is crucial for smart alarm management due to complex variable interrelationships.
• Visualizing historical alarm data correlations aids in understanding system behavior.
• Existing methods may not adequately handle noise like missed or false alarms.

Purpose of the Study:

• To develop a robust method for analyzing and rationalizing multivariate alarms using historical data.
• To enhance the identification of correlated alarms and redundancies for improved alarm settings.
• To provide a practical and effective tool for industrial alarm management.

Main Methods:

• Applied Gaussian kernel method to generate pseudo-continuous time series from binary alarm data.
• Utilized correlation color maps with reordered alarm tags to visualize variable clusters.
• Incorporated time lags and singular value decomposition for in-depth analysis within clusters.

Main Results:

• The Gaussian kernel method effectively reduces the impact of noisy alarm data (missed, false, chattering).
• Correlation color maps clearly display clusters of related alarms, facilitating analysis.
• The proposed method demonstrated superior performance compared to alarm similarity color maps in industrial case studies.

Conclusions:

• The Gaussian kernel method offers a practical and effective approach to multivariate alarm analysis.
• This technique improves the identification of alarm correlations and redundancies, leading to optimized alarm settings.
• The method provides a valuable tool for enhancing smart alarm management in industrial settings.