Cascade alarm systems: A study on singular value analysis

Area of Science:

• Signal Processing
• Statistical Process Control
• Machine Learning

Background:

• Traditional alarm systems struggle with non-Independent and Identically Distributed (non-IID) data and noise.
• Existing methods like Cumulative Sum Control Chart (CUSUM) have limitations in complex data environments.
• Need for advanced fault detection algorithms resilient to noise and adaptable to various data distributions.

Purpose of the Study:

• Introduce a novel Cascade Alarm System (CAS) for fault detection.
• Design an algorithm applicable to both Independent and Identically Distributed (IID) and non-IID variables.
• Enhance alarm system flexibility and noise resilience through a multilevel architecture.

Main Methods:

• Utilized the largest singular value of the signal for fault detection.
• Employed k alarm subsystems, with the first subsystem using singular values from the Lagged Covariance Matrix (LCM) of a sliding window.
• Leveraged Singular Value Decomposition (SVD) for a multilevel alarm architecture.

Main Results:

• The Cascade Alarm System (CAS) demonstrated satisfactory performance under various fault conditions.
• The multilevel SVD-based structure provided a filtering property, enhancing resilience to noise.
• CAS exhibited superior performance compared to median filters, moving average filters, delay timers, CUSUM, and serial methods.

Conclusions:

• The proposed Cascade Alarm System (CAS) is an effective fault detection method for both IID and non-IID data.
• The SVD-based multilevel architecture offers flexibility and improved noise immunity.
• CAS presents a promising alternative to conventional alarm system designs.