Control quality assessment using fractal persistence measures

Area of Science:

• Process Control and Automation
• Statistical Analysis
• Data Science

Background:

• Control Performance Assessment (CPA) significantly impacts industrial throughput, efficiency, and environmental footprint.
• Existing CPA methods include time-domain, Minimum Variance, and various statistical measures.
• Process industry data often exhibits non-Gaussian, fat-tailed distributions, complicating traditional analysis.

Purpose of the Study:

• To evaluate the robustness of different Control Performance Assessment (CPA) measures against signal disturbances.
• To investigate the effectiveness of fractal-based measures compared to standard statistical approaches.
• To identify reliable CPA methods for non-Gaussian process data.

Main Methods:

• Analysis of industrial production data with non-Gaussian characteristics.
• Testing Gaussian standard deviation and fat-tail distribution factors.
• Focusing on persistence measures derived from rescaled range (R/S) plots.
• Investigating the robustness of measures against disturbances with varying statistical properties.

Main Results:

• Standard statistical measures can be obscured by strong disturbances in process signals.
• Fractal measures, specifically those analyzing the R/S plot, demonstrate resilience to disturbances.
• Fat-tail distribution factors and integral indexes were also evaluated for their performance.
• Results confirm the potential of fractal measures as a robust alternative.

Conclusions:

• Fractal-based measures provide a robust approach to Control Performance Assessment (CPA).
• These methods are particularly valuable for analyzing industrial data with non-Gaussian distributions and significant disturbances.
• Fractal measures offer a reliable alternative to traditional statistical techniques in challenging process environments.