Method for Detecting Disorder of a Nonlinear Dynamic Plant

Area of Science:

• Process control and monitoring
• Machine learning applications in engineering
• Dynamic system analysis

Background:

• Accurate disorder detection is crucial for maintaining the stability and efficiency of dynamic industrial processes.
• Traditional methods like CUSUM and EWMV may struggle with complex nonlinear systems.
• Model-free approaches are desirable for real-time monitoring without requiring a reference model.

Purpose of the Study:

• To introduce a novel, model-free disorder detection method named CCF-AE.
• To evaluate the performance of CCF-AE against established methods in a nonlinear process.
• To demonstrate the efficacy of CCF-AE for complex dynamic systems.

Main Methods:

• Developed the CCF-AE method utilizing cross-correlation functions and neural network autoencoders.
• Applied CCF-AE to monitor a nonlinear pH neutralization reaction process.
• Compared CCF-AE performance with Cumulative Sum (CUSUM) and Exponentially Weighted Moving Variance (EWMV) control charts.

Main Results:

• CCF-AE achieved a superior true detection rate compared to CUSUM and EWMV.
• The proposed CCF-AE method exhibited a significantly lower false alarm rate.
• CCF-AE demonstrated enhanced capabilities in detecting disorders within complex nonlinear processes.

Conclusions:

• CCF-AE is an effective model-free method for disorder detection in dynamic plants.
• The method shows significant advantages over traditional control charts for nonlinear systems.
• CCF-AE offers a robust solution for real-time process monitoring and anomaly detection.