Kernel Generalized Likelihood Ratio Test for Fault Detection of Biological Systems

Area of Science:

• Biotechnology
• Process Engineering
• Data Science

Background:

• Monitoring nonlinear biological processes requires robust fault detection (FD) techniques.
• Traditional methods struggle with multiscale data common in complex biological systems.
• Kernel principal component analysis (KPCA) offers a data-driven approach for nonlinear process monitoring.

Purpose of the Study:

• To develop an improved fault detection technique for nonlinear biological processes.
• To enhance process monitoring by addressing multiscale data characteristics.
• To introduce a multiscale kernel generalized likelihood ratio test (MS-KGLRT) detection chart.

Main Methods:

• Combined kernel generalized likelihood ratio test (GLRT) with multiscale wavelet representation.
• Utilized KPCA for model computation in feature space.
• Developed a multiscale kernel GLRT (MS-KGLRT) detection chart for fault identification.
• Applied the MS-KGLRT chart to synthetic data and a biological Cad System in E. Coli (CSEC) model.

Main Results:

• The MS-KGLRT chart demonstrated effectiveness in detecting small and moderate shifts (offset, bias, drift).
• The proposed method successfully enhanced fault detection in the CSEC model.
• Key variables like enzymes, lysine, and cadaverine in the CSEC model were effectively monitored.

Conclusions:

• The MS-KGLRT approach significantly enhances fault detection capabilities for nonlinear biological systems.
• This technique offers improved monitoring of complex biological processes with multiscale data.
• The study validates the MS-KGLRT chart's performance on both synthetic and real biological data.