Neural Component Analysis for Key Performance Indicator Monitoring

Area of Science:

• Process monitoring and control
• Machine learning for chemical engineering
• Data analysis and modeling

Background:

• Partial Least Squares (PLS) is a standard method for Key Performance Indicator (KPI) related performance monitoring.
• Traditional PLS struggles with nonlinear features inherent in many industrial processes.
• Existing methods like Kernel PLS (KPLS) and Neural Component Analysis (NCA) have limitations in handling complex process data.

Purpose of the Study:

• To develop an advanced process monitoring method that effectively handles nonlinear relationships in data.
• To introduce Neural Component Analysis (NCA)-Partial Least Squares (NCA-PLS) as a superior alternative to existing techniques.
• To improve the extraction of components strongly correlated with KPIs for better data reconstruction and analysis.

Main Methods:

• Proposed NCA-PLS, integrating NCA principles with PLS by introducing a novel loss function and principal component selection mechanism.
• Re-derived gradient descent formulas for efficient neural network training within the NCA-PLS framework.
• Validated the method using simulation tests on a mathematical model and the Tennessee Eastman process.

Main Results:

• NCA-PLS successfully extracts components with high correlations to KPI variables, enabling effective data reconstruction.
• Simulation results demonstrate NCA-PLS's capability to handle complex nonlinear relationships in process data.
• NCA-PLS significantly outperformed standard PLS, KPLS, and NCA in performance evaluations.

Conclusions:

• NCA-PLS is a robust and effective method for process monitoring, particularly in the presence of nonlinearities.
• The proposed approach offers enhanced performance for KPI-related analysis compared to conventional methods.
• NCA-PLS provides a valuable tool for improving the accuracy and reliability of industrial process monitoring.