Neuro-ICA Based Process Monitoring Strategy for Fault Detection in Steel Billets Manufacturing Unit

The proposed monitoring strategy addresses the challenges associated with non-linear process behavior through the application of the Neural Network Fitting (NNF) Model technique, while non-Gaussian process characteristics are effectively managed using the I2 control chart. The study focuses on developing a monitoring strategy that employs statistical techniques while accounting for nonlinear as well as non-normal or non-Gaussian data. The proposed statistical monitoring strategy employs a multivariate I² control chart for non-normal data, while data nonlinearity is addressed using a Neural Network Fitting model. Quality characteristics of steel billets are initially processed through a Neural Network model to mitigate nonlinear patterns. The fully or partially linearized data are then evaluated using the I² control chart, with its control limits determined through the Bootstrap procedure. Observations identified as out-of-control are classified as faults, and their detection prompts the implementation of suitable corrective measures.