Process modeling and identification, Monitoring and performance assessment, Neural networks in process control, Fault detection and diagnosis, Condition Monitoring
Abstract
This work proposes an algorithm based on an evolving neural autoencoder for process monitoring. The proposed algorithm performs learning from scratch from an available data stream and all model parameters are updated iteratively. Furthermore, monitoring functions and associated control limits are defined to identify events that exceed the control limit in the monitored process. Also, contribution indices of the process variables are established to relate the variables to the events, allowing the root cause to be identified. The proposed algorithm is validated in a three-coupled tank system.
