Environmental statistical process control using an augmented neural network classification approach

Shifts in the values of monitored environmental parameters can help to indicate changes in an underlying system. For example, increased concentrations of copper in water discharged from a manufacturing facility might indicate a problem in the wastewater treatment process. The ability to identify such shifts can lead to early detection of problems and appropriate remedial action, thus reducing the risk of long-term consequences. Statistical process control (SPC) techniques have traditionally been used to identify when process parameters have shifted away from their nominal values. In situations where there are correlations among the observed outputs of the process, however, as in many environmental processes, the underlying assumptions of SPC are violated and alternative approaches such as neural networks become necessary. A neural network approach that incorporates a geometric data preprocessing algorithm and identifies the need for increased sampling of observations was applied to facilitate early detection of shifts in autocorrelated environmental process parameters. Utilization of the preprocessing algorithm and the increased sampling technique enabled the neural network to accurately identify the process state of control. The algorithm was able to identify shifts in the highly correlated process parameters with accuracies ranging from 96.4% to 99.8%.