A Gaussian process approach for monitoring autocorrelated batch production processes

Abstract

AbstractIn statistical process monitoring, statistical control tools are used to identify deviations from normal operating conditions. In many industrial processes, such as batch production processes, multiple process variables must be monitored as they play a key role in the quality of the final product. Several monitoring tools are found in the literature that deal with the case of multiple process variables, but a few of them deal with the case of autocorrelated data. Existing tools that are used to monitor autocorrelated process variables have two main drawbacks. First, they are run offline. That is, monitoring is performed at the end of the production cycle and hence no corrective actions can be made. Second, these tools utilize dimensionality reduction techniques to solve for computational complexity issues. A Gaussian Process–based modeling approach is proposed in this work to monitor batch production processes online. The proposed monitoring approach takes into consideration of both the correlation between process variables and the within variable autocorrelations. A simulated and two real data sets were used to evaluate the performance of the proposed modeling approach. Model performance metrics showed that the proposed modeling approach has similar performance to the optimal case of Shewhart type control charts. Process status classification accuracy of about 92–98% were achieved for the cases considered.