(39c) Some Challenges in Analyzing Manufacturing Data for Process Optimization

In recent years, manufacturers collect and store more data than was previously feasible due to the drop in the costs of both sensor technologies and data storage. With the significant advance of analytics techniques, valuable information could be extracted by analyzing manufacturing data to provide insight into process optimization. This presentation discusses several challenges in data pre-processing and statistical modeling.

With hundreds or thousands of process variables showing different patterns, there are risks in developing scripts to automate the data cleaning process. In addition, switches between products and process maintenance cause difficulties in performing time series analysis to investigate cyclical patterns in the process. Furthermore, the partial least squares (PLS) method can be used to identify relatively important variables for root cause investigation when there are high correlations between process variables. However, with hundreds of process variables being important in the PLS model, extra effort is required to judge the impact of these process variables.