The key to process manufacturing success is a mixture of knowledge and experience supported by mastery of data. A presentation I recently attended put this into sharp focus. A major paper manufacturer was faced with the challenge of maintaining paper smoothness during production. They approached this problem in a way that gave them enormous insight into their process, the ability to control it in real time, and ultimately lead to cost savings and maintained quality. There were also a few added benefits, including the ability to spot, diagnose and solve problems in real time.
Challenges of this nature generally involve the demanding, but very rewarding, process of diving into historical data. The production team gathered data for the previous year, covering the production of more than 500 paper reels at one-minute resolution, including over 250 process variables, a number of quality variables, such as smoothness, plus a few calculated variables. They then refined the data by using Multivariate Data Analysis batch modeling methods to focus on within-reel variations that were correlated to smoothness.
The team used the data to develop a predictive model that gave a very good explanation of variation in smoothness and could tolerate changes in the process. They could build on this by using PLS and OPLS techniques to identify process variables that are highly correlated to quality and performance. The team confirmed that the model was good for prediction and also for control by demonstrating that the correlations were causal. This gave them the confidence needed to build the model into SIMCA-online to estimate the effect of process adjustments on smoothness. Based on knowledge and experience, the team identified a handful of key parameters that could be controlled to improve smoothness and could close the circle by showing that the change achieved by adjusting a parameter indeed matched the prediction from the software.
This approach ultimately helped the production team to gain control over the process, based on their knowledge and experience, and a deeper understanding their data. They could move from reactive action on historical data to proactive action in real-time. The multivariate approach also provided the team with a way to transfer expert knowledge into the process to achieve sustainable improvements.