Using Advanced Analytic Tools to Gain Real-Time Insights

In Chapter 4, we presented how classifying the raw data using operational modes provides real-time insights with no additional software tools. We showed the value of creating a template that used basic production and consumable variables to characterize the operation. The EIDI is capable of aggregating the data for overall production effectiveness evaluation by unit and by shift.

The running OK mode data has shown to be extremely valuable in generating predictive analytics of quality variables based on the operational data captured from this data subset. Additional information generated for the running OK data reduces the complexity and dimensionality of the problem so it is easy to understand.

Principal component analysis (PCA) is a statistical tools method for dimensionality reduction. The multivariable operational data set, consisting of several laboratory variables and process variables, are treated with PCA. In process applications, the data are usually multivariate, collinear, noisy, and typically missing. This makes it difficult or impossible to use full- rank statistical methods for modeling and analysis, that is, multiple regression, discriminant analysis, analysis of variance, as well as neural networks. However, projection methods such as PCA and projection of latent squares (PLS) are based on realistic assumptions about the variables (collinear, noisy, etc.), making them suitable for modeling and analysis of complicated processes and other data. PCA and PLS analysis results can be displayed graphically, including multivariable control charts and multivariable statistical process control charts (MSPCs), for example, Shewhart, exponentially weighted moving average (EWMA), and cumulative sum (CUSUM) charts. These MSPC charts, combined with charts based on the model residuals, provide tools for early fault detection and the identification of drift, mean shifts, and so forth. Additional plots indicate the variables that are likely related to process problems upsets and other process events. Garrigues et al. (2000) shows additional ways to treat the data. And many others have provided valuable descriptions and examples in the chemical, iron, steel, and mineral processing industries such as MacGregor and Kourti (1998) and Dudzic (1998). These EIDI online diagnostic tools can also be implemented for both operations.

 
Source
< Prev   CONTENTS   Source   Next >