Predictive Process Monitoring Research Articles

Predictive Business Process Monitoring with Tree-based Classification Algorithms

Predictive business process monitoring is a current research area which purpose is to predict the outcome of a whole process (or an element of a process i.e. a single event or task) based on available data. In the article we explore the possibility of use of the machine learning classification algorithms based on trees (CART, C5.0, random forest and extreme gradient boosting) in order to anticipate the result of a process. We test the application of these algorithms on real world event-log data and compare it with the known approaches. Our results show that.

On-Line Industrial Implementation of Process Monitoring / Control Applications Using Multivariate Statistical Technologies: Challenges and Opportunities

The global steel industry is stnvmg to improve product quality through excellence in operation. To support this, significant investments have been made in upgrading instrumentation, data acquisition and computing infrastructures. The expectation is that with more process and product data readily available, useful information and better process knowledge can be gained in a timely fashion. The problem that has developed is that with the large volumes of data available, the associated data analysis and modeling have become increasingly complex. As a result, much of the data is either not used or summarized / heavily compressed. This means that a significant amount of the information and knowledge resident in the data is lost, diminishing the returns from the investment made in the information technology infrastructure. A class of technologies that Dofasco has used to meet this data challenge is multivariate statistics (MVS), with a primary focus on Principal Components Analysis (PCA) and Projection to Latent Structures (PLS). These methods have been successfully applied to analyze data for a variety of purposes, which includes the development of online predictive models and process monitoring systems. Since 1993, Dofasco has been involved with over 70 off-line / on-line applications of this technology at our steel facility in Hamilton, Ontario, Canada. Through these applications, significant fmancial returns to the company have been generated.

On-line applications of multivariate statistics at Dofasco

There are many complexities involved in being able to accurately analyze and model production processes in a fully integrated steel facility. The drive for improved steel product quality has contributed to significant upgrades in instrumentation and data acquisition infrastructure with the hope of developing more useful information and better process knowledge. While this increased level of instrumentation has made more data available to the analyst, the associated data analysis and modeling problem has become more complicated due to the large and ever-increasing volume of process and product quality data. A class of technologies that Dofasco has used to support improved process modeling, feedback and statistical control in its automation applications is multivariate statistics with a primary focus on Principal Components Analysis (PCA) and Projection to Latent Structures (PLS). These methods have been successfully applied at Dofasco since 1993 to analyze data for a variety of purposes to develop on-line predictive models and process monitoring systems. Two on-line applications are described to illustrate the application of this technology. The first example is an on-line monitoring system that is used to observe the operation of the casting process at the mould area of Dofasco’s number one continuous slab caster. The second example is a control system for determining the optimal amount of reagent needed to accurately remove sulfur from pig iron.