System cultivation : process monitoring and control using the historical database

Process computers are playing an increasingly important role in the chemical industry. They not only provide a system in which to implement modern control concepts but also a framework in which large amounts of process data can be collected and stored cheaply and efficiently. Today, at the fingertips of process engineers lies a vast amount of historical data, containing a wealth of information about the operation of the process. Unfortunately the engineer is often overwhelmed by the volume of data. Subtleties which could be helpful in improving the operation of the process systems can be easily lost in the vastness of numbers and records. A recent perspective in process control, system cultivation focuses on the day-to-day analysis and evaluation of the operation of the process and its control system. As today's processes become more complex and the control systems for these processes become more sophisticated, this perspective is becoming increasingly important. The often fragile nature of these complex designs demands the ability to monitor, detect, and diagnose problems quickly. System cultivation is also concerned with studying the normal operation of the process for making continual improvements. There should always be room for improvement. Analyzing the historical data base properly will provide clues and directions which would improve even the normal conditions of operations. This dissertation works towards three objectives. First, it refines two existing cultivation tools: hypothesis feedback modeling and time-shifted distribution analysis. Second, it introduces pattern recognition and neural net works to the cultivation framework. Finally, the primary objective develops a framework that incorporates several cultivation tools for effectively detecting and identifying subtle operational changes. Two simulated unit operations, a heat exchanger and a continuous stirred-tank reactor, aid in developing and demonstrating these cultivation concepts.