Time dependent Pinch analysis with sensor data and unsupervised learning algorithms

Abstract

Abstract Heat integration is essential for energy efficiency and operational cost reduction in energy-related applications. Pinch Analysis is a heat integration model that has attracted a lot of attention over the past few decades. Although the method has been successfully applied to many industrial processes, it has also been shown to have many limitations. One of these limitations is that the method only considers the average values of the physical quantities of a process, and therefore is not suitable for studying dynamic processes that undergo variations in their operating conditions. Attempts have been made to improve Pinch Analysis to circumvent this limitation, but they have been found to oversimplify and result in unrealistic solutions. The purpose of this work is to use sensor data and unsupervised learning algorithms to extend Pinch Analysis to dynamic processes. Specifically, we use time series segmentation to detect changes in the process operating conditions and clustering to group similar segments. The approach is applied to a real industrial use case. We compare the performance of heat exchanger network with and without data analysis. The solutions obtained with our methodology improves waste heat recovery by 20% while reducing the cost of the heat exchanger network by 31%.