Simplification of Data Acquisition in Process Integration Retrofit Studies Based on Uncertainty and Sensitivity Analysis

Abstract

Process integration methodologies proved to be effective tools in identifying energy saving opportunities in the industrial sector and suggesting actions to enable their exploitation. However, they extensively rely on large amounts of process data, resulting in often overlooked uncertainties and a significant time-consumption. This might discourage their application, especially in non-energy intensive industries, for which the savings potential does not justify tedious and expensive analysis. Hereby a method aimed at the simplification of the data acquisition step in process integration retrofit analysis is presented. Four steps are employed. They are based on Monte Carlo techniques for uncertainties estimation and three methods for sensitivity analysis: Multivariate linear regression, Morris screening, and Variance decomposition-based techniques. Starting from rough process data, it identifies: (i) non-influencing parameters, and (ii) the maximum acceptable uncertainty in the influencing ones, in order to reach reliable energy targets. The detailed data acquisition can be performed, then, on a subset of the total required parameters and with a known uncertainty requirement. The proposed method was shown to be capable of narrowing the focus of the analysis to only the most influencing data, ultimately reducing the excessive time consumption in the collection of unimportant data. A case study showed that out of 205 parameters required by acknowledged process integration methods, only 28 needed precise measurements in order to obtain a standard deviation on the energy targets below 15 and 25% of their nominal values, for the hot utility and cold utility respectively.

Identification of optimal measurement points for energy monitoring of industrial processes: The case of milk powder production

“…With this goal in mind, the identification of important measurements was performed according to the procedure firstly presented by Bergamini et al [18] and later extended by the same authors [20], named Required Data Reduction Analysis. Key of the method is to recognise that the ultimate purpose of measurements is to provide knowledge on the measured parameters, which is quantifiable in terms of uncertainty.…”

“…This paper demonstrates how to use the approach formulated by Bergamini et al [18], named Required Data Reduction Analysis (RDRA), to systematically answer three recurring questions of industrial companies in the field of energy monitoring systems: (i) what should be measured? (ii) how precise should this measurement be?…”

“…This feature can be further exploited for the identification of optimal measurement points for energy monitoring purposes. The approach [18] combines one of the most renowned PI tools (i.e. Pinch Analysis [22]) with uncertainty and sensitivity analysis methods and optimisation routines.…”

“…The probability distribution was centred in the parameter nominal value (µ), while its standard deviation (σ) was assumed based on the type of parameter to be measured, distinguishing between (i) temperature, (ii) volume flow rate, (iii) mass flow rate, (iv) density, (v) pressure, and (vi) other. The standard deviation of the roughly estimated parameters was set to high values according to what was recommended by Bergamini et al [18]. Finally, the minimum obtainable uncertainty was set according to the minimum inaccuracy of marketed measurement sensors, as reported by Liptak [37].…”

A Surrogate Modeling Approach for the Development of Biorefineries

“…Boukouvala and Floudas (2017) presented an algorithmic optimization framework that combines variable selection, bounds tightening, and constrained sampling techniques to develop accurate surrogate representations of unknown equations. Bergamini et al (2019) introduced a method to identify systematically process parameters that need precise measurements and reduce the data acquisition time. Hao et al (2020) integrated simulation data and monotonicity knowledge through Bayesian optimization to reduce the required number of computer simulations for finding the optimal design solution.…”