Heat exchangers have become critical components in different applications, as they are capable of achieving desired temperatures in either a system or a process. Typically, heat exchangers work differently under different operating conditions, and an accurate estimation of their performance is useful in their design, selection and control. However, due to the complexity of these systems, it is difficult to predict their behavior from a first-principles perspective. To overcome this problem, semi-empirical models (e.g., correlations equations) built from experimental prototypes have been proposed and used to formulate the behavior of heat exchangers into two main heat transfer coefficients for easy determination of the heat rate.
Unfortunately, models based on the correlation equations produce errors that compromise the system functionality. Such errors have been extensively studied in the literature. Research findings show that the process of classifying the data, from which semi-empirical models are derived, is a key element of promising approaches for minimizing the errors associated with heat transfer correlations in particular, and semi-empirical models in general. Notably, correlation equations are generally constructed from experimental data that have previously been classified into the appropriate operating conditions. The ease of the classification depends on the working fluids and the physical conditions under which these fluids are used to operate the thermal system. For instance, the visual classification is relatively simple for systems operating with working fluids under single-phase conditions and difficult for those undergoing phase changes. Despite technological advances and improved identification methods for reducing uncertainties in the classification process, it is still recommendable to eliminate any external intervention and – instead – directly classify the characteristic behavior of the system. An algorithm-based identification process that is less susceptible to external intervention can potentially achieve this goal.
On this account, Professor Arturo Pacheco-Vega from California State University and Dr. Gabriela Avila from Universidad Autónoma de San Luis Potosí proposed a new algorithmic identification and classification of experimental data from condensing heat exchangers. In particular, their main aim was to extract the operation regimes of heat exchangers working under possible condensation conditions, which could enable the construction of more accurate empirical models of these systems. Their research work which has been supported by the National Science Foundation is currently published in the journal, Numerical Heat Transfer Part A: Applications.
In their approach, the authors commenced their research by exploring the heat exchanger data derived from the literature. Their study comprised two parts. The first part described the classification method based on clustering techniques, while the second part involved applying artificial neural networks (ANNs) to assess the results of the cluster analysis-based algorithms independently. A Gaussian mixtures clustering (GMC) algorithm was applied to calculate the number of data groups and to classify such data into the respective clusters based on a maximum likelihood decision rule. Also, an ANN-based classification technique was proposed and applied to the same experimental data. Finally, the algorithmic classification was validated by comparing the cluster analysis results to those from the ANN and to a typical visual procedure.
The results, visually classified as film-, drop- and dry-surface condensation, demonstrated that the GMC-based methodology could accurately identify the data corresponding to all the system conditions, both simple and complex. This was confirmed via the independent assessment based on the novel ANN algorithm procedure. Additionally, the ANN-based allocation method depended on data patterns to allocate the into pre-established groups. The ANN approach exhibited a remarkable agreement with the clustering technique for the same data.
In summary, the study designed an algorithmic-based methodology to extract the operation regimes of heat exchanger data. It comprised a GMC algorithm to obtain the number of groups directly from the data and a machine learning decision rule to classify the data into the clusters accurately. The excellent results confirmed the possibility of accurately using algorithmic classification to determine different operation regimes of condensing heat exchangers as well as classification of data corresponding to those regimes. In a statement to Advances in Engineering, Professor Pacheco-Vega explained that algorithmic classification represents a promising alternative to visual-based procedures for analyzing thermal systems.
Pacheco-Vega, A., & Avila, G. (2020). Algorithmic performance-data classification of condensing heat exchangers. Numerical Heat Transfer, Part A: Applications, 79(1), 1-20.