Fouling Severity Classification in Industrial Heat Exchanger using Probabilistic Neural Network

Data-driven artificial intelligence-based approaches are gaining more popularity to solve industrial problems, which have been investigated in this work. Industrial Heat Exchangers (HE) are subjected to fouling and it is essential to know its severity to ensure safe and reliable operation. Design of a Probabilistic Neural Network (PNN) based fouling severity classifier has been investigated. Naphtha cooler, one of the industrial HE widely used in petroleum refineries, is considered in this work. Measurement data under various Fouling conditions are acquired from the Naphtha cooler. Fouling Resistance (FR) guided data labeling technique is adopted to label the fouling severity of the acquired data. It minimizes manual overhead and ensures accurate labeling. The spread value of the PNN classifier is also analyzed to eliminate the overfitting problem and to improve the classification performance. Confusion matrix is used to validate the classifier performance. Experimental results illustrate about 98.5% (training dataset) and 91.0% (testing dataset) of classification accuracy for the spread of 0.16. Other performance metrics are also evaluated and the proposed classifier is observed to be reliable.