Fouling still remains one of the most difficult problems for the use of heat exchangers. A methodological process of advanced analysis of experimental data on heat exchangers fouling allowing building predictive models is necessary to determine the fouling degree. Here, three different methods were used to predict the fouling resistance from some easily measurable variables of the system which are: Kern and Seaton, Partial Least Squares (PLS) and Artificial Neural Networks (ANN). Indeed, the fouling resistance was estimated according to the inlet and outlet temperature of the cold fluid, the temperature of the hot fluid, the density and the volume flow rate of the cold fluid and time for three types of heat exchangers, i.e. tubular stainless-steel and graphite blocks (Supplier (A) and Supplier (B)).The best modeling was determined by maximizing certain statistical accuracy indices. Results show that modeling by the use of Artificial Neural Networks is very performing compared with modeling by Partial Least Squares regression and Kern and Seaton. One of the key features of ANN model is their small levels of error in comparison with other models.

对于热交换器的使用，结垢仍然是最困难的问题之一。为了确定结垢程度，需要对热交换器结垢的实验数据进行高级分析的方法论过程，允许建立预测模型。这里，使用三种不同的方法根据系统的一些易于测量的变量来预测结垢阻力：科恩和西顿，偏最小二乘（PLS）和人工神经网络（ANN）。实际上，根据三种类型的热交换器（即管状不锈钢热交换器）的冷流体的入口和出口温度，热流体的温度，冷流体的密度和体积流率以及时间来估算抗结垢性。钢和石墨块（供应商（A）和供应商（B））。通过最大化某些统计准确性指标来确定最佳建模。结果表明，与使用偏最小二乘回归和Kern和Seaton进行建模相比，使用人工神经网络进行建模的效果非常好。ANN模型的主要特征之一是与其他模型相比，它们的错误水平小。