Electrical Resistance Tomography-Based Multiphase Level Measurement Methods: A Comparison Study

The electrical resistance tomography (ERT) is an effective tool for imaging and analyzing the complex multiphase mixtures that are commonly encountered in various industrial processes. In the flotation column, the level measurements of pulp-froth interface could be made by combining the ERT with the linear sensor. Usually, the ERT images could not present clear phase boundaries; therefore, requiring a proper method to extract the interface level. In this article, two methods have been proposed, i.e., the image reconstruction method and neural network (NN)-based method. The image reconstruction method relies on the accurate image reconstruction results from which the interface is calculated. The NN-based methods do not require image reconstruction and can calculate the results directly from the raw measurements. The artificial NN (ANN) and radial basis function network (RBFN) methods have been investigated. In the numerical simulations, the image reconstruction method has demonstrated the relative error of ±6%. Comparatively, the relative error of ANN method is ±1%, while the relative error of RBFN method is ±0.5%. Experiments have also been carried out with a linear sensor of 760-mm height. In the experiments, similar accuracy levels have been achieved. The relative error of the image reconstruction method is ±8%, while the relative errors of the ANN method and RBFN method are ±1% and ±0.5%, respectively. It could be concluded that the NN-based methods could achieve higher accuracy as compared with the image reconstruction method but at the cost of longer calculation time. In both the simulations and experiments, the NN-based methods have achieved the same relative errors, suggesting their good capability in dealing with measurement noises in real industrial applications. As for the NN-based methods, the RBFN method requires less calculation time than the ANN method. It has been suggested that the image reconstruction method could be considered if fast measuring rate is needed. At the same time, the NN-based methods could achieve better accuracy and reliability, but usually at the cost of longer processing time. It might be useful to develop the hardware system that embeds the NN processing unit.