Research On Soft Sensor For Component Content Of Rare Earth

China is known to have the largest reserve of rare earth resources in the world. Therare earth extracted separation technology of China has reached the most advanced levelin the world. How to obtain maximum yield and purity of the rare earth products in theshortest time as possible is the ultimate goal. Rare earth cascade extraction includeshydrometallurgy and automatic control technology and the on-line testing of rare earthcomponent content is the key technology. In the world there is no appropriate sensor totest the rare earth comnent content because of the strong corrosive in the extraction slot.In most Chinese rare earth companies use off-line testing methods to acquire thecomponent content. The testing process is hysteretic so limit to the closed-loop controland the production efficiency is also influenced. Realizing the on-line testing is aproblem to be solved and has very important value from theory research to practicalapplication.The article adopts neural network soft sensor model to realize the on-line testingafter analysing the mechanism of rare earth extract separation. The paper deeplyanalyses various methods of neural network from the network structure, nonlinear andconvergence speed and finally choose the RBF neural network to create soft sensormodel. The main works are as follows:1、Analyse the extract reaction mechanism and find the main factors that influencethe rare earth component content.2、Analyse all kinds of neural network modeling theory, confirm one suitablemodeling method.3、Create the soft sensor model, determine the auxiliary variables, the hidden nodenumber and the output weight matrix.4、Use the matlab tool to simulat and validate the model and analyse the testingresults.5、Finish the software and hardware design of real-time monitoring system to showthe craft parameters of rare earth production line.The simulation results indicate that the soft sensor model meets the requirementsof rare earth extraction process and the model can be used to estimate the componentcontent of rare earth.