Improvement of the ANN inversion based soft-sensing method and its application in bioleaching process

The soft-sensing method based on assumed inherent sensor inversion (AISI) is presented in our previous work where the AISI as a soft-sensor is constructed only by using the information of directly-measurable state variables. In this paper, the AISI based soft-sensing method is greatly improved by the following two means: firstly, the directly-measurable variables used to construct the AISI are extended from the state variables to the so-called functional variables that are the nonlinear functions of states; secondly, the previous modeling algorithm used to construct the AISI is also improved. These improvements not only significantly increase the probability of the successful construction of the AISI, but also make it possible to lower the orders of the derivatives of the directly measurable variables used to construct the AISI, thereby facilitating its practical use. In addition, a static artificial neural network (ANN) is used to approximate the AISI and then the ANN AISI is obtained, which overcomes the difficulty in constructing the AISI by analytic means, therefore making it more practical in engineering uses. Finally, the improved ANN AISI is applied to the bioleaching process, and the on-line soft-sensing (or estimation) of the directly-immeasurable state variables is achieved. The simulation results show that the estimation values of the ANN AISI well approximate to the actual ones, which verifies the validity of the ANN AISI.