GA-BORN ADAPTIVE UNIT

This section introduces the concept of GA adaptive unit for BORN. Both a simple architecture for the GA-BORN adaptive system and simulation results of a hydrocyclone modeling problem are presented here.

GA-BORN Adaptive System Architecture

In the classic application of the BORN algorithm, once the BORN algorithm is executed, the NN connectivities cannot be changed. In order to alter the NN connectivities after BORN optimization, a GA is implemented as an adaptive unit for further optimal connection selections. Figure 11.11 shows the simple GA-BORN Adaptive System Architecture Diagram. In this figure, an NN is trained as a general controller with BORN optimization for the system. Once simulation starts, the Performance Estimator checks the NN controller output error. The GA adaptive unit should be executed only when the Performance Estimator detects the unacceptable NN output for the system due to sudden changes in input patterns, or the environment. For example, the NN is trained and optimized by BORN for input U=sin x as a general controller. In the case of input frequency and phase changed to U = sin(2x + π/2) for some reason during operation, the GA adaptive unit will look for more optimal connectivities for the NN controller. The advantage of this method is that since only the NN connectivities changed, and NN weights, which is the memory part of the NN, will not be changed by GA, the NN can retain its memory which was previously optimized by BORN. In other words, in order to regain the original NN after the GA optimization, the previous connectivities can be applied without further NN training. Decode and fitness functions are essentially the same as described in previously in this chapter.

Figure 11.11  GA-BORN adaptive system architecture diagram.

Hydrocyclone Modeling Problem

Hydrocyclone modeling problem is used for the performance evaluation for GA-BORN adaptive unit. Hydrocyclones are commonly used for separating slurries in the mineral processing industry [11], and are used extensively to perform separations in other industries as well. Hydrocyclones utilize centrifugal forces to accelerate the settling rate of particles. Since their mechanical structure is simple, durable, and relatively inexpensive, hydrocyclones are one of the most popular mineral separation devices found in industry. They are used in closed-circuit grinding, de-sliming circuits, de-gritting procedures, and thickening operations. Figure 11.12 shows a schematic of a typical hydrocyclone. The lower part of the hydrocyclone has a conical vessel shape with an opening at the apex that allows the coarse or heavier particles to be removed. The top of the hydrocyclone is a cylindrical section that is closed with the exception of an overflow pipe called a vortex finder. This vortex finder prevents the mineral sample from going directly into the overflow while allowing the fine particles to remain in the hydrocyclone. The actual mineral separation occurs in this cylindrical section due to the existence of a complex velocity distribution that carries the fine particles out the top and the coarse particles to the apex.

Figure 11.12  A typical hydrocyclone.

There have been numerous efforts in the area of modeling hydrocyclones. Plitt [12] developed a statistical model to predict the split size of the hydrocyclone, D₅₀. The split size is that size particle that has a 50 percent chance of exiting in either the overflow or the underflow. The Plitt’s model has proven to be quite robust and is often cited in the literature. Despite the numerous efforts to model hydrocyclones, none of the models has been universally adapted. Most of the models are either too computationally intensive or are only applicable to a limited range of hydrocyclone designs. To facilitate this hydrocyclone modeling effort, hydrocyclone performance data has been acquired from the U.S. Bureau of Mines. This data pertains to a wide range of designs used on a variety of mineral samples including iron, phosphate, and copper. A computer model of a hydrocyclone must consider several input parameters to compute a value of D₅₀. Figure 11.13 is a schematic of a hydrocyclone model where:

Figure 11.13  Hydrocyclone computer model diagram.