A rapid prototyping co-simulation environment for neural control

A flexible approach to the development of control systems containing Artificial Neural Networks (ANNs) is to construct a co-simulation involving the two development packages of choice, where each provides the appropriate system components and they execute concurrently. This has advantages over the small number of control system development packages which support ANNs, in that the ANN components are provided by a specialist package which can offer more comprehensive facilities than would otherwise be available. This cc-simulation approach has been implemented on a Unix system using two fully-featured packages which have a large user base in their respective areas of applicability. One control scheme which has been investigated is indirect adaptive control and the implementation of this scheme within the co-simulation is described. Two examples of its operation are given; one based on a simple linear control problem, and another based on a nonlinear helicopter mathematical model. Construction of a neural control simulation environment by integration of two proprietary packages appears to be highly flexible approach. It enables a potentially wide range of different control strategies to be examined, without recourse to explicit programming, and by using ANN code generation, allows rapid prototyping of ANN controllers to take place. In principle, this approach can be utilised for cost-effective investigation of other promising technologies where inadequate support for control system simulation exists..