Abstract: A tool, and the method of making the tool, for process system identification that is based on the general purpose learning capabilities of neural networks. The tool and method can be used for a wide variety of system identification problems with little or no analytic effort. A neural network is trained using a process model to approximate a function which relates process input and output data to process parameter values. Once trained, the network can be used as a system identification tool. In principle, this approach can be used for linear or nonlinear processes, for open or closed loop identification, and for identifying any or all process parameters.

Abstract: An automatic tuner for control systems that produces, as output values, parameters of an arbitrary controller. The controller is in a control loop so that its output effects changes in actuators and regulates a physical process. The controller has either linear or nonlinear controller components, or a combination of both. The tuner has a nonlinear approximator that has been optimized off-line. The off-line optimization is done without supervised learning so that desired outputs of the nonlinear approximator do not need to be available, and separate optimization to generate the desired outputs is not necessary. The off-line optimization can also rely on arbitrary criteria. Such optimization ensures robustness of generated controller parameters so that the input process characteristics do not need to be highly accurate. The inputs to the nonlinear approximator consist of two sets of input parameters, either of which may be empty. A first set of input parameters can relate to process characteristics.

Abstract: An apparatus for identifying the structure and estimating the parameters in the structure of a controlled system is described. The apparatus uses a structure identifier, in a preferred embodiment, employing a neural network, to match the relationship of the system inputs and outputs to a mathematical relation. This identified structure is communicated to a parameter estimator. The parameter estimator may employ a neural network to give an initial parameter vector which can be used to generate, iteratively, better parameter estimates. This increases the efficacy of the parameter estimator. The apparatus can be inserted into controllers for various systems to improve performance. Using a neural network for parameter estimation as shown can be generalized to any nonlinear optimization problem.

Abstract: The method of making the tool, for process system identification that is based on the general purpose learning capabilities of neural networks. The method can be used for a wide variety of system identification problems with little or no analytic effort. A neural network is trained using a process model to approximate a function which relates process input and output data to process parameter values. Once trained, the network can be used as a system identification tool. In principle, this approach can be used for linear or nonlinear processes, for open or closed loop identification, and for identifying any or all process parameters.

Abstract: A closed loop neural network based autotuner develops optimized proportional, integral and/or derivative parameters based on the outputs of other elements in the loop. Adjustments are initiated by making a step change in the setpoint which may be done by a user or automatically. A Smith predictor may also be employed.

Abstract: A controller based on a neural network whose output is responsive to input signals that represent user or designer defined control system parameters which may include process parameters, control parameters and/or disturbance parameters. The neural network can be "trained" to mimic an existing controller which may or not receive inputs of control system parameters. The trained neural network controller may have advantages of faster execution and reduced code size. The neural network can also be trained to result in a nonlinear controller that is more powerful than an existing controller.

Abstract: PID controllers form a large proportion of controllers in use in many controlled systems today. This application describes how to use a neural network which receives PID inputs to be a controller and operate as a PID controller to save on retraining and provide other efficiencies in control. Also shown is the user selectability between PID conventional controllers and Neural Network controllers.

Abstract: The disclosure relates to the use of genetic learning techniques to evolve neural network architectures for specific applications in which a general representation of neural network architecture is linked with a genetic learning strategy to create a very flexible environment for the construction of custom neural networks.

Abstract: A method for scale and rotation invariant pattern recognition. The method involves the recogniziing of patterns, such as visual images of objects, irrespective of scale or two-dimensional orientation. A prior art format for representing data known as a polar exponential grid or PEG format is utilized. For storage, patterns are centered, transformed into a PEG representation, and the centered PEG representation is stored in an associative memory. For recall, patterns are likewise centered and transformed into a PEG representation. They are then first scaled up until the outer layer of the PEG grid is activated, and then incrementally rotated until either a desired level of match is found with some stored pattern, or until a 360 degrees rotation has been completed without a match. Hardware implementation for the PEG transformation and for the incremental rotation and scaling are described.

Abstract: A neural network associative memory which has a single layer of primatives and which utilizes a variant of the generalized delta for calculating the connection weights between the primatives. The delta rule is characterized by its utilization of predetermined values for the primitive and an error index which compares, during iterations, the predetermined primative values with actual primative values until the delta factor becomes a predetermined minimum value.

Abstract: The invention relates to a neural network centering scheme for translation-invariant pattern recognition. The scheme involves the centering of a pattern about its centroid to prepare it for subsequent subjugation to an associative match. The scheme is utilized in a camera assembly of the type used for image acquisition. Movement of the camera assembly is controlled in accordance with the scheme to effect the centering of a pattern in the field of view window of the camera assembly.