PURPOSE: To use a general-purpose technique for various system discriminations without requiring a mathematical effort by adopting a process delay discrimination technique as an open loop discrimination technique based on a learning neural network approach.

CONSTITUTION: When a delay estimate is required, a control loop is broken through a switch 54, and a disorder of step input is applied to a process 56 through a step input generator 59 by a unit 52. The response of the process 56 to this disorder is sampled and is stored in a buffer 60. When the sufficient number of samples are received, a vector of samples is used as the input to a trained neural network 30. The output of the network is processes by a postprocessor 62 to obtain the delay estimate. When the delay estimate is inputted to a Smith predictor, the switch 54 is closed again to return the process to the close loop control.