Sensor Failure Detection Based on a Powell Neural Network Method

It is essential for the control system of a large sized thermodynamic system to detect sensor failures and then take pertinent measures to ensure the successful implementation of the control process. The authors have come up with a new type of failure detection method based on Powell neural network. Under this method a neural network observer is set up for each sensor of the thermodynamic system, which at first received an off line training. On this basis, failure detection and on line training were conducted simultaneously. The neural network training by the use of the Powell method features a rapid and stable training process. It has the merits of the ability to perform on line learning and diagnose the failure of a multiple of sensors. Actual tests on boilers show that the above method is highly effective.