Detection of blade contamination in turbine flowmeters using neural networks

The accumulation of contaminating materials on the rotor blades of a gas turbine flowmeter is a condition of concern to the manufacturers, since it can lead to loss of calibration. Over time, a deposit of tar and heavy oil mixed with sand and dust can build up on the rotor blades, which causes the meter factor to change from its original value of calibration. A novel neural network technique is described for the detection of blade contamination. The method uses a C language implementation of the multi-layer perceptron (MLP) neural network, which uses a cumulative back-propagation (BKP) algorithm to train and be tested on blade passing time vectors (BPT).