Infrared monitoring and identification of lubricating oil pipe on a gas turbine based on faster RCNN

Abstract

The reliable work of gas turbine depends heavily on the good operation of the lubricating oil system and monitoring the temperature of lubricating oil removing heat from the heated end components can indirectly reflect the working state of the gas turbine, so the research on the infrared monitoring and identification of lubricating oil pipe on a gas turbine is of great significance to analyze the working state of the gas turbine. The method of infrared detection and recognition based on deep neural network is proposed in this paper, three different pipes were identified in a certain field of vision. The network obtained was tested by testing data sets and was found to be unable to give a unique label to an object in the image, and it was not effective to identify the infrared characteristics of the lubricating oil pipe. In order to accurately identify the three pipes, the pre-training AlexNet, Vgg16, Vgg19 and Resnet50 networks were used for training and testing the pipes by Faster RCNN algorithm. The results show that the method can accurately identify the lubricating oil pipe and other parts on the gas turbine, and the identification accuracy are more than 80%. Comparing with the training and testing results of four kinds of the transfer networks, the detection accuracy of Resnet50 transfer network for the oil pipe was lower than that of the other three transfer networks in the case of background interference, and the accuracy of the method was 89.5%.