Study Of Thermal Infrared Fault Diagnosis Technique Of Diesel Engine Based On BP Neural Network

Thermal infrared imaging technology began in fifties the last century, because ofgood concealment, high safety index, not subject to electromagnetic interference; itdevelops rapidly, it has become a newly developed technology recent years. Thetechnology was first used in the field of military, but, due to the development of theinfrared technology, it has been widely used in every field.The diesel engine is a common and widely used power equipment. Its durabilityand reliability are very important during the operation process. Because of internalstructure strengthening, the fault appearing is very possible when working in the harshconditions. In order to ensure the engine working normally, improve the safety andreliability, which must strengthen the administration of the engine working and preventthe fault early. Giving the problem of how to diagnose the fault early, there is a way ofnon-contact type testing, fault diagnosis based on the thermal infrared imagingtechnology.The technology of fault diagnosis has some advantage such as imaging visually,not subject to electromagnetic interference, non-contact type probing, being safe andreliable, judging accurately, being efficient, high-speed testing, especially applying tothe high-speed operating power equipment. The infrared detector can receive theimage information of the thermal field, when the diesel engine is operating rapidly andradiating. Then the potential fault point will be detected by the analysis of theabnormal phenomenon of the thermal field.The pretreatment of the thermal infrared image and feature extraction. Thepretreatment is very important for the next step, because it can eliminate the noiseinformation, strengthen the effective information, by median filter, smoothing filter,sharpening filter, and strengthen the image information. Then it will divide the imageby operator like Roberts, Sober and so on. And the feature extraction is the key of theimage recognition, it will also used in the ANNs. The methods of the feature extractioninclude the image moments, the information entropy of the image, coefficient transformation characteristics, edge extraction and so on.Back Propagation. It uses the error inverse algorithm as its learning rule forsupervised learning feedforward networks, it realize the multilayer network learning.When a given network input, the signal from input layer to the hidden layer,processed by hidden layer unit, then sent to the output layer, after processed, outputlayer unit to produce an output mode, if the output response and the expected outputmode has some differences, the error will be sent along the connection path backpropagation, at the same time to amend each layer connection weights, The featuresextracted from infrared image can be input BP network to train, finally judging thefeatures input by the output.