Engine Mechanical Fault Diagnosis System Classifier Design

The abnormal vibration occurred when the engine works failure, so you can get the vibration signal by the sensor directly to the engine vibration signal and use the appropriate method of spectral analysis and classification by the computer. The purpose is achieve with the engine mechanical system fault diagnosis task by the equipment to replace the people. In this paper, the research work will develop a set of mechanical systems which can be used for the engine fault-line detection, identification and diagnostic instrumentation systems of classifier.The work, discuss the engine fault diagnosis strategy at first. By inference rules for engine fault diagnosis and production practice on the inference rules of a comprehensive analysis of constraints, the inference engine of modeling task mathematical abstraction and analysis, and analysis the five kinds of typical reasoning model to determine use the neural network modeling as the route of the classifier and decomposed the fault diagnosis process by fault discriminant and fault locate.In accordance with the above-mentioned technical route the paper which discussed the problem of modeling the fault discriminant classifier. The classifier is a linear classifier model based on two types of classifier, and its function is based on the sample characteristic parameters of the tested samples quickly determine the engine is failure or not. To avoid the noise of the sample is very sensitive to low generalization accuracy of the limitations of discriminant classifier of sensor criteria which can only solve linearly separable problems, this work using the mean square error as the cost function to guide the iterative process and the establishment of LMS iterative algorithm based on fault discriminant classifier model. This classifier model, can be used non-linear separable cases, you can converge to the minimum mean square error of the satisfactory solution, to effectively solve the engine fault diagnosis fault reasoning tasks.Secondly, the paper discusses fault location of classifier modeling. The classifier is a three-layer BP neural network based on the topology of multi-class classifier and its function is based on the sample characteristic parameters of the tested samples of failure mode for quick positioning. For the engine fault diagnosis of the problem domain, this paper discuss the design of hidden layer neurons, output layer neurons, learning function design, the expected distortion and network topology design and so on. Also discussed the characteristics of normalization, sequence and batch training and training with the algorithm design problems. On this basis, the classifier model that can be used for on-line and real-time engine fault diagnosis fault location.Discrimination and fault location based on fault classifier, paper work to establish a unified model of engine fault diagnosis. The model can be based on the approximate area of the theory and a small amount of training sample (partially covered) to achieve learning and training, the establishment of the initial general assumptions; while it supports reasoning in the latter part of the learning process compared with the data (new/old), and domain knowledge consistent with the new general assumptions on the machine level model of reasoning automatically adjusted.Based on the above work to develop the relevant algorithms C-language application program. The test by the unit and a joint test results show that the algorithm and applications to meet the design requirements.