Research On Sensor Fault Diagnosis Method Based On LMD And SVM Coupling

In the automatic control system,the sensor is an essential part.Its working conditions are closely related to the equipment's normal operation conditions.If a failure occurs,it is bound to cause problems for security equipment.Because the number of sensors is too large and they are more difficult to place in the actual situation; The messy, harsh application conditions and some other reasons will inevitably lead to failure during operation.Normal pressure sensor can easily be affected by temperature changes. Especially in the environment of high pressure and low temperature,Resistivity, piezoresistive coefficient, Poisson's ratio and elastic modulus and other factors of strain pressure sensor will change,which directly lead to that the test results have great difference with the actual situation.The fault signal output form is mainly divided into the following categories:shock, deviation, noise, periodic interference, output constant value and drift.Concerning several common sudden failure of self validating pressure sensor,the author proposes an adaptive LMD(Local means algorithm) and support vector machine model.It firstly uses the LMD method to decompose the sensor output signal into several PF components which equal the envelope signal and pure FM signal multiplied.Then use the support vector multi-classification machine for fault identification.Compared with the Matlab programming simulation results of Wavelet packet transform and EMD(Empirical Mode Decomposition) algorithm,the LMD algorithm's feature vector extraction is superior to the above two.The integration of LMD algorithm and SVM algorithm in the diagnosis on the pressure sensor fault,not only combines that the SVM algorithm can solve the small sample, nonlinear and high dimensions and characteristics of local minimum,but also combines that the LMD algorithm is not affected by the Fourier transform limit,signal can adaptively selected band according to the vibration, and signal analysis can be optimized and so on.Effectively achieve the recognition of the pressure sensor's 7 states,and improve the accuracy of fault diagnosis.