Research On Faiure Prognosis Technology For Key Circuit Of SMPS

The SMPS is the critical part of much CNC equipment,especially in the automatic wheel-set assembly machine.It not only affects the precision of hydraulic system and sensor,but also influences the accuracy of assembly and the reliability of the control system.Once the circuit appears degradation,failure or sudden fault,a slight case is abnormal operation resulting in stagnation of production.Its serious failure will lead to fault of control system and cause safety accidents.In view of this,we found DC-DC converter circuit is the key to determine its service life in the study.So we choose forward converter as a typical object in this paper.Through the research that based on the failure prognosis method of HSMM and improved technique,we can evaluate degradation degree of the device.At the same time,the occurrence time of failure can be predicted.The main research contents as follows:(1)This paper summarizes the background and research status of failure prognosis for SMPS.Through comparison and analysis,HSMM is identified as main method.Research is starting from the key weak points of the circuit--power devices.We analyze the failure factors and manifestations of electrolytic capacitors,MOSFET,inductors and power diodes one by one,and establish the degradation models of device related parameters with time(T)or temperature(T_K).According to this,the standard of varying parameters of 20%-200%is set as the threshold of failure.It provides a basis for identifying degradation status and predicting the time of failure.(2)Through qualitative analysis,the influence of device degradation on output is determined.Then the fault characteristic parameters,which can characterize the degradation process of the device,are selected--output voltage and inductor current.According to the parameters of the device,we used MATLAB/Simulink simulation to collect the corresponding fault signals and provide the data basis for the model establishment.In this paper,the four layer decomposition of Db5 wavelet in wavelet packet decomposition is applied to realize the fault feature extraction.It solve the problem of the soft fault features change is not obvious.(3)According to the gradient and irreversibility of the soft fault of the device,we choose the left-right two transfers type Markov chains to describe the degradation process.The observation probability matrix distribution B and the duration of states distribution D are described by mixed Gauss distribution and single Gauss distribution respectively.And the basic structure of the HSMM model is established.The generalized forward and backward algorithm,Baum-Welch algorithm and Viterbi algorithm are derived and improved respectively in the model.In view of the problem that the device degradation is not obvious,the problem of training with multiple observation sequences and the problem of parameter initialization are optimized.Finally,the reliability theory is introduced to realize the accurate estimation of RUL,which makes up for the shortcomings of the conventional calculation method.(4)In this paper,we designed a converter circuit test platform and data acquisition and process system.The data preprocessing and training processes are studied,and we get four device state classifiers and the whole life cycle prediction model of electrolytic capacitor through training.Then we use the simulation data and the experimental data to test the accuracy of two kinds of models.The state recognition ratio of the electrolytic capacitor reached 90.83%and 84.17%.The prognosis error of RUL was only 5.68%,which is 50.64%higher than that of the conventional method.The state recognition ratio of other devices remained above90%,and the model in general reached a high precision.