Research On RUP Prognostics For Analog Circuits

Fault diagnosis and failure prediction for analog circuits is the research hot sopt in terms of prognostics and health management(PHM) for electronic systems. Compared with the fault diagnosis, the failure prediction can better avoid fault, so the research of failure prediction is very important. The failure prediction of analog circuit has two difficulties: first, in case of failure of one component of analog circuit, the failure component still works, so it is hard to conduct accurate failure prediction under the state of working; second, the failure time of analog circuit is always not more than the whole remaining life(RL) of component, so the failure time of circuit cannot be predicted according to the component whole life. The researches of the failure of analog circuit are deficiencies to a large extent. First, the calculation methods of fault indicator(FI) in the existing research of failure prediction are imperfect and the existing calculation methods of FI are complex and lacks of reasonableness; second, the methods which predict the remaining life of each component in the circuit under the circumstance of ensuring the circuit performance are not perfect. Considering the above-mentioned problems, some researches to prediction methods of remaining useful performance(RUP) of analog circuits are carried out. In this paper, the main researches to analog circuits are methods of processing degradation data for some special analog filters rather than degradation causes and failure types for all kinds of analog circuits. In order to improve the prediction accuracy and simplify the process, different RUP prediciton methods are presented: RUP prediction method of analog circuit based on white Gaussian noise(WGN) estimate; RUP prediction method of single component of analog circuit based on complex field modeling; RUP prediction method of single component of linear analog circuit based on voltage features. At the same time, to increase the circuit state monitoring ability of the RUP prediction of analog circuit, an analog circuit test generation algorithm based on extreming learning machine is presented. The innovations of this research are as follows:1. Considering the complexity of FI calculation in the existing RUP prediction method of analog circuit, a RUP prediction method based on WGN estimate is presented. This method obtains a new FI calculation method and FI degradation model by acquiring the time domain waveform of initial state and degradation state at any moment, adding the assisted noise into the waveform of initial state with Kalman filter and comparing them with that of degradation state at any moment to put forward an improved power updating model for particle filter(PF). The test proves that this RUP prediction method contains more information of features as it acquires the overall time domain waveform and increases the prediction accuracy. At the same time, it simplifies the FI calculation process as it does not need feature fusion.2. Considering the problems that the FI calculation in the existing RUP prediction methods of analog circuit are lacked and correlated to the structure of analog circuit to affect the RUP prediction accuracy, a RUP prediction method of single component of analog circuit based on complex field modeling is presented. This method presents a component degradation analysis method of analog circuit that combines the three-point circle drawing with the parameter sweeping by introducing the existing fault diagnosis method of the cycle model of analog circuit combined with the circuit component degradation model. According to this method, a new FI calculation model is obtained. Considering the features that the cycle model of partial components in the analog circuit is degraded into line model, a FI calculation method of line model and the distinguishing method of line model and cycle model are presented, so as to perfect the FI calculation method based on complex field modeling in the analog circuit prediction. Finally, different PF updating models are given according to the different FI calculation models. The experiment verifies that this method is superior performance in terms of RUP prediction accuracy.3. Considering the problems that the measuring points in the existing RUP prediction method of analog circuit are the final output point so that architecture of fault diagnosis and failure prediction cannot use the measuring point optimization and nonfixed point fault diagnosis methods and that the existing RUP prediction method needs to acquire the frequency features so that the pure resistance circuit cannot be of RUP prediction, a RUP prediction method based on voltage features is presented. This method proposes to calculate FI according to the Euclidean distance by acquiring the voltage of any double measuring points by means of introducing the existing slope method for circuit analysis and combined with the circuit component degradation model, so as to build the FI degradation model. The experiment verifies that this method can effectively simplify the complexity of FI calculation and increase the RUP prediction accuracy, with better effects in RUP prediction of pure resistance circuit.4. State monitoring is an important technological way of PHM and the test generation algorithm is an economical and effective state monitoring method. Considering such problems of existing test generation algorithms as complex structure, unstable accuracy upon sample compression and long test time, a test generation algorithm of analog circuit based on the extreming learning machine(ELM) is presented. According to the classification hyperplane obtained by sample classification training of ELM, a test signal generation model based on ELM is presented to improve the original signal generation method and test structure. The experiment verifies that this method reduces the test structure by 50% and the original consumed time by 28%. In addition, under the state of sample compression, the test accuracy is obviously improved without additional parameter or sample optimization.