Based On The Characteristics Of The Analog Circuit Fault Prediction

With the rapid development of electronic technology, analog circuits are increasingly complicated, and test parameters increasingly diverse. Since each test point includes the system’s health status information, in such case it can’t maximum represent fault information of circuits by extracting single test point or single characteristic feature. In order to improve the accuracy of failure prediction, it is necessary to use a variety of fault characteristic features on the fault prediction and condition monitoring. At home and abroad the research about the fault diagnosis of circuit is abundant, however, the research about analog fault prediction is rare. Since predicted values which getting by the prediction methods can’t directly associate with the health status of the analog circuit, converting predicted values to quantity which can directly response to the health status of the circuit is necessary, providing the basis for maintenance and further fault location.Based on the above reasons, the main research works of this dissertation are shown as follows:1. Study on multi-feature extraction of the analog circuit. Feature extraction is the premise of fault prediction, in order to make the fault prediction right and reliable, we always want to select the most effective features which obtain the most fault information of circuit as possible, therefore the research of multi-feature extraction is crucial. A feature extraction method based on the contribution rate of fault information is proposed, extracting the characteristic quantities which provide the large amount of fault information for the fault type, and removing characteristic quantities which have no or small contribution rate, so it can reduce dimensionality of the characteristic quantities and at the same time ensure the optimality of the characteristic quantities taken. Combined with a specific circuit, the effectiveness of the method is proved.2. Study on multi-dimensional prediction model. The quality of prediction method determines the forecasting accuracy of the circuit’s health status, traditional prediction method only use a single fault characteristic, which can’t take full advantage of the association between the different characteristics. Therefore the multi-dimensional autoregressive prediction model and multi-dimensional grey prediction model are researched, the parameters of the model are optimized by using the particle swarm optimization algorithm. Using the specific circuit, the effectiveness of multi-dimensional prediction model is proved.3. Study on health status assessment of analog circuit. The purpose of analog circuit’s fault prediction is to predict the health status of the circuit, so it can provide the basis for the further fault diagnosis and localization. But the value of the characteristic quantities which getting by using the prediction method can not be an intuitive response to the health status of the analog circuit, therefore, a health status assessment method based on mahalanobis distance is proposed. Considering the different importance of each characteristic, the weighted mahalanobis distance based on the weighted average of sensitivity is proposed. Experimental results show that the proposed method can correctly assess the health status and apply to the monitoring of the early failure, fault detection rate is high and it provides a new solution way for the weight of the weighted mahalanobis distance.