Study And Implementation Of Fault Prediction Technology Based On Multi-feature For Analog Circuit

With the rapid development of electronic technology, the structure and function of electronic equipment becomes more perfect and complex, which makes the demands of the electronic system reliability become higher and higher.The fault prediction of electrical system is to estimate the changing trend of the system’s future state by making full use historical data, current state and algorithm processing. Fault prediction can provide basis for on-condition maintenance, reduce the probability of system failure, improve the system reliability, and reduce support cost effectively. Extracting appropriate fault features and designing state evaluation and fault prediction algorithms with high performance is the key of fault prediction. Due to the above reasons, the major contents in this paper are summarized as follows:(1) Multi-feature extraction of the analog circuit. Due to the increasing complexity of the circuit structure, multiple features are chosed to characterize the health status of the circuit under test. Firstly, this paper describes the extraction of common fault features, and discussed the extraction of the wavelet features and statistical features in detail.(2) Assessment of the analog circuit health status based on the multi-feature. Although the health status of the circuit under test can be reflected by a single feature parameter, the change trend of various features are different and even opposite, which result in the inaccurate and nonintuitive outcome.Therefore, a fault indicator based on the improved Mahalanobis distanceis through the principal component analysis is proposed in this paper.And this approach improves the ability of fault indicator to recognize early and weak failures.(3) Research on methods of fault prediction. This paper describes the basic theory of ELM prediction model and Kalman prediction model.At the same time, advantages and disadvantages of two prediction models are discussed.In order to improve generalization ability of the traditional ELM model, the input weighting matrix and deviation of hidden layer nodes are optimized by differential evolution algorithm.AR model is used as a state model for Kalman filter, and the order of the model is preferred by particle swarm algorithm. Finally, a actual circuit is chosen to verify the effectiveness of the two prediction models.(4) Design and implement of the fault prediction software system.Combining with the theoretical research, this paper developes a set of open fault prediction software. The software integrates a variety of functional modules, including database system, fault feature extraction, condition assessment, fault prediction, man-machine interface and other functional modules. In order to meet the actual demand, the software allows users to upgrade and loade of key algorithms online.The software provides a necessary foundation to promote fault prediction technology towards engineering practice. Finally this paper verifies the validity and practicability of the software is proved based on a practical application.