Abstract
In the soft fault diagnosis of nonlinear analog filter circuits, the single feature can't maximally reveal the behaviors hidden in signals. In order to overcome such shortcomings, a fusion algorithm weighted feature from multi-group is proposed. This method use reliefF algorithm to optimize canonical correlation analysis combines support vector machine(RCCA-SVM) for diagnosis. The fault characteristics used in this method are extracted from the time-domain, statistical features and frequency-domain by wavelet packet transform (WPT). And then the CCA algorithm is used to improve the correlation of features according to the weights of the features. Finally, the fusion features are dimension reduced by principal component analysis(PCA), support vector machine(SVM) is the classifier of the diagnosis. The simulations show that the proposed method has a good diagnostic effect on circuit fault diagnosis of non-linear analog circuits.
Highlights
With the substantial increase of the circuit integration, electronic circuit systems become more larger, more complex and more difficult to diagnosis
The first aspect is the comparison of the results of RCCA-SVM and single feature diagnosis
The second aspect is the comparison of RCCA-SVM and non-weighted multi-groups feature diagnosis
Summary
With the substantial increase of the circuit integration, electronic circuit systems become more larger, more complex and more difficult to diagnosis. Based on the above reason, it is difficult to extract the fault features of nonlinear analog circuits and the diagnosis results are unsatisfactory. There are a great deal of extraction methods for analog circuit faults were developed including (a) no preprocessing at all [10], [15], (b) wavelet transform feature [16], [17], (c) statistical features (range, mean, standard deviation, kurtosis, and entropy) [18]–[20], and (d) frequency-domain features [21], [22]. The method is based on features weighting fusion to solve the aforementioned problems This method applies wavelet transform features(energy), statistical features (Kurtosis) and frequency features(central frequency, bandwidth, gain) to form the original fault feature sets.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
