BIT-Based Intermittent Fault Diagnosis of Analog Circuits by Improved Deep Forest Classifier

In order to reduce the high built-in test (BIT) false alarms of analog circuits caused by intermittent faults, a BIT-based intermittent fault diagnosis method for analog circuits by improved deep forest (DF) classifier is proposed. First, the local mean decomposition and multiscale entropy (LMD-MSE) are employed for multiscale time-frequency analysis since it can handle the data nonlinearity and eliminate redundant information. Second, the particle swarm optimization (PSO) algorithm is adopted in optimizing the multiscale factors to form the feature sets. Then, the feature sets are used to train the DF classifier and the intermittent faults of the analog circuits are diagnosed by the classifier. Meanwhile, in order to improve the diagnostic accuracy of the DF classifier for intermittent faults, the classifiers of each level of DF are replaced by extreme random forests and rotation forests. The optimized characteristic of DF improves the diagnosis accuracy and can locate the intermittent faults to the circuit branch with intermittent faults. The method is evaluated with the four-opamp biquad high-pass filter circuit. Compared with other common methods, it is shown by the given extensive comparative experiment test results that the proposed approach has achieved better diagnostic results, exhibiting greater advantages in intermittent fault diagnosis with small sample data.

Automated summary

This study proposes a method for diagnosing intermittent faults in analog circuits using a deep forest classifier. The method employs local mean decomposition and multiscale entropy for feature extraction, and optimizes the multiscale factors using particle swarm optimization. By replacing the classifiers, the diagnostic accuracy is improved, and the method can locate the faults.