Fault Diagnosis for Analog Circuits by Using EEMD, Relative Entropy, and ELM

Jian Xiong, Shulin Tian, Chenglin YangComput. Intell. Neurosci 2016 · DOI

TL;DR

This paper introduces a novel fault diagnosis method for analog circuits that combines Ensemble Empirical Mode Decomposition (EEMD) for signal processing, relative entropy and kurtosis for feature extraction, and Extreme Learning Machine (ELM) for classification. The proposed method achieves high accuracy and efficiency in identifying faults in analog circuits, outperforming existing techniques.

What Problem Does This Paper Address?

Diagnosing faults in analog circuits is significantly more challenging than in digital circuits. This difficulty stems from the continuous nature of component parameters, the inherent complexity and variability of analog circuit structures, and the impact of nonlinearities and component tolerances. Existing methods often struggle with limited fault coverage, require extensive simulation data, and may not adequately account for real-world component variations. This paper aims to address these limitations by developing a robust and accurate fault diagnosis approach that can handle these complexities.

Methods

The core of the proposed method lies in a multi-stage process:

  1. Data Acquisition: Nominal and faulty response waveforms of the circuit under test (CUT) are measured.
  2. Signal Decomposition (EEMD): The acquired waveforms are decomposed into Intrinsic Mode Functions (IMFs) using Ensemble Empirical Mode Decomposition (EEMD). EEMD is chosen for its ability to handle non-linear and non-stationary signals, effectively separating different oscillatory modes within the signal. This step is crucial for extracting meaningful information from complex circuit responses.
  3. Feature Extraction (Kurtosis and Relative Entropy):
    • Kurtosis: This statistical measure quantifies the "tailedness" of a probability distribution, helping to identify deviations from normal behavior caused by component changes.
  4. Relative Entropy: This measures the difference between two probability distributions. In this context, it quantifies the divergence between the probability distribution of a faulty circuit's response and that of a fault-free circuit, making it sensitive to parameter variations. These two metrics are calculated for each IMF, comparing faulty IMFs against nominal (fault-free) IMFs.

  5. Feature Vector Construction: The calculated kurtosis and relative entropy values from selected IMFs are combined to form a feature vector for each fault. Normalization techniques are applied to these feature vectors to improve classifier performance.

  6. Classification (ELM): An Extreme Learning Machine (ELM) classifier is trained using these feature vectors. ELM is selected for its proven generalization performance and low computational cost, making it suitable for efficient fault diagnosis.

Datasets

The paper uses two benchmark analog circuits for validation:

  • Second-order Sallen-Key bandpass filter circuit: This circuit consists of resistors, capacitors, and an operational amplifier. Faults are simulated by varying component values within specific ranges, considering ±5% tolerance for resistors and capacitors. Monte Carlo simulations are used to generate a large dataset of impulse response waveforms for both fault-free and faulty conditions.
  • Leapfrog filter circuit: This circuit is used to evaluate the method's performance on multifault diagnosis.

The simulation data is generated using OrCAD/PSpice and then processed in Matlab.

Key Insights and Ideas

  • EEMD's Power for Analog Signals: The paper highlights EEMD's effectiveness in decomposing complex analog circuit responses into meaningful IMFs, which serve as a foundation for robust feature extraction.
  • Synergy of Kurtosis and Relative Entropy: Combining kurtosis and relative entropy proves to be a powerful approach for capturing subtle changes in signal characteristics caused by component faults, offering complementary information.
  • ELM's Efficiency: The choice of ELM as a classifier demonstrates its capability to achieve high diagnostic accuracy with significantly reduced training and testing times compared to traditional methods like SVM.
  • Handling Component Tolerances: The method's ability to account for component tolerances and continuous parameter variations is a significant advantage over methods that rely on discrete fault models.

Key Results

  • High Accuracy: The proposed method achieved an average test accuracy of 99.4% for single faults on the Sallen-Key bandpass filter circuit. For multifault diagnosis on the leapfrog filter, the average accuracy was 98.8%.
  • Superior Performance: The method outperformed existing techniques, including wavelet-based methods and those using kurtosis and entropy alone, in terms of diagnostic accuracy.
  • Reduced Time Cost: The ELM-based classification significantly reduced the time cost compared to SVM-based methods, with an example showing a reduction from 11.2 seconds to 0.0289 seconds for wavelet coefficients.
  • Effective Multifault Diagnosis: The method demonstrated strong performance in diagnosing multiple simultaneous faults in the leapfrog filter circuit.

Limitations and Future Work

  • Simulation-Based: While the results are promising, they are based on simulated data. Further validation with real-world hardware is a natural next step.
  • Test Point Limitations: The paper acknowledges that for actual analog circuits, test point limitations can be a challenge. The proposed method relies on output voltage waveforms, which might not always be accessible in all scenarios.
  • Future Work: The authors suggest that the method's effectiveness in handling multifaults and its efficiency make it a strong candidate for practical analog circuit fault diagnosis. Further research could explore its application to more complex circuits and different types of faults.

Summary updated

BibTeX
@article{xiong2016fault,
  author = {Jian Xiong and Shulin Tian and Chenglin Yang},
  title = {{Fault Diagnosis for Analog Circuits by Using EEMD, Relative Entropy, and ELM}},
  journal = {Comput. Intell. Neurosci.},
  year = {2016},
  doi = {10.1155/2016/7657054},
}
Figure