Neural Network-Based Prediction Intervals for Uncertainty Quantification in VI Curves

The voltage-current (VI) curve often exhibits uncertainties due to component tolerances and variations in measurement techniques, potentially compromising fault analysis accuracy. This paper proposed a fault analysis approach utilizing neural network-based prediction intervals to quantify uncertainty in circuit board VI curves. Firstly, a neural network is designed to precisely model the VI curves, employing a mean squared error loss function to ensure the model's fidelity in capturing the curves' distinctive features. Secondly, statistical methods are applied to estimate the uncertainty of the modeled VI curves, depicting the fluctuation range of the VI curve. For the fault analysis phase, two critical indicators: coverage rate and curve similarity are incorporated to enhance the fault detection process. Finally, simulation experiments conducted on experimental circuit boards confirm the efficacy of the proposed approach.