This research presents a failure-integrated assessment procedure and
structure for energy accumulator packs using an enhanced Convolutional Neural
Network (CNN). The proposed approach involves wavelet packet decomposition
processing of voltage change and State of Charge (SOC) signals from a lithium
accumulator to extract energy values as input features. The assessment network
performs a preliminary failure assessment on the energy accumulator pack, followed by
evaluating whether the preliminary assessment result satisfies the assessment
confirmation condition. If met, an assessment result for the energy accumulator pack is
obtained. Otherwise, an auxiliary assessment using a CNN network is conducted for
further analysis. The primary assessment result and auxiliary assessment result are then
fused using the D-S evidence theory procedure to generate a comprehensive integrated
assessment result. Finally, the integrated assessment result is evaluated, and the
ultimate assessment result is determined. The proposed procedure improves the
assessment accuracy of energy accumulator packs by enhancing the structure of the
CNN network, determining the optimal size of the convolution kernel based on the
Bayesian Information Criterion (BIC), and incorporating auxiliary assessment networks
for enhanced accuracy and integrated assessment.
Keywords: Assessment confirmation, Auxiliary assessment, Accuracy improvement, Convolutional neural network (CNN), D-S evidence theory, Energy accumulator pack, Failure assessment, Integrated assessment, State of charge (SOC), Wavelet packet decomposition.
