A Hybrid Fault Diagnosis Framework for High-Voltage Circuit Breakers: NRBO-Optimized ICEEMDAN and CPO-Enhanced CNN-SVM

High voltage circuit breakers exhibit nonlinear and nonstationary characteristics in their mechanical vibration signals, which pose significant challenges such as modal aliasing and parameter selection difficulties for signal decomposition and feature extraction in fault diagnosis. This paper proposes a novel fault diagnosis framework that integrates an improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) and a convolutional neural network-support vector machine (CNN-SVM) model, jointly optimized by a Newton-Raphson-based optimizer (NRBO) and a crested porcupine optimizer (CPO). Specifically, the NRBO is employed to dynamically optimize the ICEEMDAN parameters, effectively suppressing modal aliasing and enhancing signal decomposition accuracy. Subsequently, highly discriminative intrinsic mode function components are selected for feature extraction, and the CPO is used to optimize the hyperparameters of the CNN-SVM model, thereby improving deep feature representation and fault classification performance. Experimental results validate the effectiveness of the dual-optimizer strategy, demonstrating superior performance in terms of convergence speed, solution accuracy, and computational efficiency, achieving a fault diagnosis accuracy of 98.0%, which is 2.0%–6.0% higher than existing hybrid models. Furthermore, computational complexity analysis confirms the framework’s real-time capability, aligning well with the stringent requirements of high voltage circuit breaker fault diagnosis.