To achieve accurate monitoring of electrical faults in propane dehydrogenation units, an automated noise-exceedance detection technique based on multi-channel acoustic-signal denoising is proposed. Using an Advanced Reduced Instruction Set Computer(ARM) platform, the system assesses noise levels within the units in real-time while capturing sound signals via a multi-channel acquisition scheme. An improved Ensemble Empirical Mode Decomposition(EEMD) method, aided by information fusion, is employed for multi-channel denoising.Wavelet-packet analysis is then applied to extract acoustic features, which are subsequently normalized. A multi channel Convolutional Neural Network(CNN) is constructed to model electrical-fault detection for the propane dehydrogenation units. Experimental results demonstrate that the proposed technique effectively monitors and denoises electrical noise, raising automatic monitoring accuracy to above 0.8.