The automated slot wedge inspection based on acoustic signal is an effective way for the generator's fault diagnosis. A portable inspecting device is convenient and efficient for the slot wedge inspectors. However, it is a paradox to keep both the diagnosis precision and energy efficiency at a high level. In this paper, a compressive sensing based close-loop sampling structure is designed for the analysis of the slot wedge of large generators. The characteristics of the acoustic signal is analyzed and the effective part of the signal is selected. Moreover, a model between inspection precision and sampled data quantity is built to minimize the transmission energy. This system can maintain the inspecting precision and energy efficiency at a high level. Finally, an experimental platform is built to prove that the system can decrease the transmission energy by up to 75% while maintaining the diagnosis precision to 99%.