This paper proposes an online monitoring method for the deposition height of arc additive manufacturing (WAAM) based on arc voltage sensing and long Short-Term memory (LSTM) deep learning. Aiming at the lag problem existing in the traditional offline detection methods, by constructing a software control system including an arc voltage acquisition unit, a robot welding platform and Labview programming, the mapping relationship between the arc voltage signal and the arc length under different currents was systematically studied. It was proved that the arc voltage fluctuation characteristics showed a significant correlation with the change of deposition height. Through the preprocessing method of wavelet threshold denoising, the voltage signal characterizing the deposition quality was effectively extracted. On this basis, a time series prediction model based on the LSTM algorithm was constructed. This network builds the input matrix through the sliding window of the time series and predicts the forming height at the next moment according to the input multi-step arc voltage signal. Experimental verification shows that in the step experiment of wire feeding speed, the mean square error between the predicted height of the model and the actual measured value is less than 0.01, and the fitting determination coefficient exceeds 0.9. This research provides a high-precision and low-cost online quality monitoring solution for the metal additive manufacturing process, which is of great significance for achieving closed-loop control.