values and thresholds were set. And then data preprocessing was performed using linear interpolation. The proposed model learns based on the steady-state data of manufacturing facilities. Then, the input vector of preprocessed data was sampled using a hybrid long short term memory (H-LSTM) circulatory neural network model and used for learning. In order to verify the proposed method, data were collected based on two fault conditions and the experiments were performed based on the two fault conditions. The degree of abnormality is expressed by measuring the root mean square error(RMSE) between the output of each state data and the prediction result. The experiments verified the accuracy of the proposed failure prediction technique.