CNN-LSTM-CVA Based Quality Monitoring of Waviness for Float Glass Manufacturing Process

The float electronic glass production process is characterized by complex and variable conditions, numerous parameters, and strong correlation, which leads to unstable quality of its ripple degree and is challenging to diagnose. In this paper, a quality monitoring method based on CNN-LSTM-CVA is proposed considering the spatial and temporal characteristics of the data. Firstly, the spatial features of process variables are extracted using CNN, and then the temporal features of data are extracted by the LSTM layer and output from the hidden layer to get the spatiotemporal features of data after dimensionality reduction. Secondly, the CVA algorithm is used to maximize the correlation between spatiotemporal factors and quality variables and to construct monitoring statistics and control limits. Finally, real samples from an electronic float glass production process of a company are used to verify the effectiveness of CNN-LSTM-CVA and applied to the production process. The algorithm has better sensitivity and robustness than the traditional CVA method.