Rubber Calendering Width Prediction Based on Data Denoising and CNN-BiLSTM

The rubber calendering process is critical in tire production, with the width of the rubber strip during calendering significantly impacting the final tire quality. This process is characterized by high noise, nonlinearity, temporal dependency, and pronounced dynamic features, which complicate the development of accurate models using traditional mechanistic modeling methods. To address this issue, a predictive model based on data denoising and a CNN-BiLSTM framework is proposed. The method employs improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) to decompose rubber calendering width data into multiple intrinsic mode functions (IMFs). Wavelet threshold denoising is then applied to high-frequency noise-containing IMFs, and the denoised IMFs are reconstructed to form a clean data series. Subsequently, a deep learning model combining convolutional neural network (CNN) and bidirectional long short-term memory (BiLSTM) networks is developed to forecast the denoised rubber calendering width sequence. Comparative experiments demonstrate that this method effectively handles complex time-series data, with the model showing favorable evaluation metrics, highlighting signif-icant advantages in predicting rubber calendering width.