Novel Deep Learning Based on Data Fusion Integrating Correlation Analysis for Soft Sensor Modeling

Abstract

Accurate soft sensing modeling of complex industrial processes can provide operation guidance for improving the product quality. However, most modeling methods cannot mine the process data sufficiently, which leads to low prediction accuracy and generalization performance. Therefore, a novel soft sensing method based on dilated convolution neural network (DCNN) combining data fusion and correlation analysis is proposed. The fused data can be obtained by the sliding window approach, with window sizes of 1 day, to eliminate noise caused by uncertain working conditions. Then, the correlation analysis method is used to analyze the relevance of the fused data to reduce redundant variables. Moreover, the processed variables and target values are taken as inputs and outputs of the DCNN to build the soft sensing model. Finally, the proposed method is applied in a polypropylene production system to predict the melt index. Compared with the extreme learning machine, the convolution neural network, the DCNN, the DCNN based on data fusion, and the DCNN based on correlation analysis, the proposed soft sensing method has achieved state-of-the-art prediction accuracy and generalization ability, which can improve the production efficiency.

Novel Gray Orthogonal Echo State Network Integrating the Process Mechanism for Dynamic Soft Sensor Development

“…(2) The catalyst particles put into the reactor have the same particle size. (3) The heat transfer between the catalyst and polypropylene is neglected, and it is assumed that the catalyst particle rupture in the early stage of polymerization is completed in an instant. (4) The chain transfer effect of the catalyst and poison pollution is neglected.…”

“…The melt index (MI), which is defined as the mass of polypropylene melt passing through the specified standard pore diameter every 10 min under certain temperature and pressure, is one of the most important quality indexes of polypropylene products. 3 It is mainly used to divide different product grades, so as to determine different uses of polypropylene products. At present, the off-line measurement method of the MI is still adopted in the field, but there is a lag time of 2−4 h according to the specific process production conditions, which introduces discontinuity and significant delays.…”

“…The melt index (MI), which is defined as the mass of polypropylene melt passing through the specified standard pore diameter every 10 min under certain temperature and pressure, is one of the most important quality indexes of polypropylene products . It is mainly used to divide different product grades, so as to determine different uses of polypropylene products.…”

Challenges and Opportunities of Implementing Data Fusion in Process Analytical Technology—A Review

“…Several deep learning solutions can be found in the literature for DF in different industrial processes but not yet for pharmaceutical processes. For example, convolutional neural networks (CNN) could be used for fault diagnosis [ 147 , 148 ] or soft sensing in the production of polypropylene [ 149 ]. It has also been demonstrated that support vector machines, logistic regression, and CNNs could be used to fuse laser-induced breakdown spectroscopy (LIBS), visible/NIR hyperspectral imaging, and mid-IR spectroscopy data at different levels [ 119 ].…”

Adaptive Data Fusion Method of Multisensors Based on LSTM-GWFA Hybrid Model for Tracking Dynamic Targets

“…To improve the performance of data fusion in wireless sensor networks, Pan [ 31 ] proposed a data fusion algorithm that combines a stacked autoencoder (SAE) and a clustering protocol based on the suboptimal network powered deep learning model. Aiming at the problem that soft sensing modeling methods of most complex industrial processes cannot mine the process data resulting in low prediction accuracy and generalization performance, Wu [ 32 ] proposed a soft sensing method based on an extended convolutional neural network (DCNN) combined with data fusion and correlation analysis. However, most data fusion research is the application of traditional filtering architecture to seek fewer errors and accurate target state estimation in the field of tracking.…”