Data-driven soft sensors have been widely used in industrial processes for over two decades. Industrial processes often exhibit nonlinear and time-varying behavior due to complex physical and chemical mechanisms, feedback control, and dynamic noise. Lately, variational autoencoder (VAE) has arisen as one of the most prevalent methods for unsupervised learning of intricate distributions. Despite being successful in deep feature extraction and uncertain data modeling, it still suffers from instability and reconstruction error due to random sampling in the latent subspace representation of original input space. In this article, to deal with those limitations, constrained VAE (CVAE) is proposed by utilizing input sample information. Enthused by parallel interaction mechanism between the ventral and dorsal stream of the human brain in object recognition, parallel interaction spatial-temporal CVAE (PIST-CVAE) is proposed to extract spatial and temporal features from input samples. Lower dimensional nonlinear features extracted from PIST-CVAE are used to build the soft sensor. The effectiveness of CVAE and PIST-CVAE is demonstrated in an industrial case study, a polyester polymerization process. The obtained results demonstrate that CVAE is able to reconstruct inputs with higher accuracy and the proposed PIST-CVAE-based soft sensor yields more accurate estimations for the melt viscosity index of the polymerization process.

Control valve stiction is an industrial problem that often causes oscillations in process control loops. Oscillating control loops are not capable of maintaining key process variables near or at their desired values, thus yielding low‐quality products, inducing economic loss, and increasing environmental impacts. Therefore, it is of vital importance to detect stiction in industrial control valves. In this regard, the present work proposes a new method based on the Markov transition field and convolutional neural network (CNN) to identify sticky control valves in industrial control loops. The Markov transition field is employed to convert process variable (PV) and controller output (OP) into two‐dimensional images, which are then utilized by CNN to learn to distinguish stiction induced oscillations from oscillations brought out by a non‐stiction condition. A transfer learning strategy is adopted to improve the stiction detection capability of the proposed method. Its performance is evaluated via its application to benchmark control loops taken from the chemical, paper, mining, and metal industries. Results demonstrate that the proposed method obtains the correct verdict for the majority of the control loops studied.

Polymerization process produces industrially important products; hence, its monitoring and control are of paramount importance. However, the non-availability of real time (on demand) measurement of quality variables gives rise to difficulties in achieving effective monitoring and control. To overcome this hurdle, a novel multi-output soft sensor algorithm is proposed for simultaneous estimation of four quality variables (rate of esterification, degree of polymerization, average molecular weight and melt viscosity index) of the industrial polymerization process. The proposed soft sensor is established on canonical correlation analysis with the help of deep learning techniques. It is tested through process data collected from a real polyester plant. The results of the proposed soft sensor are compared with those of the soft sensors constructed using state-of-the-art machine learning algorithms. It is found that the proposed soft sensor shows superior prediction accuracy. The proposed soft sensor has advantages such as ability to extract complex feature extraction, capable of dealing with overfitting and offering quick estimations for the quality variables of interest. In addition, both statistical analysis and sensitivity analysis are carried out on the proposed soft sensor.