In modern industrial processes, soft sensors have played an important role for effective process control, optimization, and monitoring. Feature representation is one of the core factors to construct accurate sensors. Recently, deep learning techniques been developed high-level abstract feature extraction in pattern recognition areas, which also great potential sensing applications. Hence, stacked autoencoder (SAE) introduced sensor this paper. As output prediction purpose, traditional...

Soft sensor has been extensively utilized in industrial processes for prediction of key quality variables. To build an accurate virtual model, it is very significant to model the dynamic and nonlinear behaviors process sequential data properly. Recently, a long short-term memory (LSTM) network shown great modeling ability on various time series, which basic LSTM units can handle nonlinearities dynamics with latent variable structure. However, hidden variables unit mainly focus describing...

Industrial process data are naturally complex time series with high nonlinearities and dynamics. To model nonlinear dynamic processes, a long short-term memory (LSTM) network is very suitable for soft sensor development. However, the original LSTM does not consider variable sample relevance quality prediction. In order to overcome this problem, spatiotemporal attention-based proposed modeling, which can, only identify important input variables that related at each step, but also adaptively...

Deep learning is a recently developed feature representation technique for data with complicated structures, which has great potential soft sensing of industrial processes. However, most deep networks mainly focus on hierarchical the raw observed input data. For sensor applications, it important to reduce irrelevant information and extract quality-relevant features from quality prediction. To deal this problem, novel network proposed in article, based stacked quality-driven autoencoder...

The growth of data collection in industrial processes has led to a renewed emphasis on the development data-driven soft sensors. A key step building an accurate, reliable sensor is feature representation. Deep networks have shown great ability learn hierarchical features using unsupervised pretraining and supervised fine-tuning. For typical deep like stacked auto-encoder (SAE), stage unsupervised, which some important information related quality variables may be discarded. In this article,...

This article studies the performance monitoring problem for potassium chloride flotation process, which is a critical component of fertilizer processing. To address its froth image segmentation problem, this proposes multiscale feature extraction and fusion network (MsFEFNet) to overcome weak edge characteristics images. MsFEFNet performs simultaneous at multiple scales automatically learns spatial information interest each scale achieve efficient fusion. In addition, process multistage...

Soft sensors have been increasingly applied for quality prediction in complex industrial processes, which often different scales of topology and highly coupled spatiotemporal features. However, the existing soft sensing models usually face difficulties extracting multiscale local features multicoupled process data harnessing them to their full potential improve performance. Therefore, a attention-based CNN (MSACNN) is proposed this article alleviate such problems. In MSACNN, convolutional...

Industrial process plants are instrumented with a large number of redundant sensors and the measured variables often contaminated by random noises. Thus, it is significant to discover general trends data latent variable models in probabilistic framework before soft sensor modeling. However, traditional such as principal component analysis mostly static linear approaches. The dynamics nonlinearities have not been well considered. In this paper, novel weighted dynamic system (WLDS) proposed...

Just-in-time learning (JITL) is a commonly used technique for industrial soft sensing of nonlinear processes. However, traditional JITL approaches mainly focus on equal sample sizes between process (input) variables and quality (output) variables, which may not be practical in processes since are usually much harder to obtain than other variables. In order handle unequal length dataset with only few labeled data, novel semisupervised framework proposed sensor modeling processes, based...

In industrial processes, inferential sensors have been extensively applied for prediction of quality variables that are difficult to measure online directly by hard sensors. Deep learning is a recently developed technique feature representation complex data, which has great potentials in soft sensor modeling. However, it often needs large number representative data train and obtain good deep network. Moreover, layer-wise pretraining causes information loss generalization degradation high...

Deep learning has been recently introduced for soft sensors in industrial processes. However, most of the existing deep networks, such as stacked autoencoder, are pretrained a layerwise unsupervised way to learn feature representations raw input data itself. For sensors, it is necessary extract quality-relevant features quality prediction. Thus, supervised pretraining framework proposed extraction and sensor modeling this article, which based on encoder-decoder (SSED). In SSED, hierarchical...

Probabilistic principal component regression (PPCR) has been introduced for soft sensor modeling as a probabilistic projection method, which is effective in handling data collinearity and random noises. However, the linear limitation of relationships may cause its performance deterioration when applied to nonlinear processes. Therefore, novel weighted PPCR (WPPCR) algorithm proposed this paper sensing In WPPCR, by including most relevant samples local modeling, different weights will be...

Due to the existence of complex disturbances and frequent switching operational conditions characteristics in real industrial processes, process data under different subject distributions, which means there exist manifold structures broad operations. Globally, entire are distributed a multimanifold structure. Nevertheless, existing data-driven quality prediction methods do not consider relationships among manifolds just treats as single manifold. How extract effective structure feature...

Abstract Industrial processes are often characterized with high nonlinearities and dynamics. For soft sensor modelling, it is important to model the nonlinear dynamic relationship between input output data. Thus, long short‐term memory (LSTM) networks suitable for quality prediction of modelling. However, they do not consider relevance different variables variable. To address this issue, a variable attention‐based (VA‐LSTM) network proposed sensing in paper. In VA‐LSTM, attention designed...

Data-driven soft sensors have been widely used in industrial processes. Traditional are mostly shallow networks, which cannot easily describe the complicated process data patterns. In this article, a new deep learning approach is proposed for sensors, based on stacked enhanced auto-encoder (SEAE). The original (SAE) learns hierarchical features of raw observed input with unsupervised layerwise pretraining. each layer, learned from its low-level ones an AE by minimizing reconstruction error...

Soft sensors have been extensively developed to estimate the difficult-to-measure quality variables for real-time process monitoring and control. Process nonlinearities dynamics are two main challenges accurate soft sensor modeling. To cope with these problems, temporal convolutional network (TCN)-based models in this paper. With a hierarchy of convolution kernels large receptive fields, TCN is able describe long dynamic dependence variables. Thus, model first designed nonlinear sensing....

Missing values are ubiquitous in industrial data sets because of multisampling rates, sensor faults, and transmission failures. The incomplete obstruct the effective use degrade performance data-driven models. Numerous imputation algorithms have been proposed to deal with missing values, primarily based on supervised learning, that is, imputing by constructing a prediction model remaining complete data. They limited when amount is overwhelming. Moreover, many methods not considered...