A Just-in-Time Fine-Tuning Framework for Deep Learning of SAE in Adaptive Data-Driven Modeling of Time-Varying Industrial Processes

Wu, Yongjun; Liu, Diju; Yuan, Xiaofeng; Wang, Yalin

doi:10.1109/jsen.2020.3025805

Cited by 46 publications

(17 citation statements)

References 37 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Specifically, Recurrent Neural Networks (RNN) and Temporal Convolutional Networks (TCNs) [15] have been deeply applied to depict temporal dependencies; Convolutional Neural Networks (CNNs) have been widely used to depict spatial dependencies [16], [17]. Autoencoders (AEs) [18] have been widely used to depict dependencies among process variables. For example, Yuan et al [19] proposed variable-wise weighted stacked autoencoder (VW-SAE) to acquire output-related representations.…”

Section: Introductionmentioning

confidence: 99%

Modeling Task Relationships in Multivariate Soft Sensor With Balanced Mixture-of-Experts

Huang

Wang

Liu

et al. 2023

IEEE Trans. Ind. Inf.

View full text Add to dashboard Cite

Accurate estimation of multiple quality variables is critical for building industrial soft sensor models, which have long been confronted with data efficiency and negative transfer issues. Methods sharing backbone parameters among tasks address the data efficiency issue; however, they still fail to mitigate the negative transfer problem. To address this issue, a balanced Mixture-of-Experts (BMoE) is proposed in this work, which consists of a multi-gate mixture of experts (MMoE) module and a task gradient balancing (TGB) module. The MoE module aims to portray task relationships, while the TGB module balances the gradients among tasks dynamically. Both of them cooperate to mitigate the negative transfer problem. Experiments on the typical sulfur recovery unit demonstrate that BMoE models task relationship and balances the training process effectively, and achieves better performance than baseline models significantly.

show abstract

Section: Introductionmentioning

confidence: 99%

Modeling Task Relationships in Multivariate Soft Sensor With Balanced Mixture-of-Experts

Huang

Wang

Liu

et al. 2023

IEEE Trans. Ind. Inf.

View full text Add to dashboard Cite

show abstract

“…The framework involved real time monitoring, simulation, and behavioral matching for temperature uniformity control. Wu et al (2020) proposed a novel digraph‐based data knowledge‐driven method for monitoring large‐scale processes. It combines process data and knowledge to improve the capability of fault detection and diagnosis.…”

Section: Introductionmentioning

confidence: 99%

“…The proposed hybrid framework allowed for earlier fault detection than data‐driven and knowledge‐driven approaches taken in isolation or even when state estimator did not perform entirely satisfactorily. Wu et al (2020) introduced a deep learning‐based adaptive updating framework. It is based on just‐in‐time tuning of stacked autoencoder (JIT‐SAE).…”

Section: Introductionmentioning

confidence: 99%

A data‐analytic framework to monitor product density of four‐effect falling‐film evaporator for skimmed milk production

Wagh

Agashe

2023

J Food Process Engineering

View full text Add to dashboard Cite

The implementation of Industry 4.0 needs the convergence in operational technology and information technology (OT–IT) for improving the flexibility to access real‐time data and KPIs. In India, a small‐scale automation industry is full of edge devices and produces massive amount of data due to which an infrastructure may overwhelm. These industries face challenges in convergence of IT–OT like scalability. This work developed a data‐driven framework, which is an additional infrastructure between the IT–OT platforms. The framework consists of data cleaning which helps to remove the anomaly from the sensor data. The uncertainty analysis ensures the correctness of the data. Variable ranking and correlation matrix help to identify the KPIs of the process. All these stages are helpful to select the data and push this selected data to cloud for developing predictive or diagnostic models. This framework is deployed for the performance prediction of four‐effect falling‐film evaporator used to produce skimmed milk. The outlet skimmed milk density is the key performance indicator for the present case study and it is monitored here. Both linear and neural network‐based nonlinear models are used in the framework. It is noticed that linear regression model is capable of predicting the plant product density of skimmed milk with good accuracy. However, it is not capable to identify the faults in the key parameters like inlet milk flow rate. These faults can be better identified using k‐means clustering at the initial stage of the framework. Practical applications This article titled “A data‐analytic framework to monitor product density of four‐effect falling‐film evaporator for skimmed milk production” by Nivedita Wagh and Sudhir D. Agashe contains the results obtained for data‐driven framework for performance prediction and fault diagnosis for four‐effect falling‐film evaporator used for production of skimmed milk. A data‐driven framework which includes uncertainty analysis, variable ranking, various linear and neural network‐based nonlinear regression models for performance prediction, and fault detection of an evaporator is proposed. The results are presented in terms of density of the skimmed milk as a performance parameter. The objective of the work is to provide the possibility of IT–OT convergence so as to solve the scalability issue in the IT–OT convergence. The authors are able to bring out a data‐driven framework for fault identification using k‐means clustering using four‐effect falling‐film evaporator used for production of skimmed milk as a case study. We hope the results will be useful to data managers for understanding the KPIs statistics and take appropriate decision in time for the overall growth of organization.

show abstract

“…Finally, the compensation model was combined with the mechanistic model to obtain a hybrid temperature prediction model. Wu et al [8] proposed a stacked auto-encoder deep learning method based on just-intime learning, and applied it to the modeling of industrial hydrocracking processes. Yang et al [9] combined a mechanistic model of the smelting process with a data-driven approach using artificial intelligence technology.…”

Section: Introductionmentioning

confidence: 99%

A Hybrid Modeling Method for Aluminum Smelting Process Based on a Hybrid Strategy-Based Sparrow Search Algorithm

et al. 2022

View full text Add to dashboard Cite

In the production of aluminum, the regenerative aluminum smelting process is an important part for energy efficiency and product quality. Aluminum liquid temperature is a significant variable in the aluminum smelting process, and it is costly to measure timely because it requires protective temperature sensor. To handle this problem, a kind of modeling framework which combine a mechanism model with multi-scale kernel technique is proposed. First, the mechanism model is built for the aluminum liquid temperature by the energy conservation law and heat transfer mechanism. Since the mechanism model is based on some assumptions, it often results in unknown variables. Thus, the multi-scale kernel technique is used to obtain the unknown variables. Finally, a hybrid temperature prediction model is built by combining the multi-scale kernel and the mechanism model. The parameter identification of the hybrid model is described as an optimization problem, and a hybrid strategy-based sparrow search algorithm (HSSA) is proposed to solve this problem. The experiment results show that HSSA has higher convergence accuracy and stronger global search ability than the original sparrow search algorithm (SSA), and the proposed hybrid model can correctly estimate the aluminum liquid temperature.

show abstract

A Just-in-Time Fine-Tuning Framework for Deep Learning of SAE in Adaptive Data-Driven Modeling of Time-Varying Industrial Processes

Cited by 46 publications

References 37 publications

Modeling Task Relationships in Multivariate Soft Sensor With Balanced Mixture-of-Experts

Modeling Task Relationships in Multivariate Soft Sensor With Balanced Mixture-of-Experts

A data‐analytic framework to monitor product density of four‐effect falling‐film evaporator for skimmed milk production

A Hybrid Modeling Method for Aluminum Smelting Process Based on a Hybrid Strategy-Based Sparrow Search Algorithm

Contact Info

Product

Resources

About