An integrated computational intelligence technique based operating parameters optimization scheme for quality improvement oriented process-manufacturing system

https://doi.org/10.1016/j.cie.2020.106284Get rights and content

Highlights

  • A multistage modeling mode is proposed to overcome the time delay problem.

  • Two interrelation modes between two adjacent stages are proposed and compared.

  • The integrated CIs is adopted to determine the optimal operating parameters.

  • MGGP is employed to build correlation model with explicit formulation.

  • The method can determine the quality specification for intermediate product.

Abstract

The analysis and improvement of product quality for process industry is an increasing concern for academia and industry. As the outputs of a manufacturing system mainly depend on corresponding input conditions, so it is of high significance to develop an optimization scheme to actively and accurately determine operating parameters to obtain desired quality. However, the widely employed single-model modeling mode for whole production process neglects the natural characteristics within process manufacturing system such as multistage manufacturing and hysteresis. Additionally, the popular data-driven modeling techniques in current works, especially black-box machine learning models have been restricted to satisfying the requirements regarding excellent approximation capability and explicit mathematical expression simultaneously. To fill up above research gap, it is meaningful to develop a new data-driven optimization scheme in this work to effectively and accurately determine the optimum operating parameters considering the abovementioned characteristics and requirements. Firstly, two different connecting strategies are discussed to determine the more accurate and feasible quality propagation mode between adjacent stages. Then, two computational intelligence (CI) techniques, i.e., Multi-Gene Genetic Programming (MGGP) and Multi-objective Particle Swarm Optimization (MOPSO) algorithm are exploited to construct correlation model with explicit mathematical expression and derive the optimal operating parameters, respectively. Afterwards, the fuzzy Multi-criteria Decision Making (FMCDM) method is further proposed to select the optimal solution from the obtained Pareto solutions sets. The application of the proposed scheme in a coal preparation process indicates that the proposed scheme is promising and competitive on prediction accuracy and optimization efficiency over baseline methods, and can significantly improve the final product quality comparing with initial parameters setting. Moreover, the feasible quality specification for intermediate product can also be obtained by our proposed scheme which is beneficial for early detection of quality abnormality and timely parameters adjustment.

Introduction

Process industry refers to elementary raw material industries such as petroleum-chemical engineering, coal preparation and papermaking. Obviously, these sectors are fundamental industries for the national economy, and are important in supporting momentum for the sustainable growth of the economy of the world’s manufacturing power (Wu et al., 2019, Qian et al., 2017). Although significant developments have been made in process industry including manufacturing techniques, equipment and automation in recent decades, a gap still exists between the discrete industry and process industry in terms of production effectiveness, especially for quality analysis and improvement capability (Chai, Qin, & Wang, 2014). Meanwhile, the excess production capacity and intense market competition impose traditional large-scale process industry to produce higher quality product (Gao et al., 2020, Rodriguez-Ramos et al., 2018). Besides, it has been demonstrated that a small improvement in product quality may bring large profits for process industry (Jia et al., 2018, Zhang et al., 2019). Generally, the final product quality is affected by outputs quality of various operating units. And all the final and intermediate product quality level heavily depends on various in-process operating parameters. Hence, how to effectively and accurately derive the optimal operating parameters to achieve desired final product quality and obtain feasible quality specification for intermediate product during the production process becomes vital for complex manufacturing process.

Basically, the operating parameters optimization methodology consists of two key issues, i.e., correlation model development and optimization method selection. The correlation model is used to fit the relationship between operating parameters and quality characteristics, and then the established model is regarded as objective function to inversely derive the optimal working parameters with the given quality requirement using feasible optimization algorithm (Rao et al., 2017, Sadati et al., 2018, Yang et al., 2018).

The existing methods regarding correlation model construction for the process manufacturing system broadly fall into two main categories, which are fist-principle models (FPM) and data-driven models (DDM) (Li et al., 2019a, Li et al., 2019b). FPMs are built on the deep knowledge of process physicochemical backgrounds, which is usually very time-consuming and difficult to obtain (Yuan, Huang, Wang, Yang, & Gui, 2018). Even if the FPM can be obtained, the mechanism expression is fixed and cannot be adaptive with real-time condition (Wang & Liu, 2018). What’s more, the FPMs always express the visible mechanism of single production stage, but the mechanism model of entire manufacturing system is difficult to describe (la Fe-Perdomo, Beruvides, Quiza, Haber, & Rivas, 2019). Hence, the optimum setting of operating parameters oriented whole manufacturing system by the FPMs is cumbersome. To alleviate the limitations of FPMs, the data-driven predictive models have been widely established as alternative solutions in complex production system. Additionally, the DDMs can be further generalized into two sub-categories including traditional statistical model and computational intelligence-based model. Among them, the multivariate polynomial regression (MPR) model coupled with analysis of variance (ANOVA) and parameter estimation method using experimental data is the mainstream for statistical modeling (Najarian et al., 2018, Nayak et al., 2018). However, these methods usually rely on some assumptions (i.e., pre-definition of structure of the model and non-correlated residuals) and build on the entire data base without consideration of testing of method on the test data samples (Garg, Lam, & Panda, 2017). Moreover, the established MPR models are limited in low-dimensional and simple-nonlinear correlation development, which cannot describe the characteristic of actual engineering problems such as high nonlinearity and multidimensionality. Thus, the traditional statistical modeling methods are often not applicable for complex manufacturing system. Fortunately, with the extensive use of sensor technology, computing systems and distributed control systems (DCS), large amounts of sensor data can be recorded. This provides the opportunity to use a computer algorithm to capture hidden knowledge from data and to use the learned knowledge for training ‘‘intelligent machine” to make complex decisions without human intervention (Casalino, 2018). As a result, the computational intelligence (CI)-based predictive modeling becomes applicable in practice for complex industrial process. Specifically, CIs is the generic term of intelligent algorithms, which represents a large growing slice of soft computing techniques. The most used CIs such as artificial neural network (ANN), support vector regression (SVR) and extreme gradient boosting (XGBOOST) do not require statistical assumptions and prior knowledge. They can build the predictive models completely based on process historical data (Nascimento et al., 2019, Yang et al., 2019). In addition to the abovementioned shallow machine learning models, the deep learning methods, especially deep neural network (Liu, Guo, Wang, Du, & Pang, 2019) including convolutional neural network (CNN) and recurrent neural network (RNN) (Xia et al., 2020, Xing and Lv, 2019, Zhang and Ge, 2019) have been researched and exploited to construct end-to-end prediction model to capture the characteristics hidden in complex industrial process recently. Although these CIs have achieved excellent approximation ability and overcome the limitations of statistical models, all of them are black-box models and difficult to give detailed and explicit function expression.

Obviously, the conventional optimization methods such as linear programming and nonlinear programming (Rao et al., 2017) are not suitable to cope with the intelligent prediction models based optimization formulation due to the their inherent limitations and complexity of obtained objective function. In the past two decades, the meta-heuristic algorithms, mainly including evolutionary and swarm intelligence based algorithms, have been provided to search the feasible solutions for these complicated predictive models. Typically, the product quality is evaluated by multiple indicators in practical production, which gives rise to the need to formulate the optimization problem considering multiple conflict objectives simultaneously (defined as multi-objective optimization (MOO) problem). After reviewing current works on operating parameter optimization for process industry, we found that most researchers tend to convert the MOO into a single objective function optimization problem and ultimately give a unique optimum solution. Generally, this measure is denoted as priori approach (Najarian et al., 2018, Ong et al., 2018, Wu et al., 2019).

On above basis, it can be investigated that although the methodology in term of operating parameters optimization has gained extreme development due to the computation power improvement and industrial data availability, there still exist several issues that need to be further explored:

  • (a)

    Single-model mode problem. The modern process manufacturing process has become larger-scale and included more operating units to produce higher additional value and qualified product (Yue & Shi, 2018). The above characteristics increase the difficulty in building the correlation model using traditional single-model mode. The difficulty is mainly caused by two detailed problems: One is non-corresponding time stamp for various variables at different operating units and the other is the fold of quality changing process along the operating units. If the abovementioned problems are not considered, the developed models will have poor practical effectiveness and be difficult to interpret. The detailed analysis will be presented in Section 3.

  • (b)

    Black-box correlation model problem. As aforementioned, although the widely reported and used CIs could alleviate the disadvantages of traditional FPMs and statistical models, these models are like a “black box” without explicit mathematical expression and difficult to be interpreted.

  • (c)

    Prior MOO problem. The widely used prior methods for MOO problem heavily depend on the assigning weights, which may cannot reflect the industrial practices and lead to unfeasible solution. Besides, these methods are computationally inefficient because they require several optimization runs to achieve the best parameter setting.

To deal with the abovementioned problems, we attempt to propose a new operating parameters optimization framework for process manufacturing systems to improve the product quality. Fist, a multistage and multi-model modeling mode is proposed to build the correlation model for whole production system after data preprocessing. Herein, the Multi-gene Genetic Programming (MGGP) based intelligent modeling method within operating stage and connection strategy among stages are elaborated, respectively. Then, the Multi-objective Particle Swarm optimization (MOPSO) algorithm and the fuzzy multi-criteria decision-making (FMCDM) is employed to solve the obtained response model (nested equation) formulated optimization problem and determine the optimum solution. Finally, a real industrial case with actual operating data is introduced and applied to verify the effectiveness and superiority of the proposed scheme.

The main contributions of this work can be summarized as follows:

  • (a)

    A multistage and multi-model modeling mode is proposed to construct the correlation model. This mode can overcome the limitations that exist in single-model and maintain excellent fitting ability because it considers the characteristics of the process manufacturing system, i.e., multistage manufacturing, hysteresis and multiple responses. Moreover, the developed model is easy to interpret the evolution process of quality characteristic and can derive the feasible quality specification for intermediate product.

  • (b)

    Two connection strategies between adjacent stages: only the upstream outputs propagation and all variables (inputs and outputs) at upstream stage simultaneous propagation are discussed to select the most effective one to enhance the prediction performance.

  • (c)

    The integrated CI techniques are adopted to form the operating parameter optimization scheme with sensor data. Specifically, MGGP evolution method is used to build the correlation model at each operating unit and then the multiple nested equations based optimization formulation is solved by the MOPSO algorithm. Additionally, the hyper-parameters of MGGP model are tuned by Taguchi experimental design and analysis method instead of trial-and-error method.

The rest of the paper is organized as follows. Section 2 overviews the related research literatures. Section 3 outlines the framework of proposed scheme and elaborates the detailed application procedure. Experimental validation of the proposed method on a real dense medium coal preparation process is presented in Section 4. And the experimental results and discussions are reported in Section 5. The conclusions and future works are drawn in Section 6.

Section snippets

Literature review

In previous literature, various optimization frameworks have been designed to determine the optimal operating parameters for process manufacturing systems. Among these methodologies, the data-driven especially computational intelligence based approaches have gained increasing popularity due to the high accuracy and adaptability. Thus, this section mainly reviews the literatures focusing on data-driven intelligent optimization methods and discusses the merits and demerits of these works for our

Problem statement and the proposed optimization scheme

In this section, the problem description and proposed optimal operating parameter estimation framework are explained in detail.

Industrial case study

Our proposed optimization scheme is verified on real operating data acquired from a coal preparation process in Shaanxi Province, China. As a typical process manufacturing process, coal preparation is the most economical and effective way to reduce air pollution caused by coal-burning and to achieve efficient and clean coal utilization (Yin, He, Niu, & Li, 2018).

Experimental results and discussion

In this section, two different connecting modes between two adjacent operating stages are firstly discussed to decide which one is more effective. Then, the superiority and effectiveness of MGGP based correlation model and MOPSO based optimization results are illustrated and compared with other baseline models. Finally, the practical industrial utilization effects are further given to describe the quality change and verify the usefulness of proposed optimization scheme.

Conclusions

The operating parameters optimization plays a significant role for process manufacturing system to actively design and improve final product quality. However, the complex characteristics of process industry and limitations of previous research works make the optimal operating estimation problem challenging. In this work, a new operating parameters optimization scheme is proposed based on integrated computational intelligence techniques and deep analysis of production process characteristics.

CRediT authorship contribution statement

Xianhui Yin: Conceptualization, Methodology, Writing - original draft, Writing - review & editing, Software. Zhanwen Niu: Conceptualization, Supervision. Zhen He: Conceptualization, Methodology, Project administration, Writing - review & editing, Funding acquisition. Zhaojun (Steven) Li: Conceptualization, Supervision, Writing - review & editing. Donghee Lee: Methodology, Supervision.

Acknowledgements

This research work was supported by the National Natural Science Foundation of China (NSFC) [grant numbers 71661147003 and 71532008], and the first author was sponsored by the China Scholarship Council (CSC) under grant 201806250063.The authors would also like to give sincere gratefulness to the editors and reviewers for their constructive comments on our paper. Finally, we gratefully acknowledge the support in real industrial data collection and production experiment development by Shandong

References (55)

  • S. Ibrahim et al.

    Sensitivity analysis of artificial neural networks for just-suspension speed prediction in solid-liquid mixing systems: Performance comparison of MLPNN and RBFNN

    Advanced Engineering Informatics

    (2019)
  • A. Jeang et al.

    Robust parameters determination for ergonomical product design via computer musculoskeletal modeling and multi-objective optimization

    Computers & Industrial Engineering

    (2018)
  • S. Jia et al.

    Energy modeling method of machine-operator system for sustainable machining

    Energy Conversion and Management

    (2018)
  • Z. Li et al.

    Prediction of Material Removal Rate for Chemical Mechanical Planarization Using Decision Tree-Based Ensemble Learning

    Journal of Manufacturing Science and Engineering-Transactions of the Asme

    (2019)
  • Z.X. Li et al.

    Prediction of surface roughness in extrusion-based additive manufacturing with machine learning

    Robotics and Computer-Integrated Manufacturing

    (2019)
  • Z.C. Liu et al.

    A multi-objective service composition recommendation method for individualized customer: Hybrid MPA-GSO-DNN model

    Computers & Industrial Engineering

    (2019)
  • N.A. Menad et al.

    Modeling temperature-based oil-water relative permeability by integrating advanced intelligent models with grey wolf optimization: Application to thermal enhanced oil recovery processes

    Fuel

    (2019)
  • H. Mofid et al.

    Enhancing the performance of a parallel nitrogen expansion liquefaction process (NELP) using the multi-objective particle swarm optimization (MOPSO) algorithm

    Energy

    (2019)
  • F. Najarian et al.

    Multi-objective optimization of converting process of auxetic foam using three different statistical methods

    Measurement

    (2018)
  • M. Nayak et al.

    Artificial intelligence driven process optimization for cleaner production of biomass with co-valorization of wastewater and flue gas in an algal biorefinery

    Journal of Cleaner Production

    (2018)
  • R. Prasad et al.

    Designing a multi-stage multivariate empirical mode decomposition coupled with ant colony optimization and random forest model to forecast monthly solar radiation

    Applied Energy

    (2019)
  • F. Qian et al.

    Fundamental theories and key technologies for smart and optimal manufacturing in the process industry

    Engineering

    (2017)
  • R.V. Rao et al.

    A multi-objective algorithm for optimization of modern machining processes

    Engineering Applications of Artificial Intelligence

    (2017)
  • A. Rodriguez-Ramos et al.

    An approach to fault diagnosis with online detection of novel faults using fuzzy clustering tools

    Expert Systems with Applications

    (2018)
  • N. Sadati et al.

    Observational data-driven modeling and optimization of manufacturing processes

    Expert Systems with Applications

    (2018)
  • V. Saxena et al.

    Multi-objective optimization of modified nanofluid fuel blends at different TiO2 nanoparticle concentration in diesel engine: Experimental assessment and modeling

    Applied Energy

    (2019)
  • M. Shafiullah et al.

    Low-frequency oscillation damping in the electric network through the optimal design of UPFC coordinated PSS employing MGGP

    Measurement

    (2019)
  • Cited by (0)

    View full text