1. Introduction
Vanadium nitrogen alloys (VNs) are important additives in industrial production, valued for their high electrical conductivity, hardness, and melting point. The industrialized production level of VNs represents the direction for the development of the global vanadium industry, making product quality a top priority in vanadium and nitrogen production. Quality control, achieved by adjusting the process parameters of key production equipment, is central to the entire lifecycle of VN production. As VNs need to be prepared in a confined space, the kiln chamber is sealed so that operators cannot judge what is happening inside the kiln. As a result, traditional visibility-based methods of quality prediction methods do not perform effectively for process parameters. For this problem, mainstream scholars mainly regulate the process parameters of equipment by using intelligent modeling and optimization methods. Garg et al. [
1] analyzed the heat transfer process of a natural circulation boiler through numerical simulation and used the resulting data to train a neural network model to establish a natural circulation boiler performance prediction model, which is capable of predicting key boiler parameters with high accuracy. Chaibakhsh et al. [
2,
3] took the boiler machine as the research object, used a numerical simulation method to analyze its thermal characteristics, established the boiler and mathematical model, and used a genetic algorithm to optimize the model, and experimentally proved that this method better improves the model accuracy. Yadav et al. [
4] achieved the optimization of heat source control parameters for heat treatment furnaces through an inverse technology approach combining a neural network and a simulated annealing algorithm. Rosso et al. [
5] proposed a structural health monitoring approach that integrates artificial neural networks with subspace-based damage indicators. Zhang et al. [
6] constructed a robust model for temperature prediction of a shuttle kiln furnace by using the method based on an integrated random vector function linkage network. Zheng et al. [
7] modeled the cement calcination process by a modified Just-In-Time Gaussian mixture regression method, which effectively reduced NOX emissions. He et al. [
8] used a self-organizing TS fuzzy neural network to predict the temperature of a high-temperature furnace during waste incineration, providing a theoretical basis for the control and optimization of the high-temperature furnace. Zhang et al. [
9] combined machine learning algorithms and data mining techniques to propose an intelligent prediction model based on industrial IoT-driven for the prediction of key carbon content in converters of steel mills, which can effectively guide the production process and reduce the carbon content error, and achieve the optimization of process parameters in the production line.
However, the traditional kiln production line has limited digitalization, and many monitoring techniques are neither practical nor cost-effective. This leads to serious black-box problems in process quality management, and the operators cannot promptly address failures or sudden abnormalities. Furthermore, they are unable to adjust other process parameters to regulate production when certain parameters become abnormal. These problems are obstacles to the achievement of advanced automation and intelligent control in the kiln production line. And using equipment and product data to predict production process parameters and achieve real-time control is a critical problem that enterprises need to address urgently.
The concept of Digital Twin (DT) brings a new solution to the above problems. Digital twin is a method of connecting and fusing physical and digital spaces. It is able to integrate the characteristics of multi-physical fields and different physical scales and has the ability to synchronize virtual and physical space. Additionally, it offers a unified platform for intelligent algorithms [
10]. The concept of DT was first introduced by Prof. Grieves [
11] in 2003 in his report on product lifecycle management. In 2010, NASA applied digital twins for practical use for the first time by collecting data from physical space vehicles and using digital twins to monitor the operational status of the entities. In 2015, General Electric Company enhanced its ability to maintain and analyze aero engines through the application of digital twins, improving the efficiency and reliability of engine maintenance [
12]. With the advancement of the new generation of information technology and physical digital representation modeling methods, the digital twin is rapidly evolving, and countries are integrating it into their national development strategies.
In terms of real-time monitoring of workshop production lines, Tao et al. [
13] proposed the concept of a digital twin workshop and discussed its structure, operation mode, and related technologies. They emphasized the fusion method for integrating diverse elements from the physical workshop with multidimensional models from the digital workshop using heterogeneous twin data. On this basis, the development of digital twin-related theories and applications is analyzed, a five-dimensional structural model of digital twin is proposed, and six guidelines for digital twin-driven applications are summarized [
14,
15]. Zhuang et al. [
16] proposed an intelligent control method for a complex product assembly workshop based on a digital twin, which effectively improves the production efficiency of the workshop. Guo et al. [
17] proposed a synchronous evolutionary prediction method for the production schedule of discrete manufacturing workshops using digital twin technology, which improves the accuracy of production schedule predictions. Yan et al. [
18] achieved monitoring and autonomous decision-making for the entire lifecycle of actual shop floor production by integrating physical entities, data interaction, virtual entity construction, autonomous model updating, and decision-making prediction. Pei [
19] et al. proposed an innovative online quality monitoring and control method based on digital twins. This method involves quantifying and mapping specific loss parameters that impact quality performance and feedback, which optimizes the production process control and improves the quality of solar cells. Tong et al. [
20] proposed a real-time machining data application and service based on digital twin technology, which achieves real-time monitoring, analysis, and optimization of Intelligent Machine Tools. Liu et al. [
21,
22] generated digital models representing the real-time status of process equipment and Work-In-Progress (WIP), enhancing the practicality, efficiency, and intelligence of the three-dimensional (3D) process through research on digital twin-based processing quality tracking and dynamic interactive control methods.
In terms of the prediction and optimization of process parameters, Maarif et al. [
23] conducted a predictive analysis of injection molding quality using an artificial neural network based on structural learning and forgetting, which significantly improved the accuracy of product quality prediction in the injection molding process. Liu et al. [
24] used genetic algorithms to optimize the feature selection for the multivariate support vector machine, aiming to decrease the model complexity and develop a quality pattern recognition model to achieve the real-time prediction of quality characteristic data and dynamic quality control. Olivo et al. [
25] introduced a steel exoskeleton design optimized using genetic algorithms to enhance the seismic performance of current RC structures. Li et al. [
26] achieved optimization of machining time, energy consumption, and surface roughness based on a neural network approach using process parameters, such as feed rate, as optimization variables. Liu et al. [
27] developed a surface roughness prediction model for brake discs by combining the digital twin based on a genetic algorithm and a multi-objective particle swarm algorithm to achieve adaptive optimization for process parameters. Cheng et al. [
28] proposed a process internet reference framework for intelligent manufacturing based on digital twins to achieve digital control and optimization for turbine production. Hu et al. [
29] proposed a data model for welding tasks to facilitate intelligent process planning and seamless integration of information among CAD/CAPP/CAM platforms for robotic welding. Li et al. [
30] proposed a dynamic recommendation framework for manufacturing resources applied to shop floor resource management and optimization based on digital twins to improve production efficiency and reduce manufacturing costs. Booyse et al. [
31] proposed a deep digital twin (DDT) framework based on deep generative models for learning the health distribution of assets directly from operational data to achieve detection, diagnosis, predictive maintenance and health state estimation of assets without relying on historical failure data. Zhao et al. [
32] used digital twins to construct a high-precision cutting simulation model for the analysis and optimization of cutting parameters to reduce carbon emissions during machining. Mi et al. [
33] proposed a cooperative perception and interconnection framework supported by digital twin technology to enhance the accuracy and reliability of predictive maintenance decisions, and the effectiveness of this methodology is validated through a case study involving the predictive maintenance of grinding roll bearings in a large vertical grinding machine.
The studies mentioned above utilize digital twins to bridge the gap between physical and virtual spaces, enabling prediction and control of process parameters in manufacturing to enhance product quality. However, current research mainly concentrates on physical processing scenarios like assembling and machining, and there are fewer studies on the production of VNs through chemical reactions. The special condition of the kiln production lines makes it challenging to promptly adjust the process parameters to achieve the desired product quality. How to build a digital twin framework for online control of the kiln production line, including monitoring and integrating process parameter prediction and control functions, is a pressing challenge that needs further research.
In summary, this paper proposes a method for predicting process parameters and online control in VN production based on digital twins. By collecting real-time data from both the digital twin production line and the real production line, combined with data transmission and communication technology, a system is developed for predicting and controlling vanadium nitrogen alloy production process parameters online. The system is designed to achieve real-time data-driven synchronous simulation of virtual and real production lines, dynamic prediction, and online control of process parameters. By synchronizing and regulating the entire production process, enterprises can effectively assess production line status, which improves production efficiency and achieves optimal production outcomes.
The basic structure of this paper is as follows:
Section 2 firstly represents the production processes of VNs and ensures which process parameter for control through experiment. Additionally, the architecture of the process parameter prediction and control system based on digital twins is constructed. Then, the prediction and online control of process parameters are achieved by constructing the ISSA-GRNN fusion network model. And a digital twin system is developed to achieve online control. Then, in
Section 3, the advancement of the proposed algorithm and the validation of the real-time online control are verified. Finally,
Section 4 represents the conclusion and future works.