A data-driven decision-making framework for online control of vertical roller mill
Introduction
Vertical roller mill (VRM) is a type of heavy-duty and energy-intensive grinding equipment for many industries, such as cement, steel and chemical industries. It is used to grind slag, nonmetallic ore and other block and granular raw materials into required powder materials. The fine powders can be used as raw materials for cement production. In 2015, VRM dominated over 80% market share in China in cement raw meal and slag grinding (Gao, 2017). Fig. 1 shows a structure of VRM and grinding process (Salzborn & Chin-Fatt, 1993). As the core equipment of a grinding system, VRM is multi-functional, including crushing, grinding and separation. Raw materials are normally belt-transported to a VRM through feeding inlet, and crushed by millstones in the motor-driven grinding rollers, then finished powders are carried away by airflow, separated and discharged from the outlet.
Due to the inappropriate setting of process parameters or the great difference in moisture, hardness and particle size of raw materials in different batches, VRM can suffer from frequently excessive vibration, which may even lead to emergency shutdown. This seriously damages equipment useful life, product quality, production efficiency and energy consumption (Brundiek and Poeschl, 1995, Lian et al., 2015). Therefore, how to ensure continuous and stable operations is a common concern.
In current practices, operators evaluate the operation status of a VRM mainly by monitoring parameter changes, and regulate based on experience. Although the basic principles of regulation have been established, the empirically suggested strategy is often weakly practicable.
A large number of studies have explored the operation optimization of VRMs from the perspectives of equipment structure optimization (Dou et al.,2011, Etse et al., 2018, Faitli and Czel, 2014), process flow optimization (Altun et al., 2017, Jiang and Ye, 2011, Pani and Mohanta, 2014, Reichert et al., 2015, Roy and Terembula, 2003), process control optimization (Agrawal et al., 2015, Ning et al., 2011, Shaoming and Chao, 2017). As an important indicator to judge the running stability of VRM, vibration should be controlled within a reasonable range (Brundiek and Poeschl, 1995, Çopur et al., 2014). The cause of VRM vibration has also been discussed by Simmons, Gorby, and Terembula (2005), but the conclusion is still in the qualitative stage. Çopur et al. (2014) constructed a mathematical model for a set of working parameters based on the analysis of the influence of physical parameters on vibration characteristics. In view of the unsteady vibration during the operation, Liu (2019) set up a 3D simulation model of the rocker arm-roller system and carried out an intelligent finite element analysis. However, the simplified mathematical model is difficult to deal with the changeable external factors in practical application. Although the simulation model was of high accuracy, yet validation with actual data was lacking. Some researchers have also studied the influence of specific operation parameters such as the influence of pressure difference and outlet temperature on grinding efficiency and operation state, and they also analyzed different operation modes of grinding by studying vibration characteristics (Su et al., 2008, Zhang and Thulen, 2006). But these parameters are often analyzed independently, lacking consideration of variable coupling relationship. In addition, expert system was constructed to summarize the causes of VRM vibration from the perspective of process flow (Du et al., 2017, Qin et al., 2010). However, it is difficult to acquire sufficient prior knowledge in advance, and to clarify the causal logic relationship between knowledge in the construction of expert system. From the perspective of intelligent control, some studies divided the complex process into several single loops for control. For example, Agrawal et al. (2015) provided multiple single-loop controls, such as flow control, differential pressure control and outlet temperature control in a mill. Meng, Wang, Xu, and Shi (2015) designed Bang-Bang control and fuzzy PID self-tuning control for the current circuit of VRM hoist. There are also studies that adopt the model predictive control to control grinding circuit throughput and product quality (Le Roux et al., 2016, Le Roux et al., 2014). These control methods have achieved good results for the local unit control of grinding system. However, VRM usually works with multiple variables and loops. These parameters are coupled with each other and distributed in different loops. It is difficult to achieve an overall optimization effect by only relying on one local loop.
This paper aims to make the online control decision of VRM based on running state data, and to build an overall operation optimization framework of the grinding process. As shown in Fig. 2, the data-driven online control decision-making will replace the experience-based decision-making. The decision-making then becomes autonomous based on online control system with auxiliary operator assistance.
However, the following challenges remain in realizing the decision-making of data-driven operation process control,
Firstly, the complex dynamics and multi-variable nature of the grinding process, along with its nonlinear reaction kinetics and variable feed characteristics, make a VRM process inherently difficult to operate efficiently (Altun et al., 2017). For example, hot air plays a role of drying and grinding materials and separating powder. Its volume and speed significantly affect equipment output and powder specific surface area. When the air speed is constant, the increase of air volume will lead to product quality deterioration, but output increase. On the contrary, the reduction of air volume will lead to product quality improvement, but output decrease. But in either case, equipment will become unstable. Thus, the dynamic balance of multiple parameters is essential for stable operations. However, the correlations among coupling parameters make the regulation extremely complex.
Secondly, multiple monitoring variables do associate with a same controllable variable. For instance, ventilation rate affects many monitoring variables, such as pressure difference, main motor current, powder specific surface area, output, and etc. The adjustment of ventilation rate will change the above factors differently, which will lead to many uncertain changes of equipment working condition. It is still difficult to determine a control strategy by relying solely on process knowledge and operator experience.
Thirdly, a time lag between the control operation and the response of the monitoring variables was observed for VRM (Agrawal et al., 2016). Taking vibration as an example, when an abnormal vibration occurs, it is necessary to regulate the speed and feed of the air valve. After adjusting the above parameters, the value of vibration does not change immediately, but gradually returns to normal after a certain time. Therefore, judging the running state online will cause mistakes in control by relying solely on the current status.
Then, operators often give qualitative and subjective descriptions, such as too much material in the mill, too much concentration in the mill or too little ventilation, which lead to different understandings, control decisions and regulatory effects. Such ambiguous descriptions easily cause confusion in the regulation. Furthermore, the big data accumulated during continuous operation is worth exploring, so that to form a closed-loop framework that links offline data mining and online decision-making.
To address these challenges, this paper aims to find the key indicators for stability judgment through feature extraction, simplify the complexity of regulation and control, and at the same time, use clustering and association rule analysis to obtain reliable control strategies from historical data, so as to enhance the interpretability of regulation. In addition, a window-based rolling prediction model is constructed to support the real-time predictive control of VRM and enhance the predictability of the control results.
The remainder of the article is organized as follows. Section 2 demonstrates the proposed framework. In Section 3, a case study is described, while the results are compared and discussed in detail in Section 4. Finally, the limitations and future work are given in Section 5.
Section snippets
Online control decision-making framework
The running state data of equipment contains stable operation information, which is of great significance for the online control decision-making. In this paper, a framework for online control decision-making of VRM production system based on running state data is constructed. Data mining methods are used for intelligent control.
Depicted in Fig. 3, the proposed framework consists of four components, namely data sensing and acquisition, data preprocessing, data mining and knowledge
Experiments
In order to validate the effectiveness of the proposed framework, the VRM control experiments were conducted in a commercial slag VRM (Model MLK2650). The enterprise had arranged 60 measuring points in advance, which covered the parameters related to production process and equipment operation. They were distributed in different control loops and could reflect the VRM operation in the process of feeding, grinding, dust separation, drying, and so on. Data was collected every 2s for 15 days and
Discussions
In the study, a data-driven online control decision-making framework is proposed to address the high coupling and time delay in VRM stability control, and explain how to achieve the online status judgment and regulation of an unstable state.
As shown in the overall framework, this study makes full use of the offline and online status data of VRM. This is different from building mathematical models or physical structure models in existing studies (Çopur et al., 2014, Etse et al., 2018).
Taking VMS
Conclusion and future work
In this paper, an online control decision-making framework for VRM based on historical running state data is proposed. It takes VMS as the response variable to judge the stability status, and concludes that VMS is influenced by MPD, TML, and MOT. These four monitoring variables are used together as stability indexes in the status definition of a VRM. Meanwhile, a time series algorithm based on ARIMA model and an association rule algorithm based on the Apriori principle are used to perform
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
The authors are grateful for the support of the National Natural Science Foundation of China (No. 51975521).
The authors would like to thank Hubei Gucheng Longtai Cement Products Co., Ltd. and Jiangyin Xingcheng Special Steel Co., Ltd. for providing the real data of equipment operation status.
References (43)
- et al.
Review of control and fault diagnosis methods applied to coal mills
Journal of Process Control
(2015) - et al.
Operational parameters affecting the vertical roller mill performance
Minerals Engineering
(2017) - et al.
Completion of wind turbine data sets for wind integration studies applying random forests and k-nearest neighbors
Applied Energy
(2017) - et al.
An improved random forest classifier for multi-class classification
Information Processing in Agriculture
(2016) - et al.
The use of ARIMA models for reliability forecasting and analysis
Computers & Industrial Engineering
(1998) - et al.
Hour-ahead wind power forecast based on random forests
Renewable Energy
(2017) - et al.
Throughput and product quality control for a grinding mill circuit using non-linear MPC
Journal of Process Control
(2016) - et al.
Optimal control of grinding mill circuit using model predictive static programming: A new nonlinear MPC paradigm
Journal of Process Control
(2014) - et al.
A survey on feature selection
Procedia Computer Science
(2016) - et al.
Soft sensing of particle size in a grinding process: Application of support vector regression, fuzzy inference and adaptive neuro fuzzy inference techniques for online monitoring of cement fineness
Powder Technology
(2014)
Research of iron ore grinding in a vertical-roller-mill
Minerals Engineering
Experimental investigation of vibration signal of an industrial tubular ball mill: Monitoring and diagnosing
Minerals Engineering
Machine learning on high dimensional shape data from subcortical brain surfaces: A comparison of feature selection and classification methods
Pattern Recognition
Application of K-NN regression for predicting coal mill related variables
A roller mill for cement and blast furnace slag in theory and practice
Fault diagnosis expert system for large vertical mill with alarming judgement based on measured data
Mining & Processing Equipment
Foundations design of multiple vertical mills focused on vibration mitigation
Cited by (12)
A process knowledge-based hybrid method for univariate time series prediction with uncertain inputs in process industry
2024, Advanced Engineering InformaticsChronicle knowledge-based multi-level response prediction for predictive control by forest models in process industry
2024, Engineering Applications of Artificial IntelligenceEnergy consumption mode identification and monitoring method of process industry system under unstable working conditions
2023, Advanced Engineering InformaticsProcess knowledge-based random forest regression for model predictive control on a nonlinear production process with multiple working conditions
2022, Advanced Engineering InformaticsCitation Excerpt :Vertical roller mill (VRM) system is a crucial component of process industries such as cement production, steel industry, and mining engineering. It is characterized by grinding the multi-component block solid material into power with the assistance of high-temperature hot air generated in the blast stove [54]. The control of outlet temperature in the VRM system is known to be challenging due to its nonlinear and time-varying characteristics with multiple working conditions, and the temperature has a significant impact on system’s vibration and the product quality [54,55].
An integrated decision support system for multi-target forecasting: A case study of energy load prediction for a solar-powered residential house
2022, Computers and Industrial EngineeringCitation Excerpt :When the data set is multi-sourced, heterogeneous, or inadequate, single data-driven model may lead to convergence issues or poor model accuracy. Zhu et al. (2020) developed an online control decision-making framework based on the running state data. Verstraete et al. (2020) proposed a forecasting framework that automatically selects the relevant variables and predicts future sales.
A new predictive method supporting streaming data with hybrid recurring concept drifts in process industry
2021, Computers and Industrial EngineeringCitation Excerpt :A VRM system is an important grinding production line in the process industry; it can grind block and granular raw materials into powder. During grinding, the outlet temperature of VRM is an important index to measure operational consumption and stability (Zhu et al., 2020). Therefore, accurate TSP of the outlet temperature of VRM is an essential requirement for industrial practice.