(362a) Virtual Manufacturing System for Refinery Process | AIChE

(362a) Virtual Manufacturing System for Refinery Process


Yang, M. - Presenter, East China Universtiy of Science and Technology
Wang, X., East China Universtiy of Science and Technology
Zhong, W., East China Universtiy of Science and Technology
Long, J., East China Universtiy of Science and Technology
Fan, C., East China Universtiy of Science and Technology
Du, W., East China University of Science and Technology

With the rapid development of refinery
industry in China in recent years, the better operation of these large-scale
industrial plants has become an imperative and significant requirement. Because
of the very large processing capacity (usually more than 10 millions of tons
per year) and long time lag of the process, a tiny fluctuation in operation
will cause great influence and maybe loss in economic profits. This leads to
enormous risks if a trial-and-error method is applied on actual process aiming to
find the best operation pattern. Thus, a virtual manufacturing system based on
accurate mechanism models will provide a good platform for improving the
understanding and operation of the process.

Fig.1 Architecture of
the virtual manufacturing system

Taking a typical refinery factory in
China as an example, the virtual manufacturing technique is developed, applied
and validated. The technique structure is divided into three layers, i.e. the visualization
layer, the application layer and the model layer. In the visualization layer,
all the main units in the actual factory are rendered by three-dimensional
modeling, which provides a virtual reality environment for real experiencing
and training purpose. Through this 3D layer, the application layer is accessed
for operators and engineers by clicking a specific 3D unit. The application
layer consist of graphical user interface (GUI), OPC-based data communication
and core application software implementing several core functions. A
user-friendly GUI is developed following the DCS interface for respecting the
operator’s operating habits. Then, a data coordinator is developed being responsible
for data transmission between GUI and application software. This coordinator is
based on industry-standard OPC protocol, which makes it flexible to be
connected by a client implementing extended applications, such as the real-time
optimization. It guarantees the openness of the virtual manufacturing system
for application of higher level applications. In this refinery example, the
application layer implements four subsystems, which are the (1) plant-wide
simulation and validation, (2) sensitivity analysis and case studies, (3) unit operation
optimization and (4) plant-wide production planning optimization, covering all
necessary requirements from operators for improving operation level.

The architecture of the refinery virtual
manufacturing system is shown in Fig.1. As the footstone for higher level
applications, mechanism models of primary units in the refinery process were
developed first. These models include crude distillation units, catalytic
cracking, catalytic reforming, hydrocracking, hydrotreating,
delayed coking and S-ZORB, etc. In the first subsystem for providing a reliable
and accurate application basis, these models are running parallelly
with the actual units and validating in real-time. This means the data of feedstock
properties (such as the true boiling point curve, specific gravity, content of
sulfur, etc.) and operating conditions (such as reaction temperature, pressure,
etc.) is collected from actual process (from DCS, Real-Time Database, or
Laboratory Analysis Database) and processed before inputting into the simulator
and getting the simulation results. This scheme is executed every 20 minutes
automatically. The simulation results and actual outputs (such as product
properties, product yields, etc.) are recorded and compared. When a significant
predictive bias is observed, the calibration module is triggered to calibrate
key parameters of the model, which make it possible to follow and represent the
change of unit characteristics. The historical comparison data can be downloaded
for reproduction of the conditions and further error analysis. Based on this
subsystem, the model accuracy is validated, which guarantees the reliability
for the following applications, especially the optimization.

The second subsystem is for sensitivity
analysis and case study. After the initialization using actual conditions at
present or anytime in the past, the operator can change the operating condition
or feedstock property to evaluate the effect of these changes. This will give
the operator an intuitive and profound understanding of the process
characteristics. Also, a reliable direction can be obtained for improving the
performance of the process from current operating condition. This subsystem is
especially useful for urgent adjustment of operating conditions. When some
critical situations occur (such as significant change in feedstock properties),
the operator may have not encountered before and had no experience in how to
change the operating conditions to keep the process operating normally and
profitably. Thus, this subsystem will provide a platform for the operator to
evaluate different operating conditions without risks in damaging the units. A
detailed report covering all input and output information of different
conditions is available to the operator for further comparison and analysis.
But it may be very low in efficiency when the operator wants to find an optimal
operating condition with some objectives (such as economic profit of the unit)
in mind, because he has to try and compare many different conditions. Thus, the
next subsystem is provided for operation optimization based on mechanism models
and optimization algorithms.

The third subsystem is for unit operation
optimization. As the basis for optimization, an algorithm library covering
stochastic algorithms, direct search methods, etc. is developed and integrated
into the platform. According to the requirements of operators, all variables
which can be manipulated in the unit are made available to the GUI and
algorithms. One can choose several variables from the manipulating variable set to construct a user-defined optimization
problem with maximization or minimization of a specific objective. There are
several predefined optimization objectives (such as economic profits, total
liquid yields, energy consumption, etc.) which can be chosen. A constraint set
is also integrated which makes it possible to select different production modes
(e.g. maximizing the yield of a specific product). With the selection of
optimization algorithms, optimal operating conditions can be obtained for
further evaluation and implementation. This will provide guidance for the
operation of the units and is easy to make it real online implementation benefit
from the openness of the system. The constraint set can also be used to take
production indices from planning layer into account to find an optimal operating
conditions meeting the product requirements while maximizing the economic

The last subsystem is for optimal
production planning which is at the highest level in the automation hierarchy and
is managing global resource allocation in the factory. Traditionally, planning optimization
is based on linear process model and linear programming. Because a linear
process model cannot reflect the actual nonlinear characteristics of the
process, this cannot guarantee the feasibility on actual process though linear
programming problem is easy to solve. Thus a successive linearized model from
mechanism model is included and updated in the planning optimization. The
planning in this subsystem comprehensively considers the market requirements,
resource supply and actual units characteristics to generate feasible and
optimal planning solution. Also, to consider the uncertainties in the actual
process (such as the inaccurate true boiling point curves used in the
planning), a receding horizon optimization scheme is applied to update monthly plan
and to divide the plan into weekly plan. This subsystem generates planning
reports and guides the production in monthly scale or weekly scale. The
progress of plan implementation is monitored and accessed.

The system in this case fully considers
the requirements of operators and is aiming at improving the understanding and
the operation levels of the refinery process. As the footstone of this system,
mechanism models of primary units were developed and validated online using
real-time operating data. For understanding of the characteristics of the
process and some special conditions which are not frequent encountered, the
sensitivity analysis and case study subsystem provides a real but undamaging
operation environment. The unit operation optimization subsystem helps to find
the best operation conditions of an unit considering production indices from
upper level, while the planning optimization subsystem optimize the plant-wide
resource allocation to maximize the global economic profits. The virtual
manufacturing system is a profound practice for smart process manufacturing and
is playing an important role in the refinery production.