Background to Process Simulation

Process simulators are used to simulate processes in the process industries such as,

• Oil & Gas
• Chemicals
• Utility (fuel treatment and steam, electricity, water, compressed air, etc. generation, distribution & consumption) plant
• Metals and Mining

Process simulators predominantly fall into one of two categories, either steady state or dynamic, and, broadly, the method of solving the simulation ranges from fully equation based to fully modular based, but is generally a combination of the two.

Steady State Simulation

In a steady state simulator, the model is run until the calculated results for the current iteration do not significantly differ from the previous one. Real world features that cause the process to be dynamic, (i.e. the thermal mass of the equipment, equipment volumes, controllers) are excluded. The user does not care how long it takes the process to steady out, only that it does.

Steady state simulators are used for process

• design (what process conditions are required to produce the required product)
• equipment design (to size the equipment required to produce the required product)
• design optimisation, (to determine the optimum configuration of equipment and maximise energy recovery)
• optimisation, (to determine changes to the current operating conditions that can either reduce operating costs or increase production)

Steady state simulators are typically equation based models. The mathematical equations that describe how the process equipment operates, (e.g. pumps, valves, heat exchangers, vessels, …) respond to changes in the operating conditions are coded into "modules" that correspond to the physical item of process equipment. The equations and/or correlations implemented have typically been developed based experiments. These are often generalisations and may not scale well.

As the simulation modules are developed from mathematical models that describe the under lying process, equation based models are often called "first principles models". An alternative is to use regressed process data as polynomials or neural networks. However, these models are not valid outside the range of the process data used, they can not be used for prediction.

Steady state process models do not require process control. The simulation engineer specifies the values of key parameters, (typically model boundaries, e.g. product production rate and purity, feed flow, …). The simulator solves the model to determine all the intermediate process values that will result in the specified values being met. The engineer can further constrain the model by specifying limits on the intermediate process values, (e.g. maximum temperatures, pressures).

Dynamic Simulation

Dynamic simulators are used for process design, operator training and for optimum process control. Process design and operator training dynamic simulators are generally built using first principles models, while optimum process control models are often implemented using Laplace transform models obtained from "stepping" the actual process.

Dynamic process design simulators are the "next stage up" from steady state models, with factors such as equipment thermal mass and volumes are added. Process stream specifications are removed and key operating conditions are maintained using PID controllers. The simulators are used to study the response of the process to sudden changes in operating conditions, (e.g. if a boiler trips will the steam header pressure drop to the steam turbine trip point?).

The level of detail incorporated into an Operator Training Simulator is significantly higher.

The basic dynamic process model, (e.g. includes thermal mass of the system; equipment volumes; indicated liquid levels related to level tappings, equipment geometry and liquid density), is extended;

• simple PID controllers are replaced by the actual process control system configuration, either emulated or by stimulation of the actual vendor's hardware
• the actual process SIS (Safety Instrumented System) configuration, either emulated or by stimulation of the actual vendor's hardware is incorporated, (dynamic process simulations for dynamic response studies either do not incorporate, or have simple Cause and Effect matrices implementing the overall effect of the ESD (Emergency Shut-Down) system, but not consider the operator's interaction with the logic
• if not included in the SIS or ESD system, the Burner Management System (BMS) will be incorporated
• key proprietary vendor controllers, such as anti-surge controllers and Gas Turbine & Steam Turbine controllers and protective logic may be emulated or stimulation of the actual vendor's hardware incorporated
• (rotating) machine monitoring (e.g. vibration, axial displacement) systems are typically not included but their inputs to the SIS can be used to simulate equipment failure
• controls that allow the instructor or process engineer to alter the environmental conditions affecting the operation of the simulated process, (e.g. ambient air temperature, cooling water temperature, fuel gas composition, feed pressures), are incorporated
• the ability for the instructor or process engineer to simulate the failure of process equipment, (e.g. motor, drive shaft or belt, transmitter, valve actuator failure; valve, heat-exchanger leak; utilities (electrical power, Instrument Air, cooling water) failure) is incorporated
• the ability to enable the instructor to simulate the failure of process equipment, (e.g. motor, drive shaft or belt, transmitter, valve actuator failure; valve, heat-exchanger leak or fouling) and utilities (electrical power, Instrument Air, cooling water) failure is incorporated
• the simulator user (instructor) interface incorporates tools to evaluate the performance of the trainee

However Fire & Gas Systems are not included.

Operator Training Simulators are used for training operators, typically;

• new operators to the process plant
• refresher courses on start-up, shutdown and trip procedures for experienced operators
• operator certification and, or qualification
• review, dissection and corrective action review and training based on abnormal operating events that have occurred

Operator Training Simulator can also be used by commissioning teams during the planning of an initial start-up of a new chemical process plant. Their use allows the teams to reduce the start-up time of the new chemical plant by enabling the team to;

• test the operation of the DCS control scheme with dynamic process values
• test the operation of the SIS logic scheme with dynamic process values
• verify the communication between the DCS and the SIS and vendor package PLC with the DCS and or SIS
• confirm operating procedures and the available equipment and pipelines
• confirm if start-up is possible with the available process feed

This allows detection of incorrect configuration, missing control loops, equipment and process lines and the identification of the required additions or modifications to be made prior to plant start-up.

The major problem for operator training simulator implementers is the vast number of calculations that are required, and which have to be done in real time, as the simulator must have the look and feel of the real process to the trainee and respond to their changes via the control system. Therefore for large models, or models making heavy use of complex physical property calculations, i.e. flash / distillation columns, the accuracy of the physical property prediction routines must be reduced. Of course the side effect is that more "tuning" of the model is required and it may need to verified against a steady state model.

Like steady state simulators, dynamic simulations can be built using process data, rather than creating a first principles model. The process data is fitted to a Laplace Transform. This has proved highly successful in the field of optimum process control because,

• Laplace transforms can represent process dead time, which is hard to do with algebraic equations. (Process delay has a major effect on process controllability when using PID controllers)
• Because the identification of the process dynamics is done in the modelling package, the development and maintenance of the model is easier for the "casual" user than the equivalent modular based model
• The parameters of the equations that represent the process are fitted based on the process response and so the physical properties of the process are implicitly included. Because the models don't use physical property calculations, the model is capable of running fast enough to predict the resulting steady state of the process and the path of the process variables to the steady state, within the execution time-frame of the controller. (Regular (PID) DCS controllers have scan rates in the region of 250 milli-seconds).
  Being able predict the path of the process value to the predicted optimum value enables the optimiser to find the optimal path to the optimum operating point, while ensuring the process remains within operating constraints, both maximum and minimum values and rates of change

However,

• the models are not valid outside the range of the process data used, they can not be used for prediction of abnormal events, whether for validating process and control system design, or training of operators

Equation vs. Sequential Modular Based Simulators

This refers to how the simulation engine solves the equations that characterise the physical process equipment being simulated.

In equation based simulators, the simulation engine implements the mathematical equations that describe the physical process in an equation solver that uses appropriate techniques to solve the equations simultaneously.

In modular based process simulators the mathematical equations that describe the physical process are coded into modules, these modules are then solved sequentially by the simulation engine.

The benefits and drawbacks of sequential modular simulation engines are,

• When the module's configuration, or the inlet conditions create an infeasible result, it is easy to identify the module causing the problem
• Simulations can restart from invalid conditions

• Can not handle recycle streams. A technique called
  tearingA module is added in the recycle line, that makes a guess at the recycle line's conditions. When the simulation has solved, the actual recycle line conditions are compared against the guessed conditions and if different, a new guess is made. This is repeated until the guessed conditions converge to the calculated conditions.
  has to be used
• Not all properties can be directly specified, for example product line density. Simulators include some kind of "control" modules that allow the engineer to specify a target value for a property and a parameter to adjust to achieve the target value
• Handling multiple flow setting devices in series is difficult
• In dynamic simulators solving for properties that are tightly coupled, (a change in one causes a near instantaneous change in the other, for example flow and pressure, or electrical power and electrical frequency), can be very difficult. Although higher CPU speeds mean that shorter time steps can be used to reduce this
• Execution order is important for the model to solve quickly. This is generally set by the simulator engineer when the simulation is configured using configuration files. It is normally handled by the simulation system when the simulation is built using a Graphical User Interface

The benefits of equation based simulation engines are,

• Can handle recycle streams without the need to add modules to implement the maths required, i.e. modules to implement the tear and do not represent anything in the real world
• There are no properties that must be set and any property can be made a specification, e.g. the liquid density from a flash drum can be specified if the pressure specification, (normally must be specified for a sequential modular solver)
• Can solve problems where properties are tightly coupled, e.g. flows and pressures, electrical network power balance and frequency
• There is no execution order

The drawbacks of equation based simulation engines, which are mostly due to equation based solvers solving all the equations simultaneously, are,

• Difficult to find the cause, which module or configuration data, if the solver is unable to find a solution. All values are "garbage"
• Every equation required is in the solver and any one can cause the entire solution to be "garbage"
• It is easy to create an infeasible solution, due to constraints, e.g pressures must be greater than 0 absolute; valve position limits, 0 ≥ IVP ≥ 100% etc.;the matrix becomes singular
• It's "impossible" to solve a simulation that has "garbage" initial values. The simulation will have to be reloaded from the last good values
• programming and debugging of the modules is easier than analysing sets of equations and variables in a matrix

Hybrid Simulation Engine

Some process simulators combine equation based and sequential modular based techniques, a hybrid simulation engine. The equation based solver is used to solve properties that are tightly coupled, or where execution order has an effect on the speed of solution. For example,

• In a steady state simulator, solving the flow rates using an equation solver allows the solution to be found faster when intermediate and product flows are specified, i.e. if the product flow is specified, in a sequential solver, all the modules from the feed to the product would be checked but not calculated as the flow specification wouldn't been known until the product module (the last module) is processed. Then all the modules would be calculated in reverse order
• In a dynamic simulator, the flow and pressures in the simulation will be solved simultaneously using an equation solver as equipment pressure drop is strongly affected by the flow rate, and pressure drop affects the flow

However with dynamic simulators it is possible for there to be a mismatch between values calculated by the equation solver and what the module calculates during the sequential pass, based on the values provided by the equation solver. The situation occurs in equipment with a volume. The equation solver will be based on the assumption that the pressure of the equipment will only change based on the flow difference. But if there is a phase change in the outlet, e.g. the vessel goes empty, for the same volumetric flow the flow for the new phase will be different, and because the vapour holdup is much lower than the liquid holdup, there can be large changes in vessel pressure.