In choosing a dynamic simulator for operator training systems, how does the user determine the level of simulation accuracy needed?
While this may seem like a simple question, the answer can be complicated. When deciding upon the accuracy requirements of a dynamic simulator you need to consider the following:
- How do you want to measure accuracy?
- What is your basis for determining the actual process value?
- Are you more concerned about accuracy at steady-state conditions or during dynamic changes?
To follow is a concise summary of how we view simulation accuracy requirements and how we account for it when we develop Mimic Simulation Software.
Accuracy of the calculated model value for a chosen unit operation
Mimic Advanced Modeling Objects provide easy-to-implement unit operations models based upon first principles chemical engineering methods. When we test the Advanced Modeling Objects we check the results the model provides versus the result generated from manually calculating the method. If the difference is greater than normal error due to rounding errors and precision, the object must be reviewed and the source of the inaccuracy addressed before release. This test ensures that the calculations used in the Advanced Modeling Objects are accurate in our implementation in the Mimic environment.
Accuracy of Mimic values versus process design predictions of other software packages
This comparison can be very problematic. There is a misperception in the industry that the results calculated by some process design software modeling packages are without error. A knowledgeable user of these packages will tell you that you can change the results calculated by the steady-state models significantly by how you configure the models. For instance, changing the thermodynamic properties package used by the model can dramatically change the calculated results. Comparison to a design package does not ensure that the model will accurately represent the actual process.
Accuracy of Mimic values versus actual plant data
In our opinion, the process performance data extracted from the plant process historian is the only valid basis for assessing simulator accuracy. However, because no process responds exactly like the first principles equations the dynamic models need to be tuned using empirical plant or design data. Tuning and testing of the models will require good comparison test cases derived from actual plant historical date. Care must be taken to ensure that the data for comparison is gathered under conditions and operating states that are identical to the configuration and current operation of the simulator in order to get a valid basis. In general, you should be able to achieve a steady-state 5% accuracy of physical variables (pressure, flow, temperature, level) with a reasonable amount of tuning. Greater steady-state accuracy can be achieved with more attention to tuning. However, it is important that the user determine the requirements of the system up front to avoid wasting excessive time tuning values to achieve the goal of a “perfect simulator”.
Accuracy at steady-state versus dynamic conditions
ANSI/ISA specification 77.02-1993 (R2005), Fossil Fuel Power Plant Simulators – Functional Requirements, section 6.1 states the following performance criteria for steady state accuracy:
“As a minimum, the simulator-computed value of critical parameters for steady-state, full power operation with the reference plant control system configuration shall be stable and shall not vary more than 2% of the measuring instrument range as observed in the reference plant.”
The specification makes the following statement in section 6.2 for dynamic performance defined as during transient operation.
“Transient operations include malfunctions, abnormal operations, and any non-steady-state plant condition. Simulation performance under transient conditions shall meet the following criteria:
- Where applicable, it shall be the same as the plant start-up test procedure acceptance criteria.
- The observable change in the parameters shall correspond in direction to those expected from a best estimate for the simulated transient and shall not violate the physical laws of nature.
- The simulator shall not fail to cause an alarm or automatic action if the reference plant would have caused an alarm or automatic action, and, conversely, the simulator shall not cause an alarm or automatic action if the reference plant would not have caused an alarm or automatic action.
- The overall system transient characteristics’ time shall be within 20% of the reference plant when under the same operating conditions.”
The requirements for dynamic performance are much looser than steady-state performance for good reason. When measuring simulator performance during transient or dynamic conditions it can be difficult to create an accurate environment of comparison. Getting all the factors that influence the models exactly the same as the factors that influence the real process is problematic. Test cases developed for dynamic response need to be practical and not focused on imposing steady-state simulator accuracy.
Performance criteria for evaluating dynamic simulators
The goal of investing in a dynamic simulation is to train the plant operator on the safe and effective operation of the plant during malfunctions, abnormal operations, and startups /shutdowns. In addition, we want to provide a system that allows process control strategies to be developed and tested in a safe environment isolated from the plant operations. Dynamic, real-time response of the process models is much more important to meet these goals than steady-state modeling accuracy.
The operations manager who wants a system that will help him meet his operational goals should ensure that the simulation solution proposed will provide dynamic, real-time performance.
Other resources for learning more about dynamic simulator performance
- Business Case for the Virtual Plant Using Mimic Simulation Software
- Understanding Simulation Fidelity
I look forward to your comments, questions, or suggestions.
Hope to hear from you soon.