Process modeling software usually plays a key role in estimating operating conditions, evaluating unit and equipment performance, and designing and modifying plants. Advances in computing power and software ease-of-use have enhanced capabilities. Nevertheless, reports of improperly designed equipment and processes remain common. So, let’s look at one human factor and some technical issues that can cause simulation failures.
[pullquote]
Perhaps the biggest human factor is a penchant to decrease design margins because of the improving capability to model processes. Yet, the overall accuracy of the model may not change. Today’s cars with anti-lock breaking systems (ABSs) provide an everyday example of the danger. ABSs have reduced accident rates — but not as much as expected. As drivers became used to ABSs, some modified their driving habits, e.g., crowding the car ahead. The process plant equivalent shifts improvements in analysis to decrease design margins to reduce capital requirements.
Technical reasons for differences between the computer model and plant include:
[javascriptSnippet ]
• Thermodynamic and property methods. We must predict thermodynamics and physical properties for systems. Data underpin the calculation methods. Original data measurement may include errors that can make the original thermodynamic or property methods inaccurate. Another source of error is extrapolating far outside the range for which a correlation was developed (see: “Don’t Model in a Vacuum,” http://goo.gl/3C0MXR). Bizarre calculations may result. Other differences stem from using plainly wrong methods. Ion-ion interactions require complex thermodynamics. Vapor phases also can have complicated chemistry. Some molecules have dimer and other structures in the vapor phase. Failing to take this into account can lead to dramatically wrong results.
• Data accuracy. Temperature, pressure, flow, composition and all other measurements have accuracy limits. If you haven’t checked the possible consequences of errors in data measurement, how do you know if a result is reasonable? Do you know what statistical analysis is available to determine measurement accuracy? Far too few plant and design engineers know the answers, even when the information is available.
• Improper model setup. The most common models use steady-state analysis. However, more-complex plants never may operate in a steady-state mode. In one recent example, a simulation gave three times the actual purge rate because it assumed a pump was operating 24h/d when it only was intended to run during one shift (8 h/d). The opposite problem can occur when using a model to size a pump. If the model has the correct “average” pump size, operating only on one shift requires a pump that’s three times larger.
• Non-equilibrium conditions. Models often assume instantaneous approach to equilibrium. Real plants have limits based on surface area, phase separation, time and other factors. Tanks exemplify this; their contents can stratify based on density or temperature. The plant never may encounter the average composition of a tank. Liquid in drums has limited surface area to reach equilibrium with the vapor above. Reactions may not proceed to equilibrium due to reaction rates.
• Simplifying assumptions. To get useful results in a reasonable amount of time, all models simplify the physical world in ways that range from minor to critical; all simplifications should be checked. Most engineers setting up simulations ignore piping between units and elevation changes. In some cases, the plant elevation itself may be important because of the effects on air pressure. Nearly everyone ignores heat losses from equipment. In one extreme case, the heat loss from a furnace exceeded the process heat absorbed. Even in non-extreme situations, understanding heat loss is essential for getting the last 5–6% of performance from fired heaters.
At its best, simulation software properly captures our understanding of nature and how we control it, and provides accurate results. At its worst, the software can have outright errors and give highly inaccurate values. Experience and diligence in checking the model can make the difference. A user must understand the limitations of basic knowledge and how the software uses that basic knowledge, as well as the simplifications employed.
