Simulation pitfalls p302023
- 1. Most Common Pitfalls in Process Simulation Abstract ID# 302023 Authors: Richard Pelletier and Mike Donahue, Invensys Operations Management Process simulation is a tricky skill to master. Even with years of experience, a typical user is constantly facing new problems that they need to address. There are many issues that can lead to unexpected and unnecessary errors with the use of process simulation tools. Despite what simulation program you are using, many of these issues are common from one program to another. In this presentation, we will discuss the “Most Common Pitfalls in Process Simulation“. For each topic, we will discuss various techniques and methods to fix these troublesome problems and prevent their occurrence. Topics to be discussed include: 1. Bad data entry 2. Simulation defaults 3. Bad plant data 4. Thermodynamic method selection 5. Simulation tolerances and convergence 6. Recycle identification and handling 7. Model complexity 8. Difficult or impossible specifications 9. Assay pseudo-component development 10. Assay property blending These topics are based on decades of experience and feedback from end users to technical support representatives for process simulators. They often represent the most commonly asked questions or areas that most users find to be problematic. While the specific examples highlighted in each topic come from SimSci-Esscor software, the intent of this presentation is to be broad based and should be applicable for both new and experienced users of process simulation programs. Also, no topic is more or less significant than another. The numbering of each topic is meant solely for organizational purposes, not for priority or importance.
- 2. #1: Bad Data Entry Bad data entry is the easiest mistake to make when doing process simulation. Most often, units of measurement are the largest source of errors that occur. For example, in the specification below, we are incorrectly dividing a mass flowrate by a molar flowrate. Sometimes these incorrect units of measurement are minor mistakes that play no impact on our modeling. More often than not however, a mistake in units of measure can be a major problem. A perfect example of this is the Mars Climate Orbiter (MCO) 1 . Launched in 1998 by NASA to study the climate, atmosphere, and surface changes on Mars, communication with the MCO was lost after the MCO passed behind Mars, and the result was a permanent loss of communication and $328 million. When investigated, it was found that the use of English units of measurement instead of Metric units of measurement caused the MCO to have an underestimated spacecraft trajectory. To ensure that errors like this do not occur, we must use many different ways to help us ensure that our data entry is correct. As NASA’s associate administrator for space science at the time of the MCO, Edward Weiler, once put it, “People sometimes make errors. The problem here was not the error, it was the failure of NASA’s systems engineering, and the checks and balances in our processes to detect the error. That’s why we lost the spacecraft.” Here are some ways that you can help ensure proper data entry in your process simulation files – 1. Checks and balances: Check the data that you entered after you entered it. Make sure that it matches what you intended to enter, paying close attention to the units of measurement. 2. Organize your work: Have all of your information ready for entry, in the units of measurement that you need for data entry. Keep everything in an organized manner so you can go through everything easily and double check as you go. 3. Minimize distractions: Try not to perform tasks like data entry without staying focused. Turn off distractions, like radios, TVs, etc. Close out of other windows, like web browsers or other pieces of work you were working on. Try to keep side conversations with co-workers, friends, or family to a minimum or eliminate them all together. Anything you can do to keep focused will help. 4. Use “alternate forms” of data input: Copy/paste information from one source to another. Import data from a data historian. Use data links, such as defines in Pro II, specifications, etc to ensure that the proper data is being entered into the data entry field at all times. 5. Don’t ignore warning messages: While most warning messages are often there just to keep you informed in case it is something that you should be worried about, others (especially error messages) often tell you of a major problem in your simulation. Be sure to always review any messages, warnings, and/or errors the program gives you. 1 – Mars Climate Orbiter Official Website. http://mars.jpl.nasa.gov/msp98/orbiter/
- 3. #2: Defaults Defaults are methods, values, and conditions that are automatically chosen for you by the program when you begin a new simulation model. These defaults will affect the results calculated by your model and every simulation program has them. Most defaults within simulation programs fall under these categories – 1. Convergence Criteria: How the program will converge (tolerances, calculation methods for various properties like pressures or flowrates, recycle handling conditions, etc) 2. Thermodynamic Method Selections: When you select a thermo method, you are usually selected a grouping of defaults on how to calculate all of the various properties in the program. 3. Units of measurement: Most programs already have the units of measurement selected for you when you start a new model. 4. Characterization Methods: Information like assay processing methods or pipe segmentation data Defaults within simulation programs are designed to make modeling quicker and easier. There is usually no reason for a user to have to specify every unit of measure for the model or all of the convergence criteria that the program will use for trying to solve the model every single time you begin a new model. Because of this, simulation programs come with defaults already selected to make your life easier. The problem with defaults is when you didn’t realize a default has had an unwanted affect on your model. Take the time when you begin working with a new simulation program to learn the defaults for the program and how those defaults will affect the results you will eventually obtain from the program. If you do not agree with the defaults selected by the program, most simulation programs will allow you to change the default to something else.
- 4. #3: Bad Plant Data Plant data changes over time. What you enter into a simulation program when you initially build a model will not be the same information that is valid in the future. Your feed information may be different, especially in the case of assay data which always changes over time. Your unit operation’s process conditions may also be different now than they were when you built the model. Not only does plant data change over time, but measurement information is often very inaccurate. This could be due to faulty or poorly calibrated measurement equipment, leaks in lines, sensor failures, and other issues. Because of these changes and inaccuracies over time, you always have to validate the information that is entered in your model to ensure that you are obtaining the proper results. A simple mass balance around the plant below would show that something does not add up properly! There are programs out there that can help with this data validation and reconciliation. ROMeo is SimSci-Esscor’s online optimization program. With it, you can connect your model to your data historian to constantly pull up data measurements into your model and then reconcile those measurements to determine when problems arise within your plant and where those problems are most likely occurring. After running data reconciliation, you can manually transfer and use your reconciled plant data as feed information to our other software, like PRO/II and PIPEPHASE. If you do not have a data reconciliation program to determine the quality of the plant data that you are given, you can use your engineering judgment to determine if the data is valid. For example, in the process below, depending on the mass balanced performed, we have 3 different flowrates for stream S4. If you look at just the measurement value, you will get 30.6. If you perform a mass balance around node C using the measured values for streams S3 and S5, you obtain a flowrate of 29.3. If you perform an overall mass balance for the process using the measured values given for S1, S5, and S6, you get a flowrate of 28. For the most complete evaluation, if you do reconciliation around all nodes (all streams with MS# shown; S1, S2, and S3 provide enough information to determine S7 = 59 and S8 = 37.9, which then allow you to measure S6 = 21.1), you get a flowrate of 29.3 for S4, same as the balance around node C. Four different mass balances providing different flowrate results for S4. Which is correct? As an engineer, it is up to you to decide, but knowing that this is a possibility and checking for this is how you help ensure good plant data instead of bad. In this situation, most likely 29.3 is correct and this would indicate that the plant data obtained from measurements MS6 and MS4 are incorrect.
- 5. #4: Thermodynamic Method Selection Proper selection of thermodynamic methods for your simulation is the most important decision that you make when modeling a process. The method you end up choosing could have a drastic change on your results. In the example below, using Peng-Robinson instead of Grayson-Streed for your column would result in a higher duty for your column’s condenser and an increased reflux ratio. If you did not know that this was wrong, you could end up designing and purchasing a column much larger than you needed. Normally, any number of thermodynamic methods are suitable for a given application. It is up to the user to determine the best method to use based on knowledge, personal experience, and best judgment. When making a decision, there are several factors that you will want to consider – 1. What components are present in the simulation? Take particular consideration for components that are permanent gases, hydrocarbons, water, and other chemical moieties present. 2. How much of each component do you have? Composition plays a large factor on selecting a thermo method. More or less of some components could make the difference between one thermo method and another. 3. What are you operating conditions (temperature/pressure)? Only use methods that apply to the temperature and pressure range you are operating at. Oftentimes, the thermo method you have selected will probably still need some additional input from you as a user for it to most accurately model your system, even if you have selected a thermo system that most closely matches the type of process you are modeling. Most thermo methods do not contain all of the binary interaction parameters for all components in a process simulator. Any time a thermo method does not have the interaction parameters available for a specific pair of components, it treats those components as behaving ideally, which is probably not accurate. You will want to provide any missing interaction parameters to the software rather than letting the program treat these as ideal. In instances where you have multiple thermo methods selected and used in your model, you are going to want to make sure that you use similar methods or use a thermo-reset unit operation to make sure that none of your mass or energy balances turn out unbalanced. PRO/II provides a thermo-reset unit operation through the SimSci Add- ons Unit Operation on the PFD palette. Here are some ways to find help when selecting a thermo method – 1. Manuals: Most software comes with manuals that are installed on your computer when the software is installed. These manuals usually have guidelines on how to select the proper thermo method and a detailed description of technical equations and information used for each thermo method in the program. PRO/II for example has a Thermodynamic Keyword Manual and a Reference Manual (Volume 1) which both provide descriptions of different thermo and property methods. 2. Application Library Examples and Other Examples: Most software programs come with already built sample files that you can use as a reference for thermo selection. Most examples are named after the process that they model, making it easy to find one that is similar to your process. Also, if you have internal models available to you within your company, usually they will be the best way to determine an applicable thermo method for a new model. 3. Training: Many companies offer training on thermodynamic selection throughout the year in specific courses related to thermo and physical property calculations. In SimSci training courses, we also briefly discuss thermo selection in most of our basic Intro courses for our software. 4. Hotline Support: If you are still unsure of which thermo method to select, technical assistance is usually available for most companies. For SimSci technical support, please call 1-800-SIMSCI-1 (1-800-746-7241) or email [email protected]. We have a thermo expert on staff with over 30 years of thermodynamic experience who can assist you.
- 6. #5: Tolerances Many of the calculations in process simulation are iterative. When these calculations iterate, the program needs to compare the results calculated from one iteration to the results calculated in the next iteration in order to determine whether it is converged to a solution and can stop calculating or if it needs to keep going. This way of comparing results is the basic formula for tolerance values, and simulation programs use tolerances to determine the stopping point. The diagram below shows how a typical tolerance calculation is performed, where N = the value of the current iteration number, N-1 = the value of the previous iteration number, and Ep is the tolerance value. This example is for a pressure calculation. There are tolerances for many different properties in simulation programs. Temperatures, pressures, duty, and mass balances are just a few examples of the type of calculations that are dictated by tolerance values. Not only do you have different properties, but different aspects of a model can be dictated by different tolerance values on the same calculation type. For example, you could have an overall temperature tolerance for your flowsheet equal to 0.1 F. You could then set an even tighter tolerance on recycle temperature calculations equal to 0.01 F. You could even go a step further and specify a 0.001 F tolerance for calculations within a given unit operation. Tolerance values can come in two forms – relative or absolute. A relative tolerance is a percentage comparison (i.e. – 0.01 would be 1% difference between current and previous iteration values). An absolute tolerance means that the difference between the current and previous iteration values must be equal within an exact range (i.e. – 1 would mean that the current and previous iteration must be within a value of 1 from one another, like 101 and 100). Here is a mathematical representation of these, again using pressure as our property – One of the major mistakes when modeling a process is not understanding or using the correct tolerance values. A looser tolerance may make the file easier and quicker to solve, but also reduces the accuracy of the model. A tighter tolerance may be more accurate, but might be impossible for the model to solve or increase the convergence time exponentially. As an engineer using process simulation, you need to investigate when you have reached the correct level of tolerance. A common suggestion is that when you have reached a converged solution with your model, tighten the tolerances within the model and solve it again. If you notice that your answers have changed significantly, you need to keep tightening the tolerance values until your results do not change much from one iteration to another. Then you will know that you have reached an accurate solution for your model. Note: When tightening the tolerance values for calculations within your model, you will most likely also need to increase the number of iterations for the model to accommodate for the more iterative calculations that the program will need to perform in order to solve the model to the tighter tolerance.
- 7. #6: Recycles By definition, recycles in process simulation are any time when you are returning to a previously calculated unit operation or operations to iteratively recalculate the unit due to a change made elsewhere in the model. They are very common and can often be difficult parts of a model to converge. When working with a model, it is important to identify and fully understand the recycles that are in your process. Recycles can come in many forms within process simulation. The most common form is the compositional recycle. A compositional recycle is a recycle that returns flow from one unit operation to a previously calculated one. The key identifier for a compositional recycle is the fact that the flow is mixing together with a previous flow and the compositions are changing due to this remixing. Another kind of flow recycle is a thermal recycle. In this case, flow returns to a previously calculated unit operation only to transfer heat and does not actually mix with a previously calculated flow. A good example of this would be a hot stream recycling back to feed the other side of a heat exchanger. In this case, it is transferring heat to the other stream, but there is no mixing of flow or changing of composition. The last type of recycle that you can have in a simulation is a controlled recycle. This non-flow recycle occurs when a unit operation, like a controller or calculator, makes a change to a process condition elsewhere in the model and causes the simulation to recalculate to achieve a specific purpose. While these recycles do not involve flow in their calculations, they can oftentimes be the most difficult and time consuming recycles to converge. Examples of all 3 types of recycles can be seen below – Compositional Recycle (Blue) and Thermal Recycle (Orange) Controlled Recycle Once you have identified and understand your recycles and how they affect your model, you can begin troubleshooting their convergence. Here are some suggestions on troubleshooting recycle convergence – 1. Eliminate any recycles whenever possible. A thermal recycle is almost always avoidable if process conditions (temperature, pressure, flowrate) are known for the streams within the recycle. 2. Try solving your model without the recycles to see how it behaves, and then manually implement the changes necessary to get the model to where you want it to be with the recycle. 3. Avoid using complex unit operations within recycles. Move these calculations outside of the loop if possible and use reference streams to obtain their feed data. 4. Ensure all components have a way in and out of the recycle. Perform mass balances and use purge or makeup streams where necessary. 5. Use tighter tolerance values on unit operations within recycles than the total recycle tolerances themselves. Remember to increase the number of recycle iterations if you tighten the tolerance values. 6. Provide initial estimates to recycle streams. 7. Define your own calculation sequence and recycle loops. 8. If obtaining a converged solution takes a long time, use acceleration options (Wegstein and Broyden instead of direct substitution). These will help to achieve a solution faster. The chart below is a good example of this. Wegstein achieves convergence in 9 iterations and Broyden in 25.
- 8. #7: Model Complexity There are 2 ways to designing a process simulation model – 1. Make it so simple that there are obviously no deficiencies anywhere in the model. 2. Make it so complex that there are no obvious deficiencies anywhere in the model. These 2 methods are polar opposites and could mean the difference between obtaining a converged model and saving significant amounts of time or not getting convergence and spending all of your time troubleshooting why it didn’t converge. The preferred method is number 1. The general rule of thumb when modeling is to start simple and add complexity a little bit at a time as you go. Each time you add something, converge and save the model before moving on. Make sure you justify any additions that you make to the model. You can most likely model what you are trying to accomplish without adding every single unit from your PFD into the simulation. When modeling, you should be thinking about what the goal of your simulation is and not about literally translating every real-world piece of equipment into your simulation model. Make the judgment call on what is most important and what the purpose of every unit you add is to your simulation. For example, if your real-life process shows a heat exchanger, a pump and a splitter, but the only thing that’s really necessary is that the temperature, pressure and separation are achieved properly, consider using a flash drum to accomplish all three functions instead of using a separate unit operation for each. The flash drum can set the temperature, the pressure and perform your separation all at once. Finally, think about whether you can use sub-flowsheets or multiple simulation files, if you can reduce or simplify recycles, and if you can eliminate controllers. These will all help reduce the level of complexity of your model. In summary, don’t start off doing this - Build up to this a little bit at a time!
- 9. #8: Difficult or Impossible Specifications Specifications within process simulation are everywhere. Every stream or unit operation you place in your model contains specifications. In the simplest form, a specification is a constraint that you place on the model in which the model needs to meet the constraint in order to successfully converge. For streams, these are usually temperatures, pressures, flowrates and compositions. For unit operations, they range from process conditions, like temperatures and pressures, to product conditions, like an overhead composition in a column. Specifications can be categorized as either an absolute specification or a relative specification. An absolute specification means that the result calculated must be equal to the value that you specified. For example – Column Overhead Methane Flowrate = 1000 lb/hr In this example, the column must solve the overhead methane flowrate to be equal to 1000 lb/hr in order for the flowsheet to converge. This is a valid solution as long as there is at least 1000 lb/hr of methane coming in from the feed. When there is not 1000 lb/hr coming in from the feed however, this specification will always fail. A better solution here would be to use a relative specification. A relative specification provides a little bit more flexibility in achieving a solution. With a relative specification, you tell the program to be within some range of the solution you are trying to achieve. For example – (Column Overhead Methane Flowrate) / (Total Methane Feed Flowrate) = 99 wt% In this example, the column has the flexibility to solve to a converged result where the overhead methane flowrate may be a different value every iteration depending on how much methane feed is coming into the column. With this kind of specification, the program should always be able to come to a converged solution no matter how much methane is coming in the feed. As shown above, many absolute specifications have the ability to become infeasible over time. There are other pitfalls however that you can run into with specifications. Here are a few to consider – 1. Infinite solutions: Many times specifying 0 or 100% of something can lead to infinite solutions for the convergence algorithms and make it so the model does not converge. 2. Azeotropes in columns: Be careful not to specify a composition specification in a column which is above an azeotrope that you can not break. 3. High purity specifications: Very difficult to converge. Often better to provide impurity specifications instead. 4. Multiple specifications may conflict with one another: Be aware of what it is that you are specifying. Sometimes specifying one thing makes it extremely difficult or impossible to meet another specification. For example – a. Compositional and product flow specifications often clash with one another. Free one up and specify something else. b. Specifying all product flows constricts the material balance and gives no room to find a solution. Allow at least 1 product flowrate to be adjusted by the program. c. Two compositional specifications for the same product are difficult to converge. Also difficult is having the same component specification in different products.
- 10. #9: Assay Pseudo-Component Development Petroleum assays are sets of distillation data that are used to characterize all of the components that reside in a given crude sample. When you enter the distillation data into a process simulator, the program needs to take this data and convert it into components that it can use when modeling your process. The components that are generated are called pseudo-components, and each pseudo-component represents all of the unknown components that would be within a given boiling point range in the crude. In order to generate these pseudo-components, the program needs to do several things – 1. Convert the data that you have entered from the basis you entered it in (D86, D1160, or D2887) to TBP equivalent data. 2. Distribute the assay curve into separate cuts. Each cut represents a different pseudo-component. 3. Determine moles, mass, and volume of each cut. 4. Process any light ends 5. Determine the average normal boiling point, gravity, and molecular weight of each cut. 6. Characterize the rest of the component information for each cut. Each step in this process is a possible point where errors could occur. Luckily, most process simulation programs provide you many options for processing methods and ways to enter the data into the program that can ensure that the pseudo-components that are developed are the most accurate pseudo-component possible for your assay. Here are some things you will want to ensure when properly developing your assay components – 1. Did you enter your assay data correctly? Always make sure to check that you used the correct distillation curve basis (TBP, D86, D1160, D2887), % distilled basis (weight or volume), entered the correct NBPs and gravity data vs % distilled, and so forth. 2. Did you use enough pseudo-component cuts? View the graph of your curve and see how your current cutpoint set distributes the curve into components. Notice any areas with large % distilled gaps? You probably need more cuts within the temperature range of that % distilled range. 3. Are you using the correct characterization methods? There are different options for how the program fits a trend curve through the distillation curve data you enter, how it converts to and from TBP based on how you entered your data, how it calculates the gravity or molecular curves, how it calculates where to make the pseudo-component cuts, ways to help it fit the trend properly through end points, and more. Make sure you are using the most applicable methods for the type of assay you have, and if you are not sure, try switching up the methods to see if they give you a better fit or better results.
- 11. #10: Assay Property Blending When you have more than 1 assay in a simulation model, you need to be concerned with assay blending and how it will affect the results of your file. Assay blending is the mixing of 2 or more groups of components (i.e. – sets of cuts/pseudo-components) in order to calculate averaged component properties for cuts that are in the same boiling point range as one another. Once these new averaged component properties are determined, the program will use these averages from there and on in the simulation model any time it needs to access component properties for a given pseudo-component. All component properties are blended, including normal boiling points, specific gravity, molecular weight and refinery inspection properties, and usually the blended component properties are calculated using a weighted average, like the equation below for normal boiling point – The below example demonstrates how this blending occurs – As you can see from the example, given the 3 separate assay curves above, for the boiling point range between 205 and 215 F, the component properties will be blended together and new component properties are calculated. Instead of using the original component property values for this NBP component, it will now use the average properties. This affects all 3 assays. In order to retain the original assay curve properties, you will need to use different cut point sets for each curve. In PRO/II you would add more sets by going to the Assay Characterization window defining new sets. You can then go into each assay stream and choose that cut point set in the Flowrate and Assay data entry window, shown below. Choosing to use another cut point set will result in more total pseudo-components being generated for the model. There is a second option in this paragraph that also allows you to include or exclude the assay stream in the blending calculations. By choosing to exclude the assay from blending, the stream will not contribute its pseudo- component property information to the blending calculations, however, this does not necessarily mean that it will retain its individual pseudo-component properties. The stream itself will still be affected by the blending of other components within the same cutpoint set; it just won’t contribute its own property data to the blending.