Anda di halaman 1dari 9

a) Draw the flowchart of seven-steps systematic modelling procedure introduce in this unit.

1) Probelm Definition

2) Identify Controlling Factors

3) Evaluate the problem data

4) Construct the model

5) Solve the model

6) Verify the model solution

7) Validate the model

b) What are the three types of boundary conditions for partial differential equations? Give a process engineering example for each. Three types of boundary conditions for PDE. ( ) 1) Dirichlet: The value of the state variable at boundary is known, Example: Far from the interface, the concentration is the free stream value. 2) Neumann: The gradient of the state variable normal to the boundary is known, ( ) Example: Symmetry of concentration profile at centre of catalyst 3) Mixed (Robin): The value and gradient of the state variables at the boundary are related according to: ( ) ( ) ( ) Example: Diffusion rate of A to surface is balanced by first order surface reaction.

c) What does the Sergent-Westerberg Algorithm do? When the algebraic equations are more complicated, Sargent-Westerberg algorithm can help to generate partitions and their precedence order. Partitions are sets of equations that need to be solved together.

d) What is the difference between verification and validation? Briefly outline some verification methods and some validation methods. Verification is the process of cheking that you get the right numerical solution to your intended model equations, whether or not they are right. Verification process is something to do with the solution of your model equations. Validation is the process of testing that a model represents reality to a degree that is acceptable for the purpose for which it is being built. Plant data is needed for validation process. Validation is something to do with the model.

e) For the purposes of degrees of freedom analysis and setting of specifications, variables in process model should be placed into one of four categories. What are the four categories? Give an example of a variable that would be in each category. The four categories are States variables, Constants variables, Parameters variables, and algebraic variables. States variables are differential variables which are those in d( )/dt terms. For example, water level in the water level modelling process, i.e how the water level changes with time. Constants variables are variables which their quatities that are known and unchanging within the context of the broad modelling scenario. For example, Gas Constant, R. Parameters variables are variables which their quantities are fixed during process simulation, but may be altered by a designer / process engineer to optimize the process. For example, efficient heat transfer area, A. Algebraic variables are variables which do not fall into the above mentioned categories, for example, flow rate of an upstream.

f) Whats the difference between script and function m-files in Matlab? A script m-file is an external file that contains a sequence of Matlab statements. A function m-file is a m-file that adds a function definition to Matlabs list of functions.

g) Model can be classified on a spectrum from white box to black box. What do these terms mean and why are most process models grey box? White box model is model that based entirely on understanding the phenomena acting in the system and expressing them using physical laws. All parameters values involved in white box are derived from prior independent experiments. Black box model is model that derived directly from correlating real worl input and output data, with no underlying physical basis or understanding. It uses statistical methods, including the design of experiments, to derive the model form and estimate parameter values. Grey box model is the model that often is a mechanistic framework based on conservation principles and independently-measured parameters for well-understood systems parts. It is a combination of some empirical termas and parameters fitted using process data for lesserunderstood or very complicated relationships.

h) What is the DAE index and why is it relevant to process modelling? DAE index is the minimum number of differentions of the system which would be required to solve and ODE (i.e to obtain an explicit ODE). Pure ODE systems have DAE index of 0 and for a DAE systems, if you can do an assignment, it has an index of 1. However, if you can not do an assignment, the index will be higher than 1, which often lead to a so-called high index problem and it cannot be solved by normal commercial solving tool. It must be avoided generally. Any process modelling has ODE, PDE and so on generated. Determination of the DAE index can help in developing the model equations to be solvable by normal solving tool.

i) What is the mathematical form of a first-order, semi-explicit differential algebraic equation systems? Write down the equations and define all the functions and variables. A first order and semi-explicit differential algebraic equations systems are those ODE system with contraints. Mathematical form, ( ( ) (differntial equations) ) (algebraic equations)

j) Why is the model building process iterative? When do you stop? A model building process is interative because the checkings of verification process and validation process do not pass. It stops when both of the verification and validation processes return favourable results.

k) Two of the methods for solving DAEs are i) using a DAE solver, & ii) exploiting the algebraic equation structure and using an ODE solver. What are the essential, in-built Matlab functions needed to calculate the solution for method i) and method ii)? Briefly explain what they do. Using a DAE solver: First, the equations are needed to be written in standard DAE form. Function: 1) opts = odeset(MassSingular, yes, Mass, M); This function sets the singular mass matrix to be brought to ode15s. 2) [t,x]=ode15s(@yderiv,[interval], x0, opts); solve the transformed DAE (that it is in ODE form) with the initial vector and Mass singular matrix. Explaination: At first, a mass matrix M is needed to be specified. Then initial vector x0=[y0 z0]... and Write the function to evaluate dy/dt for the differential equations and the residuals of the algebraic equations. Call the DAE solver (i.e ode15s ... with singular mass matrix... and etc) DAE state vector, i.e initial vector DAE to be expressed like an ODE-like form. ( ) ( )

M is mass matrix, with 1 on the main diagonal for each differential variable and zeros the rest. It is non-invertible which indicates it must be a singular matrix. Nevertheless, another issue is consistent initialisation, the x0 you supply must be valid solutions to 0=g(y0,z0,0) or otherwise close to it.

Exploiting the algebraic equation structure and using an ODE solver: Functions: 1) Fsolve; to solve the AEs.

2) Ode45; to solve the combined ODEs with substitute AEs. Explanation: All the AEs are to be substituted into ODEs to form a pure ODE system. At each time step, solve AEs, the evaluate ODEs with ODE solver.

l) What is the key difference between solving a set of partial differential equations by a full finite difference (FD) scheme and compared to a method-of-line (MOL) scheme? FD is the oldest approach and most common. It is easy to be set up compared to most other methods. FD transforms PDE into a set of AEs. MOL is similar to FD and the lumped approach to deriving a distributed model. It uses finite difference approximations too for the spatial derivatives, but keeps continuous time derivatives. MOL transforms a PDE into a set of ODEs.

m) Statistician George E.P Box famously quoted, All models are wrong, but some are useful! Why are all models wrong and how do you judge if a model is useful? All models cannot 100% be fitted into the real-world problem, i.e the simulation of a model will not exactly predict the results as the experiment results. This is the reason why we need model validation process. By that mean, we can judge if a model is useful. Normally, the model is useful when the inputs varying within their nominated ranges, and the outputs can be predicted to the required accuracy, and nevertheless, all any other requirements have been met!

n) Five classes of constitutive equations were outlined in this unit. Name 4 of them and write down an equation example of each. Use conventional notation and do not define the symbols. Five classes of constitutive equations: 1) Transfer rate equations. Conduction heat transfer: 2) Reaction rate equations Rate law: ra=kCa (first order elementary reaction, AB) 3) Physical Property relations Ideal Gas Law: pV=nRT 4) Balance volume relationships

Vin-Vout+Vgeneration=Vaccumulation 5) Control System equations P controller: Oc=B+Kc o) What is meant by a stiff-set of equations? What is its relevance to numerical solution of models? A stiff-set of equations is a set of differential equations for which certain numerical methods for solving the equations are numerically unstable, unless the step size is taken to be extremely small. For a stiff-set of equations, numerical solution that implements small step size is applicable to solve the model equations that are stiffed.

p) Name two built-in Matlab functions that are used to solve algebraic equations. Briefly indicate the types of algebraic equations for which they are used. roots : Find the zeros of a polynomial equation, and hence solve it. f(x)=c1xn+c2xn1...+cnx+cn-1 fzero: Find a zero of a continuous non-linear function of one variable, f(x)=0

q) Write down the mathematical form of typical least squares optimisations problem for parameter estimation, assuming that there are two parameters to be estimated, and . Define any other variables and functions you introduce. Given: (xexpt,i, yexpt,i), i=1...n Proposed model form (with parameters): ymodel= f(,x) ; ymodel= f(,x) Find: ( Where ymodel= f(,xexpt,i) ( Where ymodel= f(,xexpt,i) ) )

r) Briefly outline some good practices for data acquisition and screening. Data Acquisition: Continuous record without missing values be generated from randomised trials to reduce the effect of uncontrolled or unmeasured variables. Sampling trial should last 3 to 4 times the longest time scale of interest. Never try to overcome poor primary data with sophisticated statistics. Data Screening: Use interpolation method such as linear interpolation or cubic splines to fill in the missing data, if any. Use pattern matching, i.e find a similar-looking section of valid data and us it to patch in the missing point(s). Develop a time series model and us it to attempt predict the data that would have been there.

s) Four methods for solving DAEs were discussed in this unit. Briefly describe 2 methods and compare some of their advantages and disadvantages. ( Method: 1) Substitute the algebraic equations (g) into the ODEs (f) to form a pure ODE system (dy/dt=f). Solve with an ODE solver. 2) Write the equations in standard DAE form and solve using a dedicated DAE solver. ) ( ) ( )

Method 1

Advantages Disadvantages Once the pure ODE It might not be system is determined, possible to rearrange can be easily solved all the AEs into by using any ODEs explicit form solver Substituting the Aes into the ODEs can quickly yield unmanageably large equations Some information is hidden or lost in the process Very flexible that you Initialisation must be just need to supply consistent. the DAE in standard At first, standard form and solver does DAE form is to be the work acquired.

t) What are the two elements of a modelling problem definition statement? A modelling problem definition statement includes a description and a modelling goal. A description is a statement that to describe and tell the situation that to be modelled. The modelling goal consists of a list of input variables and their anticipated ranges and also a list of output variables to be predicted by the model and the required prediction accuracy. Often, other requirements and criteria that gauge the success of the model might come together with the modelling goal.

u) What is the best way to enter 5.4 8 into a programming language such as 10 Matlab? 5.4e8. For example, in general, a 10b = a e b, where a an b are any integral.

v) In the context of comparing a model to data, what is a residual? Sketch and label the key features of two residual plots: 1) where there is a good match between the model and the data, and 2) where there is a systematic difference between the model and data. Residual is an observable estimate of statistical error. The simplest test to assess the fit of a parameter estimate is to check whether the residual ri=yi-yiM, here, yi is the experimental data and yiM is value generated from model. This is normally presented in Residual plots, by plotting ri vs input variables u, the independent variables xi, or predicted values from model yiM. Simple visual inspection of residual plots is a powerful way of detecting model deficiencies (comparing the model to data).
r r Good residual plot with good match between model and data Residual plot with modelling problems

Zero mean Even scatter No detectable trends

Non zero mean Uneven scatter Detectable trends, i.e swift upward with a positive gradient of slope

Where x can be either input variables, independent variables or predicted values. w) List four controlling factors that you would need to consider in developing a dynamic model for an open neutralisation tank fitted with a stirrer, acid feed line and control loops for pH and level. The feed to the neutralisation tank is a hot liquor stream and contains suspended particles. 1) Agitation speed 2) Concentration of feeds 3) Volume of feeds 4) Temperature of feeds

x) Briefly explain the difference between distributed parameter and lumped parameter models and give an example of each. Distributed parameter: Quantities in Balance Volume vary in 1, 2, or 3 spatial dimensions as well as time. Relatively complex. Example: Non-ideal Batch Reactor modelling or unsteadystate cooling system with heat transfer varies in different directions. Lumped parameter: Zero Dimensions, well-mixed, uniform and spatially homogenous. No spatial variation in the balance volume, but only time variation. Relatively simple. Example: Highly ideal CSTR modelling or lumped capacitance heat transfer.

y) What are the key parts of a modelling goal statement? Give a brief example of each. See question (t) above.

Last update: 08/06/2012, 8.50 pm.

Good luck!

Anda mungkin juga menyukai