Development of Empirical Dynamic

Models from Step Response Data

Black box models
• step response easiest to use but may upset
the plant manager (size of input change?
Chapter 6

move to new steady-state?)

• other methods

impulse - dye injection, tracer

random - PRBS (pseudo random binary sequences)
sinusoidal - theoretical approach
frequency response - modest usage (incl. pulse testing)
on-line (under FB control)

1
 Some processes too complicated to model using
physical principles

• material, energy balances

• flow dynamics
• physical properties (often unknown)
Chapter 6

• thermodynamics

Example 1: distillation column

50 plates
•For a 50 plate column, dynamic models have
many ODEs that require model simplification; and physical
properties must be known; e.g., HYSYS

•black box models (only good for fixed operating conditions) but
requires operating plant (actual data)

•theoretical models must be used prior to

2
plant construction or for new process chemistry
6.11

Figure 6.2

3
4
Chapter 6

Figure 6.1

•Need to minimize disturbances during a plant test

5
 Simple Process Models
- s 1st order system with gain K,
Ke
G( s)  dead time  and time
 s 1 constant ; 3 parameters to
be fitted.
Chapter 6

Step response:

y(t )  KM (1  e (t  ) /  ) t  y(t )  0 t 

6
 For a 1st order model, we note the following
characteristics (step response)
(1) The response attains 63.2% of its final response
at one time constant (t =  ).
(2) The line drawn tangent to the response at
Chapter 6

maximum slope (t = ) intersects the 100%

line at (t =  ). [see Fig. 6.2]
K is found from the steady state response for an input
change magnitude M.
There are 4 generally accepted graphical techniques
for determining first order system parameters , :
1. 63.2% response
2. point of inflection
3. S&K method
4. Semilog plot ln(1  yi / KM ) vs. ti   7
 Fitting first order models using step test

First order model:

when the system is initially

at steady state with u(0)=0 and
y(0)=0. if the input u is abruptly changed
from 0 to M at time t=0 then step response result as: Summary: The intercept of the
tangent at t=0 with the
horizontal line y/KM=1,
occurs at t=τ, as an alternative, τ can
be estimated from step response plot
Using the value of t at which the
response is 63.2% complete.

After rearranging and evaluating the limit

at t=0, the initial slope of the normalized step response is 8
Chapter 6

(θ = 0)
  speed of response 9
 Figure 6.4
Example 6.2

Figure 6.4
10
 Summary
Chapter 6

Inflection point hard to find

with noisy data.
Figure 6.4
11
 S & K Method for Fitting FOPTD
Model
• Normalize step response
(t = 0, y = 0; t →∞, y = 1)
Chapter 6

• Use 35 and 85% response times (t1 and t2)

 = 1.3 t1 – 0.29 t2
 = 0.67 (t2 – t1)
(based on analyzing many step responses)
K found from steady state response

• Alternatively, use Excel Solver to fit  and  using

y (t) = K [1 – e –(t-)/t] and data of y vs. t
12
 Fitting an Integrator Model
to Step Response Data

In Chapter 4 we considered the response of a first-order process

to a step change in input of magnitude M:

 
Chapter 6

y1  t   KM 1  et /τ (4-18)
For short times, t < , the exponential term can be approximated
by t
et /τ  1 
τ
so that the approximate response is:
  t   KM
y1  t   KM 1  1     t (6-22)
  τ  τ
(straight line with slope of y1(t=0))
13
 is virtually indistinguishable from the step response of the
integrating element
K2 M
G2  s   U (s)  (6-23)
s s
In the time domain, the step response of an integrator is
Chapter 6

y2  t   K 2 Mt (6-24)

  t   KM
Comparing with (6-22), y1  t   KM 1  1     t (6-22)
  τ  τ

K
K2  (6-25)
τ
matches the early ramp-like response to a step change in input.

14
Chapter 6

Figure 6.10. Comparison of step responses for a FOPTD

model (solid line) and the approximate integrator plus time
delay model (dashed line).
15
 Graphical Techniques for second-order models
Chapter 6

Figure 6.6

(  0)
16
Chapter 6

Figure 4.8

17
 Smith’s method: Find t20 and t60 and ratio of t20/t60

For overdamped Figure 6.7

second order

1      2  1

2      2  1

18
Exampled 6.3

6.8

19
 Chapter 6

20
Chapter 6

Figure 6.8
21
 Smith’s Method
20% response: t20 = 1.85
60% response: t60 = 5.0
t20 / t60 = 0.37
from graph
Chapter 6

t60 1      2  1
 2.8  = 1.79

 = 1.3 2      2  1
Solving,  1  3.81
 2  0.84
1 2 Sum of squares
S 3.81 0.84 0.0757
NLR (θ = 0 ) 2 . 9 9 1 . 9 2 0 . 0 0 0 0 2 8

F O P T D ( θ = 0 . 7 ) 4 . 6 0 - 0 . 0 7 6 0

22
 Using Excel Solver to Fit Transfer
Function Models
• use y (data) vs. y (predicted)
Chapter 6

• column 1 is data (taken at different times), or y1

• column 2 is model prediction (same time values as
above), or y2
• target cell is S (y1 - y2)2 , to be minimized
• specify parameters to be changed in reference cells
(e.g. 1 = 1, 2 = 2)
• open solver dialog box to check settings
• click on < solve > (calls optimization program) 23
 Chapter 6

24
 Chapter 6

25
 Neural Network Models
Chapter 6

26