Re: raw meal homogenization

Hello,

I think something misunderstood here.

We have homo and stock silos for raw meal preparetion. They are big silos, about capacity of 3000 ton. Raw meal from silo to air slide, then elevator and then small bunker i mentioned. The bunker i mentioned is not main silo. This is just a small bunker above weighfeeder.

Homogenization job is completed in homogenization silos.

We would like to improve homogenization by a new system i mentioned.

We are thinking about that put a pipe, feed back pipe from half heigh of bunker to elevator.

I think so it is now more clear.

Thank you.

Reply

Know the answer to this question? Join the community and register for a free guest account to post a reply.

Re: raw meal homogenization

If I understand correctly, you want to start a cycle 20t bunker -> elevator -> 20 t bunker.

Due to the very low mass of meal contained in this part of the system and residence time involved, frankly, I do not think it will improve your homogenization in any visible way.

By the way, a standard deviation of 0,75 on LSF of the fresh feed is already excellent (provided this is calculated on spot samples that is)

Reply

Re: raw meal homogenization

For the fun again, here is what happens if there is an autocorrelation in A(t):

V(B) = 
x²*V(A) + y²*V(B) +
2*x*y*sqrt(V(A)*V(B)) * C(A,B,T)

where

C(A,B,T) = <A(t)*B(t-T)>/sqrt(V(A)*V(B))

is the correlation of signals A and B
and assuming the correlation does not depend on t (stationarity)

Since the variance V and the standard deviation s are related by:

V(X) = s²(X)

we can write the first equation in another way:

s²(B) =
x²*s²(A) + y²*s²(B) + 2*x*y*s(A)*s(B) * C(A,B,T)

When A and B are perfectly correlated,
that is when C(A,B,T)=1,
we get:

s²(B) = (x*s(A) + y*s(B))²

or

s(B) = x*s(A) + y*s(B)

and this leads to:

s(B) = x/(1-y)*s(A)

or

s(B) = s(A)

In this way we just see the obvious!
When A and B are perfectly correlated, there is no variance reduction!
If A(t) and A(t-T) are perfectly correlated, then A and B are also.

As Mr ovancantfort  stated, this occurs for low frequencies, frequencies lower than 1/T.
Those frequencies cannot be reduced by this recycling process.

And we are back to the main question: which frequencies to you want to reduce?
Do you want to reduce the day-to-day variance, the hour-to-hour varaince, the year-to-year variance?

Some variance reduction can be obtained by other means than homogeneisation: this is the purpose of the raw mix control system based on samples analysis and feedback. Clearly for very low frequencies (like slow drifts) this is more appropriate than homogeneization.

Reply