is it a Usual DBN?

[bahman]

Hi
i want to make this DBN.but as you can see in the Parameters i highlighted in the table 1; the node kt=Kt-1 and similarly Mt=Mt-1; it mean M and K are not change during time steps because these are parameter of meterals such as metal or steel. anyway and node D have formula that is not conditional on previous time step and it change only with t parameter in it's formula.

dbnn.jpg (171.14 KiB) Viewed 2785 times

so with this information i think this network is not a usual DBN because not have markov order arc like others i have seen.
during make this network in software i defined m0 node and then i put Mt=M0 for node Mt and Mt_1=Mt and ... for time stpe t . and similar for node K. and for node D i change t in formula manulally .
I have attached my file too

DBN.xdsl

(8.94 KiB) Downloaded 291 times

would you please help me and say that is this way i doing is true ?

thank you

[marek [BayesFusion]]

Hi Bahman,

When I look at the table of the variables in this model, I am also getting confused: Something is funny here, even though it does seem to be a DBN. We are quite surely dealing with N different temporal models (different devices?) that don't interact with each other outside of the initial parameters alpha (and possibly N, which seems to have a temporal link from another model?). It seems that the only two variables that changes in time are ln C_it and D (the table shows the value of Di0 and formula (10) shows the values at later times) and all other variables should be outside of the temporal plate. This, however, is contradicted by the t index in M and K, which suggests that they do change with time. It is possible that they could change in time but the author made an assumption that they don't, i.e., M_{i,t}=M_{i,t-1}. What are the variables Z, E and N? They are not in the table and they seem to be changing in time as well. Nodes that do not change in time should be placed outside of the temporal plate and it will be a better solution than entering equations of the form Mt=M0.

The figure seems to be showing an unrolled network. You will find GeNIe's graphical notation much more intuitive, although as I answered elsewhere (a related question), it does not yet support equation nodes in temporal plate, so you will have to work with an unrolled network.

If what I wrote above is true, you need temporal arcs, you need to add the variable C to your model with the only parent being M. I assume that dt represents variable D. You will need an arc from C to dt (it is a temporal arc; of course in an unrolled model it is going to be an arc between two neighboring slices).

My suggestion is that you clarify the issues with the author of the original network, as the combination of the figure and the table is not very easy to understand.

I hope this helps,

Marek

[bahman]

HI Marek,

you said right; network is for many metal components that are parallel to each other and D is the depth of the weld crack (mm) during time. other nodes are these are

lsf.jpg (95.77 KiB) Viewed 2778 times

and node Nf= cumulative number of component with Ec=1=fail
i was wondering to ask you about this nodes too;
how i can define a conditional node? i mean how i can get inverse of the node mean of node D ? because i need to check whether it is bigger than 50 or not !! i defined for node EC >>> ec=if(d>=50,1,0) but im not sure it is true
Nf is cumulative number Ec node for ec=1 for example 5 or 4 or ... it is deterministic value and it is not a probability node.

my big problem is in the Node Es; it is a reliabilty problem with limit state function. it can be solvel with Reliability methods like FORM and SORM methods but in paper it said "we solve this equation with DBN" and after that use inference for updating the network which i asked you about it in this topic. viewtopic.php?t=4788
i really afraid that is this node really can be solve with this dbn or not???
as you know Reliability methods are very comlex and do something on variables and transform them and ...
i really enthusiastic if you give me another idea about this.

thank you

[marek [BayesFusion]]

Hi Bahman,

Generally, a continuous (equation) variable can contain any function from the list of functions supported by GeNIe (please look at the manual or the tree of functions on the right-hand side in node definition. Conditional functions include If(), Choose() and Switch(). I believe that you want an If() in the case that you mentioned. I am somewhat confused by your reference to the mean of another node. This is, I'm afraid, impossible as there are theoretical problems with this -- you would like the definition to refer to the results and the result relies on the definition. So, unless I am misunderstanding your intentions, this will not work.

It is really hard for me to guide you through this problem, as I am not an expert in reliability methods. Has contacting the authors of the paper not worked for you? I suspect that if they used a DBN to solve this problem, then this problem can be solved using a DBN. Possible theoretically but very unlikely that they are misrepresenting the truth or are confused.
Cheers,

Marek