The blocks between Add1 through to Transfer Fcn: set any one of them to explicitly have an output width of 1.
The difficulty you are having is that with your loop the way you have it set up, it would be valid for you to be having a vector of any size being propagated. For example if the vector were of length 5 out of Add1, then that would go into the PID, which would create an output of length 5, which would go into Gain5, which would create an output of length 5, which would go into Gain4, which would create an output of length 5, which would go into Add2 which would produce an output of length 5, which would go into the Transfer Fcn, which would produce an output of length 5, which would go into the DeMux. But the Demux has not been told how to split up that input of length 5 because you have not told it how many values go to which output. If we suppose the bottom one were to get 5 outputs, then that would feed back in to Add1 and the loop would be complete and there would not be any problem with that part -- so a vector output is not inconsistent with any of those blocks. And that in turn means that Simulink cannot deduce what size it is supposed to be using. If you were to fix the output size from any one of the blocks, especially the Add1 block, then Simulink would be able to figure out all of the down-stream sizes.
However, look again at the Demux block. In order to demux, you have to be splitting the input up into multiple paths, so each of the outputs must be smaller than the inputs. So if you were sending a scalar through those Add1 and so on blocks, you would have a scalar out of the transfer fcn, going into the demux. Now you have at least two outputs from the demux but only one input, and that is a problem.
Demux is for splitting vectors. I don't think you want to be splitting the output of the transfer function: I think you want the same scalar output to be going to Add1 and the scope block. Which does not require any block: just connect both to the same input.
