Two types of injection-compression used for thin walled high aspect ratio parts. The first uses a moving side core or secondary cylinder that leaves the cavity spaced and upon injection clamps forward to compress the part stress free. This usually requires a specific capability of the press. The second method sometimes referred to as coining closes the mold under reduced clamp pressure and the cavity is filled.
In this design it is also typical to have roughly a 7 degree draft angle on the parting/vent line to avoid flashing during the injection-compression stroke depending on the material and criteria for edge definition. keep in mind other critical control parameters are melt viscosity, plastification, pellet size uniformity as the cushion is typically much lower than with standard straight injection to avoid shot to shot variance.
Scientific Molding to me is a bit different than having a whole lot of data studies. To me, and to many of my peers, scientific molding is an disciplined elimination of multiple variables, which yields an range of conditions/settings/protocol/things that you must do. This will yields the highest volume of acceptable parts with the least amount of cost. I.E., when you do this, and you do that - the press operates thus and you get money. Going thru the methology checklist eliminates the variables in a disciplined fashion resulting in an updated group of setting, updated methods of handling material change outs, setting up a mold, scheduling jobs, etc. and reduction of scrap. Happy customers, happy plant operators, more money.
The simulations results have a qualitative accuracy for this reason many companies and the academia are working in this area. The reason is that all models use strong assumptions that can be weak depending on the material, geometry and process conditions. I can give an example for fiber filled injection molded composites. For short fiber composites in a center-gated disk test sample, we have found that the actual models fail to predict the orientation near the gates and up to r/H =39. The actual model implemented in most software is reasonably good above 39, but still prediction of fiber orientation near the walls is challenging. Additionally, the quality of the predictions reduces noticeably as you increase the fiber content and the thickness of the parts. These two conditions reduce the validity of the assumption used in the simulation packages. The area of interest of many software developers is the prediction of fiber orientation of long fibers. This is an area of opportunity because we are still unable to predict a reliable fiber orientation due to the semi-flexibility of the fibers.
Specific decisions for the best approach (material, process, design detailing) can only be made with a detailed knowledge of all the constraints (structural, dimensional and financial) and all the must-have, nice-to-have and absolute must-not-have features and attributes. Is there not a web address where we can get more detailed information? Another process (actually, a hybrid process) came to mind and this might suit the combination of requirements that I think you have. This is offered by a couple of machine manufacturers. You'd need to contact the machinery manufacturers direct for contact details of processors offering this service. A thin skin (can be vacuum-formed) is normally inserted in the relatively low-cost mold, to give high-quality surface. The robotic head dispenses PUR foam with reinforcing fibers - that are chopped in the head also - into the open mold. Once the mold is closed, the "B" surface is formed, complete with any ribbing and fixing points, etc. However, the total thickness will tend to be significantly well above the 5mm maximum that you have defined. Rigidity is very impressive. Impact strength will depend on surface material and density/reinforcement variables in the backing material.