Getting a result, even an improbable one, is better than getting an Unknown Error. What change did you make to flip between these two outcomes?
An incremental approach is to break the full model into smaller pieces. For example, suppose the full model is made up of 20 parts, and those parts are grouped into 4 subassemblies that are connected with bolts to a base part. In Workbench, under the Components section of the Toolbox, drag out a Mechanical Model and name that Assy1. Attach the geometry file (that should be organized with 4 subassemblies), and suppress 3 assemblies so you can build this first model with just the base part and assy1. Drag and drop a Static Structural on this Assy1 Mechanical Model. Now you can support the base part and apply loads to interfaces on assy1. Get that model working.
Next drag out another Mechanical Model. Import the geometry of assy2 and get that meshed and connected internally. Drag and drop a Static Structural onto Assy2 and add some supports and loads to just that to make sure it is working.
Drag out another Mechanical Model and link the Model cells of Assy1 and Assy2 to it. Drag and drop a Static Structural model onto this Mechanical Model.
Continue adding each Assy and testing it out before linking it to the Full Model where the connections between them are defined. In this way, you establish assemblies that solve and give reasonable results.