Friction stir welding is not used for hard alloys because of premature tool failure. A scheme is created that exploits the physical three-dimensional heat and mass flow models, and implements them into a fast calculation algorithm, which, when combined with damage accumulation models, enables the plotting of tool durability maps that define the domains of satisfactory tool life. It is shown that fatigue is an unlikely mechanism for tool failure, particularly for the welding of thin plates. Plate thickness, welding speed, tool rotational speed, shoulder, and pin diameters and pin length all affect the stresses and temperatures experienced by the tool. The large number of these variables makes the experimental determination of their effects on stresses and temperatures intractable and the use of a well-tested, efficient friction stir welding model a realistic undertaking. An artificial neural network that is trained and tested with results from a phenomenological model is used to generate tool durability maps that show the ratio of the shear strength of the tool material to the maximum shear stress on the tool pin for various combinations of welding variables. These maps show how the thicker plates and faster welding speeds adversely affect tool durability and how that can be optimized.