Automatic Tuning of Injection Molding Using Stochastic Search

Injection molding processes cannot be represented accurately by physical models. As such, methodical specification of process inputs that will yield acceptable products is generally not possible. For these processes, the input variables are assigned either by trial and error, based on heuristic knowledge of an experienced operator, or statistical design of experiments methods which require a comprehensive empirical model between the inputs and part quality attributes. In this paper, an iterative method of input selection (tuning) referred to as Virtual Search Method (VSM) is introduced to efficiently tune the process. VSM uses measurements of part quality attributes (e.g., size tolerances and surface integrity) as feedback and incorporates learning to improve the quality of future parts through changes in the process inputs. Input selection in VSM is performed by relying on an approximate input-output (I-O) model which speculates how prospective process inputs will affect the part quality attributes. This allows the VSM to utilize the I-O model as a ‘virtual’ process and perform the major part of the search for the appropriate inputs independent of the process. The I-O model is initially formulated from the available knowledge of the process to provide only a cursory representation of the process. This model is subsequently updated based on measurements of part quality as they become available during the iterative application of the process. The utility of VSM is demonstrated experimentally in tuning an injection molding process.