A computer-implemented method identifies an operation parameter of an industrial process as a candidate for modification so that the process can continue even of the material at the input of the process changes. In simulation instances, the computer receives representations of the material and of operation parameters and provides a representation of the would-be product. The computer classifies the instances into first and second quality classes. The computer continues by clustering—separated by parameters—the instances according to parameter attributes and according to the first and second quality classes. The computer repeats the simulation with variations that are related to significant differences, and identifies the candidate for modification.