In customized productions, i.e. manufacturing processes with low batch quantities, control systems need to be integrated into the process with relatively little adaption being required. To do this, a module-based system technology has been developed that focuses on the general conditions of inline quality assurance - so-called low runner processes. Quality forecasting is generally conducted synchronous with the cycles in tristate form (OK, check, reject). Starting with a minimum quantity of memorized process data, the process model of the system then teaches itself on the job as required.
A main feature of the system technology is the integration of additional sensors into the process tool. The signal sequences recorded in this way are specifically edited depending on the task at hand. Based on these sequences and on automatically-determined parameters with adaptive models, conclusions are then drawn about the quality of parts. The datasets used to model the components are automatically modified by the system with the aid of special training functions. At the start of the adaption process, the lowest possible quantity of datasets is first generated with statistical test plans. Later on, these are gradually extended parallel to the production process. Practical trials with injection-molding processes have classified 98% of components correctly. Process states that had not been included in the training data were reliably detected and their quality also correctly determined once the model had been adjusted.