Fuzzy supervision in statistical process control

Concerns the design of a DSS to improve the efficiency of fuzzy supervision of SPC. SPC is widely used in industry to control manufacturing processes, and therefore to control product quality. Using control charts, two statistical parameters (mean and standard deviation) are regularly checked to maintain the process under control and to improve its performance. However, the theoretical and practical limitations of SPC have led us to propose a fuzzy-rule-based supervision system. Actually, this system models the scheme of analysis, diagnosis and decision of an expert who controls a process by means of control charts. It is divided into three modules and at present time, delivers two gradual visual alarms. The first alarm, characterizing the current centering of the process, is given by the first supervisor. The objective of the second supervisor is to alert the operator of potentially abnormal variations of the process. A third one decides whether it is necessary to recalculate upper and lower control limits when it detects a new stability state of the process; it then delivers these parameters to the other supervisors. This system has been tested in simulation with real data and seems to induce better control compared to classic charts. Indeed, the fuzzy supervisor makes it possible to have a good anticipation of the process´s behavior and therefore helps the operator take decisions more easily.