|M.Sc Student||Zohar Kobi|
|Subject||Reliability Assessment of High Pressure Vessel by Fuzzy|
|Department||Department of Quality Assurance and Reliability||Supervisor||Dr. Yefim Haim Michlin|
|Full Thesis text - in Hebrew|
The proposed model encompasses all relevant factors, with a view to an optimal DM system up to the statistical errors allowed for in the design phase. The system was specifically intended for classification of high-pressure vessels in three clusters [qualified (compliance); disqualified (non-compliance); retest], but is appropriate for any product / component / system subject to such processing prior to its release. In the original case, the verdict is based on NDT data (corrosional & mechanical defects), age (number of undergone pressure cycles) and general visual-inspection findings.
Because of its inherent complexity, a soundness / serviceability evaluation process dictates a reliabilistic approach in which all aspects relevant to the decision are accounted for. Under conditions of uncertainty, a suitable tool is Fuzzy Logic in a Matlab ambience. FL is the overall term for theories in which the principles of rational thinking are applied to domains where the two basic laws of classical logic are inoperative, mainly where subjective or non-unequivocal conclusions have to be drawn. The essential difference between the two logics is that in CL any logic statement is either "true" or "false", whereas FL involves "degrees of truth" quantifiable as numbers.
The FL system comprises three blocks: entry, exit and central. These in turn comprise, respectively: the factors governing the DM process; the decision versions; and the set of rules linking the first two (the "brain" of the system).
Once the DM system has been established, limits have to be defined. In the present case of three alternatives, there are two such limits-a lower one between "disqualified" and "retest", and an upper one between "retest" and "qualified". For the limits to be judiciously chosen, a function has to be introduced, specifying in penalty for each of the two statistical errors associated with each limit. A penalty is minimized by finding the local minimum of the response surface. With a target function constructed incorporating all (error x coefficient) products, the decision limits can be chosen so as to create a balance between high and low-penalty cases.
The system ensures consistency in the DM process. It is unaffected by noise factors and adaptable to new requirements due to novel tools and to fresh information which introduces further parameters to the monitored. Finally, there is the major benefit of a greatly reduced need for the services of outside experts, thus preserving possibly classified or confidential data within the organization.