FMEA automation on computer log data and solution approaches with alternative methods

Since the terms of system and process have emerged, increase of quality of these and efforts for continuously improving them have become significantly important topics. In the literature and the studies that have been carried out up to now, quality and continuous improvement have been handled by using different methods by the researchers and authors. When technological opportunities were very limited, improvements were generally based on the efforts of experts. Along with the technological developments, operations and error records during processing can be stored in almost all the computer based systems. However, operations and records may reach massive sizes and thus new statistical methods have to be developed to obtain useful information from the raw data. Failure Modes and Effects Analysis (FMEA) is a powerful and proactive quality tool for defining, detecting, and identifying potential failure modes and their effects. An FMEA team consisting experts from different departments is organized in carrying out FMEA using traditional methods. This team firstly determines the failure modes. Afterwards, the effects of determined failure modes are analyzed and the analyzed failure modes are scored for different risk metrics. Finally, for each failure mode, a Risk Prioritization Number (RPN) is obtained by multiplying the score values. RPN values are evaluated for the effects of failure modes and the most important failure modes in the system are determined by ranking the RPN values. FMEA can be applied in any stage of a system. If FMEA is applied during the design stage, it enables prevention of possible failures. If it is applied at later stages of the system, it provides methods for analyzing and decreasing the failures which have been revealed. However, traditional FMEA consists of many deficiencies and inadequacies in a lot of ways. The leading and the most significant one of these deficiencies is that varying scores caused by different experts, and thus subjective decisions. Former studies were mainly focused on RPN and there was not much done research to decrease subjectiveness of scores decisions. In this thesis, computer log data domains are considered as application area. It is aimed to automate the stages of the FMEA and thus to remove both subjective decisions as well as making the analysis easier. Initially, the data obtained from the computers were pre-processed to make them fit for an FMEA study. At this stage, noisy and repeat data were removed. Failure modes were determined automatically from this data set instead of subjective expert decisions. For evaluating each failure mode, data based different objective risk metrics were suggested along with traditional methods and the suggested metrics were analyzed in detail. After that, by using Gray Relational Analysis (GRA), TOPSIS, VIKOR methods, and a traditional FMEA based method, risk rankings were ascertained. Results and methods are compared.