| dc.description.abstract | The master thesis has the aim of predicting the remaining useful lifetime (RUL) for an aging compressor drive system (CDS). Several objectives where established to achieve the aim. Among them are a review of the aging process, aging of the CDS, selection of a degradation model, prediction given condition and parameter estimation.
The CDS is located at the process plant Kollsnes, which is an onshore center for treatment of gas. Kollsnes supplies 8% of the European gas demand, and the fact that the CDS is aging makes the modeling of RUL of interest. The CDS at Kollsnes consist of 4 main components, where a synchronous electrical motor has been the main focus in this thesis.
Aging is a process that will gradually lead the motor towards an unacceptable condition due time and operation. The most relevant aging process for this thesis is functional degradation, which is a physical process that reduces the motors ability to function and/or perform as required. Aging stresses will cause degradation mechanisms to occur, which eventually will result in a motor failure. Based on several motors studies the insulation is identified as the part that limits the motors lifetime. A reason for this is its lack of maintainability, in addition to its vulnerability to the degradation mechanisms. Partial discharge (PD) are closely linked to these degradation mechanisms, where PD is the best variable to represent the aging process. The PD level is therefore used as a health indicator for the insulation condition, thus the motor condition.
The prediction of the RUL of the motor is based on its current condition and how the condition of similar motors has developed historically. There are several probabilistic models that can be used for this prediction. For this thesis a Markov process is selected, due to its ability of modeling degraded states with increasing failure rates, while it in addition gives a good overview of the degradation process. The Markov process is modeled with fictional failure rates since no real failure data were available. The degraded model is presented with several survival measures, like the RUL, and plots. The motor as a single component is the main focus of the modeling, however models on a system level considering season differences is presented in addition
The model used in the thesis is based on parameters, which is why a method for parameter estimation from interval censored data is presented. Actions to reduce the estimation error are identified, where they are related to inspection frequency and the number of samples included in the study. | |
