Predictive Maintenance of Circuit Breakers
Leung , Tat Wai (Alan)
MetadataShow full item record
For predictive maintenance of circuit breakers, a number of variables must be considered in order to assess the genuine working condition of a circuit breaker [CB]. This thesis selects vibration signatures obtained on the operating mechanisms and arcing chambers as a source of monitoring breaker conditions. The task of analyzing the behavior of a circuit breaker is perennial and difficult but the thesis has an attempt to tackle this problem. Experiments have been devised to monitor CBs; however, these have limitations details of which will be discussed. For example, each circuit breaker has its own unique vibration signature and the shape of the vibration may be different even though breakers confront similar problems. CBs have decades-long service life spans and failure rates are relatively low. Those that fail are not necessarily saved and there have been relatively few samples to base evidence upon. There are different vibration analysis algorithms available including Dynamic Time Warping [DTW], Resolution Ratio [RR], Discrete Envelope Statistics [DES], event time extraction, Chi-square based shape methods, and fractal theory. Some of these algorithms are based on acoustic properties of materials and rely on assessing extracted time component and the frequency components are extracted. This research applies multi-resolution analysis [MRA] to decomposed signals to in order to assess different sub-wave levels so that wave features may be captured and modeled. There are many ways to analyze the waves. This thesis uses optimizing fuzzy rules with genetic algorithm [GA] as the proposed method. The simuation part of the thesis uses spring performance as an example of how vibration signature analysis may be implemented. Spring vibrations are evaluated by two classification algorithms: Dynamic Time Warping [DTW] and multi-resolution analysis [MRA] with optimizing fuzzy rules with genetic algorithm [GA]. The first method is competent to identify the faulty cases from the normal ones by looking at the deviation of the vibration signature frequency content. In contrast, it is not capable to identify the degree of how bad it performs from looking at the frequency variation. For the second method, it is capable of not only classifying the abnormal cases from the normal cases, but also distinguishing the vibration signatures into different category so that the spring condition can be retrieved immediately. Fuzzy rules is capable of classify a new case to a category and genetic algorithm is an effective tool to minimize the applicable fuzzy rules. The accuracy of the identification is very satisfactory, which is over 90%. Consequently, the proposed algorithm is very useful for asset management purpose of breaker since the lifespan of the spring is known. Diagnostic technicians are able to make decision on the replacement scheme of the spring. There are some areas that this research uncovered that suggests further study is mandated. For example, there are other parameters that can be monitored and compared other than spring constant such as valve position in trip coil and close coil, acceleration parameter in changeover valves, damping in hydraulic cylinders and mechanical linkages, gas pressure in primary contacts and breaker resistance in line system.