DC-link 필름 커패시터에 대한 수명 시험 기간 최적화 연구

Abstract

전기 자동차(electronic vehicle: EV) 또는 하이브리드 전기 자동차(hybrid electronic vehicle: HEV)의 인버터 시스템에는 DC-link 커패시터(direct-current link capacitor)가 있다. 이 DC-link 커패시터는 무효전력(reactive power)을 제공하며, 리플 전류(ripple current)를 약화시키며, 전자파간섭을 줄이고, 누설 인덕턴스(leakage inductance)와 스위칭 전환으로 일어나는 역써지전압(voltage spikes)을 막는다. 필름 커패시터(film capacitor)는 고전력 시스템에서 DC-link 커패시터로 사용된다. 필름 커패시터는 다른 커패시터보다 높은 수명과 좋은 성능을 가지고 있다. 하지마 필름 커패시터의 성능은 시간이 지남에 따라서 낮아지게 된다. 필름 커패시터 성능의 감소는 배터리에서 차량의 전력 시스템에 에너지를 전달함에 있어서 문제를 야기할 수 있다. 그러므로, 적절하게 필름 커패시터의 교체 시기를 알기 위해 필름 커패시터의 수명을 아는 것이 필요하다. 즉, 필름 커패시터의 수명 예측은 전기 자동차와 하이브리드 자동차의 중요 요소 중 하나이다. 이런 이유에서 필름 커패시터의 수명과 신뢰성은 자동차 인버터 시스템의 안정성을 보여주는 중요 요소이다. 이전 연구에서는 주로 아레니우스 식(Arrhenius equation), 등가직렬저항(equivalent series resistance: ESR), 주파수의 식 등과 같은 화학식을 이용하여 필름 커패시터의 수명을 예측하였다. 하지만 이들 연구는 불확실성에 대한 신뢰성에 문제를 가지고 있다. 본 논문에서는 통계적인 방법을 이용하여 필름 커패시터의 신뢰성에 대한 새로운 예측 모델을 제안한다. 필름 커패시터의 통계적인 모델을 소개하기 전에, 커패시턴스 감소(capacitance loss)을 고장의 지표로써 설명한다. 커패시턴스 감소는 두 가지 감소로 구분되며 이는 각각 자연 감소(natural loss)와 급진 감소(sudden loss)로 구분된다. 자연 감소는 지속적으로 감소하는 특성을 가지고 있고 급진 감소는 순간적으로 감소하는 특징을 가지고 있다. 그러므로 이 커패시턴스 감소의 두 가지 종류에 대해 각각의 모델을 구성할 필요가 있다. 자연감소의 모델로는 확률 변수 저하 모델(random variable deterioration model)이 사용된다. 이 확률 변수 저하 모델은 오직 단순하고 지속적인 감소만을 고려하는 모델이다. 급진 감소의 모델로는 감마 과정 저하 모델(gamma process deterioration model)이 사용된다. 이 감마 과정 저하 모델은 단조적으로 축적되는 감소를 고려하는 모델이다. 각각의 감소에 대한 모델을 설명한 후에, 이 두 가지 모델을 합친 새로운 모델이 제안된다. 그 후 각 모델에 대한 변수 추정에 대해 설명한다. 확률 변수 저하 모델의 변수 추정은 커브 피팅 알고리즘(curve fitting algorithm)을 이용하여 구한다. 감마 과정 저하 모델의 변수 추정은 EM 알고리즘(Expectation-Maximization algorithm)을 이용하여 구한다. 이들 모델과 추정기법을 이용하여 필름 커패시터의 수명을 예측하게 된다. 예측한 결과와 실제 데이터의 비교를 통해 새로 제안된 모델이 좋은 성능을 가지고 있음을 증명하였다. |In electronic vehicles (EV) or hybrid electronic vehicles (HEV), an inverter system has a direct-current-link capacitor (DC-link capacitor) which provides reactive power, attenuates ripple current, reduces the emission of electromagnetic interference, and suppresses voltage spikes caused by leakage inductance, and switching operations. A film capacitor has been used as the DC-link capacitor in high level power system. The film capacitor has longer lifetime and better performance than other capacitors. However, the performance of the film capacitor has deteriorated over operating time. The decreasing performance of the film capacitor may make a problem to supply and deliver energy from battery to the vehicle’s power system. Therefore, it is necessary to know the accuracy lifetime of film capacitor to find proper replacing period of the film capacitor. In other words, the lifetime prediction of the film capacitor could be one of critical factors in the EV and HEV. For this reason, the lifetime and reliability of the film capacitor are key factors to show the stability of the vehicle inverter system. A lot of methods to predict the lifetime of the film capacitor using physical or chemical equations such as Arrhenius equation, considering ESR, temperature, or frequency equations have been researched. However, these previous researches have a problem about robustness with respect to uncertainty.

In this paper, a new prediction model about the lifetime and stability of the film capacitor are guaranteed by stochastic methods is proposed. Before making the stochastic model of the film capacitor, capacitance loss is introduced as a failure detecting indicator. The capacitance loss can be separated as two types: natural and sudden loss. The natural loss has steady losing character and the sudden loss has random losing character. Therefore, it is need to make two models for two types of capacitor loss. For the natural loss, random variable deterioration model is introduced and used. The random variable deterioration model considers only sample and steady loss. For the sudden loss, gamma process deterioration model is introduced and used. The gamma process deterioration model is ideally suited to model gradual damage that monotonically accumulates over time. After making two models about the natural loss and the sudden loss, a new model combined by the random variable deterioration model and the gamma process deterioration model to consider the natural loss and the sudden loss of the film capacitor is proposed. The parameters of the random variable deterioration model are obtained by using curve fitting algorithm based on real experiment data. For estimating the parameters of the gamma process deterioration model, Expectation-Maximization algorithm (EM algorithm) is used. Using these models and algorithms, the lifetime and the stability of the film capacitor is calculated and predicted. Comparing the predicting result and the real experiment result shows that the new proposed prediction model gives a good performance.; In electronic vehicles (EV) or hybrid electronic vehicles (HEV), an inverter system has a direct-current-link capacitor (DC-link capacitor) which provides reactive power, attenuates ripple current, reduces the emission of electromagnetic interference, and suppresses voltage spikes caused by leakage inductance, and switching operations. A film capacitor has been used as the DC-link capacitor in high level power system. The film capacitor has longer lifetime and better performance than other capacitors. However, the performance of the film capacitor has deteriorated over operating time. The decreasing performance of the film capacitor may make a problem to supply and deliver energy from battery to the vehicle’s power system. Therefore, it is necessary to know the accuracy lifetime of film capacitor to find proper replacing period of the film capacitor. In other words, the lifetime prediction of the film capacitor could be one of critical factors in the EV and HEV. For this reason, the lifetime and reliability of the film capacitor are key factors to show the stability of the vehicle inverter system. A lot of methods to predict the lifetime of the film capacitor using physical or chemical equations such as Arrhenius equation, considering ESR, temperature, or frequency equations have been researched. However, these previous researches have a problem about robustness with respect to uncertainty.

In this paper, a new prediction model about the lifetime and stability of the film capacitor are guaranteed by stochastic methods is proposed. Before making the stochastic model of the film capacitor, capacitance loss is introduced as a failure detecting indicator. The capacitance loss can be separated as two types: natural and sudden loss. The natural loss has steady losing character and the sudden loss has random losing character. Therefore, it is need to make two models for two types of capacitor loss. For the natural loss, random variable deterioration model is introduced and used. The random variable deterioration model considers only sample and steady loss. For the sudden loss, gamma process deterioration model is introduced and used. The gamma process deterioration model is ideally suited to model gradual damage that monotonically accumulates over time. After making two models about the natural loss and the sudden loss, a new model combined by the random variable deterioration model and the gamma process deterioration model to consider the natural loss and the sudden loss of the film capacitor is proposed. The parameters of the random variable deterioration model are obtained by using curve fitting algorithm based on real experiment data. For estimating the parameters of the gamma process deterioration model, Expectation-Maximization algorithm (EM algorithm) is used. Using these models and algorithms, the lifetime and the stability of the film capacitor is calculated and predicted. Comparing the predicting result and the real experiment result shows that the new proposed prediction model gives a good performance.