用於提升電子系統可靠度之貝氏失效診斷法

Abstract

隨著汽車智慧化及駕駛自動化的發展，電子系統的可靠度對汽車安全性有著越來越大的影響力。電子系統的運用能協助偵側危險的駕駛行為同時也能適時的輔助一般駕駛者，但這些提升駕駛安全的功能往往因為電子系統在嚴苛的環境下行駛導致可靠度降低進而影響汽車安全性。 現存的可靠度分法如故障樹分析及失效模式與影響分析可以協助了解系統破壞的因果關係；馬可夫模型及可靠度分配透過可靠度的計算及量測幫助我們量化失效模式的發生機率。 儘管有許多可靠度相關的研究及分析方法，但對電子系統這種失效原因及結果相對複雜的系統而言，要找出每一種失效背後的原因仍是相當困難的，同時現實中對系統失效原因的診斷往往會因為高昂的量測成本而無法達成。 因此本研究欲提出一個找出電子系統最有可能的失效原因的診斷方法：透過物理破壞的角度，分析當焊接點受到環境及人為不確定因素產生電阻偏差值如何影響系統可靠度並找出客觀的失效原因排序；設計者可透過此分析結果進行量測，並透過貝氏更新法快速更新診斷結果，經由慢慢增加量測樣本直到診斷結果及結論得以建立。 本研究使用增壓器及變壓(頻)器系統做為範例演示整體診斷過程，過程中透過考量環境溫度及電子原件本身的發熱，可以得知電子系統的可靠度會因為開發階段不同的電路板設計和電子原件位置配置而受到影響，透過更改電路的原件配置，可將系統可靠度大幅提升。 本研究將分析診斷結果和`失效模式與影響分析表格(FMEA)'做整合，提供一個明確的系統可靠度改善方向；透過本研究提出的方法可以幫助電子系統在設計開發階段重新檢驗並提升可靠度，並使使用大量複雜電子系統的產品如汽車能在實際使用實能有更高的可靠度。

The reliability of electrical systems on modern vehicles has an increasing impact on the on-road safety with then increase of smart technology implementations. These electrical systems help detecting dangerous driver behaviors, alleviate driving errors, prevent unintended actions, as well as provide alternatives to internal combustion engines. The goals of having more efficient and safer vehicles could be undermined by the low reliability of electrical systems at severe driving environment. Existing reliability assessment methods, such as fault tree analysis and failure mode and effect analysis, focus on the cause and effects of component failure on system faults. On the other hands, Markov-chain based methods and reliability allocation techniques quantifies how these failure modes propagates within a complex system using reliability measure. Albeit abundant research activities, diagnosing the true origin of a system fault among all possible causes can be challenging. Incorporating these results for reliability improvements requires measurements that are costly in product development. Therefore in this research we develop a method to identifying the most likely origin of a electrical system fault with incremental data using Bayesian concept. We incorporate physics of failure in the welding joints of each components and consider the performance variations within each components. The result will be a subjective probability measure to rank the relative likely cause of a fault under varying environmental conditions. Designers can then use this result to sequentially identifying or measuring the health of each components until a conclusion is made. To deal with reliability with limited measurement samples, a reliability evaluating and updating scheme via Bayesian inference is established. We demonstrate the validity of the proposed method via a boost converter and an inverter. Consider the product development stage results in a electric layout that is to be realized. We show that due to the ambient temperature gradient and the rise of component temperature in operation, the reliability of a circuit rely on its configuration. We also use the examples to show the effect of the sample size on our probability measure. Combined with FMEA, the proposed method can help re-examine the electrical system in the earliest design stage toward high reliability target. For modern vehicles with a large number of complex electrical systems, our method can help improving the final reliability in real operation.