故障樹分析軟體INERFT測試驗證之結果分析

Benchmark Tests of the Fault Tree Analysis Software Package INERFT

核能研究所故障樹分析（Institute of Nuclear Energy Research Fault Tree, INERFT）套裝軟體是核能研究所開發的本土化、中文化故障樹分析套裝軟體；為落實國內非核能領域的推廣應用，有需要進行周詳嚴謹的測試驗證，以確認INERFT的計算結果為可信。INERFT的測試驗證方法係利用學術期刊或學術報告的現成案例，並採用國際知名商業化故障樹分析軟體的結果，比對、審查並確證計算所得的最小失效組合、頂端事件機率值。測試驗證結果顯示，學術報告的案例測試驗證正確無誤；學術期刊文章告知4896個最小失效組合是錯的，而INERFT經多方面的測試證實是對的。核電廠的故障樹案例考量元件失效機率的計算模式及外接故障樹的狀況，INERFT三種展開方式及WinNUPRA均得到8949個最小失效組合，並且彼此完全吻合。INERFT航太工業版也是經多方面周詳嚴謹的測試驗證，正確無誤。頂端事件機率值的測試驗證結果顯示，一些案例的近似值仍遠偏離其正確值；至於頂端事件機率的正確值，在最小失效組合個數小於10的案例中，INERFT計算結果與手算方式的實際結果是吻合的；在最小失效組合個數上千的測試案例中，INERFT 頂端事件機率正確值的計算結果是可信的；INERFT 頂端事件機率正確值的計算所花時間相當短，有的甚至比二階近似計算所花時間更短。總而言之，INERFT 經周詳嚴謹的測試驗證，可確定INERFT的計算結果是可信的。

INERFT is a fault tree analysis (FTA) software developed by the Institute of Nuclear Energy Research. To extend the FTA software applications in the domestic non-nuclear energy field and enhance user confidence, it is necessary to verify the reliability of INERFT. The verification methods include the use of fault tree analysis examples cited in academic journals and technical reports and the use of well-known commercially available FTA software packages for comparison. The verification methods are designed to check the accuracy of all minimal cut sets (MCSs) and to review the top event probability in various ways. The verification results show that INERFT provides comparable results for the fault tree examples cited from technical reports. However, for the example in the academic journal, the reported result of 4896 MCSs is inaccurate and INERFT is verified correctly in various ways. The nuclear power plant example considers equipment failure probability model and external transfer branches. The INERFT, using three different expansion approaches, and the WinNUPRA all result in 8949 MCSs and the MCSs are all consistent one by one. The aerospace industrial version is also verified. The verification of the top event probability calculation shows that the approximation is unacceptable in some cases. As for the exact top event probability, the results for the cases with less than 10 MCSs are consistent with the results obtained by hand calculation. For cases with more than 1000 MCSs, the results are acceptable. The computing time for the exact top event probability calculation is rather short, and in some cases, even shorter than the time needed for the second approximation. Above tests clearly demonstrate the validity and applicability of INERFT.