以座艙通話紀錄器資料及座艙環境資訊進行飛安事故調查之操作開關辨識

分析座艙語音紀錄器（Cockpit Voice Recorder, CVR）的資料對於大部分的飛航事故調查來說是最重要的一個步驟。透過語音或是其他音訊的辨識，人為以及非人為的因素如通訊溝通錯誤、違反標準操作程序、機艙損壞以及警報聲響的種類都能被判斷出來。然而，要辨別出位於不同位置的相同種類開關是很困難的。傳統的聲源定位方式需要最少四支麥克風所測量出來的抵達時間差（Time Difference of Arrival, TDOA）才能進行定位，但是在座艙內通常只有三支麥克風的聲音會被錄進座艙語音紀錄器之中。因此在此提出了兩種方法，利用對座艙環境的已知來越過由麥克風數目所造成的限制。Multilateration with insufficient sensors (MLATIS)利用trilateration求得由抵達時間差及麥克風位置算出的兩個雙曲面的交線，再以此交線和以知的開關分布平面求交點，則此交點就是MLATIS所辨識出的聲源位置。Source identification by location lookup table (SI-LLT)事先對於每個開關的抵達時間差的建立數據表，在以此和測得的抵達時間差做比對以辨識可能的開關位置。接著以蒙地卡羅模擬(Monte Carlo simulation)來分析取樣頻率與駕駛員所配戴的麥克風的位置不確定性對於辨識準確度的影響。最後，以初步實驗的結果驗證模擬所預測出來的趨勢的確與實驗相符合。

關鍵字

聲源定位、座艙通話紀錄器(CVR) ； trilateration ；抵達時間差 (TDOA)

並列摘要

Analysis of the cockpit voice recorder (CVR) data retrieved from black box flight recorders is critical to most aviation accident investigations. By speech or signal recognition techniques in some form, anthropogenic and non-anthropogenic factors such as vocal communication errors, violation of standard operating procedures, fuselage damage, and alarm classification can be inspected. However, identification of activated switches in a cockpit is difficult because there are same types of switches at different locations. Conventional source localization algorithms cannot be used because they require at least four time-difference-of-arrival (TDOA) sensors while usually there are only three microphone signals recorded in a CVR. In this thesis, two methods to overcome this constraint by exploited a priori cockpit geometry information are proposed. Multilateration with insufficient sensors (MLATIS) estimates source locations by intersecting a two-hyperboloid curve derived from estimated TDOAs and modified trilateration with a known switch distribution surface. Source identification by location lookup table (SI-LLT) identifies an activated switch by comparing the estimated TDOA to a TDOA table based on known switch locations. Impacts of sampling rates and uncertain headset microphone positions on identification accuracy are analyzed by Monte Carlo simulation techniques. Preliminary experimental results verify some of the trends predicted by simulations.

並列關鍵字

sound source localization ； cockpit voice recorder (CVR) ； trilateration ； time difference of arrival (TDOA)

參考文獻

[7] Stearman R O, Schulze G H and Rohre S M 1997 Aircraft damage detection from acoustic and noise impressed signals found by a cockpit voice recorder J. Acoust. Soc. Am. 101 3085

[8] Xiao X, Yao H and Guo C 2009 Automatic detection of alarm sounds in cockpit voice recordings Proceedings of the IITA ICCASE 00 599-602

[9] ARINC 2009 767-1 Enhanced airborne flight recorder

[13] Fang B T 1986 Trilateration and extension to global positioning system navigation J. Guidance 9 715-7

[14] Manolakis D E 1996 Efficient Solution and Performance Analysis of 3-D Position Estimation by Trilateration IEEE Trans. Aerosp. Elect. Syst. ASSP-35 8

國際替代計量

以座艙通話紀錄器資料及座艙環境資訊進行飛安事故調查之操作開關辨識

全文下載

主題瀏覽