An Accelerometer-based Fall Portent Detection Algorithm for Construction Tiling Operation
|關鍵字:||營造安全;墜落職災;墜落前兆;加速規;階層式門檻值演算法;Construction safety;Fall accident;Fall portent;Accelerometer;Hierarchical threshold-based algorithm|
|摘要:||墜落意外為營造業主要職業災害類型，佔各國營造業重大職災比例約30%–50%。營造業工地範圍大、高處作業多、作業環境變異性大且多為臨時性結構，同時多層承攬及勞工流動性高導致工地現場安全管理困難。國內墜落意外預防以查核安全設施及設備為主，然其僅能減少墜落意外的傷亡，並不能避免其發生，防災成效有限。營造業為重體力負荷產業，勞工身心狀態不良導致失去平衡或失去警覺為意外發生的主要原因之一。若能偵測墜落意外前的身心狀態特徵(墜落前兆)，應能有效減少墜落意外。加速規能反映身體晃動程度，並已廣泛應用於老人及病人的遠距醫療照護及跌倒偵測。本研究即應用加速規建立墜落前兆偵測系統，並開發偵測演算法及預警機制。本研究同時建構一擬真施工架工作環境，並設計磁磚鋪貼作業任務以驗證系統預警績效。墜落前兆偵測受到作業垂直性動作(彎腰、蹲下、起立)影響，極容易產生誤報。本研究提出一階層式門檻值演算法(Hierarchical threshold-based algorithm)，其搭配多個加速規能有效減少誤報，提升預警準確率(79.13%)。其中，受試者身心狀態不良時(酒醉、睏睡)，作業動作幅度較小，演算法準確率表現(89.17%、79.31%)較正常狀態(67.31%)佳；受試者在睏睡狀態時，墜落前兆伴隨較明顯的晃動，演算法偵測率表現最佳(87.50%)。分心、恍神及注意力不集中等墜落前兆，若無伴隨明顯晃動，則較難以加速規進行偵測，未來可考慮整合加速規訊號及其他生理訊號偵測(如腦波訊號)。本研究雖僅針對磁磚鋪貼作業進行驗證，但演算法之建立過程可提供其他作業參考。此外，在動作辨識上，本研究所開發之演算法除可偵測墜落前兆動作，提供現場管理者勞工安全重要資訊外，亦可應用於勞工生產力監測，辨識勞工作業活動狀態。|
Fall accidents in the construction industry have been studied and identified as a common hazard and the leading cause of fatalities for several decades. Approximately 30%–50% of fatal accidents are caused by falls. Given the strenuous nature of construction work, workers are prone to loss of awareness and balance, increasing the safety risk and number of fall accidents. Thus, monitoring the mental and balance conditions of workers may help identify fall portents, and thus prevent falls from happening. This research developed an accelerometer-based fall portent detection system with a hierarchical threshold-based algorithm. To evaluate the warning effectiveness of the proposed system and algorithm, we designed experiments involving multiple tiling tasks. The participants performed the tasks under three different statuses (i.e., normal, inebriation, and sleepiness) on a scaffold, while carrying four accelerometers attached to their chest, waist, arm, and hand. The results show that work-related motions (e.g., standing, stooping, and squatting) had a limited impact on the proposed algorithm, which exhibited an acceptable accuracy of 79.13%. In addition, the algorithm exhibited a higher accuracy under the inebriation and sleepiness statuses than under the normal status because of the decrease of regular motions. Furthermore, the algorithm exhibited the lowest miss rate under the sleepiness status because of the portents of more obvious sways produced by the participants under that status. However, some portents related to loss of awareness were difficult to detect by using accelerometers; future research could apply brain wave monitoring to detect such portents. The proposed system and algorithm can be applied in various working operations exposed to falling hazards, and provide jobsite managers with valuable information on whether a worker’s physiological status is suitable for work. In addition to safety monitoring, the system can also be applied for productivity monitoring and the distinction between different work postures.
|Appears in Collections:||Thesis|