基於多重感測器融合與長短期記憶模型預測可穿戴式感測器剩餘使用壽命

可靠的設備功能對於醫療診斷步態分析必不可少。預測和健康管理策略可通過傳感器監控設備的剩餘使用壽命及識別故障設備,以最大程度地減少潛在的致命誤診。剩餘使用壽命的概念在航空航天工程,製造工程和電氣工程領域得到了很好的研究。然而,其在生物力學中的應用仍未得到開發。這項研究的目的是通過使用基於機器學習的特徵重要性機制,通過對各種可穿戴式步態傳感器(IMU、壓力傳感器) 進行融合以得到有指標性的傳感器數據,來構建生物力學傳感器系統, 進而增強設備的剩餘使用壽命的數據表示能力。並且,使用 LSTM 的模型架構將該數據用於預測剩餘使用壽命。實驗是根據從候選對象獲得的數據進行的,結果顯示,與基於壓力傳感器的模型架構相比,RMSE 和加權分數分別提高了 25.68%和 49.62%

關鍵字

剩餘使用壽命；長短期記憶模；多傳感器

並列摘要

Equipment reliability is crucial for medical diagnostic gait analysis. Prognostics and health management strategies may be employed, by monitoring the remaining useful life of an equipment through sensors, and identifying faulty equipment to minimise potential fatal misdiagnosis. The concept of remaining useful life is well studied in the field of aerospace engineering, manufacturing engineering, and electrical engineering; however, its application for biomechanics remains underexplored. The purpose of this study is to construct a biomechanical sensor system, by performing fusion on varied wearable gait sensors (IMU, pressure sensor) to leverage unique sensor data, using a machine learning- based feature importance mechanism, whereby an enhanced data representation of the equipment’s remaining useful life degradation is obtained. Then, this data is utilized to predict the remaining useful life using an LSTM architecture. The experiment is con- ducted on data obtained from human candidates, and results indicate RMSE and score improvement of 25.68% and 49.62% respectively, when compared to a pressure sensor based architecture.