利用穿載式傳感訊號辨識營建施工人員作業行為之機器學習模型

Improvement of construction productivity depends on accurate and effective productivity measurement. Conventional productivity measurement approach has limited uses due to its demand for high level of labor effort. This research develops a machine learning model based on LSTM, which identifies construction worker's operations. An experiment involving 55 recruited subjects is conducted to allow the model to learn from collected data, and to test the accuracy for typical activities including rebar assembly, brick laying, wheel barrow moving, and resting. The result shows that the model has great identification accuracy in terms of determining whether subjects are working or resting (96.67% ~ 99.14%), performing upper-limb operations, lower-limb, or static operations (96.51% ~ 100%). However, when the objective is to identify detail operations, the accuracies are only good for identifying wheel-barrow moving and resting. The accuracies are considered not good enough to be used in the field for identify the rest of operations.