Monitoring the fatigue of construction equipment operators (CEOs) is critical for preventing accidents and ensuring precision construction occupational health and safety (COHS). However, there exists a theoretical dilemma between centralized technical efficiency and decentralized data privacy. Thus, this study introduces smart work package learning (SWPL), a decentralized deep learning approach to monitor CEOs’ fatigue without privacy exposure risks. To illustrate the feasibility of SWPL as the fatigue classifier, this study implements fatigue monitoring through noninvasive facial images, and SWPL merges the updated parameters of the model from each smart work package (SWP). These updates are then validated by SWPs in the blockchain network and stored on the blockchain. More than 356 videos were derived from 124 operators. The results present that SWPL on decentralized SWP networks outperforms the deep learning model on individual SWP. The computational novelty is SWPL's dynamic parameter aggregation mechanism to avoid parameter exposure in centralized or fixed aggregators. The proposed SWPL will open up advanced developments in precision COHS.

中文翻译：

---

通过面部图像进行分散式疲劳监测的智能工作包学习

监测施工设备操作员 (CEO) 的疲劳度对于预防事故和确保精确的施工职业健康与安全 (COHS) 至关重要。然而，中心化的技术效率和去中心化的数据隐私之间存在着理论上的困境。因此，本研究引入了智能工作包学习 (SWPL)，这是一种分散的深度学习方法，可以在没有隐私暴露风险的情况下监控 CEO 的疲劳度。为了说明 SWPL 作为疲劳分类器的可行性，本研究通过无创面部图像实现疲劳监测，SWPL 合并来自每个智能工作包（SWP）的模型更新参数。这些更新随后由区块链网络中的 SWP 验证并存储在区块链上。超过 356 个视频来自 124 个操作员。结果表明，去中心化 SWP 网络上的 SWPL 优于单个 SWP 上的深度学习模型。计算新颖性是 SWPL 的动态参数聚合机制，以避免在集中式或固定聚合器中暴露参数。拟议的 SWPL 将开启精密 COHS 的先进发展。