While the 5G is being rolled out in different industrial sectors, the 6G is expected to implement data-driven ubiquitous machine learning for industrial information integration environment. The machine learning algorithms show great potential in predictive maintenance where data is collected over time. However, the privacy issues become increasingly challenging since the huge amount of data are exchanged in the machine learning based solutions. On the other hand, it is very important to lightweightise the machine learning by leveraging effective machine learning models without restriction of privacy concerns. The homomorphic encryption techniques make it possible to perform machine learning model over encrypted data to address above challenges. This work introduced a lightweight privacy-preserving predictive maintenance technique based on machine learning in 6G-IIoT scenarios, which can help to quantify the risk of failures for industrial assets or systems in any moment. Binary neural networks (BNNs) were introduced to train predictive maintenance model, that can be implemented with homomorphic encryption circuits to guarantee the privacy of all participants. Experimental results demonstrate the effectiveness of the proposed solution.

中文翻译：

---

支持 6G 的 IIoT 中的轻量级隐私保护预测维护

在5G在不同工业领域落地的同时，6G有望为工业信息集成环境实现数据驱动的泛在机器学习。机器学习算法在随着时间的推移收集数据的预测维护方面显示出巨大的潜力。然而，由于在基于机器学习的解决方案中交换大量数据，隐私问题变得越来越具有挑战性。另一方面，在不限制隐私问题的情况下，利用有效的机器学习模型来轻量化机器学习是非常重要的。同态加密技术使得对加密数据执行机器学习模型以解决上述挑战成为可能。这项工作在6G-IIoT场景中引入了一种基于机器学习的轻量级隐私保护预测维护技术，可以帮助量化工业资产或系统在任何时刻发生故障的风险。引入二元神经网络（BNN）来训练预测维护模型，该模型可以通过同态加密电路来实现，以保证所有参与者的隐私。实验结果证明了所提出解决方案的有效性。