面向传感器突发故障的监测模型自适应快速调整技术研究

针对多传感器监测场景中传感器突发故障导致监测模型失效的问题，本文提出了一种堆叠自编码器深度学习模型的自适应快速调整方法。根据传感器故障时采集数据的分布特点自适应调整原始数据集，采用调整后的数据集正向传播和监测误差反向传播微调模型更新模型的权重和偏置，实现监测模型自适应快速调整。以机械加工过程中刀具磨损状态监测为例，采用加州大学伯克利分校的BEST实验室提供的刀具数据集验证了自适应调整方法的有效性。结果表明，该方法可解决当传感器突发故障时，在实时监测不中断的情况下，自适应调整后的监测模型可以准确地对刀具状态进行监测。

Study on Adaptive Fast Adjustment of Monitoring Model for Sensor Sudden Fault

To solve the sensor fault leading to failure of monitoring model for multi-sensor monitoring scenarios, an adaptive fast adjustment method of stacked autoencoder (SAE) deep learning model is proposed in this paper. The original data set was adjusted adaptively according to the distribution characteris-tics of the data collected during sensor fault, the weight and bias of the model are updated by using the adjusted data set forward propagation and the monitoring error back propagation to fine-tune model, and the adaptive fast adjustment of the monitoring model was realized. Taking tool wear monitoring in machining as an example, the effectiveness of the proposed method was verified with the data set provided by the BEST lab at UC Berkeley. The results show that the adaptively adjusted monitoring model can accurately monitor the tool state under the condition that the real-time monitoring is not interrupted when the sensor breaks down suddenly.