当前位置: X-MOL 学术Measurement › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Intelligent monitoring and diagnostics using a novel integrated model based on deep learning and multi-sensor feature fusion
Measurement ( IF 5.6 ) Pub Date : 2020-06-12 , DOI: 10.1016/j.measurement.2020.108086
Xingwei Xu , Zhengrui Tao , Weiwei Ming , Qinglong An , Ming Chen

Monitoring and diagnostics are vitally important in smart manufacturing systems since early detection can reduce downtime, protect environment, improve work efficiency, and save cost. The current work for monitoring and diagnostics mainly process the system condition data from multisensor by some mainstream machine learning and deep learning methods. However, these methods’ performance is limited by the following weaknesses: (1) The multi-sensor information are not well used, and its feature fusion is not considerd. (2) Current advanced methods, such as convolution neural network (CNN), long short-term memory neural network (LSTM), are still facing some problems due to their inherent structures. CNN does not consider the sequential and temporal dependency; LSTM does not consider spatial correlation. Thus, a novel integrated model based on deep learning and multi-sensor feature fusion is proposed. The developed parallel convolutional neural network (PCNN) in the integrated model can achieve multisensory feature fusion to overcome the first weakness. The integrated CNN, deep residual networks (DRN), LSTM, can solve the second point. Specifically, the signals collected from multiple sensors are turned into multi-channel images, and the PCNN is designed to extract and fused the features of the converted images. Then, DRN and LSTM are developed to accept the extracted high-dimensional features by the designed CNN and generate the prediction results by fully connected neural networks. Two experiments, including cutting tool monitoring and bearing fault diagnosis, are conducted to validate the superiority and robustness of the proposed method. Compared with the state-of-the-art algorithms, the results show that the proposed model is more robust and accurate.



中文翻译:

使用基于深度学习和多传感器特征融合的新型集成模型进行智能监控和诊断

监视和诊断在智能制造系统中至关重要,因为及早发现可以减少停机时间,保护环境,提高工作效率并节省成本。当前的监视和诊断工作主要通过一些主流的机器学习和深度学习方法来处理来自多传感器的系统状态数据。但是,这些方法的性能受到以下缺点的限制:(1)多传感器信息使用不充分,没有考虑其特征融合。(2)卷积神经网络(CNN),长短期记忆神经网络(LSTM)等当前的先进方法由于其固有的结构而仍然面临一些问题。CNN不考虑顺序和时间依赖性;LSTM不考虑空间相关性。从而,提出了一种基于深度学习和多传感器特征融合的集成模型。在集成模型中开发的并行卷积神经网络(PCNN)可以实现多传感器特征融合,从而克服第一个缺点。集成的CNN,深残留网络(DRN),LSTM可以解决第二点。具体而言,将从多个传感器收集的信号转换成多通道图像,并且PCNN旨在提取和融合转换后图像的特征。然后,开发DRN和LSTM以接受设计的CNN提取的高维特征,并通过完全连接的神经网络生成预测结果。进行了两个实验,包括刀具监控和轴承故障诊断,以验证该方法的优越性和鲁棒性。

更新日期:2020-06-12
down
wechat
bug