11 March 2022 in IEEE Transactions on Instrumentation and Measurement
Industrial robots are one of the most typical machines in smart manufacturing systems. Their joint bearing faults account for a significant portion of failures. Data-driven bearing fault diagnosis methods, especially deep learning methods, have become a research hotspot due to the development of the industrial Internet of Things and big data. However, the varying working conditions of industrial robots, such as the continuous changing of load and speed, challenge the existing data-driven methods. Although adversarial-based domain adaptive methods are promising for solving this problem, they still face an equilibrium issue in the model training process. Therefore, a novel deep perceptual adversarial domain adaptive (DPADA) method is proposed for fault diagnosis of industrial robot bearings under varying conditions in this article. Here, a novel perceptual loss is proposed to force the target domain and the source domain to have the same distribution, which helps to improve the stability of adversarial training. Correspondingly, a timestamp mapping-based vibration signal screening method is proposed to improve data preprocessing efficiency for fault diagnosis of industrial robots. Extensive experimental results show that the accuracy of DPADA is superior to convolutional neural network (CNN) and conditional domain-adversarial network (CDAN)-based methods. A comparison is further performed on transfer tasks in three classical transfer scenes of industrial robots.
30 May 2024 in Journal of Materials Science & Technology
1 December 2023 in Machine Learning: Science and Technology
Although Transformer-based methods have achieved remarkable performance in the field of long-term series forecasting, they can be computationally expensive and lack the ability to specifically model local features as CNNs. CNN-based methods, such as temporal convolutional network (TCN), utilize convolutional filters to capture local temporal features. However, the intermediate layers of TCN suffer from a limited effective receptive field, which can result in the loss of temporal relations during global feature extraction.To solve the above problems, we propose to combine local features and global correlations to capture the overall view of time series (e.g. fluctuations, trends). To fully exploit the underlying information in the time series, a multi-scale branch structure is adopted to model different potential patterns separately. Each pattern is extracted using a combination of interactive learning convolution and causal frequency enhancement to capture both local features and global correlations. Furthermore, our proposed method,multi-scale local-global feature learning network (MLGN), achieves a time and memory complexity of O(L) and consistently achieve state-of-the-art results on six benchmark datasets. In comparision with previous best method Fedformer, MLGN yields 12.98% and 11.38% relative improvements for multivariate and univariate time series, respectively. Our code and data are available on Github at https://github.com/Zero-coder/MLGN.
4 May 2026 in Expert Systems with Applications