In practical engineering applications, the vibration signals collected by sensors often contain outliers, resulting in the separation accuracy of source signals from the observed signals being seriously affected. The mixing matrix estimation is crucial to the underdetermined blind source separation (UBSS), determining the accuracy level of the source signals recovery. Therefore, a two-stage clustering method is proposed by combining hierarchical clustering and K-means to improve the reliability of the estimated mixing matrix in this paper. The proposed method is used to solve the two major problems in the K-means algorithm: the random selection of initial cluster centers and the sensitivity of the algorithm to outliers. Firstly, the observed signals are clustered by hierarchical clustering to get the cluster centers. Secondly, the cosine distance is used to eliminate the outliers deviating from cluster centers. Then, the initial cluster centers are obtained by calculating the mean value of each remaining cluster. Finally, the mixing matrix is estimated with the improved K-means, and the sources are recovered using the least square method. Simulation and the reciprocating compressor fault experiments demonstrate the effectiveness of the proposed method.

Yanyang Li, Fuli Luo, Runxin Xu, Songfang Huang, Fei Huang, Liwei Wang. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2022.

Due to the non-linear and non-smooth characteristics of reciprocating compressor vibration signals and the influence of environmental noise, the current online feature extraction methods are not suitable for direct fault diagnosis of reciprocating compressors. Hence, there is an urgent need to find out the characteristic parameters with low noise sensitivity for reciprocating compressor fault monitoring. In this paper, multi-scale fuzzy entropy is used to analyze the vibration signal at the main shaft of the reciprocating compressor motor, and the multi-scale fuzzy entropy is standardized to determine the operating status according to its change range. The field application of the compressor enables timely detection of wear faults at the main bearing of the motor, which is more stable and efficient than the traditional feature extraction method and can detect early faults.

Pre-trained language models (PLMs) like BERT have made significant progress in various downstream NLP tasks. However, by asking models to do cloze-style tests, recent work finds that PLMs are short in acquiring knowledge from unstructured text. To understand the internal behaviour of PLMs in retrieving knowledge, we first define knowledge-baring (K-B) tokens and knowledge-free (K-F) tokens for unstructured text and ask professional annotators to label some samples manually. Then, we find that PLMs are more likely to give wrong predictions on K-B tokens and attend less attention to those tokens inside the self-attention module. Based on these observations, we develop two solutions to help the model learn more knowledge from unstructured text in a fully self-supervised manner. Experiments on knowledge-intensive tasks show the effectiveness of the proposed methods. To our best knowledge, we are the first to explore fully self-supervised learning of knowledge in continual pre-training.