A vast amount of text data is recorded in the forms of repair verbatim in railway maintenance sectors. Efficient text mining of such maintenance data plays an important role in detecting anomalies and improving fault diagnosis efficiency. However, unstructured verbatim, high-dimensional data, and imbalanced fault class distribution pose challenges for feature selections and fault diagnosis. We propose a bilevel feature extraction-based text mining that integrates features extracted at both syntax and semantic levels with the aim to improve the fault classification performance. We first perform an improved X 2 statistics-based feature selection at the syntax level to overcome the learning difficulty caused by an imbalanced data set. Then, we perform a prior latent Dirichlet allocation-based feature selection at the semantic level to reduce the data set into a low-dimensional topic space. Finally, we fuse fault features derived from both syntax and semantic levels via serial fusion. The proposed method uses fault features at different levels and enhances the precision of fault diagnosis for all fault classes, particularly minority ones. Its performance has been validated by using a railway maintenance data set collected from 2008 to 2014 by a railway corporation. It outperforms traditional approaches.
28 August 2020 in IEEE Transactions on Reliability
Degradation modeling is a critical and challenging problem as it serves as the basis for system prognostics and evolution mechanism analysis. In practice, multiple sensors are used to monitor the status of a system. Thus, multisensor data fusion techniques have been proposed to capture comprehensive information for prognostic modeling and analysis, which aims at developing a composite health index (HI) through the fusion of multiple sensor signals. In the literature, most existing methods use a linear data-fusion model for integration of multisensor data to construct the HI, which is insufficient to model nonlinear relations between sensing signals and HI in a complicated system. This article proposes a novel data fusion method based on deep learning for HI construction for prognostic analysis. A pair of adversarial networks is proposed to enable the training procedure of neural networks. To guarantee the stability of the algorithm, we propose a root mean square propagation (i.e., RMSprop)-based sampling algorithm to estimate model parameters. A set of simulation studies and a case study on a set of degradation signals of aircraft engines are conducted. The results demonstrate that the proposed method has a significant improvement on remaining useful life prediction compared to existing data fusion methods.
The complex infrastructure and harsh conditions of heavy-haul railways result in frequently and rapidly deteriorating rail surface defects. Accurate detection of these defects is essential. To solve the problem of low detection precision caused by complex background interference, significant variation in defect scales, and similar features between different types of defects, a high-precision rail surface defect detection method for heavy-haul railways based on an improved YOLOv8 is proposed. First, the original grayscale images are preprocessed to reduce background noise interference. Then, the designed scale variation adaptation module is introduced to mitigate the impact of significant scale variations in the target defects. Additionally, a bidirectional feature pyramid network is incorporated to enhance feature fusion effectiveness. Furthermore, a small target detection head is introduced to improve the detection performance of small-scale defects. Lastly, network performance is optimized by replacing the original loss function with wise-intersection over union. Experimental results demonstrate that the improved model achieves a mean average precision at 50% intersection over union (mAP50) value of 0.975, representing a 4.13% improvement in precision and a 7.75% increase in recall compared to the baseline model. The improved model effectively detects typical defects such as spalling, shelling, and corrugation, providing valuable technical support for field maintenance personnel.
With advancements in sensor technology, high dimensional signals such as functional curves and images are typically collected from multiple sensors to characterize the degradation of a system. Data fusion methods are employed to integrate multisensor signals generated from the system into a scalar health index (HI) to understand the degradation status of the system. This paper develops sparse group LASSO-principal component analysis (SGL-PCA), a method that constructs HIs for image and profile data. First, we remove the smooth background from each sensor signal. Then, we solve the degradation patterns and the degradation paths through a rank-one matrix approximation problem, with the consideration of the sparsity of the measurements related to the degradation process and the monotonicity of the degradation paths. Results from a simulation study and a case study illustrate that the HI constructed by the proposed method outperforms the benchmark methods in identifying the measurements subject to the degradation process and predicting the remaining useful life of the system.
24 August 2021 in IEEE Transactions on Automation Science and Engineering
Heterogeneous but complementary sources of data provide an unprecedented opportunity for developing accurate statistical models of systems. Although the existing methods have shown promising results, they are mostly applicable to situations where the system output is measured in its complete form. In reality, however, it may not be feasible to obtain the complete output measurement of a system, which results in observations that contain missing values. This article introduces a general framework that integrates tensor regression with tensor completion and proposes an efficient optimization framework that alternates between two steps for parameter estimation. Through multiple simulations and a case study, we evaluate the performance of the proposed method. The results indicate the superiority of the proposed method in comparison to a benchmark.