China 
Africa 
Explore chapters and articles related to this topic
A combined kernel-PCA with clustering analysis for bridge damage detection under changing environmental conditions
Published in Airong Chen, Xin Ruan, Dan M. Frangopol, Life-Cycle Civil Engineering: Innovation, Theory and Practice, 2021
The proposed method is able to predict the behavior of the dynamic parameter variations under undamaged and damaged conditions underhanging environmental conditions. In addition, the misfit both for the validation data in undamaged condition as for the training data are lower. This is because the 3500 training data points entered in the model cover a wider range under normal environmental factors. Therefore, the normal variability of modal parameters is better predicted using the kernel-PCA and also increases the damage detection capability of the model. Likewise, when a new artificial damage is imposed to the bridge, the prediction error grows very significantly, in this way, these undesirable effects can be clearly and easily detected, either visually or with a novelty detection algorithm. In this study, a novelty detection technique based on machine learning algorithms was utilized as presented in the next section.
An Introduction to Kernel-Based Learning Algorithms
Published in Yu Hen Hu, Jenq-Neng Hwang, Handbook of Neural Network Signal Processing, 2018
Klaus-Robert Müller, Sebastian Mika, Koji Tsuda, Koji Schölkopf
In unsupervised learning, only the data x1, …, xn ∈ ℝN are given, i.e., the labels are missing. Standard questions of unsupervised learning are clustering, density estimation, and data description [10, 29]. As already outlined, the kernel trick can be applied not only in supervised learning scenarios, but also in cases of unsupervised learning, given that the base algorithm can be written in terms of scalar products. The following section reviews one of the most common statistical data analysis algorithm, PCA, and explains its “kernelized” variant: kernel PCA [119]. Subsequently, single-class classification is explained. Here, the support of a given data set is being estimated [93, 114, 120, 134]. Recently, single-class SVMs have frequently been used in outlier or novelty detection applications.
Precast segmental bridge construction in seismic zones
Published in Fabio Biondini, Dan M. Frangopol, Bridge Maintenance, Safety, Management, Resilience and Sustainability, 2012
Fabio Biondini, Dan M. Frangopol
The outlier analysis performed above illustrates the fact that novelty detection is feasible without prior manipulation of data to remove the influence of environmental and operational conditions. In this case, the daily variations in frequency caused by environmental and operational conditions, illustrated in Figure 3, have been incorporated into the definition of the normal condition of the structural response. However, although this analysis is able to detect anomalies, as evidenced by the fact that a corrupted signal is detectable, the incorporation of the daily variations into the normal condition may render some potential performance anomalies undetectable; a performance anomaly that produces a variation in frequency smaller than that of the daily variation caused by environmental/operational conditions will certainly not be detected using the above approach. If one is interested in such performance anomalies, steps must be taken to account for environmental and operationally induced variation before novelty detection is attempted.
Virtual metrology as an approach for product quality estimation in Industry 4.0: a systematic review and integrative conceptual framework
Published in International Journal of Production Research, 2022
Paul-Arthur Dreyfus, Foivos Psarommatis, Gokan May, Dimitris Kiritsis
Manual outlier detection is the simplest and most used approach. Features can be plotted independently or not (Chen et al. 2020) for visualising potential outliers. This method is purely offline, which is an important limitation for VM implementation. Statistics-based methods are also highly popular as they are the simplest online approach. A specific clustering method named ART2 was developed for VM outlier detection (Huang et al. 2014). ART2 is widely used for VM as it is included in the automatic VM framework (Cheng et al. 2016). Lastly, a ML field called novelty detection proposes advanced solutions for detecting outliers. Some of those solutions have been tested successfully for VM (Chou, Wu, and Chen 2010; Kang, Kim, and Cho 2014; Kim et al. 2015). The main advantage of novelty detection is that it can deal with data streams. Indeed, whereas the other solutions assume that the distribution they learned is fixed, a data stream accepts that it will evolve with time, adding a new challenge – whether it is ‘an aberrant point, an outlier, or just the evolution of the normality.’ Detecting aberrant points is not performed simply to increase the accuracy but also – and especially – to protect the quality estimator from the time-variation of its environment. This is a major challenge that should be prioritised more.
Abnormal Usage Sequence Detection for Identification of User Needs via Recurrent Neural Network Semantic Variational Autoencoder
Published in International Journal of Human–Computer Interaction, 2020
Finally, we apply a novelty detection method to calculate the vector representation of the entire usage sequence. We utilize three novelty detection methods: LOF, FCM, and GMM. In LOF, the local deviation of the density of the usage sequence with respect to its neighbors is measured. It is a local detection method in that the anomaly score depends on the extent of the isolation of the object with respect to the surrounding neighborhood. By comparing the local density of a sample with the local densities of its neighbors, sequence data having a substantially lower density than their neighbors can be identified as constituting an abnormal usage sequence. We extract a top-n abnormal usage sequence using the lower density score of the usage sequence, as applied in previous studies (Algur & Bhat, 2016; Xu, Lei, & Zhou, 2018). In the FCM approach, the embedded sequences are clustered using FCM clustering, which allows one piece of data to belong to two or more clusters. Further, we calculate the sum of the membership grades of the embedded sequence dataset for all clusters and declare the cluster with the smallest sum as an abnormal sequence. Similar to the clustering step, GMM fits the usage sequences with a mixture of Gaussian distribution, and the data having low probability density function values are considered abnormal sequences. After detecting the abnormal sequences, we examine the sequences and, with the assistance of domain experts, derive the corresponding user implied needs.
Subspace Clustering for Situation Assessment in Aquatic Drones: A Sensitivity Analysis for State-Model Improvement
Published in Cybernetics and Systems, 2019
Alberto Castellini, Manuele Bicego, Domenico Bloisi, Jason Blum, Francesco Masillo, Sergio Peignier, Alessandro Farinelli
In this work, we focus on the problem of detecting, modeling and interpreting aquatic drone states from a data-driven point of view. We aim at using statistical learning methods to develop interpretable models of drone states from traces of sensor data acquired during water-monitoring missions. Unsupervised methods, such as clustering and time series segmentation, are ideal tools for detecting data patterns in the considered scenario since they optimize internal performance measures (Arbelaitz et al. 2013). Using these tools, similar observations are grouped together into clusters whose parameters represent the state models. Another advantage of such methods is that they avoid manual labeling of sensor traces that is often expensive, time consuming, and impracticable in some cases. Moreover, models generated by these methods are abstract descriptions of drone states that can be interpreted and validated by experts. Finally, being unsupervised, these techniques allow novelty detection.