Search

This article presents a fault diagnosis algorithm for rotating machinery based on the Wasserstein distance. Recently, the Wasserstein distance has been proposed as a new research direction to find better distribution mapping when compared with other popular statistical distances and divergences. In this work, first, frequency- and time-based features are extracted by vibration signals, and second, the Wasserstein distance is considered for the learning phase to discriminate the different machine operating conditions. Specifically, the 1-D Wasserstein distance is considered due to its low computational burden because it can be evaluated directly by the order statistics of the extracted features. Furthermore, a distance weighting stage based on neighborhood component features selection (NCFS) is exploited to achieve robust fault diagnosis at low signal-to-noise ratio (SNR) conditions and with high-dimensional features. In detail, the NCFS framework is here adapted to weight 1-D Wasserstein distances evaluated from time/frequency features. Experiments are conducted on two benchmark data sets to verify the effectiveness of the proposed fault diagnosis method at different SNR conditions. The comparison with state-of-the-art fault diagnosis algorithms shows promising results.

Detecting stress in computer users, while technically challenging, is of the utmost importance in the workplace, especially now that remote working scenarios are becoming ubiquitous. In this context, cost-effective, subject-independent systems are needed that can be embedded in consumer devices and classify users’ stress in a reliable and unobtrusive fashion. Leveraging keyboard and mouse dynamics is particularly appealing in this context as it exploits readily available sensors. However, available studies are mostly performed in laboratory conditions, and there is a lack of on-field investigations in closer-to-real-world settings. In this study, keyboard and mouse data from 62 volunteers were experimentally collected in-the-wild using a purpose-built Web application, designed to induce stress by asking each subject to perform 8 computer tasks under different stressful conditions. The application of Multiple Instance Learning (MIL) to Random Forest (RF) classification allowed the devised system to successfully distinguish 3 stress-level classes from keyboard (76% accuracy) and mouse (63% accuracy) data. Classifiers were further evaluated via confusion matrix, precision, recall, and F1-score.

This paper proposes a specific domotic sensor network to measure the well-being of elderly people in private home environments through Machine Learning (ML) algorithms trained with daily surveys. The tests have been conducted in 5 apartments lived by 8 older people where the non-obtrusive sensor network is installed. Two ML algorithms are compared, Random Forest (RF) and Regression Tree (RT), such that to verify whether the users’ well-being is encoded in behavioural patterns obtained from the domotic data. These data are used to measure users’ well-being and compared with three reference indices obtained through a daily survey: a physical (Phy), a mental (Mind) and a general health index (Avg). The extracted indices from the daily survey are used to train ML algorithms in the estimation of user’s well-being for users that live alone (single-resident) or with others (multi-resident). Single-house and multi-house procedures are tested, both to extract a user-specific behaviour, and assess whether the model is able to generalise across different users and environments. Results show that the RF algorithm provides better performance than the RT algorithm in predicting the level of well-being with a Mean Absolute Error in the multi-house procedure of 32%, 13% and 17% for the Avg, Mind and Phy indices, respectively.

Condition-based monitoring of rotating machines requires robust features for accurate fault diagnosis, which is indeed directly linked to the quality of the features extracted from the signals. This is especially true for vibration data, whose quasi-stationary nature implies that the quality of frequency domain extracted features depends on the signal-to-noise ratio (SNR) condition, operating condition variations, and data segmentation. This paper presents a novel statistical spectral analysis, which leads to highly robust fault diagnosis with poor SNR conditions, different time-window segmentation, and different operating conditions. The amplitudes of spectral contents of the quasi-stationary time vibration signals are sorted and transformed into statistical spectral images. The sort operation leads to the knowledge of the empirical cumulative distribution function (ECDF) of the amplitudes of each frequency band. The ECDF provides a robust statistical information of the distribution of the amplitude under different SNR and operating conditions. Statistical metrics have been adopted for fault classification, by using the ECDFs obtained from the spectral images as fault features. By applying simple statistical metrics, it is possible to achieve fault diagnosis without classifier training, saving both time and computational costs. The proposed algorithm has been tested using a vibration data benchmark: comparison with state-of-the-art fault diagnosis algorithms shows promising results.

This technical note solves the problem to detect and isolate faults in linear periodic discrete-time systems. A residual generator is designed through a suitable unknown input observer with the requirement that each residual should be sensitive only to one fault and, simultaneously, insensitive to the other faults that can affect the system, and with the additional requirement of ρ-detectability, that is the transition matrix of its relative error system has all eigenvalues smaller than ρ, in modulus, for a given positive p ≤ 1. The solution is based on geometric tools, and it makes use of the outer observable subspace notion.

In this paper the state estimation problem for linear discrete-time systems with non-Gaussian state and output noises is treated. In order to obtain a state optimal quadratic estimate with a lower computational effort and without loosing the stability, only the observable part of the second-order power system will be considered. The novelty of the proposed algorithm is to provide a method to compute, in a closed form, the rank of the observability matrix for the quadratic system. Considering a new augmented state-space built as the aggregate of the actual state vector and the observable components of the system squared state, and defining a new observation sequence composed of the original output measurements together with their square values, we will be in a condition to use Kalman filtering that, in this case, produces a suboptimal quadratic stable state estimate for the original system. The solution is given in closed form by a recursive algorithm.

This paper proposes an actuator fault-tolerant control scheme, composed of the usual modules performing detection, isolation, accommodation, designed for a class of nonlinear systems, and then applied to an underwater remotely operated vehicle (ROV) used for inspection purposes. Detection is in charge of a residual generation module, while a sliding-mode-based approach has been used both for ROV control and fault isolation, after the application of an input decoupling nonlinear state transformation to the ROV model. Finally, control reconfiguration is performed exploiting the inherent redundancy of actuators. An extensive simulation study has been also performed, supporting the effectiveness of the proposed approach.

Using the geometric approach, a solution is given to the problem of detecting faults in linear periodic discrete-time systems. The solution consists on developing an unknown input observer. The existing conditions are provided in terms of new geometric notion of outer observable subspace that is the dual notion of inner reachable subspace. Making use of the new notion, a residual generator is developed, such that the residual is made independent of unknown input disturbance, and its output-free response goes asymptotically to zero.