The paper proposes a free dataset, available at the following link1, named KIMORE, regarding different rehabilitation exercises collected by a RGB-D sensor. Three data inputs including RGB, Depth videos and skeleton joint positions were recorded during five physical exercises, specific for low back pain and accurately selected by physicians. For each exercise, the dataset also provides a set of features, specifically defined by the physicians, and relevant to describe its scope. These features, validated with respect to a stereophotogrammetric system, can be analyzed to compute a score for the subject’s performance. The dataset also contains an evaluation of the same performance provided by the clinicians, through a clinical questionnaire. The impact of KIMORE has been analyzed by comparing the output obtained by an example of rule and template-based approaches and the clinical score. The dataset presented is intended to be used as a benchmark for human movement assessment in a rehabilitation scenario in order to test the effectiveness and the reliability of different computational approaches. Unlike other existing datasets, the KIMORE merges a large heterogeneous population of 78 subjects, divided into 2 groups with 44 healthy subjects and 34 with motor dysfunctions. It provides the most clinically-relevant features and the clinical score for each exercise.

This paper deals with the problem of fault detection and diagnosis of induction motor based on motor current signature analysis. Principal component analysis is used to reduce the three-phase current space to a 2-D space. Kernel density estimation (KDE) is adopted to evaluate the probability density functions of each healthy and faulty motor, which can be used as features in order to identify each fault. Kullback-Leibler divergence is used as an index to identify the dissimilarity between two probability distributions, and it allows automatic fault identification. The aim is also to improve computational performance in order to apply online a monitoring system. KDE is improved by fast Gaussian transform and a points reduction procedure. Since these techniques achieve a remarkable computational cost reduction with respect to the standard KDE, the algorithm can be used online. Experiments are carried out using two alternate current motors: An asynchronous induction machine and a single-phase motor. The faults considered to test the developed algorithm are cracked rotor, out-of-tolerance geometry rotor, and backlash. Tests are carried out at different load and voltage levels to show the proposed method performance.

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.

Background: Brain-Computer Interface (BCI) is becoming more reliable, thanks to the advantages of Artificial Intelligence (AI). Recently, hybrid Deep Learning (hDL), which combines different DL algorithms, has gained momentum over the past five years. In this work, we proposed a review on hDL-based BCI starting from the seminal studies in 2015. Objectives: We have reviewed 47 papers that apply hDL to the BCI system published between 2015 and 2020 extracting trends and highlighting relevant aspects to the topic. Methods: We have queried four scientific search engines (Google Scholar, PubMed, IEEE Xplore and Elsevier Science Direct) and different data items were extracted from each paper such as the database used, kind of application, online/offline training, tasks used for the BCI, pre-processing methodology adopted, type of normalization used, which kind of features were extracted, type of DL architecture used, number of layers implemented and which optimization approach were used as well. All these items were then investigated one by one to uncover trends. Results: Our investigation reveals that Electroencephalography (EEG) has been the most used technique. Interestingly, despite the lower Signal-to-Noise Ratio (SNR) of the EEG data that makes pre-processing of that data mandatory, we have found that the pre-processing has only been used in 21.28% of the cases by showing that hDL seems to be able to overcome this intrinsic drawback of the EEG data. Temporal-features seem to be the most effective with 93.94% accuracy, while spatial-temporal features are the most used with 33.33% of the cases investigated. The most used architecture has been Convolutional Neural Network-Recurrent Neural Network CNN-RNN with 47% of the cases. Moreover, half of the studies have used a low number of layers to achieve a good compromise between the complexity of the network and computational efficiency. Significance: To give useful information to the scientific community, we make our summary table of hDL-based BCI papers available and invite the community to published work to contribute to it directly. We have indicated a list of open challenges, emphasizing the need to use neuroimaging techniques other than EEG, such as functional Near-Infrared Spectroscopy (fNIRS), deeper investigate the advantages and disadvantages of using pre-processing and the relationship with the accuracy obtained. To implement new combinations of architectures, such as RNN-based and Deep Belief Network DBN-based, it is necessary to better explore the frequency and temporal-frequency features of the data at hand.