Predictive maintenance for power electronic converters has emerged as a critical area of research and development. With the rapid advancements in deep-learning techniques, new possibilities have emerged for enhancing the performance and reliability of power converters. However, addressing challenges related to data resources, physical consistency, and generalizability has become crucial in achieving optimal strategies. This comprehensive review article presents an insightful overview of the recent advancements in the field of predictive maintenance for power converters. It explores three paradigms: model-based approaches, data-driven techniques, and the emerging concept of physics-informed machine learning (PIML). By leveraging the integration of physical knowledge into machine-learning architectures, PIML holds great promise for overcoming the aforementioned concerns. Drawing upon the current state-of-art, this review identifies common trends, practical challenges, and significant research opportunities in the domain of predictive maintenance for power converters. The analysis covers a broad spectrum of approaches used for parameter identification, feature engineering, fault detection, and remaining useful life estimation. This article not only provides a comprehensive survey of recent methodologies but also highlights future trends, serving as a resource for researchers and practitioners involved in the development of predictive maintenance strategies for power converters.
Photoacoustic (PA) signals are short ultrasound (US) pulses typically characterized by a single-cycle shape, often referred to as N-shape. The spectral content of such wideband signals ranges from a few hundred kilohertz to several tens of megahertz. Typical reception frequency responses of classical piezoelectric US imaging transducers, based on PZT technology, are not sufficiently broadband to fully preserve the entire information contained in PA signals, which are then filtered, thus limiting PA imaging performance. Capacitive micromachined ultrasonic transducers (CMUT) are rapidly emerging as a valid alternative to conventional PZT transducers in several medical ultrasound imaging applications. As compared to PZT transducers, CMUTs exhibit both higher sensitivity and significantly broader frequency response in reception, making their use attractive in PA imaging applications. This paper explores the advantages of the CMUT larger bandwidth in PA imaging by carrying out an experimental comparative study using various CMUT and PZT probes from different research laboratories and manufacturers. PA acquisitions are performed on a suture wire and on several home-made bimodal phantoms with both PZT and CMUT probes. Three criteria, based on the evaluation of pure receive impulse response, signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) respectively, have been used for a quantitative comparison of imaging results. The measured fractional bandwidths of the CMUT arrays are larger compared to PZT probes. Moreover, both SNR and CNR are enhanced by at least 6 dB with CMUT technology. This work highlights the potential of CMUT technology for PA imaging through qualitative and quantitative parameters.
Five to ten percent of school-aged children display dysgraphia, a neuro-motor disorder that causes difficulties in handwriting, which becomes a handicap in the daily life of these children. Yet, the diagnosis of dysgraphia remains tedious, subjective and dependent to the language besides stepping in late in the schooling. We propose a pre-diagnosis tool for dysgraphia using drawings called graphomotor tests. These tests are recorded using graphical tablets. We evaluate several machine-learning models and compare them to build this tool. A database comprising 305 children from the region of Grenoble, including 43 children with dysgraphia, has been established and diagnosed by specialists using the BHK test, which is the gold standard for the diagnosis of dysgraphia in France. We performed tests of classification by extracting, correcting and selecting features from the raw data collected with the tablets and achieved a maximum accuracy of 73% with cross-validation for three models. These promising results highlight the relevance of graphomotor tests to diagnose dysgraphia earlier and more broadly.
Finding a way to combine ultrasound and fluorescence optical imaging on an endorectal probe may improve early detection of prostate cancer. The ultrasound provides morphological information about the prostate, while the optical system detects and locates fluorophore-marked tumors. A tissue-mimicking phantom, which is representative of prostate tissues both on its optical (absorption and diffusion ) and its ultrasound properties, has been made by our team. A transrectal probe adapted to fluorescence diffuse optical tomography measurements was also developed. Measurements were taken on the prostate phantom with this probe based on a pulsed laser and a time-resolved detection system. A reconstruction algorithm was then used to help locate and quantify fluorescent inclusions of different concentrations at fixed depths.
We describe a novel method for localizing a fluorescent inclusion in a homogeneous turbid medium through the use of time-resolved techniques. Based on the calculation of the mean time of the fluorescence curves, the method does not require a priori knowledge of either the fluorescence lifetime or the mean time of the instrument response function since it adopts a differential processing approach. Theoretical expressions were validated and experiments for assessing the accuracy of localization were carried out on liquid optical phantoms with a small fluorescent inclusion. The illumination and detection optical fibers were immersed in the medium to achieve infinite medium geometry as required by the model used. The experimental setup consisted of a time-correlated single-photon counting system. Submillimeter accuracy was achieved for the localization of the inclusion.
Fluorescence enhanced diffuse optical tomography (fDOT) is envisioned to be useful to collect functional information from small animal models. For oncology applications, cancer-targeted fluorescent markers can be used as a surrogate of the cancer activity. We are developing a continuous wave fDOT bench intended to be integrated in systems dedicated to whole body small animal fluorescence analyses. The focus is currently put on the reconstruction of non immersed small animals imaged by a CCD camera. The reconstruction stage already corrects the tissue heterogeneity artifacts through the computation of an optical heterogeneity map. We will show how this formalism coupled with the determination of the animal boundaries performed by a laser scanner, can be used to manage non contact acquisitions. The time of reconstruction for a 10 × 9 laser source positions, 45 × 40 detector elements and 14 × 11 × 14 mesh voxels is typically 10 minutes on a 3GHz PCs corresponding to the acquisition time allowing the two tasks to be performed in parallel. The system is validated on an in vivo experiment performed on three healthy nude mice and a mouse bearing a lung tumor at 10, 12 and 14 days after implantation allowing the follow up of the disease. The 3D fluorescence reconstructions of this mouse are presented and the total fluorescence amounts are compared.