Fabrice Morganti

Time series classification (TSC) is a fundamental problem in supervised learning, with applications in domains ranging from healthcare and finance to industrial monitoring. Recent advances based on random convolution kernels, such as ROCKET, MiniRocket, MultiRocket, and HYDRA, have demonstrated that large sets of randomly generated convolutional features combined with simple linear classifiers can achieve state-of-the-art accuracy at very low computational cost. These methods typically rely on fixed input representations and a small number of pooling operators, most commonly the proportion of positive values (PPV), to extract features from activation maps. However, the optimal choice of pooling operator and representation can vary substantially across datasets. In this work, we first provide an empirical analysis on the UCR archive showing that using a fixed pooling operator does not generally yield optimal performance, and that the best choice depends on the data. Motivated by this observation, we propose SelF-Rocket (Selected Features Rocket), a novel extension of MiniRocket that dynamically attempts to select the most appropriate combination of input representation and pooling operator during training. SelF-Rocket incorporates a feature selection module that adapts to each dataset, allowing it to select the most appropriate pooling operator and representation. Extensive experiments on the UCR benchmark demonstrate that SelF-Rocket achieves competitive or superior accuracy compared with existing random convolution kernel methods.

The increase of the operating temperature of electrical machines is a major challenge in the context of the large-scale use of electricity in many industrial or individual applications such as mobility. Indeed, working at higher temperatures makes it possible to place electrical actuators in critical areas – such as aircraft engines, for example – and/or to significantly improve the power-to-weight ratio of the machine. For several applications (deep pumping, ventilation, etc.), the use of the induction machine is still preferred due to cost, simplicity and robustness essentially. The objective of this study is to estimate the possibilities of making a high-temperature squirrel cage induction machine. A significant increase in the operating temperature of a machine would indirectly allow us to envisage an increase in the current density in the active conductors and thus a significant improvement in the power-to-weight ratio. However, this is at the sacrifice of efficiency. The aim of this study is to evaluate this decrease in efficiency correlated with the increase in the temperature to find the best compromise according to the relevant applications. An analytic sizing model has been proposed for the prediction of losses, which will be coupled with a thermal model to predict the temperature in different parts of the machine. This study presents essential information on the influence of temperature on the parameters important for the performance of an induction machine.

This paper presents a novel method to determine the electromagnetic torque of grid-powered induction machines (IM). The torque determination constitutes a key point for the energy diagnosis and for choosing the well adapted motor in industrial processes. The proposed method is non-invasive as it is based on the measurement of the current and the magnetic external flux around induction machines. The torque estimation protocol is explained and experimental validations are made with an 11 kW IM. The impact of supply voltage variations on the accuracy of the proposed method is also analyzed.

This study presents an analytical and experimental method for detecting internal turn-to-turn short circuits both in the rotor or in the stator in salient pole synchronous generator machines (SSGM) connected to local power grid. Two main measurements are proposed to confirm or discard these electric faults: external magnetic field and external housing vibrations (EHV). In the event of a short-circuit winding fault, a correlation between external magnetic field and EHV is done to reinforce the diagnosis. Furthermore, the electric power injection dependence is analyzed and a noninvasive technique is proposed to determine the state of the SSGM whatever its load. Thus, it is shown that only noninvasive and online equipment can be used to diagnose SSGM. A diagnostic tool has been developed to show the viability of an operative wireless monitoring system on a 76-MW SSGM in an industrial environment.

Purpose: This paper aims to study the influence of supply voltage variations on the external magnetic field emitted by grid-powered induction machines (IMs). Design/methodology/approach: Two models are developed in the paper to analyse, for different supply voltage values, the influence of the variations of the magnetizing voltage for which there is a link with the tangential component of the external flux. The first is an analytical model based on the IM single-phase-equivalent circuit with variable magnetizing reactance to take into account the saturation of the magnetic circuit. The second is a numerical finite element simulation to model the same phenomenon. Results of both models are analysed with experimental measures of the external flux. Findings: The study shows that the amplitude of the external field strongly depends on supply voltage values. Research limitations/implications: The investigation is mainly focused on the tangential component of the external magnetic field which is of high importance concerning the applicability of non-invasive methods of diagnosis, as electromagnetic torque estimation developed by the authors or internal fault determination. Originality/value: The originality of the paper concerns the characterization of the external flux with the supply voltage for IMs. It is shown that the magnetic circuit radiates external flux differently with the load and with the supply voltage.

During the manufacturing process multi-strand conductors are subject to compressive force and rotation moments. The current distribution in the multi-strand conductors is not uniform and is controlled by the transverse resistivity. This is mainly determined by the contact resistance at the strand crossovers and inter-strand contact resistance. The surface layer properties, and in particular the crystalline structure and degree of oxidation, are key parameters in determining the transverse resistivity. The experimental set-ups made it possible to find the dependence of contact resistivity as a function of continuous working stresses and cable design. A study based on measurements and numerical simulation is made to identify the contact resistivity functions.

This paper presents a non invasive method for estimating the energy efficiency of induction motors used in industrial applications. This method is innovative because it is only based on the measurement of the external field emitted by the motor. The paper describes the sensors used, how they should be placed around the machine in order to decouple the external field components generated by both the air gap flux and the winding end-windings. The study emphasizes the influence of the eddy currents flowing in the yoke frame on the sensor position. A method to estimate the torque from the external field use is proposed. The measurements are transmitted by a wireless module (Zig-Bee) and they are centralized and stored on a PC computer.