The fault diagnosis of rotating electrical machines has received an intense amount of research interest during the last 30 years. Reducing maintenance costs and preventing unscheduled downtimes, which result in losses of production and financial incomes, are the priorities of electrical drives manufacturers and operators. In fact, both correct diagnosis and early detection of incipient faults lead to fast unscheduled maintenance and short downtime for the process under consideration. They also prevent the harmful and sometimes devastating consequences of faults and failures. This topic has become far more attractive and critical as the population of electric machines has greatly increased in recent years. The total number of operating electrical machines in the world was around 16.1 billion in 2011, with a growth rate of about 50% in the last five years [1].

There has been an increasing interest in the design of intelligent diagnostic systems for industrial applications. The key requirement in the design of practical diagnostic systems is the ability for decision making in high-dimensional feature spaces, where the prior knowledge about the system states in terms of labels is very limited. Moreover, the problem of diagnosing simultaneous defects is rarely addressed on real industrial applications. This work aims to develop a semi-supervised deep-learning scheme for diagnosing multiple defects including simultaneous ones in a gearbox directly connected to an induction machine shaft. This scheme consists of two main modules: information fusion and decision making. The former integrates captured multiple sensory streams into a very high dimensional feature space. The latter uses a semi-supervised deep learning procedure to minimize the human interaction during the training and maximize the diagnostic efficiency. This scheme facilitates learning and diagnosing defects under harshest conditions (i) where only a few number of labeled samples are collected together with a large number of unlabeled samples, and (ii) in a very high-dimensional feature space. Several state-of-the-art semi-supervised and supervised learners have also been included in the scheme, enabling a comparative experiment for diagnosing simultaneous defects.

Monitoring of induction motor faults based on stray flux measurement has been investigated by many researchers due to its potential benefits in cost and simplicity. Although it was shown that flux based monitoring can provide sensitive fault detection comparable to that of motor current signature analysis (MCSA), the lack of “remote” monitoring capability has limited its practical use. The performance and reliability of stray flux-based detection of induction motor rotor cage faults is evaluated in this paper. It is shown for the first time in this work that spectrum analysis of the radial stray flux can provide reliable detection of rotor faults immune to the influence of rotor axial air ducts, which is the most common cause of false rotor fault alarms. The reliability and sensitivity of stray flux based rotor fault detection is demonstrated through experimental testing on laboratory and 6.6 kV field motors.

This article presents a method based on vibration signals to diagnose rotor mixed eccentricity faults in three-phase squirrel cage induction motors (SCIMs). Rotor eccentricity is a typical fault in SCIMs that significantly affects the performance of electrical machines and results in unwanted vibrations. The time synchronous averaging (TSA) signal, which consists of the harmonics components generated by the fault, is obtained via the characteristic frequency of the eccentricity fault. TSA is decomposed into two signals: TSA-regular and TSA-difference, which are respectively similar to approximate and details of the wavelet transform. Based on the fault index, which is the maximum FFT value of the TSA-difference signal, the healthy and faulty motors can be distinguished. The proposed approach is evaluated on an experimental test rig equipped with a 1.5 kW SCIM supplied directly from the power grid under various load conditions. The results illustrate the effectiveness of the proposed approach in detecting the rotor mixed eccentricity faults.

In this paper, an approach for tooth localized fault detection in the sun gear of a planetary gear, using the measured mechanical torque and the stator current of a wound rotor induction generator (WRIG), is proposed. A theoretical background is developed to demonstrate that the faulty sun gear produces periodic fault signatures in the mechanical torque and consequently fault-related frequencies in the stator current of the WRIG. The employed technique needs knowledge of mechanical system torsional natural frequencies, which aid in identifying the most sensitive fault characteristic harmonics in both mechanical torque and stator current spectra. It is illustrated by simulation and experimental results that the amplitude of these harmonics can be used as a fault index for diagnosis of tooth localized defect in planetary gears. A test rig based on a 5.5 kW WRIG, shaft connected to the sun gear of a planetary gear through a high-speed shaft and couplings, was utilized for validation of the proposed technique.

The aim of this paper is the monitoring of a gearbox by using the stator current signature analysis in the driving induction machines. The detection of mechanical faults using the vibration signal analysis has been well studied before. Recently, some works have been performed in order to observe the vibration components of the mechanical part in the stator current spectrum for an induction machine but without any theoretical detail. This paper proposes a theoretical framework based on torque oscillations due to the characteristic torsional vibrations in a gearbox and the source of mesh and rotating frequency components in the stator current is presented. In order to verify the theoretical development, a test-bed based on a 5.5 kW three-phase squirrel-cage machine connected to a gearbox has been used.