SVM practical industrial application for mechanical faults diagnostic

A large percentage of the total induction motor failures are due to mechanical faults. It is well known that, machine’s vibration is the best indicator of its overall mechanical condition, and an earliest indicator of arising defects. Support vector machines (SVM) is also well known as intelligent classifier with strong generalization ability. In this paper, both, machine‘s vibrations and SVM are used together for a new intelligent mechanical fault diagnostic method. Using only one vibration sensor and only four SVM’s it was achieved improved results over the available approaches for this purpose in the literature. Therefore, this method becomes more attractive for on line monitoring without maintenance specialist intervention. Vibration signals turns out to occur in different directions (axial, horizontal or vertical) depending on the type of the fault. Thus, to diagnose mechanical faults it is necessary to read signals at various positions or use more them one accelerometer. From this work we also determined the best position for signals acquisition, which is very important information for the maintenance task.     

Controlling the parallel layer perceptron complexity using a multiobjective learning algorithm

This paper deals with the parallel layer perceptron (PLP) complexity control, bias and variance dilemma, using a multiobjective (MOBJ) training algorithm. To control the bias and variance the training process is rewritten as a bi-objective problem, considering the minimization of both training error and norm of the weight vector, which is a measure of the network complexity. This method is applied to regression and classification problems and compared with several other training procedures and topologies. The results show that the PLP MOBJ training algorithm presents good generalization results, outperforming traditional methods in the tested examples.     

A review of machine learning approaches to Spam filtering

In this paper, we present a comprehensive review of recent developments in the application of machine learning algorithms to Spam filtering, focusing on both textual- and image-based approaches. Instead of considering Spam filtering as a standard classification problem, we highlight the importance of considering specific characteristics of the problem, especially concept drift, in designing new filters. Two particularly important aspects not widely recognized in the literature are discussed: the difficulties in updating a classifier based on the bag-of-words representation and a major difference between two early naive Bayes models. Overall, we conclude that while important advancements have been made in the last years, several aspects remain to be explored, especially under more realistic evaluation settings.     

Incipient fault detection in induction machine stator-winding using a fuzzy-Bayesian change point detection approach

In this paper the incipient fault detection problem in induction machine stator-winding is considered. The problem is solved using a new technique of change point detection in time series, based on a two-step formulation. The first step consists of a fuzzy clustering to transform the initial data, with arbitrary distribution, into a new one that can be approximated by a beta distribution. The fuzzy cluster centers are determined by using a Kohonen neural network. The second step consists in using the Metropolis–Hastings algorithm for performing the change point detection in the transformed time series generated by the first step with that known distribution. The incipient faults are detected as long as they characterize change points in such transformed time series. The main contribution of the proposed approach is the enhanced resilience of the new failure detection procedure against false alarms, combined with a good sensitivity that allows the detection of rather small fault signals. Simulation and practical results are presented to illustrate the proposed methodology.     

Fault induction dynamic model,suitable for computer simulation: Simulation results and experimental validation

The study of induction motor behavior under not normal conditions and the ability to detect and predict these conditions has been an area of increasing interest. Early detection and diagnosis of incipient faults are desirable for interactive evaluation over the running condition, product quality guarantee, and improved operational efficiency of induction motors. The main difficulty in this task is the lack of accurate analytical models to describe a faulty motor. This paper proposes a dynamic model to analyze electrical and mechanical faults in induction machines and includes net asymmetries and load conditions. The model permits to analyze the interactions between different faults in order to detect possible false alarms. Simulations and experimental results were performed to confirm the validity of the model.     

Immune inspired Fault Detection and Diagnosis: A fuzzy-based approach of the negative selection algorithm and participatory clustering

This paper describes an immune-inspired system based on an alternate theory about the self–nonself distinction theory, which defines the negative selection process as a mechanism of a fuzzy system based on the affinity between antigen and T-cells. This theory may provide a decision making tool which improves the generation of detectors or even define new data monitoring in order to detect an extreme variation of the system behavior, which means anomalies occurrences. Through these algorithms, tests are performed to detect faults of a DC motor. Upon detection of faults, a participatory clustering algorithm is used to classify these faults and tested to obtain the best set of parameters to achieve the most accurate clustering for these tests in the application being discussed in the article.     

A recursive algorithm for estimating multiple models continuous transfer function with non-uniform sampling

A multiple model recursive least squares algorithm combined with a first-order low-pass filter transformation method, named λ-transform, is proposed for the simultaneous identification of multiple model orders continuous transfer functions from non-uniformly sampled input–output data. The λ-transformation is shown to be equivalent to a canonical transformation between discrete z-domain and δ-domain using the negative value of the λ-transform filter time-constant instead of the sampling interval parameter. The proposed algorithm deals with oversampling, sampling jitter or non-uniform sample intervals without the need for extra digital anti-aliasing pre-filtering, downsampling or interpolation algorithms, producing multiple models with a cost function that facilitates automatic selection of best-fitted models. Besides, measurement noise is noted as beneficial, bringing up an inherent bias toward low-order models. Simulated examples and a drum-boiler level experimental result exhibiting non-minimum phase behaviour illustrate the application of the proposed method.     

Sliding mode observer for on-line broken rotor bar detection

The major task for any monitoring system is to detect upcoming faults as early as possible. Rotor failures are responsible for a large percentage of total induction motor failures. Thus, a new nonintrusive and in-service approach has been proposed in this paper to detect one broken rotor bar in induction motor using only input quantities information. The method is not affected by the type of load and other asymmetries and it is capable of providing reliable fault diagnose without maintenance specialist intervention. The proposed sensorless technique is based on two real time state space discrete models that are used to estimate the flux of both the stator and the rotor and analyzes the differences obtained in torque when the two models are employed. One of the observers was designed for rotor resistance disturbance rejection. Firstly, simulation results have been conducted and sensitivity and robustness also have been checked for the proposed method. Secondly, an experimental setup has been constructed to implement the new technique in on-line model. The results obtained validate the proposed model and show the reliability of the new method.     

Thermal modeling of power transformers using evolving fuzzy systems

Thermal models for distribution transformers with core immersed in oil are of utmost importance for transformers lifetime study. The hot spot temperature determines the degradation speed of the insulating paper. High temperatures cause loss of mechanical stiffness, generating failures. Since the paper is the most fragile component of the transformer, its degradation determines the lifetime limits. Thus, good thermal models are needed to generate reliable data for lifetime forecasting methodologies. It is also desired that thermal models are able to adapt to cope with changes in the transformer behavior due to structural changes, maintenance and so on. In this work we apply an evolving fuzzy model to build adaptive thermal models of distribution transformers. The model used is able to adapt its parameters and also its structure based on a stream of data. The proposed model is evaluated using actual data from an experimental transformer. The results suggest that evolving fuzzy models are a promising approach for adaptive thermal modeling of distribution transformers.