José Valente de Oliveira

Supervised training algorithms for B-Spline neural networks and neuro-fuzzy systems

Complete supervised training algorithms for B-Spline neural networks and fuzzy rule-based systems are discussed. By introducing the relationships between B-Spline neural networks and Mamdani (satisfying certain assumptions) and Takagi-Kang-Sugeno fuzzy models, training algorithms developed initially for neural networks can be adapted to fuzzy systems. The standard training criterion is reformulated, by separating its linear and nonlinear parameters. By employing this reformulated criterion with the Levenberg-Marquardt algorithm, a new training method, offering a fast rate of convergence is obtained. It is also shown that the standard Error-Back Propagation algorithm, the most common training method for this class of systems, exhibits a very poor and unreliable performance.

Attribute clustering using rough set theory for feature selection in fault severity classification of rotating machinery

A novel algorithm is proposed for unsupervised feature selection.The algorithm efficacy is evaluated through the accuracy of several classifiers.Adequate attributes are effectively selected for several case studies.The proposal presents better results than other attribute clustering algorithms.The proposal provides similar results to supervised feature selection approaches. Features extracted from real world applications increase dramatically, while machine learning methods decrease their performance given the previous scenario, and feature reduction is required. Particularly, for fault diagnosis in rotating machinery, the number of extracted features are sizable in order to collect all the available information from several monitored signals. Several approaches lead to data reduction using supervised or unsupervised strategies, where the supervised ones are the most reliable and its main disadvantage is the beforehand knowledge of the fault condition. This work proposes a new unsupervised algorithm for feature selection based on attribute clustering and rough set theory. Rough set theory is used to compute similarities between features through the relative dependency. The clustering approach combines classification based on distance with clustering based on prototype to group similar features, without requiring the number of clusters as an input. Additionally, the algorithm has an evolving property that allows the dynamic adjustment of the cluster structure during the clustering process, even when a new set of attributes feeds the algorithm. That gives to the algorithm an incremental learning property, avoiding a retraining process. These properties define the main contribution and significance of the proposed algorithm. Two fault diagnosis problems of fault severity classification in gears and bearings are studied to test the algorithm. Classification results show that the proposed algorithm is able to select adequate features as accurate as other feature selection and reduction approaches.

A statistical comparison of neuroclassifiers and feature selection methods for gearbox fault diagnosis under realistic conditions

Gearboxes are crucial devices in rotating power transmission systems with applications in a variety of industries. Gearbox faults can cause catastrophic physical consequences, long equipment downtimes, and severe production costs. Several artificial neural networks, learning algorithms, and feature selection methods have been used in the diagnosis of the gearbox healthy state. Given a specific gearbox, this study investigates how these approaches compare with each other in terms of the typical fault classification accuracy but also in terms of the area under curve (AUC), where the curve refers to the precision-recall curve otherwise known as receiver operating characteristic (ROC) curve. In particular, the comparison aims at identifying whether there are statistically significant (dis)similarities among six feature selection methods, and seven pairs of neural nets with different learning rules. Genetic algorithm based, entropy based, linear discriminants, principal components, most neighbors first, and non-negative matrix factorization are the studied feature selection methods. Feed forward perceptrons, cascade forward, probabilistic nets, and radial basis function neural nets are evaluated. Six supervised and one unsupervised learning rules are considered. Both parametric and nonparametric statistical tests are employed. A ranking process is defined to elect the best approach, when available. An experimental setup was especially prepared to ensure operating conditions as realistic as possible.

Observer-Biased Fuzzy Clustering

As generated by clustering algorithms, clusterings (or partitions) are hypotheses on data explanation which are better evaluated by experts from the application domain. In general, clustering algorithms allow a limited usage of domain knowledge about the cluster formation process. In this study, we propose both a design technique and a new partitioning-based clustering algorithm which can be used to assist the data analyst while looking for a set of meaningful clusters, i.e., clusters that actually correspond to the underlying data structure. Following an observer metaphor according to which the perception of a group of objects depends on the observer position-the closer an observer is from an image more details (s)he perceives-we resort to shrinkage to incorporate a regularization term, accounting for the observation point, within the objective function of an otherwise unbiased clustering algorithm. This technique allows our resulting biased algorithm to generate a set of reasonable partitions, i.e., partitions validated by a given cluster validity index, corresponding to views of data with different levels of granularity (levels of detail) in different regions of the data space. For the illustration of the design technique, we adopted the fuzzy c-means (FCM) algorithm as the unbiased clustering algorithm and include a convergence theorem assuring that changing the point of observation in the corresponding biased algorithm FCM with focal point (FCMFP) does not jeopardize its convergence. Experimental studies on both synthetic and real data are included to illustrate the usefulness of the approach. In addition, and as a convenient side effect of using shrinkage, the experimental results suggest that our biased algorithm (FCMFP) not only seems to scale better than the successive runs of the unbiased one (FCM) but on the average, seems to produce clusters exhibiting higher validity index values as well. In addition, less sensitivity to initialization was observed for the biased algorithm when compared with the unbiased one.

Supervised training algorithms for B-spline neural networks and fuzzy systems

Complete supervised training algorithms for B-spline neural networks and fuzzy rule-based systems are discussed. By introducing the relationships between B-spline neural networks and Mamdani (satisfying certain assumptions) fuzzy model, training algorithms developed initially for neural networks can be adapted to fuzzy systems. The standard training criterion is reformulated, by separating the linear and nonlinear parameters. By employing this reformulated criterion with the Levenberg-Marquardt algorithm, a new training method, offering a fast rate of convergence is obtained. It is also shown that the standard error-back propagation algorithm, the most common training method for this class of systems, exhibits a very poor performance.

Fuzzy determination of informative frequency band for bearing fault detection

Detecting early faults in rolling element bearings is a crucial measure for the health maintenance of rotating machinery. As faulty features of bearings are usually demodulated into a high-frequency band, determining the informative frequency band (IFB) from the vibratory signal is a challenging task for weak fault detection. Existing approaches for IFB determination often divide the frequency spectrum of the signal into even partitions, one of which is regarded as the IFB by an individual selector. This work proposes a fuzzy technique to select the IFB with improvements in two aspects. On the one hand, an IFB-specific fuzzy clustering method is developed to segment the frequency spectrum into meaningful sub-bands. Considering the shortcomings of the individual selectors, on the other hand, three commonly-used selectors are combined using a fuzzy comprehensive evaluation method to guide the clustering. Among all the meaningful sub-bands, the one with the minimum comprehensive cost is determined as the IFB. The bearing faults, if any, can be detected from the demodulated envelope spectrum of the IFB. The proposed fuzzy technique was evaluated using both simulated and experimental data, and then compared with the state-of-the-art peer method. The results indicate that the proposed fuzzy technique is capable of generating a better IFB, and is suitable for detecting bearing faults.

Observer-biased bearing condition monitoring

Bearings are simultaneously a fundamental component and one of the principal causes of failure in rotary machinery. The work focuses on the employment of fuzzy clustering for bearing condition monitoring, i.e., fault detection and classification. The output of a clustering algorithm is a data partition (a set of clusters) which is merely a hypothesis on the structure of the data. This hypothesis requires validation by domain experts. In general, clustering algorithms allow a limited usage of domain knowledge on the cluster formation process. In this study, a novel method allowing for interactive clustering in bearing fault diagnosis is proposed. The method resorts to shrinkage to generalize an otherwise unbiased clustering algorithm into a biased one. In this way, the method provides a natural and intuitive way to control the cluster formation process, allowing for the employment of domain knowledge to guiding it. The domain expert can select a desirable level of granularity ranging from fault detection to classification of a variable number of faults and can select a specific region of the feature space for detailed analysis. Moreover, experimental results under realistic conditions show that the adopted algorithm outperforms the corresponding unbiased algorithm (fuzzy c-means) which is being widely used in this type of problems.

Automatic feature extraction of time-series applied to fault severity assessment of helical gearbox in stationary and non-stationary speed operation

Abstract Signals captured in rotating machines to obtain the status of their components can be considered as a source of massive information. In current methods based on artificial intelligence to fault severity assessment, features are first generated by advanced signal processing techniques. Then feature selection takes place, often requiring human expertise. This approach, besides time-consuming, is highly dependent on the machinery configuration as in general the results obtained for a mechanical system cannot be reused by other systems. Moreover, the information about time events is often lost along the process, preventing the discovery of faulty state patterns in machines operating under time-varying conditions. In this paper a novel method for automatic feature extraction and estimation of fault severity is proposed to overcome the drawbacks of classical techniques. The proposed method employs a Deep Convolutional Neural Network pre-trained by a Stacked Convolutional Autoencoder. The robustness and accuracy of this new method are validated using a dataset with different severity conditions on failure mode in a helical gearbox, working in both constant and variable speed of operation. The results show that the proposed unsupervised feature extraction method is effective for the estimation of fault severity in helical gearbox, and it has a consistently better performance in comparison with other reported feature extraction methods.

Particle Swarm Clustering in clustering ensembles

Graphical abstractDisplay Omitted HighlightsA new consensus function based on the Particle Swarm Clustering algorithm.An alignment-free efficient representation for both disjoint and overlapping partitions.Employment of evolutionary operators for ensemble pruning. A clustering ensemble combines in a consensus function the partitions generated by a set of independent base clusterers. In this study both the employment of particle swarm clustering (PSC) and ensemble pruning (i.e., selective reduction of base partitions) using evolutionary techniques in the design of the consensus function is investigated. In the proposed ensemble, PSC plays two roles. First, it is used as a base clusterer. Second, it is employed in the consensus function; arguably the most challenging element of the ensemble. The proposed consensus function exploits a representation for the base partitions that makes cluster alignment unnecessary, allows for the combination of partitions with different number of clusters, and supports both disjoint and overlapping (fuzzy, probabilistic, and possibilistic) partitions. Results on both synthetic and real-world data sets show that the proposed ensemble can produce statistically significant better partitions, in terms of the validity indices used, than the best base partition available in the ensemble. In general, a small number of selected base partitions (below 20% of the total) yields the best results. Moreover, results produced by the proposed ensemble compare favorably to those of state-of-the-art clustering algorithms, and specially to swarm based clustering ensemble algorithms.

A fuzzy transition based approach for fault severity prediction in helical gearboxes

Abstract Rotating machinery is an important device supporting manufacturing processes, and a wide research works are devoted to detecting and diagnosing faults in such machinery. Recently, prognosis and health management in rotating machinery have received high attention as a research area, and some advances in this field are focused on fault severity assessment and its prediction. This paper applies a fuzzy transition based model for predicting fault severity conditions in helical gears. The approach combines Mamdani models and hierarchical clustering to estimate the membership degrees to fault severity levels of samples extracted from historical vibration signals. These membership degrees are used to estimate the weighted fuzzy transitions for modelling the evolution along the fault severity states over time, according to certain degradation path. The obtained fuzzy model is able of predicting the one step-ahead membership degrees to the severity levels of the failure mode under study, by using the current and the previous membership degrees to the severity levels of two available successive input samples. This fuzzy predictive model was validated by using real data obtained from a test bed with different damages of tooth breaking in the helical gears. Results show adequate predictions for two scenarios of fault degradation paths.