Feature Network Research Articles

Automated diagnosis of rolling element bearing defects using time-domain features and neural networks

Rolling element bearings are essential components in most mining equipment and mineral processing plants. Condition monitoring procedures need to be developed to prevent unexpected failure of these elements during operation to avoid costly consequences. Procedures for automated defect detection and diagnosis of bearings from vibration signals collected using accelerometer are presented in this article. Time-domain signals are processed to obtain the negative log-likelihood values and the root mean square (RMS) value, which are used as input features for an artificial neural network (ANN) classifier. Effectiveness of the proposed method is illustrated using the bearing vibration data obtained experimentally, and the results showed 100% accuracy in defect detection and diagnosis. The proposed procedures do not require preprocessing of the signal before feature extraction. The algorithm uses fewer input features, resulting in faster training. The results demonstrate the potential of negative log-likelihood values and neural networks for on-line defect detection and diagnosis of bearings.

A Novel Chaotic Neural Network With the Ability to Characterize Local Features and Its Application

To provide an ability to characterize local features for the chaotic neural network (CNN), Gauss wavelet is used for the self-feedback of the CNN with the dilation parameter acting as the bifurcation parameter. The exponentially decaying dilation parameter and the chaotically varying translation parameter not only govern the wavelet self-feedback transform but also enable the CNN to generate complex dynamics behavior preventing the network from being trapped in the local minima. Analysis of the energy function of the CNN indicates that the local characterization ability of the proposed CNN is effectively provided by the wavelet self-feedback in the manner of inverse wavelet transform and that the proposed CNN can achieve asymptotical stability. The experimental results on traveling salesman problem (TSP) suggest that the proposed CNN has a higher average success rate for obtaining globally optimal or near-optimal solutions.

An Artificial-Neural-Network-Based Integrated Regional Model for Rain Retrieval Over Land and Ocean

<para xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> An integrated regional model is proposed for rain-rate retrievals over land/ocean from the brightness temperature (Tb) values of the Tropical Rainfall Measuring Mission (TRMM) Microwave Imager (TMI). The polarization-corrected temperature <formula formulatype="inline"><tex>$(\hbox{PCT}_{85})$</tex></formula> calculated from the 85.5-GHz channels is also considered as one of the inputs along with the nine channel Tb values. This model is applicable over the region between <formula formulatype="inline"><tex>$0^{\circ}{-}30^{\circ}\ \hbox{N}$</tex> </formula> and <formula formulatype="inline"><tex>$65^{\circ}{-}100^{\circ}\ \hbox{E}$</tex></formula>. For this purpose, an artificial neural network is utilized. The collocated precipitation radar (PR) near-surface rain rates as given by a 2A25 data product is considered as a target value. The methodology consists of the separation of land and ocean pixels, the separation of stratiform and convective pixels over land/ocean, and the selection of important features (inputs) for the multilayer perceptron network by the feature selection technique for each group. For the separation of land/ocean pixels, the Tb values of the 10.65-GHz vertical channel are utilized. The <formula formulatype="inline"> <tex>$\hbox{PCT}_{85}$</tex></formula> values are utilized to separate the stratiform and convective pixels both over land and ocean. The rain retrieval from the developed model is validated with TRMM PR. Overall result shows the better agreement of the model-retrieved rain rate with the PR observation compared to the TMI (2A12) rain rate particularly over land. The rain retrieved from the developed model is further validated with Doppler weather radar. A reasonably good agreement is observed between these two estimations. </para>

Breast Mass Lesions: Computer-aided Diagnosis Models with Mammographic and Sonographic Descriptors

To retrospectively develop and evaluate computer-aided diagnosis (CAD) models that include both mammographic and sonographic descriptors. Institutional review board approval was obtained for this HIPAA-compliant study. A waiver of informed consent was obtained. Mammographic and sonographic examinations were performed in 737 patients (age range, 17-87 years), which yielded 803 breast mass lesions (296 malignant, 507 benign). Radiologist-interpreted features from mammograms and sonograms were used as input features for linear discriminant analysis (LDA) and artificial neural network (ANN) models to differentiate benign from malignant lesions. An LDA with all the features was compared with an LDA with only stepwise-selected features. Classification performances were quantified by using receiver operating characteristic (ROC) analysis and were evaluated in a train, validate, and retest scheme. On the retest set, both LDAs were compared with radiologist assessment score of malignancy. Both the LDA and ANN achieved high classification performance with cross validation (area under the ROC curve [A(z)] = 0.92 +/- 0.01 [standard deviation] and (0.90)A(z) = 0.54 +/- 0.08 for LDA, A(z) = 0.92 +/- 0.01 and (0.90)A(z) = 0.55 +/- 0.08 for ANN). Results of both models generalized well to the retest set, with no significant performance differences between the validate and retest sets (P > .1). On the retest set, there were no significant performance differences between LDA with all features and LDA with only the stepwise-selected features (P > .3) and between either LDA and radiologist assessment score (P > .2). Results showed that combining mammographic and sonographic descriptors in a CAD model can result in high classification and generalization performance. On the retest set, LDA performance matched radiologist classification performance.

Assessing the importance of features for multi-layer perceptrons

In this paper we establish a mathematical framework in which we develop measures for determining the contribution of individual features to the performance of a classifier. Corresponding to these measures, we design metrics that allow estimation of the importance of features for a specific multi-layer perceptron neural network. It is shown that all measures constitute lower bounds for the correctness that can be obtained when the feature under study is excluded and the classifier rebuilt. We also present a method for pruning input nodes from the network such that most of the knowledge encoded in its weights is retained. The proposed metrics and the pruning method are validated with a number of experiments with artificial classification tasks. The experiments indicate that the metric called replaceability results in the tightest error bounds. Both this metric and the metric called expected influence result in good rankings of the features.

A comparison of colour texture attributes selected by statistical feature selection and neural network methods

In this paper, two methods for selecting input features for a neural network used to aid iconic retrieval in an image database are presented and compared. The first method involves training the network on all the feature inputs and then analysing the weight values in an attempt to find the more important input features. The second borrows a method from statistical feature selection known as the sequential floating forward selection algorithm. Both methods improve the computational efficiency of the image database search whilst not affecting the quality of the results obtained. Being more computationally efficient, network weight analysis offers an attractive alternative to the statistical method.

Evaluation of complications of kidney transplantation using artificial neural networks

The aim of this study was to develop an artificial neural network (ANN) to differentiate between rejection, acute tubular necrosis (ATN) and normally functioning kidneys in a group of patients with renal transplants. The performance of ANN was compared with that of an experienced observer using a database of 35 patients' records, each of which included 12 quantitative parameters derived from renograms and clinical data as well as a clinical evaluation. These findings were encoded as features for a three-layered neural network to predict the outcome of biopsy or clinical diagnosis. The network was trained and tested using the jackknife method and its performance was then compared to that of a radiologist. The network was able to correctly classify 31 of the 35 original cases and gave a better diagnostic accuracy (88%) than the radiologist (83%), by showing an association between the quantitative data and the corresponding pathological results (r = 0.78, P < 0.001). We conclude that an ANN can be trained to differentiate rejection from acute tubular necrosis, as well as normally functioning transplants, with a reasonable degree of accuracy.