Debangshu Dey

An ensemble system for automatic sleep stage classification using single channel EEG signal

The present work aims at automatic identification of various sleep stages like, sleep stages 1, 2, slow wave sleep (sleep stages 3 and 4), REM sleep and wakefulness from single channel EEG signal. Automatic scoring of sleep stages was performed with the help of pattern recognition technique which involves feature extraction, selection and finally classification. Total 39 numbers of features from time domain, frequency domain and from non-linear analysis were extracted. After extraction of features, SVM based recursive feature elimination (RFE) technique was used to find the optimum number of feature subset which can provide significant classification performance with reduced number of features for the five different sleep stages. Finally for classification, binary SVMs were combined with one-against-all (OAA) strategy. Careful extraction and selection of optimum feature subset helped to reduce the classification error to 8.9% for training dataset, validated by k-fold cross-validation (CV) technique and 10.61% in the case of independent testing dataset. Agreement of the estimated sleep stages with those obtained by expert scoring for all sleep stages of training dataset was 0.877 and for independent testing dataset it was 0.8572. The proposed ensemble SVM-based method could be used as an efficient and cost-effective method for sleep staging with the advantage of reducing stress and burden imposed on subjects.

Cross-wavelet transform as a new paradigm for feature extraction from noisy partial discharge pulses

In this work a new approach based on cross-wavelet transform towards identification of noisy Partial Discharge (PD) patterns has been proposed. Different partial discharge patterns are recorded from the various samples prepared with known defects. A novel cross-wavelet transform based technique is used for feature extraction from raw noisy partial discharge signals. Noise is a significant problem in PD detection. The proposed method eliminates the requirement of denoising prior to processing and therefore it can be used to develop an automated and intelligent PD detector that requires minimal human expertise during its operation and analysis. A rough-set theory (RST) based classifier is used to classify the extracted features. Results show that the partial discharge patterns can be classified properly from the noisy waveforms. The effectiveness of the feature extraction methodology has also been verified with two other commonly used classification techniques: Artificial Neural Network (ANN) based classifier and Fuzzy classifier. It is found that the type of defect within insulation can be classified efficiently with the features extracted from cross-wavelet spectra of PD waveforms by all of these methods with a reasonable degree of accuracy.

Rough-granular approach for impulse fault classification of transformers using cross-wavelet transform

A novel approach based on information granulation using Rough sets for impulse fault identification of transformers has been proposed. It is found that the location and type of fault within a transformer winding can be classified efficiently by the features extracted from cross-wavelet spectra of current waveforms, obtained from impulse test. Results show that the proposed methodology can localize the fault within 5% of the winding length with a high degree of accuracy. The basic concepts of feature extraction using cross-wavelet transform and the method of classification of those features by rough-granular method are also explained.

Monitoring of inter-turn insulation failure in induction motor using advanced signal and data processing tools

Detection of stator winding inter-turn insulation failure at early stages is crucial for promoting safe and economical use of induction motors in industrial applications. Whereas major insulation failures involving larger percentages of winding are easily discernible from magnitude of supply current, minor inter-turn insulation failures involving less than 5% of turns often go undetected. The present contribution reports experimental results of minor faults due to inter-turn insulation failures in stator windings of induction motor under different loading conditions being analyzed using data and signal processing tools combining Parks Transform and Cross Wavelet Transform. Rough Set Theory (RST) based classifier has been used for fault severity monitoring.

Rough-Set-Based Feature Selection and Classification for Power Quality Sensing Device Employing Correlation Techniques

In this paper, we present a scheme of rough-set-based minimal set of feature selection and classification of power quality disturbances that can be implemented in a general-purpose microcontroller for embedded applications. The developed scheme can efficiently sense the power quality disturbances by the features extracted from the cross-correlogram of power quality disturbance waveforms. In this paper, a stand-alone module, employing microcontroller-based embedded system, is devised for efficiently sensing power quality disturbances in real time for in situ applications. The stand-alone module is developed on a PIC24F series microcontroller. Results show that the accuracy of the proposed scheme is comparable to that obtained in offline analysis using a computer. The method stated here is generic in nature and can be implemented for other microcontroller-based applications for topologically similar problems.

An expert system approach for transformer insulation diagnosis combining conventional diagnostic tests and PDC, RVM data

Search for a reliable and efficient insulation diagnostic tool has always been the interest of power utilities. Today a large number of methods are available that can be used for insulation condition monitoring. These methods include both traditional and newer techniques. However due to complex aging process of oil paper insulation under the influence of different types of stresses, insulation condition assessment is generally performed by experts after carefully evaluating different measurement data. Furthermore, measurement data are influenced by various factors (like conductive aging byproducts, furanic compounds, paper and oil-moisture) in addition to measurement error (if any). This makes prediction of insulation condition based on single type of measurement rather difficult. This paper presents an Expert System designed to perform insulation diagnosis. The Expert System considers measurement data obtained using both traditional and newer techniques in order to come to a definitive conclusion. The Expert System extracts insulation condition sensitive information from data obtained using different techniques and then uses these to devise an optimized insulation model. This optimized model is used to predict paper-moisture content and other insulation condition sensitive parameters. Since these values are predicted using optimized model, they are not dependent on a single type of measurement and hence are less likely to be affected by error of any specific measurement. The performance of the developed Expert System is first tested on a laboratory sample and then on several real life power transformers belonging to NTPC Ltd.

Cross-correlation aided wavelet network for classification of dynamic insulation failures in transformer winding during impulse test

Wavelet network based approach for identification of fault characteristics of dynamic insulation failure during impulse test has been proposed. The network identifies the fault characteristics using the significant features extracted from cross-correlation sequence of winding currents of no-fault as well as impulse faulted winding insulation. The required winding current waveforms to extract significant features for identification of various fault characteristics are acquired by emulating different dynamic insulation failures in the analog model of 33 kV winding of 3 MVA transformer using developed analog fault simulator. The results show that the wavelet network using cross-correlation features has successfully identified the dynamic insulation failure characteristics, viz. fault type, condition and location of occurrence of failure along the length of the winding with acceptable accuracy. The efficacy of extracted features and developed wavelet network for fault characteristics identification is also compared with artificial neural network classifier. The concept of emulation of dynamic insulation failure, cross-correlation based feature extraction and wavelet based fault characteristics identification methods are explained.

On-Line Detection of Apnea/Hypopnea Events Using SpO

This paper presents an online method for automatic detection of apnea/hypopnea events, with the help of oxygen saturation (SpO2) signal, measured at fingertip by Bluetooth nocturnal pulse oximeter. Event detection is performed by identifying abnormal data segments from the recorded SpO 2 signal, employing a binary classifier model based on a support vector machine (SVM). Thereafter the abnormal segment is further analyzed to detect different states within the segment, i.e., steady, desaturation, and resaturation, with the help of another SVM-based binary ensemble classifier model. Finally, a heuristically obtained rule-based system is used to identify the apnea/hypopnea events from the time-sequenced decisions of these classifier models. In the developmental phase, a set of 34 time domain-based features was extracted from the segmented SpO2 signal using an overlapped windowing technique. Later, an optimal set of features was selected on the basis of recursive feature elimination technique. A total of 34 subjects were included in the study. The results show average event detection accuracies of 96.7% and 93.8% for the offline and the online tests, respectively. The proposed system provides direct estimation of the apnea/hypopnea index with the help of a relatively inexpensive and widely available pulse oximeter. Moreover, the system can be monitored and accessed by physicians through LAN/WAN/Internet and can be extended to deploy in Bluetooth-enabled mobile phones.

_{\bf 2}

This paper presents a novel real-time adaptive sleep apnea monitoring methodology, suitable for portable devices used in home care applications. The proposed method identifies apnea/hypopnea events with the help of oronasal airflow signal and aimed to meet clinical standards in the assessment mechanism of apnea severity. It uses a strategically combined adaptive two stage classifier model to detect apnea or hypopnea events on the basis of personalized breathing patterns. For the detection of events, optimum set of time, frequency, and nonlinear measures, extracted from overlapping segments of typical 8 s were fed to support vector machine-based classifiers model to identify the possible origin of the segments, i.e., whether from normal or abnormal (apnea/hypopnea) episodes, and then the decision of the classifier model on the time sequenced successive segments have been used to detect an event. The performance of the proposed real-time algorithm is validated on clinical tests online. Average accuracies of hypopnea, apnea, and combined event detection when compared with polysomnography-based respective indices on unseen subjects during online tests were found to be 91.8%, 94.9%, and 96.5%, respectively, which are quite acceptable.

Signal: A Rule-Based Approach Employing Binary Classifier Models

A low cost linearizing circuit is developed, placing the NTC thermistor in a widely used inverting amplifier circuit using operational amplifier. The performance of the system is verified experimentally. A linearity of approximately ± 1% is achieved over 30 °C -120 °C. When used for a narrower span, a much better linearity of ± 0.5% is obtained. The gain of the arrangement can be adjusted over a wide range by simply varying the feedback resistance. The simplicity of the configuration promises a greater reliability, and also curtails the deterioration in the stability of performance, by reducing the cumulation of drifts in the different circuit components and devices.