Abstract:In the fault diagnosis system using empirical mode decomposition (EMD), it is important to select the intrinsic mode functions (IMFs) which contain as much fault information as possible and to alleviate the problems of mode mixing and spurious modes. An effective solution to these problems in the decomposition process can help to determine significant IMFs and to improve the performance of the fault diagnosis system. This paper describes a novel power-based IMF selection algorithm and evaluates the performance of the proposed fault diagnosis system using improved complete ensemble EMD with adaptive noise and a multi-layer perceptron neural network.

Many healthcare problems are life threatening and need real-time detection to improve patient safety. Heart attack or ventricular fibrillation (VF) is a common problem worldwide. Most previous research on VF detection has used ECG devices to capture data and sent to other higher performance units for processing and has relied on domain experts and/or sophisticated algorithms for detection. In this case, it delayed the response time and consumed much more energy of the ECG module. In this study, we propose a prototype that an embedded microcontroller where an ECG sensor is used to capture, filter and process data, run VF detection algorithms, and only transmit the detected event to the smartphone for alert and call for services. We discuss how to adapt a common filtering and scale process and five light-weighted algorithms from open literature to realize the idea. We also develop an integrated prototype, which emulates the VF process from existing data sets, to evaluate the detection capability of the framework and algorithms. Our results show that (1) TD outperforms the other four algorithms considered with sensitivity reaching 96.56% and specificity reaching 81.53% in the MIT-BIH dataset. Our evaluations confirm that with some adaptation the conventional filtering process and detection algorithms can be efficiently deployed in a microcontroller with good detection accuracy while saving battery power, shortening response time, and conserving the network bandwidth.

No abstract