21 October 2016 in IEEE Access
Systems based on wireless gas sensor networks offer a powerful tool to observe and analyze data in complex environments over long monitoring periods. Since the reliability of sensors is very important in those systems, gas classification is a critical process within the gas safety precautions. A gas classification system has to react fast in order to take essential actions in the case of fault detection. This paper proposes a low latency real-time gas classification service system, which uses a multi-layer perceptron (MLP) artificial neural network to detect and classify the gas sensor data. An accurate MLP is developed to work with the data set obtained from an array of tin oxide (SnO2) gas sensor, based on convex micro hotplates. The overall system acquires the gas sensor data through radio-frequency identification (RFID), and processes the sensor data with the proposed MLP classifier implemented on a system on chip (SoC) platform from Xilinx. Hardware implementation of the classifier is optimized to achieve very low latency for real-time application. The proposed architecture has been implemented on a ZYNQ SoC using fixed-point format and the achieved results have shown that an accuracy of 97.4% has been obtained.
10 May 2016 in IEEE Sensors Journal
Increasing the number of sensors in a gas identification system generally improves its performance as this will add extra features for analysis. However, this affects the computational complexity, especially if the identification algorithm is to be implemented on a hardware platform. Therefore, feature reduction is required to extract the most important information from the sensors for processing. In this paper, linear discriminant analysis (LDA) and principal component analysis (PCA)-based feature reduction algorithms have been analyzed using the data obtained from two different types of gas sensors, i.e., seven commercial Figaro sensors and in-house fabricated $4 \times 4$ tin-oxide gas array sensor. A decision tree-based classifier is used to examine the performance of both the PCA and LDA approaches. The software implementation is carried out in MATLAB and the hardware implementation is performed using the Zynq system-on-chip (SoC) platform. It has been found that with the $4 \times 4$ array sensor, two discriminant functions (DF) of LDA provide 3.3% better classification than five PCA components, while for the seven Figaro sensors, two principal components and one DF show the same performances. The hardware implementation results on the programmable logic of the Zynq SoC shows that LDA outperforms PCA by using 50% less resources as well as by being 11% faster with a maximum running frequency of 122 MHz.
1 August 2017 in Journal of Parallel and Distributed Computing
14 February 2022 in IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control
Deep learning is an effective method for ultrasonic crack characterization due to its high level of automation and accuracy. Simulating the training set has been shown to be an effective method of circumventing the lack of experimental data common to nondestructive evaluation (NDE) applications. However, a simulation can neither be completely accurate nor capture all variability present in the real inspection. This means that the experimental and simulated data will be from different (but related) distributions, leading to inaccuracy when a deep learning algorithm trained on simulated data is applied to experimental measurements. This article aims to tackle this problem through the use of domain adaptation (DA). A convolutional neural network (CNN) is used to predict the depth of surface-breaking defects, with in-line pipe inspection as the targeted application. Three DA methods across varying sizes of experimental training data are compared to two non-DA methods as a baseline. The performance of the methods tested is evaluated by sizing 15 experimental notches of length (1–5 mm) and inclined at angles of up to 20° from the vertical. Experimental training sets are formed with between 1 and 15 notches. Of the DA methods investigated, an adversarial approach is found to be the most effective way to use the limited experimental training data. With this method, and only three notches, the resulting network gives a root-mean-square error (RMSE) in sizing of 0.5 ± 0.037 mm, whereas with only experimental data the RMSE is 1.5 ± 0.13 mm and with only simulated data it is 0.64 ± 0.044 mm.
3 April 2018 in IEEE Sensors Journal
This paper presents the development of a wireless low power reconfigurable self-calibrated multi-sensing platform for gas sensing applications. The proposed electronic nose (EN) system monitors gas temperatures, concentrations, and mixtures wirelessly using the radio-frequency identification (RFID) technology. The EN takes the form of a set of gas and temperature sensors and multiple pattern recognition algorithms implemented on the Zynq system on chip (SoC) platform. The gas and temperature sensors are integrated on a semi-passive RFID tag to reduce the consumed power. Various gas sensors are tested, including an in-house fabricated 4×4 SnO 2 based sensor and seven commercial Figaro sensors. The data is transmitted to the Zynq based processing unit using a RFID reader, where it is processed using multiple pattern recognition algorithms for dimensionality reduction and classification. Multiple algorithms are explored for optimum performance, including principal component analysis (PCA) and linear discriminant analysis (LDA) for dimensionality reduction while decision tree (DT) and k-nearest neighbors (KNN) are assessed for classification purpose. Different gases are targeted at diverse concentration, including carbon monoxide (CO), ethanol (C 2 H 6 O), carbon dioxide (CO 2 ), propane (C 3 H 8 ), ammonia (NH 3 ), and hydrogen (H 2 ). An accuracy of 100% is achieved in many cases with an overall accuracy above 90% in most scenarios. Finally, the hardware/software heterogeneous solution to implementation PCA, LDA, DT, and KNN on the Zynq SoC shows promising results in terms of resources usage, power consumption, and processing time.
1 August 2017 in Microprocessors and Microsystems