Naveen Kumar Vaegae

Bio: Naveen Kumar Vaegae is an academic researcher from VIT University. The author has contributed to research in topics: Computer science & Convolutional neural network. The author has an hindex of 4, co-authored 7 publications receiving 22 citations.

Development of an intelligent pressure measuring technique for bellows using radial basis function neural network

Abstract: This paper presents the implementation of an intelligent pressure transmitter to measure pressure using a bellow sensor. In industrial applications, the deflection of the bellow due to applied pressure must be translated into an efficient electrical readout for monitoring, transmission and control. An inductive pick-up is used to convert the deflection of the bellow into the change in self-inductance of a coil. The signal conditioning circuit designed for the bellow sensor with the inductive coil is an inductance to voltage conversion circuit (ITVCC). The stray inductances, component tolerances and ambient factors introduce errors in the output of the ITVCC. The voltage–pressure relation exhibits a considerable nonlinearity and limits the measurement to local operations. In this aspect, we propose an artificial neural network (ANN) using a radial basis function to estimate and compensate the nonlinearity of the ITVCC. The intelligence of ANN modeling is incorporated into an embedded plug-in-module (EPIM). The output voltage of the EPIM is converted into a 4–20 mA current signal for further processing. The performance of the proposed technique is experimentally verified. The nonlinearity expressed as maximum deviation from the desired response is within ±0.6% of full scale reading. The design aspects, simulation analysis and experimental results of the technique are reported.

Sign Language Recognition Using Convolutional Neural Networks

Abstract: The lack of familiarity with sign language among non-users and the absence of a universal method for communication poses major problems for the deaf community. The digitization of communication channels can greatly benefit society. Previous attempts to develop sign language recognition (SLR) systems have posed disadvantages of low feasibility, portability, and accuracy which hamper widespread use. In this paper, a vision-based sign language recognition system using convolutional neural networks (CNN) is proposed that can efficiently and accurately recognize the American Sign Language (ASL) alphabet. The proposed method is implemented using the Keras library on Python. It is easy to implement and has attained high accuracies of over 99% with low bias and variance. The implementation has been validated on the ASL MNIST dataset and its performance is established through simulation studies.

Theoretical and Experimental Investigation of Thermal Dynamics of Steinhart–Hart Negative Temperature Coefficient Thermistors

Abstract: The temperature-dependent dynamics of a negative temperature coefficient (NTC) thermistor conducting variable electric current is modeled using the differential approach. The thermistor is assumed to follow the Steinhart–Hart resistance-temperature equation. The developed mathematical model consists of a nonlinear differential-algebraic equations system, and it was analyzed by the Adomian decomposition method (ADM) and its time-marching version known as the multistage Adomian decomposition method (MADM) as well as the Dormand–Prince (DP) numerical method. Five sets of experiments were conducted on five different NTC thermistors and the laboratory measurements were compared with the model predictions. It is demonstrated that the proposed model, when combined with the MADM, can accurately simulate the thermal behavior of the NTC thermistors. The MADM reproduces the experimental temperature dynamics of the five NTC thermistors with an average absolute relative error of about 2.601% while the corresponding errors for the DP method and the classic ADM are 8.122% and 51.255%, respectively. Also, it is shown that the MADM is highly efficient in terms of computational efficiency and it is approximately 6.5 times faster than the classic DP method, when tuned appropriately.

FPGA Implementation of Steinhart–Hart Equation for Accurate Thermistor Linearization

Abstract: This paper presents fixed point operation-based Field Programmable Gate Arrays (FPGA) implementation of Steinhart–Hart Equation (SHHE) for thermistor linearization. FPGA implementation issues of SHHE are presented and their solutions are proposed and experimentally validated in a LabVIEWTM environment. Experimental temperature calibration, performed using a M/S Fluke drywell calibrator, revealed a lowest nonlinearity of 0.11% for an industrial grade thermistor in the input temperature range from −20 °C to 120 °C. Therefore, the main contribution of this work is to demonstrate the lowest nonlinearity for a wider temperature range. This work is expected to be very useful to instrumentation engineers as it employs a time-tested technique, for thermistor linearization in FPGA, leading to the lowest nonlinearity for a larger input temperature range.

A Combination of Astable Multivibrator and Microcontroller for Thermistor-Based Temperature Measurement Over Internet

Abstract: The paper reports the development of a cost-effective Negative Temperature Coefficient thermistor-based temperature transducer for embedded systems and for the Internet of Things applications. The venture uses a 555 timer-based astable multivibrator as a first-stage linearizer in which the thermistor and a linearizing resistance in series with it modulate the threshold voltages of the internal comparators of the integrated circuit. The output is further processed by a microcontroller which performs a second stage linearization with the help of a look-up table coupled with linear interpolation. Furthermore, the microcontroller uses an Ethernet shield to transmit the temperature information to remote locations. The performance of the proposed setup has been examined for three different thermistors, over 30 °C to 120 °C. The measurement strategy has been found to outperform the competing arrangements reported in recent times, with regard to the measurement accuracy, linearity of transfer, compactness, cost, reliability, and immunity to drift.

Temperature and Nonlinearity Compensation of Pressure Sensor With Common Sensors Response

Abstract: Temperature sensitivity compensation of pressure sensor considering its nonlinearity is crucial to obtain a highly accurate measurement. Unfortunately, various software methods are known to have some drawbacks, thus a large number of calibration points is necessary to model the sensor response properly. This paper aims to present a new compensation method that features the utilization of common response of the given type of sensors and the reduction of particular sensor calibration data set, which subsequently reduces the duration of necessary experiments. Mathematical formulas, which are the basis of a solution, are derived from the sensor general model. Two cases have been considered and analyzed theoretically: iterative solution with sensor response, and noniterative with sensor reproducing function. The method is experimentally verified using two sets of sensors. There is a tenfold increase in accuracy after compensation. In addition, the practical limitation of achievable accuracy by full individual calibration in the given instrumentation is estimated. The cost of simplification increases the remaining error by about 50%. Hence, the advantage is the reduction of acquisition time more than by half. The achievable accuracy is about 0.01% full scale (FS). The results demonstrate that the proposed compensation method is valid for high-accuracy measurements. Furthermore, the comparison of various cases confirms the validity of sensors similarity assumption and correctness of a theoretical analysis presented in this paper.