Comparative assessment of modified deconvolution and neuro-fuzzy technique for force prediction using an accelerometer balance system. (February 2021)

• Record Type: Journal Article
• Authors: Deka, Sushmita; Pallekonda, Ramesh Babu; Rahang, Maneswar
• Is Part Of: Measurement. Volume 171(2021)
• Publication Date: 2021-02
• Subjects: Calibration -- Deconvolution -- Adaptive Neuro Fuzzy Inference System -- Impulse -- Accelerometer balance; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2020.108770
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: Impulse response function is calculated from resultant of measured response. Modified deconvolution can predict applied forces from resultant with good accuracy. Adaptive neuro fuzzy inference system predicts impulse forces with better accuracy. Adaptive neuro fuzzy inference system is more accurate than modified deconvolution. Abstract: This article describes the dynamic calibration of a three component accelerometer balance for impulse forces applied in three perpendicular directions and prediction of these forces using a modified deconvolution and adaptive neuro-fuzzy inference system (ANFIS) technique. The accelerometer balance is housed inside a hemispherical model and a tri-axial accelerometer is mounted to measure the accelerations in axial, normal and azimuthal directions. The experimental accelerations are compared with the accelerations obtained from finite element simulations performed using ANSYS Workbench 18.0. The predicted forces using modified deconvolution and ANFIS are compared with each other and with the actual input forces applied during the calibration experiment. Impulse force prediction using deconvolution from multiple components of acceleration is a complex task and needs excessive calculations and effort. In this study, a modified deconvolution methodology has been devised for force prediction using the resultant of the accelerations to obtain the impulse response function. This technique reduces the complexity of force prediction using multiple components of acceleration. The forces can be predicted without making complex calculations for considering the coupling effects of the accelerations in different directions. It has been observed that the modified deconvolution technique performed using the resultant of the accelerations in three directions is able to predict the forces with an average accuracy of 94.50%. However, the accuracy of forces predicted using ANFIS is slightly higher, having an average accuracy of 96.30%. Thus, a comparison of two force prediction techniques which are used in dynamic calibration has been studied in this paper. It has been observed that both modified deconvolution and ANFIS can be used for force prediction without intensive calculations, however the accuracy of ANFIS (having a maximum accuracy of 97.20%) is slightly higher than that of modified deconvolution (having a maximum accuracy of 96.80%). This confirms the ability of modified deconvolution to predict the forces in line with the other standard techniques such as ANFIS with less complexity and can be useful for accurate force prediction.