Smooth Curve from noisy 2-Dimensional Dataset

TL;DR

This paper presents methods for transforming noisy 2D datasets into smooth curves, comparing various nonparametric regression techniques to identify the most effective smoothing method for experimental data.

Contribution

It evaluates multiple smoothing techniques, including Savitzky-Golay, Lowess, Loess, and splines, to determine the best approach for noise reduction in experimental torque-twist data.

Findings

Savitzky-Golay filtering effectively smooths noisy data

Spline and Loess methods are also suitable depending on the context

The choice of smoothing method depends on the specific data and analysis goals

Abstract

In this paper we will be represent the transformation of a noisy dataset into a regular and smooth curve. We performed torsion test on 15Cr 15Ni Titanium modified austenitic stainless steel up to its rupture. We did it these torsion tests multiple strain rates varying from 0.001 to 25 per Sec and obtained huge number of data points has been obtained from with a data acquisition rate of 100 Hz. We also have few more data of 1500Hz of the same experiment on a different material e.g 316 L Austenitic Stainless Steel. Machine is actually acquiring only torque value and the angle of rotation. The torque vs. twist data itself will be having a noise, which gets multiplied when we take the first derivative of torque-twist data. The noise becomes huge and it fluctuates from desired material properties, although some serrated flow should be present but the range of serration should not be as the fluctuation observed in our curve so smoothing was necessary. We will be documenting final shear stress-strain curve we achieved through Nonparametric Regression Lowess and Loess, Savitzky Golay filtering, and also the robust Nonparametric Regressions, body splines and shape preserving curves. The most acceptable one is Savitzky Golay method for our experiment or one can choose splines or loess or lowess according to the object of interest.