Savitzky–Golay smoothing and numerical differentiation of cone calorimeter mass data

Abstract

The Savitzky–Golay (SG) smoothing filter is applied to cone calorimeter mass data in order to obtain superior estimates of both sample mass and mass loss rate as functions of time, when compared with traditional numerical differentiation procedures. The smoothing polynomial is differentiated and the derivative used to obtain mass loss rate. The smoothing procedure is first applied to synthetic data, where the noise is known, and the results are used to identify the best smoothing parameters. The procedure is then applied to real cone data and the results compared with traditional methods. In practice, it is found that the SG method is effective at reducing noise in the original mass data and more importantly produces a mass loss rate with far less noise than a standard numerical difference scheme. Furthermore, it is shown that noise in the mass data, amplified by traditional numerical differentiation schemes, swamps any error introduced into the mass loss rate by the scheme itself. Since the proposed method greatly reduces noise, the final accuracy of the mass loss rate should be improved.