(603d) Teaching Computers to Interpret Mdsc Thermograms

We present a novel algorithm for the automatic data analysis of modulated Differential Scanning Calorimetry of thermograms designed to minimize subject matter expert input reduce the man reliability.

Differential Scanning Calorimetry (DSC) and mainly modulated DSC (mDSC) analysis are widely used to support the production of amorphous solid dispersions for the identification of kinetic and thermodynamic events such as glass transition, crystallization and melting. However, interpretation of thermograms require careful analysis of subject matter experts (SMEs) which can be time-consuming and thus can be a bottleneck to, e.g., formulation development where large numbers of samples are generated are require prompt analysis for reiteration. Furthermore, often there are elusive phenomena that can sometimes go unnoticed or cause dubious interpretation from different SMEs.

The presented Python-based software tool’s computation of the derivatives and the maximums, minimums and inflections of each curve, is the base of the software. It then uses this information to search for events, based on their magnitude. One essential advantage of using an automated software is the ability to use the derivatives of the heat flow curves to improve the precision of the results. The derivatives are essential not only to the detect the exact temperature at which events occur but also to help define their onsets. In order to tackle the inherent noise of the DSC analysis and data acquisition, the proposed tool employs a low pass, Fourier transform-based filter, which enables the use of the thermogram derivatives.

We show case studies where the software analysis increases the detection rate of events that can easily be unnoticed by visual assessment due to their magnitude or shape, as they only cause very subtle variations. We also present case studies where smaller events are distorted or confounded with a nearby dominant events. However, following the digital detection of these events a more detailed visual assessment confirms their presence and often indicative of the onset of undesirable events such as phase separation or crystallization. Furthermore, the presented digital analysis tool promotes a faster and objective analysis of generated DSC thermogram data.