DeepDeg: Forecasting and explaining degradation in novel photovoltaics

Abstract

Degradation is a technical and market hurdle in the development of novel photovoltaics and other energy devices. Understanding and addressing degradation requires complex, time-consuming measurements on multiple samples. To address this challenge, we present \textit{DeepDeg}, a machine learning model that combines deep learning, explainable machine learning, and physical modeling to: 1) forecast hundreds of hours of degradation, and 2) explain degradation in novel photovoltaics. Using a large and diverse dataset of over 785 stability tests of organic solar cells, totaling 230,000 measurement hours, DeepDeg is able to accurately predict degradation dynamics and explain the physiochemical factors driving them using few initial hours of degradation. We use cross-validation and a held-out dataset of over 9,000 hours of degradation of PCE10:OIDTBR to evaluate our model. We demonstrate that by using DeepDeg, degradation characterization and screening can be accelerated by 5-20x.

Data science accelerates energy device development

“…Oviedo et al incorporated interpretable ML to lifetime prediction of thinfilm organic solar cells to gain insights into the degradation cause. 47 The authors implemented a black-box neural net model to predict solar cell lifetime from input voltage curves, as seen in Fig. 4(D), and in parallel they calculated the physical parameters from a device model to estimate the contribution of physical parameters to input voltage curves, thus identifying the main degradation driving forces (Fig.…”

“…4(E) and (F)). 47 For common energy devices, whether their purpose is to harvest, convert, or store energy, understanding the failure mechanisms is equally, if not more, important than achieving accurate lifetime predictions. Interpretable predictions are powerful as they enable scientific understanding in a high-dimensional space where correlations between inputs can be non-linear or complex.…”

“…This integration forms an end-to-end ML pipeline that is more likely to generate actionable knowledge and insights for specific energy systems. 47 (A) A diverse battery aging dataset generated by varying 6 cycling parameters 39 (B) Physics based measurement (pulse tests for resistance measurements) and model (differential voltage analysis) for parameter estimation of this dataset (C) SHAP analysis for understanding how cycling conditions and physics-based parameters affect the final battery equivalent full cycles. The extracted SHAP values can be used to construct the matrix plot for a quick visualization of the parameter impact.…”

“…(D) Time series representation of the initial cell degradation (current density-voltage) as input for neural network prediction. 47 (E) Time-regularized physical inference. A device model is used to fit the degradation dynamics to the dynamics of various physical parameters on the time axis.…”

Data science accelerates energy device development

“…This integration forms an end-to-end ML pipeline that is more likely to generate actionable knowledge and insights for specific energy systems. 47 (A) A diverse battery aging dataset generated by varying 6 cycling parameters 39 (B) Physics based measurement (pulse tests for resistance measurements) and model (differential voltage analysis) for parameter estimation of this dataset (C) SHAP analysis for understanding how cycling conditions and physics-based parameters affect the final battery equivalent full cycles. The extracted SHAP values can be used to construct the matrix plot for a quick visualization of the parameter impact.…”