Speaker
Description
Degradation of catalysts presents a considerable obstacle in the path to the commercialisation of CO2 electrochemical reduction. This is primarily due to the reduction in activity and selectivity it incurs. Despite this, the high costs associated with catalyst characterisation experiments make it difficult to produce adequate and significant data on catalyst deterioration. Machine learning (ML) models have recently shown great promise in supplanting these expensive procedures, although their lack of interpretability presents its own set of challenges. In this paper, we present a comprehensible ML framework that is capable of quickly and accurately projecting the state of the catalyst using simple linear sweep voltammetry (LSV) in under a second, while also providing valuable insight into the degradation process. The performance of a convolutional neural network trained on a dataset comprised of 5236 LSV results significantly outstripped the competition in predictions of total current and faradaic efficiency. The framework exhibited remarkable accuracy, exceeding 99%, in predicting the Faradaic efficiency of various products regardless of the conditions of operation and types of catalysts. The prediction strategy utilised by the model was made intelligible via explainable artificial intelligence (XAI) and key degradation descriptors were identified. The credibility of our proposed framework was validated through surface analyses conducted along with the interpretation provided by XAI. The proposed methodology has potential for widespread application across numerous catalytic processes, battery degradation and chemical process monitoring, thereby offering a dependable and efficient tool for performance monitoring.
References
.
| Keywords | CO2; XAI; CNN; electrochemical |
|---|
