Auditable thermal descriptors for machine-learning-assisted supercapacitor electrode design : A critical review and framework for sustainable energy storage
Machine-learning (ML)-assisted supercapacitor electrode design is limited by inconsistent descriptors, heterogeneous reporting, and scarce benchmark-ready datasets. This critical review evaluates thermogravimetric analysis (TGA), differential thermogravimetry (DTG), and isoconversional kinetic quantities as auditable candidate features for data-driven electrode research. Candidate descriptors include Tonset[jls-end-space/], Tmax[jls-end-space/], DTG morphology, residual mass, atmosphere contrast, and method-specific activation-energy profiles. Across carbons, transition-metal oxides, conducting polymers, MXenes, MOF-derived systems, and composites, these quantities can encode aspects of protocol-dependent thermal mass-loss behaviour, residue, oxidative sensitivity, conversion behaviour, and degradation heterogeneity, but their meaning is strongly material- and protocol-dependent. A structured critical search and targeted May 2026 reporting-gap audit identified seven additional papers within the scope of the 15 query families with relevant TGA and matched electrochemistry; two contained numerically extractable thermal information in text and five reported the relevant information mainly graphically. Primary-source verification further showed that apparently reusable records can fail quantitative modelling eligibility because of graph-only reporting, sample-stage mismatch, incomplete or contradictory protocols, missing paired TGA measurements, or unverified surrogate transformations. We therefore propose a five-tier descriptor hierarchy with provenance classes, metadata requirements, uncertainty controls, leakage-aware validation, and process-aware ablation against conventional descriptors. The framework treats electrochemical protocol comparability as a prerequisite for pooled targets and sustainability as a testable resource-efficiency outcome: thermal screening would be resource-efficient only when validated decision utility and avoided downstream burdens exceed the material, instrument, energy, gas, and consumable costs introduced by TGA.
| Item Type | Article |
|---|---|
| Identification Number | 10.1016/j.est.2026.125367 |
| Additional information | © 2026 Elsevier Ltd. This is the accepted manuscript version of an article which has been published in final form at https://doi.org/10.1016/j.est.2026.125367 |
| Keywords | data provenance, isoconversional kinetics, machine learning, supercapacitors, thermal descriptors, thermogravimetric analysis, renewable energy, sustainability and the environment, energy engineering and power technology, electrical and electronic engineering, open - publisher policy |
| Date Deposited | 18 Nov 2026 14:33 |
| Last Modified | 18 Nov 2026 14:33 |
