A digital solution framework for enabling electric vehicle battery circularity based on an ecosystem value optimization approach

The study addresses how to enable an effective circular economy for lithium-ion EV batteries by overcoming key barriers in end-of-life (EOL) management: forecasting and tracking EOL battery availability, estimating remaining value and state of health (SOH), optimizing reverse logistics costs, ensuring safety and regulatory compliance, and maximizing value recovery via repurposing or recycling. Contextually, EV adoption is accelerating (global spending ~USD 280B in 2021; ~16.5M EVs on roads), with projections up to 200M EVs by 2030 under IEA STEPS. EOL battery volumes are expected to surge (e.g., EVs sold in 2019 could yield ~500,000 tons of unprocessed pack waste), and current cathode material supply chains are energy-intensive with significant social and environmental risks. While EOL batteries typically retain 70–80% capacity and can be repurposed, and recycling technologies can reduce energy, water, and GHG impacts relative to primary production, practical barriers remain—especially high and variable reverse logistics costs (often a large fraction of total recycling costs) and data silos across the battery value chain. The purpose of the study is to propose and demonstrate a digital, multi-stakeholder ecosystem framework and optimization approach to manage five value drivers (safety, compliance, carbon reduction, quality, financials) and thereby improve economic and environmental outcomes of battery circularity.

The paper situates its contribution within several strands of existing work and initiatives: (1) Battery Passport initiatives (Global Battery Alliance’s Battery Passport; Germany’s BMWK Battery Pass) that provide provenance, manufacturing history, and ESG performance data and may help with regulatory requirements (e.g., U.S. IRA tax credit criteria), but currently lack end-to-end EOL management support. (2) Environmental assessments of recycling pathways: studies indicate direct recycling of NMC111 can substantially reduce energy (to ~27%), water use (~31%), and GHG emissions (~32%) compared with primary production; hydrometallurgical recycling of EV batteries can reduce carbon footprint of recovered materials by ~38% versus primary sources; additional analyses suggest second-life use prior to recycling reduces LIB carbon footprint by ~8–17% compared to direct recycling after EV use. (3) Economic viability: recycling can be viable with net values ranging from roughly −$21.43/kWh to +$21.91/kWh depending on transport distance, wages, pack design, and method; transportation averages around 41% of total recycling costs in literature reviews. (4) Behavioral dimensions: consumer biases about repurposed battery quality can hinder second-life adoption; behavioral interventions (nudges) can improve acceptance and purchase likelihood. (5) Forecasting methods: when time-series of returns are available, ARIMA and LSTM models are common baselines; however, EOL return histories may be sparse, necessitating alternative approaches based on current registration data. These sources motivate a need for integrated data platforms and analytics that extend beyond provenance to operational optimization of EOL flows.

The authors propose an ecosystem value optimization approach coupled with a digital solution framework. Value drivers include safety (incident minimization via data sharing and traceability), regulatory compliance (end-to-end tracking; adherence to hazardous materials and UL standards), carbon footprint reduction (using estimated CO2 reduction factors and sharing of actual yields/energy data), quality (addressing stakeholder biases; ensuring repurposed product and recycled material quality), and financials (optimizing reverse logistics and value recovery). Method components: (1) EOL battery availability forecasting: Using EV registration data and OEM sales, the method models the probability distribution of returns by age via binomial distributions per make/model and year of sale; distributions are merged via convolution to produce yearly return forecasts with means and standard deviations, and 95% prediction intervals. Alternatives like ARIMA/LSTM are considered when sufficient time series exist. (2) Reverse logistics simulation and planning: A multi-product dynamic stochastic simulation (using MESA) models arrivals (e.g., Poisson), chemistries/health mixes (multinomial), geospatial network of dealers/warehouses/Re* providers, cost structures (distance/weight-based transport, storage), and pricing/yield by Re* partner. Financial objective NetProfit is formulated as a sum over items of revenue minus costs, expanded to include unit prices and yields, distance-based transport cost, acquisition cost, and storage costs. Reinforcement learning is trained on forecasted arrivals to optimize operational decisions over time. (3) SOH estimation: Pack-level SOH is predicted from relaxation-phase voltage time series. A bidirectional LSTM regression model (TensorFlow/Keras) maps sequences of 14 relaxation voltage values to remaining capacity; an SVR baseline uses three statistical moments (max, variance, skewness) of the sequence. Conditional estimation methods are developed to predict expected counts and variance of good modules/cells (e.g., modules with SOH ≥ 80%) given pack SOH. (4) Offer pricing and acquisition cost optimization: A behavioral-economics-based pricing model uses logistic regression to estimate offer acceptance probability as a function of price, nudge, and their interaction, trained on two 200-participant cohorts (with/without nudge). The objective minimizes expected net acquisition cost = (price + nudge cost) × Pr(acceptance). (5) Collection route recommendation: Exact shortest Hamiltonian path (SHP) is solved via heuristic search with tailored admissible heuristics, outperforming Bellman-Held-Karp dynamic programming by ~2 orders of magnitude up to problem sizes of 16; distances are computed via geocoding libraries; payload mix optimization maximizes truck utilization. (6) Digital framework: A permissioned blockchain (Hyperledger Fabric v2.2 LTS) with channels and Private Data Collections supports trusted multiparty data sharing among OEMs, service partners, logistics providers, and Re* partners; off-chain databases (MySQL/CouchDB), document stores (e.g., S3), data integration (Apache Airflow), trusted IoT (e.g., TPM), and analytics services (TensorFlow, RLlib) underpin applications. A representative network design with 13 organizations includes service channels for initiation/inspection and EOL channels for shipping and Re* status, with PDCs for offers and quotes.

- Reverse logistics optimization: Simulation on an OEM use case (126 dealers across 5 states; annual EOL returns scenarios of 2,000; 10,000; 20,000) indicates transportation cost reductions of 11% to 44% versus baseline double-handling practice, and value recovery improvements of 52% to 60% by routing higher-value EOL batteries to second-life providers. - SOH estimation accuracy: BiLSTM achieved RMSE ≈ 0.0149 and MAPE ≈ 0.0093 (~0.93% error); SVR baseline achieved RMSE ≈ 0.0168 and MAPE ≈ 0.0127 (~1.27% error). The LSTM is ~36% better in error than SVR, with both under ~1–1.3% error. - Forecasts: Using California registration data and OEM sales, the method projects >22,500 EOL battery packs for the OEM in California by 2027. - Acquisition and pricing: Behavioral-economics-informed offer recommendations suggest potential gross profit of $507 to $667 per EOL battery (small EVs, 20–30 kWh) by optimizing acquisition offers (price and nudges) and routing. - Routing: For a demonstration subset of 11 Los Angeles County locations, the exact SHP-based route totals ~80 miles, reducing distance, costs, and emissions. The heuristic SHP solver outperforms Bellman-Held-Karp by almost 2 orders of magnitude for sizes up to 16. - Carbon accounting enablement: The framework supports tracking estimated CO2 reductions per EOL battery and sharing of actual energy/yield data from Re* and logistics partners for improved carbon footprint estimates. - Ecosystem performance: The framework supports safety and compliance through traceability (battery removal instructions, diagnostics, end-to-end tracking) and smart contracts for automated decisions (acquisition, sales, shipping).

The proposed ecosystem value optimization approach directly targets the primary barriers to EV battery circularity: fragmented data, high reverse logistics costs, uncertain EOL value, and variable quality perceptions. By integrating a permissioned blockchain for trusted multiparty data sharing with predictive and prescriptive analytics, the framework enables end-to-end visibility and decision support across collection, transport, repurposing, and recycling. Simulation results show substantial transport cost savings and value recovery gains, while accurate SOH estimation facilitates better matching of EOL packs/modules to Re* providers, thereby increasing economic returns and safety. Behavioral-economics-based pricing reduces acquisition costs by increasing acceptance probabilities at lower effective net cost. Route optimization lowers both costs and emissions. The approach extends beyond Battery Passport functionality by operationalizing EOL processes and decisions, helping OEMs and partners meet safety and compliance requirements and providing mechanisms to quantify carbon reductions from reuse and material recovery. Overall, the integrated solution advances practical circularity by aligning incentives and information flows among OEMs, service partners, logistics, and Re* providers.

The paper introduces a comprehensive digital solution framework and ecosystem value optimization approach that integrates trusted data sharing with advanced analytics to improve EV battery circularity. Contributions include: (1) a forecasting method for EOL availability using current registration and sales data; (2) a stochastic simulation and RL-based optimization of reverse logistics with demonstrated cost and value-recovery benefits; (3) high-accuracy SOH estimation via BiLSTM and conditional estimation of good modules/cells; (4) a behavioral-economics-based acquisition pricing model that balances price and nudges; (5) an exact heuristic-search SHP routing solution; and (6) a permissioned blockchain architecture with channels and PDCs tailored to EOL workflows. Future work includes expanding the library of analytical models to adapt to varied stakeholder scenarios, collecting richer EOL return data for benchmarking and validation of forecasts and decision models, scaling routing methods to larger networks, incorporating more granular carbon accounting (actual energy and yield data), and validating the framework with asset owners, OEMs, and recyclers in operational deployments.

- Data and validation: Several demonstrations rely on simulated data and a representative OEM use case; the OEM identity and certain business-sensitive datasets are not publicly available. Forecasting uses current registration and sales data due to limited historical EOL return time series; authors note the need to collect return data to benchmark forecast accuracy. - Benchmarking: Meaningful external benchmarks for the proposed forecasting method are not available; alternative time-series baselines (ARIMA/LSTM) are suggested contingent on data availability. - Modeling assumptions: Arrival processes (e.g., Poisson), chemistry/health mixes (multinomial), cost structures, and yield/pricing at Re* partners are modeled; results depend on these assumptions and parameterizations. Non-financial value drivers are quantified using proxies (e.g., penalties, carbon price), which may vary across stakeholders. - Routing scalability: The exact SHP solver is demonstrated to outperform BHK up to problem sizes of 16; performance and exactness at larger sizes may be constrained and could require heuristics or decomposition. - Generalizability: The solution is not one-size-fits-all; customization is required for different organizational contexts, geographies, and regulatory environments. - Proprietary code and partial data availability: Code is copyrighted, with only snippets available upon request; some datasets (OEM-related) are not public, limiting full reproducibility.