Reinforcement Learning-Based Adaptive Wire Gauge Selection for Zonal Automotive Harness Design Under Dynamic Driving Scenarios

This paper presents a reinforcement learning (RL) framework for adaptive wire–gauge selection in zonal automotive wiring harnesses under dynamic driving scenarios. Focusing on the centroid–to–battery (pre-distribution) segment, the method minimizes conductor mass while enforcing electrical constraints: a voltage–drop limit of 0.36V (3% of 12 V) and temperature–aware ampacity. Scenario–dependent current profiles are derived from a zonal dataset with four representative operating conditions; distances and topology are held fixed, and only wire types are optimized. A multi–head Deep Q–Network (DQN) selects a gauge per centroid using a reward that balances mass reduction, constraint compliance, and voltage–budget utilization. Validation with MATLAB scripts and a companion Simulink model shows near–identical voltage drops, margins, and validity flags across 65°C, 85°C, and 105°C. On a 12–zone High–application case at 85°C, the RL policy achieves 100% compliance with an average voltage drop of 0.30V, whereas baselines (a static–gauge design from Hossain et al. and a rule–based scenario–aware method) exhibit higher drops and lower compliance. System–level comparisons with Hossain et al. (distances unchanged) indicate total harness mass reductions of –3.62% (Low), –32.01% (Mid), and –34.51% (High). For the 12–zone High case, the RL design reduces centroid→battery mass by 12.52 kg, yielding about €108.13 copper cost savings (priced at € 8.6338/kg) and 51.95 kg CO2e lower embodied emissions per harness. Results demonstrate that scenario–aware RL provides a practical, scalable path to lighter, standard–compliant harnesses that remain robust across temperatures and load variability.