As electric intelligent vehicles advance, drive-by-wire systems are increasingly adopted, and the thermal reliability of electromechanical brake (EMB) motors—the key actuators—remains safety-critical. Under stalled-rotor operation, unequal DC currents are typically applied to the three phases, producing nonuniform winding heating. Conventional thermal models can miss the associated tangential heat-transfer effects, increasing the risk of phase-wise end-winding hot spot. This paper analyzes EMB motor thermal behavior under stalled-rotor conditions using a modular 3-D lumped-parameter thermal network (LPTN). First, a standardized tooth module with external interfaces is developed. Its internal parameters are informed by experiments and computational fluid dynamics (CFD) and identified via particle swarm optimization (PSO), allowing the module to be encapsulated for reuse. Next, based on the machine topology, a minimal motor is derived and multiple tooth modules are interconnected through common nodes to form a modular 3-D LPTN that resolves radial, axial, and tangential heat-flow paths. Finally, a stepwise, weighted PSO is applied—module level followed by system level—to calibrate the full network. The tooth-module abstraction also enables rapid network assembly, and the boundary-cooling and loss-allocation modules can be updated to accommodate different cooling architectures and heating patterns while retaining the same internal formulation. Bench tests with inhomogeneous three-phase heating, validated against three-phase end-winding thermocouple measurements, show that the proposed model predicts temperatures more accurately than existing LPTNs. These results indicate that explicitly accounting for tangential heat exchange can improve temperature prediction for EMB motors under stalled-rotor duty and provides a reusable template for other concentrated-winding machines subject to nonuniform thermal loading.