Instructional design is a cognitively demanding activity in preservice teacher education because it requires the integration of instructional objectives, learning activities, assessment strategies, and pedagogical decision-making. Conventional cognitive load studies have mainly relied on post-task self-report instruments, which are insufficient for capturing real-time cognitive fluctuations during task execution. In this study, we proposed a sensor-integrated multimodal cognitive load assessment framework for instructional design tasks in preservice elementary teachers by combining physiological sensing signals, eye-tracking sensors, behavioral interaction sensing, synchronized multimodal data fusion, subjective cognitive load scales, and expert evaluations of instructional design quality. The study included 72 participants and three instructional design subtasks: instructional objective formulation, learning activity planning, and learning assessment design. A real-time multimodal sensing architecture was established to support synchronized signal acquisition, cross-modal feature extraction, and dynamic cognitive load monitoring during instructional design processes. The results revealed significant differences among the three subtasks across all four subjective cognitive load dimensions. Learning assessment design consistently produced the highest cognitive load, followed by instructional objective formulation, whereas learning activity planning showed the lowest load level. These findings demonstrate the potential of multimodal educational sensing technologies and sensor-fusion-based cognitive monitoring for task-sensitive cognitive load assessment and smart teacher education applications. The framework also provides application requirements for future educational sensors, including wearable sensing materials and compact multimodal integration. However, the single-institution sample and limited correspondence between subjective and objective sensing indicators constrain generalizability. Future studies should improve sensor-fusion methods and validate the framework with larger and more diverse samples.