Abstract:

Within the framework of the educational innovation projects PIME/2025-26/575 and PIME/25-26/534, this work presents the validation of an automated assessment system designed to scale academic rigor in the subject "Machines and Mechanisms" within the Electrical Engineering degree at the Alcoy campus of the Universitat Politècnica de València. The study arises from a critical need identified during the 2024-2025 academic year, when an enrollment of 76 students required the development of personalized kinematic and dynamic analysis assignments for a planar mechanism, using 80 distinct parameter variants to prevent plagiarism issues observed in previous courses. While this strategy successfully eradicated plagiarism, both the preparation and manual correction of submissions entailed a workload exceeding 12 hours of cumulative evaluation time, revealing a significant vulnerability to teaching fatigue. Given the projected enrollment increase for the 2025-2026 course, now reaching 98 students, a 29% rise, the transition toward digital audit methods has become an unavoidable operational necessity.

To address this scenario, a Python script was developed to function as a "Numerical Verification Node". The algorithm's main innovation lies in its "Kinematic Awareness": the system can discern between conceptual errors and technical ambiguities inherent to the CAD software (SolidWorks Simulation). It was identified that, depending on assembly order, the software may generate complementary solutions with inverted signs that remain physically valid. The script mitigates this conflict by evaluating absolute magnitudes and implementing a "Conditional Relative Error" metric, which handles mathematical singularities at zero-velocity points (mechanism dead points), where standard relative error calculations would diverge toward infinity.

Results obtained after comparing manual corrections from the previous course with the new system are conclusive. The script not only reduces processing time from hours to seconds, enabling near-instant feedback for all 98 enrolled students, but also ensures superior evaluative fairness. During the audit phase of the prior course, the algorithm detected 1.78% of "false negatives" and 1.78% of "false positives" originally committed by the human instructor due to transcription errors or visual fatigue when consulting reference tables, while also validating as correct 34.29% of answers previously marked wrong solely due to inverted signs. Finally, a sensitivity analysis established that a 1% tolerance threshold is optimal for validating CAE model convergence against the instructor's analytical Ground Truth. It is concluded that this tool is essential to sustain the personalized teaching model in high-demand academic environments, ensuring a fully objective evaluation process free from fatigue-induced biases.

Keywords:

Assessment Automation, Python Scripting, IA Script, Mechanical Engineering, Personalised Learning, Quality Assurance, Machines and Mechanisms, Fatigue Biases.