Using AI Interactive Interfaces in Design of Machine Elements Education

Abstract With their ongoing progress, artificial intelligence interfaces are poised to profoundly impact STEM education and participation. Engineering design educators are perhaps among those at the forefront of STEM education experiencing the first tides of this change. An example of such a trend is the course Design of Machine Elements, a mainstay of MAE curricula, which embodies many algorithms that integrate a combination of scientific topics and industry protocols. Design engineers are often tasked with developing computer codes to execute these multi-step technical procedures to design different machine elements, such as power transmission shaft components. In this work in progress, we assigned a class of machine design students to write computer codes that implement several required inputs to generate design parameters for shafts used for specific power transmission parameters. The student codes will prompt users to input various factors, including projected mechanical loads (dynamic and static), shaft material, safety factors, geometric parameters (related to stress concentrations), and design criteria. Based on these decision factors, the students' codes use iterative algorithms to provide shaft design layouts that satisfy the required metrics and constraints. In the second part of the assignment, the students are asked to explore the applicability of an open artificial intelligence interface, such as ChatGPT, to help develop a multi-step design code. In particular, students are tasked to investigate the possibility of using well-thought prompts in ChatGPT to acquire a code or skeleton/pseudocode capable of providing shaft design parameters based on the entered torque and bending loads, material selection, and geometric considerations. After generating and verifying the AI-assisted design codes, students are required to evaluate their accuracy and functionality by comparing them to the codes they wrote in the first part. The requested analysis will address the following areas: (i) the role of appropriate prompting in getting applicable skeleton codes and how a knowledgeable designer iterates on the prompt to enhance the sophistication of the obtained code, (ii) the reproducibility of the AI-assisted code, based on prompt patterns, (iii) the comparison between students' code syntax and structures, and those of AI-assisted codes, and (iv) developing the general techniques that a knowledgeable engineer needs to verify AI-generated codes (critical conditions and pitfalls). Analyzing the results of our novel design assignment offers valuable insights into the interplay between the users' design expertise level and the obtained AI-assisted codes' efficacy. The results also indicate the future role of engineers as experts who can use their deep physical understanding of engineering systems to verify AI-generated algorithms and steer them away from various pitfalls. On a broader level, our study suggests that the advent of AI-assisted interfaces will provide fundamentally new learning and teaching modalities with significantly faster pace of experimentation and emphasis on conceptual mastery of STEM topics.