Source-linked AI summary
Giving Mechanical Engineers Intelligent Tools: A Project-Based AI Education Curriculum in Thermal Engineering
Changgen Li, Han Hu, Christy Dunlap, Nathaniel House, Jonathan Wai
TL;DR
Mechanical engineering students face limited preparation in electrical and computer engineering, difficult thermal modeling challenges, and constrained coursework. The paper introduces UARK’s project-based, scaffolded AI curriculum for thermal engineering, and reports student projects spanning machine-learning development and advanced thermal modeling, including models that achieved 100% precision for image-based critical heat-flux detection.
Problem
Mechanical engineering students have limited AI, electrical, computer engineering, and programming backgrounds while traditional thermal modeling presents complex challenges under constrained coursework.
Method
UARK’s MLME course uses scaffolded, project-based instruction across introductory, application, and advanced learning levels, linking thermal problem formulation to AI algorithm selection.
Results
Student-initiated projects produced machine-learning models for engineering tasks and thermal analyses for advanced device designs; four models achieved 100% precision for image-based critical heat-flux detection.
Takeaways & Limitations
The open-source curriculum gives students hands-on AI implementation experience and a comprehensive skill set for mechanical and thermal engineering applications.
Abstract
from arXiv · showhide
Mechanical engineering (ME) requires a broad knowledge base across several disciplines. However, ME students often have insufficient training in electrical and computer engineering, complex challenges in traditional thermal system modeling, and endure heavy course loads with limited class hours. To help address these challenges, this paper proposes a new curriculum that integrates artificial intelligence (AI) into ME at the University of Arkansas (UARK), with a particular emphasis on thermal problems and their interplay with electrical and computer engineering. The curriculum has introductory, application, and advanced levels, covering core and optional AI projects. Key goals are to enhance students' understanding of AI models, ability to tackle engineering tasks, and teach multidisciplinary communication skills. This curriculum offers educators and researchers valuable insights into courses that can enhance students' practical skills and creativity. This curriculum, including the syllabus, data, and codes, is available to the public in open-access repositories.
I. INTRODUCTION
Mechanical engineering spans diverse STEM domains but offers limited intersection with electrical and computer engineering. The paper responds by proposing project-based AI education to strengthen interdisciplinary problem-solving within constrained coursework.
- Mechanical engineering requires expertise across diverse science, technology, engineering, and mathematics domains.
- The paper argues that AI concepts and skills should be incorporated synergistically into thermal engineering education.
- Traditional mechanical engineering curricula offer limited intersection among electrical, computer, and mechanical engineering despite diverse, time-intensive content.
- UARK’s mechanical engineering program provides limited exposure to electrical engineering and programming, constraining cross-disciplinary preparation.
- The authors present project-based AI education for mechanical engineers as hands-on problem-solving rather than course examinations.
II. RELATED BACKGROUND OF AI FOR ENGINEERING
AI-integrated education is presented as a way to develop critical thinking and AI literacy while encouraging students to evaluate AI outputs, limitations, and ethical implications.
- AI-integrated science lessons help students develop critical thinking skills and AI literacy.
- Students learn to use AI tools and critically evaluate their outputs, limitations, and ethical implications.
B. Thermal engineering
Thermal engineering seeks to manage heat efficiently while reducing energy use and safety risks. Traditional thermal monitoring can overlook interactions among physical quantities and available design, operational, and historical data.
- Thermal engineering designs systems that manage and utilize heat efficiently to optimize energy use and reduce safety risk.
- Data-center cooling systems consume around 45% of total energy, compared with about 40% for IT equipment.
- Critical heat flux during boiling can sharply increase surface temperature and potentially damage cooled equipment.
- Traditional cooling regulation relies on selected physical quantities and cannot effectively use design parameters, operational parameters, and historical monitoring data.
- The paper identifies AI-integrated thermal engineering education as necessary for interdisciplinary skills relevant to thermal management, energy use, efficiency, and safety.
A. Curriculum goals
UARK’s Machine Learning for Mechanical Engineers course uses project-based learning to integrate AI into thermal engineering problem-solving. It emphasizes selecting and implementing algorithms for engineering tasks while developing technical and professional competencies.
- The MLME course applies project-based learning to integrate AI into engineering development, especially thermal engineering problem-solving.
- Unlike computer science or data science AI courses, MLME focuses on selecting and implementing AI algorithms for thermal engineering problems.
- Students are expected to develop, train, and test machine-learning models using Python with TensorFlow and MATLAB.
- The course targets competencies including critical thinking, communication, teamwork, technology, leadership, and professionalism.
B. Curriculum hierarchy
The MLME curriculum uses scaffolded instruction, progressing from foundational STEM knowledge and introductory AI concepts to practical application and advanced thermal-engineering topics.
- Curriculum hierarchy: Scaffolded instruction gradually reduces guidance as students gain mastery across the curriculum hierarchy.The hierarchy begins with curiosity and self-motivation, followed by foundational STEM knowledge, introduction, application, and advanced levels.
- Foundational knowledge: The foundational level reinforces matrix operations and programming syntax in MATLAB and Python within each project.Students later use pre-built computational packages so they can focus on algorithmic architectures and programming logic.
- Introduction level: The introduction level teaches ML mathematics and statistics while translating thermal problems into classification, regression, or time-series forecasting tasks.This framing helps students select AI algorithms appropriate to the modeled engineering problem.
- Application and advanced levels: The application level practices ML algorithms on public datasets and adds signal-processing topics such as Fourier transforms and time-series analysis.The advanced level then covers heat flux, boiling, bubble dynamics, and thermal experiment datasets in detail.
C. Curriculum entry requirements
MLME combines broad prerequisites with a 39.1-hour sequence of four thermal-engineering projects spanning monitoring, diagnosis, dimensionality reduction, clustering, and prediction.
- Entry requirements: High-school mathematics and physics are the minimum prerequisites, while calculus, linear algebra, probability or statistics, programming, and thermal-science knowledge are preferred.The scaffolded design teaches or reinforces essential mathematical and programming skills within the course.
- Course structure: The 39.1-hour curriculum integrates Python and MATLAB programming with linear regression, MLP, CNNs, clustering, PCA, time-series analysis, and RNNs.Four assignment projects connect these methods to mechanical-engineering applications.
- Project 1: Project 1 trains MLP and GPR models to predict boiling heat flux from temperature data while accounting for experimental non-idealities.Students report training curves and timing and use k-fold cross-validation to address overfitting.
- Project 2: Project 2 classifies pre-CHF and post-CHF boiling images with CNNs or pre-trained CNNs as a supervised two-class diagnosis task.CHF is associated with rapid heater-surface temperature increases and potential device failure.
- Projects 3 and 4: Projects 3 and 4 extend boiling analysis through unsupervised dimensionality reduction and clustering, then RNN forecasting of future vapor fractions.Project 4 varies input and output sequence lengths to evaluate their effect on prediction error.
A. Student performance assessment
MLME assesses students through four assignment projects and a final project, equally weighting technical project work and culminating deliverables.
- Assessment structure: The course grade totals 100 points, divided equally between four assignment projects and a final project.Each assignment requires a written report and code, while the final project includes staged presentations, reports, and a comprehensive submission.
- Assessment criteria: Each assignment is evaluated by model development at 40%, successful implementation at 30%, and report or presentation quality at 30%.The final project allocates 40 points to the comprehensive final report with source code and operating instructions.
B. Student performance analysis
Across three course offerings, student grades shifted upward and became less variable as the instructors provided more reference code and progressively complex examples.
- Student background: Undergraduates initially produced lower-quality projects and grades than graduate students because of weaker programming, research, and prerequisite-course experience.These challenges persisted after projects were made more manageable and example code was added in 2022.
- Grade trends: Grade variance decreased from 88.02 in 2021 to 42.00 in 2022 and 48.49 in 2023.The reported distributions also shifted toward higher grades in 2022 and 2023.
- Instructional adjustment: More reference code helped students focus on fundamental concepts and problem-solving strategies instead of programming syntax.The authors connect this outcome to the worked example effect, where fully guided solutions support better test performance.
C. Examples of Student Outcomes after the MLME Course
Student-initiated MLME projects produced published outcomes spanning AI model development for manufacturing, sensing, and data analysis, alongside thermal modeling for advanced device design.
- Student Outcomes: Student projects produced outcomes in manufacturing, sensing, and data analysis, as well as thermal modeling and analysis for advanced device design.The AI model-development examples are shown in Fig. 6, while advanced device-design outcomes are shown in Fig. 7.
- Manufacturing: MLP and GPR models predicted final hole diameters from laser-processed steel-substrate data.The datasets contained 300, 600, 900, and 1210 data points, and images were captured using an industrial camera.
- Sensing: PCA models monitored two-phase cooling systems using acoustic signals, hydrophones, and boiling images as multimodal inputs.The research results indicated that PCA was useful for two-phase cooling-system monitoring.
- Data Analysis: Four supervised models achieved 100% precision for image-based critical heat-flux detection.PCA-MLP trained fastest at 1s/epoch, whereas the transformer trained slowest at 212s/epoch; PCA-K-Means also achieved high prediction accuracy with many principal components.
- Advanced Device Design: Students proposed semi-direct cooling for silicon-carbide power devices and designed a high-performance air-cooling heat sink.The cooling setup supported 16 devices and up to 1.7 kV; the heat sink offered performance and stability comparable to standard heat exchangers with a smaller surface area.
V. CONCLUSIONS
The paper introduces a fully open-source, project-based AI curriculum for thermal engineering. Its reported outcomes include stronger AI understanding, coding performance, reduced cognitive load through scaffolding, and interdisciplinary modeling and innovation.
- Curriculum: An AI-integrated thermal-engineering curriculum uses four core and three optional projects to develop AI and engineering problem-solving skills.The curriculum emphasizes project-based AI implementation and is fully open source.
- Reported Outcomes: Project-based curricula enhanced students’ understanding of fundamental AI concepts, interest in AI tools, and coding performance for mechanical-engineering tasks.These effects are stated among the paper’s conclusions based on student scores and advanced outcomes.
- Instructional Design: Examples and scaffolded coding instruction reduced students’ cognitive load and improved learning outcomes and efficiency.
- Interdisciplinary Learning: Project-based interdisciplinary courses stimulated students’ abilities to model and innovate.The conclusion connects the courses with interdisciplinary project work in mechanical and thermal engineering.
- Open Access: The curriculum’s datasets and algorithms are publicly available through GitHub and MathWorks File Exchange repositories.