Fundaments of Machine Learning for and with Engineering Applications (MOD550)

Machine learning has recently emerged as one of the most promising resources for engineers, offering a set of powerful approaches to tackle complex engineering challenges. By employing various statistical techniques in learning algorithms, machine learning enables the development of predictive models, optimization strategies, and decision support systems that can enhance the design, analysis, and control of engineered systems. This technology empowers engineers to extract meaningful insights from vast datasets, automate repetitive tasks, and improve the efficiency, reliability, and performance of engineering processes. Its applications span diverse domains, from mechanical and chemical engineering to geo, material science, etc. making machine learning an indispensable resource for modern computational engineers for physics based and data-driven solutions to real-world problems.

Most undergraduates in engineering and science fields have little exposure to data methods, while most computer scientists and statisticians have little exposure to dynamical system control. Our goal is to provide an entry point and interface for both these groups of students.

The focus of this course is the basic knowledge and related assumptions involved in applying statistics and machine learning in spatial and/or temporal contexts. The emphasis is on providing students with knowledge of the fundamentals of statistics and machine learning most relevant for spatial or temporal data.

The core of the course is data analysis with a clear focus on engineering applications. Data sources, their quality, and consistency will be discussed to guide the selection of suitable spatial and temporal models. Simulation techniques are introduced as a means of modeling heterogeneity and uncertainty. Machine learning techniques, such as regression modeling and analysis and multivariate data analysis, will be introduced and applied. Python and other programming tools will be used for modeling, preparing spatial and temporal data, scripting statistical workflows, and constructing visualizations to communicate model outputs and analysis results.

The primary purpose of modeling is to generate decision insight. That is the criteria to assert the 'usefulness’ of a model and along which its interpretability, resolution and outcome accuracy will be evaluated in a set of given tasks.

Knowledge:

• Understanding of engineering data sources and consequent data properties, statistics, and probability distributions (from feature engineering to digital twins)
• Understanding of data analysis/machine learning approaches outcomes, method sensitivity and their applications (e.g. molecular modeling, flow simulation, geoscience).
• Understanding of deep learning principles and its advantages/limitations in engineering applications.
• Ability to construct basic predictive models (e.g. Monte Carlo simulations, language models as chatGPT)
• Data handling, data wrenching and feature engineering.

Skills:

• Data handling, data wrenching and feature engineering.
• Can develop data driven models of physical systems for chemical and mechanical engineering, material science and geology.
• Test data driven models against physical models and experimental data.
• Apply appropriate statistical methods to obtain better insights.
• Develop own programs written in Python and develop wrappers for open machine learning repositories.

General Competence:

• Can identify engineering problems, develop hypotheses, and apply mathematical models and data driven solutions.
• Can structure different statistical models in a wide range of engineering applications.
• Can connect engineers and data science solutions -both to specialists and to the general public.