Data Science for Mechanical Systems

Columbia University · MECE 4520 · Fall 2026 · Vergil

MECE 4520 introduces data science and machine learning through real engineering systems.

  • Location: 717 Hamilton Hall
  • Time: Tuesday & Thursday 10:10am - 11:25am
  • Lecturer: Changyao Chen (cc2759@columbia.edu)
  • TAs: Mai Al Shaaban (ma5070) and Vlad Pyltsov (vp2517)
  • Office hours: TBD

Course materials and source files are available on GitHub.

Course focus

Rather than treating machine-learning methods as disconnected algorithms, the course follows a practical engineering progression:

understand engineering data → build predictive models → diagnose systems → learn representations → make decisions and control systems

Students will work repeatedly with a small number of real mechanical-engineering datasets, including gas-turbine emissions and CNC milling tool-wear data. This makes it possible to connect model choices to the systems, sensors, variables, and operating conditions that produced the data.

Grading policy

The course grade is based on:

Component Weight
Homework assignments 30%
In-class quizzes 30%
Final project 40%

Students should expect approximately five homework assignments. The final project will be completed in groups and presented as a five-minute presentation on a selected engineering-data topic.

In-class quizzes

Expect a short, time-limited quiz about once each week. At the end of the semester, your 3 lowest quiz scores will be dropped.

We will use Poll Everywhere for in-class quizzes. You need to sign in Poll Everywhere with your Columbia UNI before the first quiz.

Syllabus

This syllabus may evolve as the Fall 2026 course is finalized.

Class dates (2026) Topic
Before Sep. 8 Python, NumPy/Pandas, linear algebra, and probability refresher
Sep. 8 — asynchronous
Sep. 10
Course launch, Python/data foundations, engineering EDA, and measurement variability
Sep. 15
Sep. 17 — asynchronous
Simple and multiple linear regression
Sep. 22 & 24 Statistical inference for regression, prediction intervals, and regularization
Sep. 29 & Oct. 1 Logistic regression and classification metrics
Oct. 6 & 8 Validation, leakage, grouped splits, and temporal train/test splits
Oct. 13 & 15 Decision trees, random forests, and gradient boosting
Oct. 20 & 22 PCA, dimensionality reduction, and clustering
Oct. 27 & 29 Neural networks and nonlinear function approximation
Nov. 3 — Election Day (no class)
Nov. 5
Sequential data, degradation, and optional signal processing
Nov. 10 & 12 Reinforcement-learning foundations: bandits, states, actions, rewards, and MDPs
Nov. 17 & 19 Bellman equations, dynamic programming, Q-learning, and SARSA
Nov. 24
Nov. 26 — Thanksgiving (no class)
Function approximation and deep Q-networks
Dec. 1 & 3 Policy gradients, PPO, continuous control, and RL versus classical control
Dec. 8 & 10 Final project presentations