Free Machine Learning Course for Mechanical and Electrical Engineers

A free machine learning course with certificate that suits mechanical and electrical engineers should train on sensor data, not handwritten digits. The EDWartens route does that: Python on meter logs, machine learning on motor currents and energy data, then predictive maintenance on vibration and engine data, all in Google Colab's free tier. The certificate is optional.

Why sensor data instead of MNIST

Most machine learning courses start with MNIST, a set of small images of handwritten digits. It is a good teaching set for neural networks. It is a poor teaching set for the problems an engineer is actually handed.

Plant data behaves differently. The failures you care about are rare. Readings come in time order, so tomorrow cannot be used to predict today. Rows from the same machine look alike, so a random split lets the model memorise the machine instead of learning the fault. Columns carry units, and a feature in amps and one in kelvin will mislead a distance-based model unless you scale them. Labels often come from maintenance records written after the event.

A course that teaches on this data teaches the habits that matter: grouped splits, recall on the rare class, scaling fitted on the training set only, and a baseline that does nothing. Those habits are what separates a notebook that impresses from one that survives a week on site.

Machine learning for electrical engineers

The electrical side of the route starts in Python for AI and Engineering Data. The first practice task converts 12.4 mA on a 0 to 10 bar transmitter to bar and prints it as a tag. By the end you have cleaned a year of plant energy data, computed daily kWh and monthly maximum demand, and plotted a week against a contract line.

Machine Learning with Python then fits kWh against production, ambient temperature and run hours, and prints each coefficient with its unit. You classify healthy against bearing faults on a motor dataset, print the confusion matrix, and compare recall at thresholds of 0.5 and 0.3. You encode the machine column, split the energy data by date, and list two features that would leak.

Instrumentation and control engineers get their own details along the way. One early exercise writes a function that converts a 4 to 20 mA reading to engineering units and returns nothing when the signal is outside the NAMUR limits, because a model fed a broken-wire reading as a real value learns nonsense. The optional Industrial Data with Python course then reads live values over Modbus, OPC UA and MQTT, so the data in your notebook can come from a real controller rather than a CSV.

Machine learning for mechanical engineers

The mechanical side lives in Predictive Maintenance with Machine Learning. You begin with the P-F curve and failure modes: which sensor, in which direction, sees which fault on a pump. You compute a bearing's outer-race defect frequency from its geometry and shaft speed, then look for it in the spectrum of a healthy and a faulty vibration window.

From there you build condition indicators, fit an anomaly detector on healthy data only, and train a fault classifier with the folds grouped by motor. The remaining-useful-life module uses NASA's C-MAPSS turbofan data, published on the NASA Open Data Portal. Each engine starts healthy and develops a fault; in the training set it runs until failure. The FD001 subset has 100 training and 100 test engines under one operating condition.

It runs on Google Colab's free tier

Google describes Colab as a hosted Jupyter notebook service that needs no setup and is free of charge to use. Nothing to install, and no licence to buy. Know its limits before you start a long job:

• Google says free resources are not guaranteed and usage limits fluctuate.
• Runtimes time out when idle, and in the free version notebooks run for at most 12 hours.
• Access to GPUs is heavily restricted on the free tier.

None of that gets in the way here. Tabular sensor data and scikit-learn models run on a CPU runtime in seconds or minutes. Only the optional deep learning module benefits from a GPU. Save the notebook to Drive as you go, and use Restart and run all before you submit anything, so you know it works from a clean start.

The learning route

Take the courses in order. Each one names the course it expects before it:

1. [Python for AI and Engineering Data](/free/python-for-ai-and-engineering-data) for absolute beginners. Lists, dictionaries, functions and classes, then NumPy and Pandas on engineering files.
2. [Machine Learning with Python](/free/machine-learning-with-python) needs basic Python and Pandas. Regression, classification, trees and forests, cross-validation, pipelines, and k-means and PCA for unsupervised work.
3. [Predictive Maintenance with Machine Learning](/free/predictive-maintenance-with-machine-learning) is intermediate and needs the machine learning course.
4. [Industrial Data with Python](/free/industrial-data-with-python) is optional, for reading data straight from Modbus, OPC UA and MQTT.
5. [Deep Learning with TensorFlow and Keras](/free/deep-learning-with-tensorflow) is optional, and helps with the 1-D CNN module in predictive maintenance.

The same courses are grouped on the Industrial AI engineer path and the Applied AI engineer path.

Which algorithms you will use, and why these

The route stays with scikit-learn for most of its length, on purpose. Engineering data is usually a table of readings, and for tables the classic models are hard to beat and easy to explain.

• Linear regression for energy against production, because each coefficient has a unit a plant manager understands: kWh per tonne, kWh per degree.
• Logistic regression for healthy against faulty, because it gives a probability you can threshold.
• Decision trees and random forests because a depth-3 tree can be printed as rules and argued with.
• KNN and SVM mostly to show why scaling matters when one feature is in amps and another in degrees.
• k-means and PCA to find operating modes in healthy data, before anyone labels a fault.
• Isolation forest in predictive maintenance, trained on normal data only, because real failures are too rare to learn from directly.

Deep learning comes last and is optional. A 1-D convolutional network on raw vibration windows is compared against a forest on hand-built indicators, so you see what the extra complexity buys.

What you can show at the end

Each course ends in a project you can put in front of an interviewer. Machine Learning with Python ends with a motor-fault classifier trained, tuned and reported. Predictive Maintenance ends with a model built and reported the way a maintenance manager reads it: failure modes in scope, sensors, warning time, the metric, and who acts on an alert. A notebook that runs from a clean start and a one-page report beat any certificate on their own.

The checks that matter more than the algorithm

Random forest or gradient boosting matters less than whether you evaluated honestly. Keep this list beside you on every project in the route.

A model that scores 97% accuracy on a dataset with 3% failures may have learned nothing. The do-nothing baseline tells you that at once. Recall on the failure class, with precision alongside it, tells you what the maintenance team will actually experience.

How the assessment works

Each course has one final assessment of 15 questions drawn from the whole course. The pass mark is 60%, you get three attempts, and after that there is a 24-hour wait and a fresh paper. Practice tasks in each module are ungraded preparation for it.

Start the course

Start with Python for AI and Engineering Data or, if you already write Pandas, go straight to Machine Learning with Python. All the AI courses are listed under free AI courses for engineers. Sign up for a free account so your progress is saved.

Learning is free in full. The optional EDWartens Certificate of Completion is issued after you pass the final assessment. It is a small one-off fee, from US$2.99 for a beginner course; checkout shows your price. Anyone can check it by code at edwartens.com/verification. It is not a vendor certification, a university award or an accredited qualification.

Take the free course

AI and machine learning · Beginner · Free

Python for AI and Engineering Data

AI and machine learning · Beginner · Free

Machine Learning with Python

AI and machine learning · Intermediate · Free

Predictive Maintenance with Machine Learning

AI and machine learning · Intermediate · Free

Industrial Data with Python