Automotive Monitoring

Background: Automotive data is available from OBD-II codes that are standard for engine monitoring of light duty vehicles (passenger cars) starting in year 1996. Newer vehicles may include additional sensors from other electronic systems such as the brake system, tire pressure, climate control, and transmission. Data is processed with on-board computers to monitor the health of the vehicle systems and provide warning diagnostics when there is an indication of malfunction.

Data: Data is collected from a 2014 Ford Focus (manual transmission) and a 2021 Chrysler Pacifica minivan with city and freeway driving:

Data sets 1 and 2 are collected with the ELM327 module with raw sensor data and calculated values that are collected with Car Scanner (App) from the OBD-II codes through a Bluetooth connection. Data set 3 is collected with an OBDLink MX+ OBD-II module through a Bluetooth connection.

  # warm-ups since codes cleared ()
  A/C pressure (kPa)
  Absolute load value (%)
  Absolute pedal position D (%)
  Absolute pedal position E (%)
  Absolute throttle position B (%)
  Air:fuel ratio ()
  Altitude (GPS) (feet)
  Ambient air temperature (℉)
  Average fuel consumption (MPG)
  Average fuel consumption (total) (MPG)
  Average fuel consumption 10 sec (MPG)
  Average speed (mph)
  Average speed (GPS) (mph)
  Barometric pressure (kPa)
  Calculated boost (bar)
  Calculated engine load value (%)
  Calculated instant fuel rate (gal./h)
  Catalyst temperature Bank 1 Sensor 1 (℉)
  Commanded evaporative purge (%)
  Commanded throttle actuator (%)
  Control module voltage (V)
  Distance traveled since codes cleared (miles)
  Distance traveled with MIL on (miles)
  Distance travelled (miles)
  Distance travelled (total) (miles)
  Engine coolant temperature (℉)
  Engine RPM (rpm)
  Ethanol fuel percent (%)
  Evap. system vapor pressure (Pa)
  Fuel economizer (based on fuel system status and throttle position) ()
  Fuel level input (%) (%)
  Fuel level input (V) (gallon)
  Fuel rail press. (kPa)
  Fuel used (gallon)
  Fuel used (total) (gallon)
  Fuel used price ($)
  Fuel used price (total) ($)
  Fuel/Air commanded equivalence ratio ()
  Instant engine power (based on fuel consumption) (hp)
  Intake air temperature (℉)
  Knock retard (°)
  Lambda ()
  Learned octane ()
  Long term fuel % trim - Bank 1 (%)
  MAF air flow rate (g/sec)
  OBD Module Voltage (V)
  Oxygen sensor 1 Wide Range Current (mA) (mA)
  Oxygen sensor 1 Wide Range Equivalence ratio ()
  Oxygen sensor 2 Bank 1 Short term fuel trim (%)
  Oxygen sensor 2 Bank 1 Voltage (V)
  Power from MAF (hp)
  Relative throttle position (%)
  Short term fuel % trim - Bank 1 (%)
  Speed (GPS) (mph)
  Throttle position (%)
  Throttle Position Actually (°)
  Throttle Position Desired (°)
  Timing advance (°)
  Transmission Odometer (℉)
  Vane position sensor (V)
  Variable camshaft actual advance #1 (°)
  Vehicle acceleration (g)
  Vehicle speed (mph)
  Latitude
  Longitude

Objective: Investigate driver and automobile performance. Use an OBD-II reader and collect data from a vehicle. Analyze the following:

Determine Light-off Time of the Catalytic Converter
Visualize the Speed and Elevation on a Map
Find maximum fuel economy versus speed

Example data is provided for each sample solution.

import pandas as pd
url = 'http://apmonitor.com/pds/uploads/Main/automotive.txt'
data = pd.read_csv(url)

[$[Get Code]]

Automotive Notebook

Jupyter Notebook in Google Colab

Box Plot

Parity Plot

Case Study 1: Catalyst Light-Off

The catalyst light-off temperature is a minimum temperature to initiate the catalytic reaction. Before light-off temperature is reached, the catalytic converter is less effective at reducing the pollutants. A typical light-off temperature is between 400 to 600 degrees F. The normal operating temperature is between 750 to 1,600 degrees F. The operating temperature increases with more pollutants in the exhaust.

Automotive Warmup Data

Calculate the time to catalyst light-off (500 degF) in seconds. Create an ARX model of the catalytic temperature and coolant temperature based on the vehicle speed and engine status.

Automotive Catalyst Light-Off Notebook

Jupyter Notebook in Google Colab

Case Study 2: GIS Map Visualization

Geographic Information Systems (GIS) are used to create, manage, analyze, and map data. GPS is used to identify the location of the vehicle. Car data is merged with location data in a single data file.

This exercise demonstrates mapping capabilities in Python with three options to create a map and display speed and elevation on the map.

Automotive Mapping Notebook

Jupyter Notebook in Google Colab

Case Study 3: Fuel Efficiency Regression

Fuel efficiency is affected by many factors such as speed, elevation changes, and aggressive acceleration. Calculate the fuel efficiency in miles per gallon (MPG) or Liters per 100 km (L/100 km). Use regression to develop a model of how measured factors relate to the fuel efficiency.

Automotive Fuel Efficiency Notebook

Jupyter Notebook in Google Colab

Solutions

Automotive Monitoring Solution in Python

Google Colab

Automotive Monitoring Solution with Matlab Live Script

Generative AI Learning

Use these prompts to test your understanding after completing the exercise. Direct the AI - do not let it do the analysis for you.

"Quiz me with 4 questions, one at a time, about catalytic converter light-off: why the catalyst must reach a minimum temperature (around 500 degF) before conversion begins, why the warm-up looks like a first-order dynamic response and what plays the role of the time constant and dead time, which OBD-II signals (vehicle speed, engine status, coolant temperature) drive the catalyst temperature and why, and why cold starts dominate total vehicle emissions. Grade my answers and list my misconceptions."

"I built an ARX time-series model that predicts catalyst and coolant temperature from vehicle speed and engine status: {paste your model form, coefficients, and fit metrics}. Critique it as a data-science reviewer: is the model order justified, does it extrapolate safely to drive cycles it has not seen, what does the residual pattern say, and how would I detect overfitting with a train/test split by trip? Ask me to defend the model before you suggest improvements."

"Ask me to predict, before I compute it: how the time to light-off changes for (a) a winter cold start at -10 degC, (b) an aggressive driver with early high RPM, and (c) a hybrid that shuts the engine off at stops. After I commit to predictions with reasoning, evaluate them against the physics and tell me which one my ARX model could NOT predict and why (training data coverage)."

Prompt planning: For the map and time-series visualizations, plan the prompt with the Agentic Visualization builder - specify the data source (OBD-II CSV), the message (when and where light-off occurs on the route), and the audience before asking for code.

What to Turn In

Submit a short report (PDF, 1-2 pages) that curates your results into a demonstration of what you learned. You may use Generative AI to help write the report and draft the code, but you must guide it to the correct visualizations, justifications, and assumptions. Answer these questions:

What is the time to catalyst light-off (500 degF) in seconds for the assigned data set? Include the temperature plot with the light-off point and the engine-start time marked.
Include your ARX model: structure, inputs, fit plot on training data, and validation on a separate trip or time window. Report a quantitative fit metric for both.
Does the catalyst warm-up behave like a first-order response? Estimate an approximate time constant from the data and state what physical effects the simple picture misses.
From the prediction prompt: which of your cold-start/driver/hybrid predictions was wrong or unverifiable, and what would you add to the data collection to answer it?
One claim or code snippet the AI produced that you corrected or verified, with the evidence.

Generative AI Learning

Use these prompts to test your understanding after completing the case studies. Direct the AI — you own the data analysis and the model. (If you arrived here from Process Dynamics and Control for the Catalytic Converter Light-off assignment, apply the prompts to Case Study 1.)

"Quiz me with 5 questions, one at a time, on the automotive monitoring case studies: what catalytic converter light-off means and why the time to reach 500 degF matters for cold-start emissions, what an ARX time-series model is (outputs regressed on past outputs and past inputs) and why past catalyst temperature must appear in the model, how to judge whether an ARX fit is any good beyond overlaying it on training data, why fuel economy versus speed has a maximum (aerodynamic drag at high speed, engine and transmission inefficiency at low speed), and what pitfalls to expect in real OBD-II data (dropouts, unit mix-ups, GPS glitches). Grade my answers and list my misconceptions."

"Here is my ARX model relating catalyst temperature to vehicle speed: {model orders and coefficients}. Interrogate it: is the model stable, what steady-state gain does it imply and is that physically sensible (degF per mph), and what does it predict for a cold start at zero speed? Ask me to validate on a data segment the model never saw and to report the error there vs the training segment — then help me interpret the gap. Do not just praise the fit."

Tip: The 62-channel OBD-II dataset rewards curiosity beyond the three case studies — air-fuel ratio during acceleration, oxygen sensor switching, elevation vs fuel flow. Pick one extra channel, form a hypothesis before plotting, and check it. Modern vehicle fleets run exactly this kind of monitoring continuously; the skill scales.

What to Turn In

Submit a short report (PDF, 1-2 pages) that curates your results into a demonstration of what you learned. You may use Generative AI to help write the report, but you must guide it to the correct visualizations, justifications, and assumptions. Answer these questions:

Report the time to catalyst light-off (500 degF) in seconds, with the plot and the exact criterion you used to detect it (first crossing, sustained crossing, interpolation between samples).
Include your ARX validation: predicted vs measured catalyst temperature on a segment not used for fitting, with the error quantified. Is the model good enough to predict light-off from a speed profile alone?
Include the speed vs fuel economy analysis: at what speed is this vehicle most efficient, and what physical effects explain the falloff on each side?
From the ARX interrogation prompt: what steady-state gain does your model imply, and did it survive the physical sanity check?

Machine Learning for Engineers

Automotive Monitoring

Case Study 1: Catalyst Light-Off

Case Study 2: GIS Map Visualization

Case Study 3: Fuel Efficiency Regression

Solutions

Generative AI Learning

What to Turn In

Generative AI Learning

What to Turn In