ODSC Hackathon

Timeline:

• Optional training classes (for Bootcamp passes only): starting on July 29th
  Pre conference training for ODSC Europe 2020
  Pre conference training for ODSC West 2020
• Submissions due:  August 25th
• Judging: August 26th – September 4th
• Winners announced: September 11th
• Winners recognition: At ODSC Europe 2020 and ODSC West 2020

Challenge: Improving Efficiency & Production Process of Electric Vehicles using Data Science Techniques

This challenge has been designed to provide with you hands-on understanding of data science problems in commercial EVs’ production and optimization in most advanced motor technologies used by companies like Tesla, BMW and Ford.

It is often a challenging and complex task to measure rotor and stator temperatures in commercial electric vehicles. Even if these specific tasks can be completed successfully, these testing processes cannot be classified as economical for manufacturers. Keeping in mind that the temperature data have significant importance on dynamical responses of vehicles and motors’ performances, there is an emerging need for new proposals and scientific contributions in this domain.

Consider, one manufacturer of electric cars hired you to propose an estimator for the stator and rotor temperatures and design a predictive machine learning or deep learning model. Such a model could significantly help your new company to utilize new control strategies of the motors and maximize their operational performances. If you build an accurate ML/DL model, the needs of the company for implementing additional temperature sensors in vehicles will be reduced. The potential contribution will directly result in lowering car construction and maintenance costs, and will convince the company to invest further in hiring DS experts like you.

Initial considerations

• The motors are excited by reference torques and reference velocities. These reference signals are achieved by adjusting motor currents (“i_d” and “i_q”) and voltages (“u_d” and “u_q”) within appropriate control strategy.
• Temperature estimations should be real-time, and not based on future values for current predictions. Real-time predictions shall protect the motor from overheating.
• The motor torque increases in inverse proportion to the decreased temperature.
• A steady-state of a motor can be achieved faster at lower temperatures.
• Phase currents increase with increased magnet temperature.

Dataset

• Training dataset
• Test dataset
• Solution dataset (to calculate RMSE)

Each row in the csv files represents complete measurement information from sensors in one time step and one row is recorded every 0.5 seconds. Individual measurement sessions last between 1 and 6 hours and can be identified with the “profile_id” column. The following table provides variables of interest and their short descriptions.

The above dataset is sourced from the following publications:

Kirchgässner, Wilhelm & Wallscheid, Oliver & Böcker, Joachim. (2019). Empirical Evaluation of Exponentially Weighted Moving Averages for Simple Linear Thermal Modeling of Permanent Magnet Synchronous Machines. 

Kirchgässner, Wilhelm & Wallscheid, Oliver & Böcker, Joachim. (2019). Deep Residual Convolutional and Recurrent Neural Networks for Temperature Estimation in Permanent Magnet Synchronous Motors.

Rules To Participate and Submit Solution:

1. Create a compelling notebook of your analysis and prediction that allows your manager to better understand your approach. Attach a video screencast explaining the above.
2. Submit your prediction results of your test dataset with four below variables (in csv file) and be sure to name them “predicted_temperatures“.
3. Unique IDs of these sessions are not presented in the Test dataset as they are within Training dataset, so be careful. Don’t switch the rows within the test data frame and use all the measurements in the established order.
   

   Variable	Description
   pm_predicted	Predicted rotor temperature
   stator_yoke_predicted	Predicted stator yoke temperature
   stator_tooth_predicted	Predicted stator tooth temperature
   stator_winding_predicted	Predicted stator winding temperature

4. Calculate the overall Root Mean Square Error (RMSE) by adding the RMSE of each of the examined variables with the help of the solution dataset, and name them: RMSE_pm, RMSE_stator_yoke, RMSE_stator_tooth, RMSE_stator_winding. 
5. Projects must be posted as a Github repository with your code, results of test dataset and RMSE. No pre-existing projects will be accepted. Submissions must be original work of you and your team.
6. Form a team of up to 4 people or participate by yourself.

Evaluation Criteria:

1. Code structure/quality
2. Data mining
3. Findings and explanations
4. Predictions and performance of the model