Next-Gen Battery Intelligence & Automation

This repository delves into cutting-edge advancements in battery intelligence and automation, showcasing a detailed process to highlight these innovations.

Through our work, we have achieved significant cost savings of $85K annually by automating the Python pipeline with AWS stack (Redshift, QuickSight) and data science libraries (Numpy, Pandas, Scikit-learn). This automation effectively replaced a full-time Data Analyst role and reduced analysis time by an impressive 95%, bringing it down from 40 hours to just 2 hours weekly.

Additionally, we have boosted battery life prediction accuracy by 10 percentage points, increasing it from 79% to 89%. This improvement was realized through ensemble modeling techniques (XGBoost, RandomForest, SVM) and innovative feature engineering, including the use of cell-pack resistance ratios and thermodynamic calculations.

Furthermore, our work has substantially reduced testing costs, slashing $500K annually by detecting failing batteries within their first 100 test cycles. This was achieved through statistical modeling of key electrical parameters, which reduced overall testing time by 80%.

I'm happy to discuss more about this project and the fascinating techniques employed. If you have any specific questions or need further information, feel free to reach out! 😊

Disclaimer: This is a novel work in battery intelligence, and due to a Non-Disclosure Agreement (NDA), I cannot share all minor details. This is a redacted version of how automation generally works, and how feature engineering is done in batteries using electrical parameter measurements to enhance machine learning predictive power.

Table of Contents

• Prerequisites
• Files and Directories
  • Python Files
    • parse_compact.py
    • CycloidConnector.py
    • adhoc_analysis.py
      • Methods
        • outlier_handler
        • get_discharge_energy
        • ttest
        • get_statistics_before_after_rpt
        • get_energy_cycle_gain
        • get_dcir
        • plot_dcir
        • plot_mean_voltage
        • plot_rover_metrics
        • Data Filtering Methods
        • Voltage Analysis Methods
        • Temperature Analysis Methods
  • Jupyter Notebooks
    • adhoc_worksheet.ipynb
    • dump_data_into_psql.ipynb
    • error_in_current_0.5_6A.ipynb
    • error_measurement_estimation.ipynb
    • pack_voltage_capacity_per_current_step.ipynb

Prerequisites

• Python 3.x
• Pandas
• NumPy
• Seaborn
• Matplotlib
• Amazon Web Service (AWS) Command Line Service (CLI)
• Redshift
• tqdm
• boto3
• bz2
• json
• pingouin

You can install all required dependencies using:

pip install -r requirements.txt

Files and Directories

Python Files

python files/parse_compact.py

The parse_bz2 function reads and processes compressed JSON data from Rockpi raw log messages stored in .bz2 files. It transforms the data into a structured DataFrame for analysis.

Dependencies:

• bz2
• json
• pandas

Installation:

pip install pandas

python files/CycloidConnector.py

The VWCycloid class interacts with AWS Redshift to retrieve and process cycloid data using boto3 and pandas.

Dependencies:

• boto3
• pandas
• tqdm

Installation:

pip install boto3 pandas tqdm

python files/adhoc_analysis.py

The adhoc class streamlines and automates daily ad-hoc analysis requests. Below are its key methods:

Methods

outlier_handler

Handles outliers in a pandas Series using rolling median replacement.

• Parameters:
  • series: Input Series
  • window: Window size (default: 5)
  • sigma_multiplier: Outlier boundary definition (default: 0.1)
• Returns: Series with handled outliers

get_discharge_energy

Retrieves discharge energy data with median and lower bound energy.

• Parameters:
  • df: Input cycloid DataFrame
  • window: Window size (default: 7)
  • sigma_multiplier: Outlier boundary (default: 0.1)
  • disclaimer: Print energy jump disclaimer (default: True)
  • raw: Return raw discharge energy (default: False)
• Returns: DataFrame with discharge energy data

ttest

Performs t-test for A/B testing.

• Parameters:
  • sample1: First sample
  • sample2: Second sample
  • alternative: Test hypothesis ("two-sided", "greater", "less")
  • interprete: Print interpretation (default: True)
• Returns: DataFrame with statistical parameters

get_statistics_before_after_rpt

Calculates statistics around RPT cycles.

• Parameters:
  • df: Input DataFrame
  • rpt_cycle: RPT cycle number (default: 201)
  • neighbors_cycle: Cycles to consider (default: 3)
  • window: Rolling window size (default: 5)
  • sigma_multiplier: Outlier detection (default: 0.1)
  • avoid_close_cycle: Cycles to avoid (default: 0)
• Returns: DataFrame with cycle statistics

get_energy_cycle_gain

Calculates energy gains around RPT cycles.

• Parameters:
  • df: Input DataFrame
  • rpt_cycle: Reference cycle (default: 201)
  • neighbors_cycle: Cycles to consider (default: 5)
  • avoid_close_cycle: Cycles to avoid (default: 0)
  • window: Window size (default: 5)
  • sigma_multiplier: Outlier detection (default: 0.1)
  • show_plot: Display plot (default: True)
• Returns: DataFrame with energy statistics

get_dcir

Retrieves DCIR data.

• Parameters:
  • df: Input cycloid DataFrame
• Returns: DataFrame with DCIR data

plot_dcir

Plots DCIR data over cycles.

• Parameters:
  • df: Input DataFrame
  • hue: Color grouping (default: 'instance_number')
  • label_title: Legend title (default: 'Instance Number')
  • ax: Matplotlib axes object

plot_mean_voltage

Plots mean voltage data.

• Parameters:
  • df: Input DataFrame
  • hue: Color grouping (default: 'instance_number')
  • charging_state_name: Substate to plot (default: 'CHARGE')
  • label_title: Legend title
  • raw_data: Plot raw data (default: True)
  • window: Window size (default: 5)
  • sigma_multiplier: Outlier detection (default: 0.2)
  • ax: Matplotlib axes object

plot_rover_metrics

Creates subplot dashboard of up to 6 metrics.

• Parameters:
  • df: Input DataFrame
  • x: X-axis column (default: 'elapsed_minutes')
  • charging_state_name: Charging state (default: 'CHARGE')
  • hue: Color grouping
  • suptitle: Overall title
  • label_title: Legend title
  • legend: Show legend (default: True)

Data Filtering Methods

• read_data: Reads and processes data files
• filter_by_doe: Filters by DOE values
• filter_by_instance: Filters by instance values
• filter_by_cycle: Filters by cycle numbers
• filter_by_doe_instance: Filters by DOE-instance combinations

Voltage Analysis Methods

• get_cells_voltage_std: Calculates voltage standard deviation
• plot_cells_voltage_std: Plots voltage standard deviation
• plot_cells_voltage: Visualizes cell voltages
• get_polarization_voltage: Calculates polarization voltage
• plot_polarization_voltage: Plots polarization voltage

Temperature Analysis Methods

• plot_temperature: Visualizes temperature data across cycling states

Jupyter Notebooks

jupyter notebooks/adhoc_worksheet.ipynb

Example usage of adhoc class methods:

# User input configuration
doe_number = 'MTSOWPhase2_Pack25CValidation'
instance_number = None
cycles = None
default_columns = True
override_query = None

# Fetch data
dfc = redshift.get_cycloid_data(
    doe=doe_number,
    instances=instance_number,
    cycles=cycles,
    default_columns=default_columns,
    override_query=override_query
)

# Analysis
dcir = adhoc.get_dcir(dfc)
discharge_energy = adhoc.get_discharge_energy(dfc)

jupyter notebooks/dump_data_into_psql.ipynb

Handles data dumping to Redshift database with these key components:

• DBCredentials: Dataclass for database credentials
• load_credentials: Loads credentials from JSON
• check_database_exists: Verifies database existence
• create_db_and_dump_data: Creates database and imports data

jupyter notebooks/error_in_current_0.5_6A.ipynb

Features the CapacityTimeUncertainty class for analyzing cycloid data:

• Calculates capacity and time uncertainties
• Handles data loading and parameter management
• Processes capacity bounds and uncertainties
• Implements data cleaning and outlier removal

jupyter notebooks/error_measurement_estimation.ipynb

Contains detailed data preprocessing for CapacityTimeUncertainty calculations.

jupyter notebooks/pack_voltage_capacity_per_current_step.ipynb

Explains error limit variations across different current regimes for charge time and capacity uncertainty estimation.