Source code for ampworks.ici._extract_params

from __future__ import annotations

from typing import TYPE_CHECKING

import numpy as np
import pandas as pd

from scipy.stats import linregress

if TYPE_CHECKING:  # pragma: no cover
    from ampworks import Dataset


[docs] def extract_params(data: Dataset, radius: float, tmin: float = 1, tmax: float = 10, return_stats: bool = False) -> Dataset: """ Extracts parameters from ICI data. ICI, or incrememtal current interruption, is an experiment that interrupts low-rate charge or discharge experiments with short rests. The experiments can be used to extract important parameters for physics-based models. For example, a pseudo open-circuit voltage and solid-phase diffusivity. The following protocol was used to test this algorithm: 1. Rest for 5 min, log data every 10 s. 2. Charge at C/10 for 5 min; with a voltage limit. Log every 5 s or 5 mV. 3. Rest for 10 seconds, log data every 0.1 s. 4. Stop if voltage limit reached in (2), otherwise repeat (2) and (3). The protocol assumes formation cycles have already been completed and that the cell was rested until equilibrium before starting the steps above. Implementation details are available in [1]_. This specific protocol assumes the cell starts at a fully discharged state and only includes charge pulses; however, you can similarly perform the experiment in the discharge direction or in both directions. The only change would be to step (2) where you would discharge at C/20 instead of charge. It is common to perform the ICI tests in both directions, but you must process the charge and discharge segments separately by slicing your data and calling this routine twice. Parameters ---------- data : Dataset The sliced ICI data to process. Must have, at a minimum, columns for `{'Seconds', 'Amps', 'Volts'}`. See notes for more information. radius : float The representative particle radius of your active material (in meters). It's common to use D50 / 2, i.e., the median radius of a distribution. tmin : float, optional The minimum relative rest time (in seconds) to use when fitting sqrt(t) vs. voltage for time constants. Default is 1. tmax : float, optional The maximum relative rest time (in seconds) to use when fitting sqrt(t) vs. voltage for time constants. Default is 10. return_stats : bool, optional If False (default), only the extracted parameters vs. state of charge are returned. If True, also returns stats with info about each rest. Returns ------- params : Dataset Table of parameters. Columns include 'SOC' (state of charge, -), 'Ds' (diffusivity, m2/s), and 'Eeq' (equilibrium potential, V). stats : Dataset Only returned if `return_stats=True`. Provides additional stats about each rest, including errors from the sqrt(t) vs. voltage regressions. Raises ------ ValueError 'data' is missing columns, required=['Seconds', 'Amps', 'Volts']. ValueError 'data' should not include both charge and discharge segments. Notes ----- Rests within the dataset are expected to have a current exactly equal to zero. You can use `data.zero_below('Amps', threshold)` to manually zero out currents below some tolerance, if needed. This should be done prior to passing in the dataset to this function. This algorithm expects charge/discharge currents to be positive/negative, respectfully. If your sign convention is the opposite, the mapping to 'SOC' in the output will be reversed. You must process data in one direction at a time. In other words, if you performed the ICI protocol in both charge and discharge directions you should slice your data into two datasets and call this routine twice. The default `tmin` and `tmax` values assume that rests occur for at least 10 s. You should adjust these accordingly if you use a protocol with shorter rests. Also, if a rest has fewer than two data points between the set relative `tmin` and `tmax` then the linear regression performed to find the diffusivity and equilibrium potential will return `NaN` for both. References ---------- .. [1] Z. Geng, Y. Chien, M. J. Lacey, T. Thiringer, and D. Brandell, "Validity of solid-state Li+ diffusion coefficient estimation by electrochemical approaches for lithium-ion batteries," EA, 2022, DOI: 10.1016/j.electacta.2021.139727 Examples -------- >>> import ampworks as amp >>> data = amp.datasets.load_datasets('ici/ici_discharge') >>> params, stats = amp.ici.extract_params(data, 1.8e-6, return_stats=True) >>> params.plot('SOC', 'Eeq') >>> params.plot('SOC', 'Ds', logy=True) >>> print(params) >>> print(stats) """ import ampworks as amp from ampworks._checks import _check_columns, _check_only_one from ampworks._auxiliary import _infer_state, _calc_soc, _calc_relative_time _check_columns(data, {'Seconds', 'Amps', 'Volts'}) charging = any(data['Amps'] > 0.) discharging = any(data['Amps'] < 0.) _check_only_one( conditions=[charging, discharging], message="'data' cannot include both charge and discharge segments.", ) ds = data.copy() ds = ds.reset_index(drop=True) # States based on current direction: charge, discharge, or rests _infer_state(ds) # Add in state-of-charge column to map each value to an SOC _calc_soc(ds, charging) # Count each time a rest/charge or rest/discharge changeover occurs rest = (ds['State'] != 'R') & (ds['State'].shift(fill_value='R') == 'R') ds['Rest'] = rest.cumsum() # Relative time of each rest/charge or rest/discharge step _calc_relative_time(ds, ['Rest', 'State'], col_name='StepTime') # Remove last cycle if not complete, i.e., ended on charge or discharge if ds.iloc[-1]['State'] != 'R': ds = ds[ds['Rest'] != ds['Rest'].max()].reset_index(drop=True) # Record summary stats for each loop, immediately before the rests groups = ds[ds['State'] != 'R'].groupby('Rest', as_index=False) summary = groups.agg(lambda x: x.iloc[-1]) # Store slope and intercepts (V = m*t^0.5 + b) for each rest groups = ds.groupby('Rest') regression = None for idx, g in groups: if idx > 0: rest = g[g['State'] == 'R'] pulse = g[g['State'] != 'R'] dt_rest = rest['StepTime'].max() - rest['StepTime'].min() dt_pulse = pulse['StepTime'].max() - pulse['StepTime'].min() rest = rest[ (rest['StepTime'] >= tmin) & (rest['StepTime'] <= tmax) ] x = np.sqrt(rest['StepTime']) y = rest['Volts'] if len(x) <= 1: x, y = [0, 1], [np.nan, np.nan] result = linregress(x, y) new_row = pd.DataFrame({ 'Rest': [idx], 'Eeq': [result.intercept], 'Eeq_err': [result.intercept_stderr], 'dUdrt': [result.slope], 'dUdrt_err': [result.stderr], 'dt_rest': [dt_rest], 'dt_pulse': [dt_pulse], }) regression = pd.concat([regression, new_row], ignore_index=True) stats = pd.merge(summary, regression, on='Rest') stats['dEdt'] = np.gradient(stats['Volts'], stats['Seconds']) params = amp.Dataset({ 'SOC': stats['SOC'], 'Ds': 4./9./np.pi * (radius * stats['dEdt']/stats['dUdrt'])**2, 'Eeq': stats['Eeq'], }) params.sort_values(by='SOC', inplace=True, ignore_index=True) if return_stats: return params, amp.Dataset(stats) return params