Analyzing Electric Motor Temperatures with ASAM ODS

This example Notebook shows how to implement Kirgsn example notebook EDA and Baseline on Motor Temperature Estimation from Kaggl using odsbox.

The data are collected by the LEA (Power Electronics and Electrical Drives) Department of the Paderborn University from a permanent magnet synchronous motor (PMSM) which is deployed on a test bench: Electric Motor Temperature.

The first sections are on initializing and connecting. The fun starts with “Load data”.

Dependencies for this notebook

The ASAM ODSBox contains some functionality that wraps the ODS HTTP API making using Python easier ;-)

try:  
  import odsbox
except:
  !pip install odsbox 
  
from odsbox.con_i import ConI
from odsbox.submatrix_to_pandas import submatrix_to_pandas

Electric Motor Temperature

Establish session

The ODS HTTP API is a session based API. The session ID is called conI in the ODS documentation. The ASAM ODSBox uses con_i as API object representing the session. Close this session to release the connection license. Otherwise the session will be auto closed after 30 minutes of inactivity.

conI = ConI(url='http://79.140.180.128:10032/api', auth=('Demo','mdm'))

Import Libs

import numpy as np # linear algebra
import pandas as pd # data processing
import matplotlib.pyplot as plt
from matplotlib.colors import rgb2hex
import seaborn as sns
from sklearn.decomposition import PCA

Load Data

The electric motor example data is stored in individual measurements starting with “Profile*” and we query für all (69) measurement (or more precisely their submatrices).

(We gave those measurements a random start date, which we use to order them by date…)

The mass data hasn’t been touched yet and we can examine it a bit before we actually load it - a bit like shape an steroids.

mea_lenghts = conI.query_data({
    "SubMatrix": {"measurement.test.Name": {"$like": "Profile*"}},
    "$attributes": {
        "measurement.Name": 1,
        "SubMatrixNoRows": 1
    },
    "$orderby": {
        "measurement.measurement_begin": 1
    }
})

mea_lenghts = mea_lenghts.set_index(mea_lenghts.columns[0])
print(mea_lenghts.head())

ax = mea_lenghts.plot.barh(figsize=(10, 20), title='Sample size')
_ = ax.set_xticks(2*3600*np.arange(1, 8)) # 2Hz sample rate
_ = ax.set_xticklabels(list(range(1, 8)))
_ = ax.set_xlabel('Time in hours')

                SubMatrix.SubMatrixNoRows
MeaResult.Name                           
Profile_17                          15964
Profile_05                          14788
Profile_12                          21942
Profile_32                          20960
Profile_21                          17321

ax = mea_lenghts.plot.hist(title='Sample size distribution', bins=50, figsize=(10, 5), grid=True)
_ = ax.set_xticks(2*3600*np.arange(1, 8)) # 2Hz sample rate
_ = ax.set_xticklabels(list(range(1, 8)))
_ = ax.set_xlabel('Time in hours')

Distributions

Until now, we haven’t loaded any mass data, yet.

In our data model, the individual profiles are not contained in a big single dataset. They are stored profile by profile in so called submatrices (we looked at them before a bit). So we go and fetch the first profile returned as DataFrame by using submatrix_to_dataframe.

data_matrices = conI.query_data({
    "SubMatrix": {"measurement.name": {"$like": "Profile_68*"}},
    "$attributes": {
        "measurement.Name": 1,
        "Id": 1
    },
    "$orderby": {
        "measurement.measurement_begin": 1
    }
})

# just pick the first one
dm_id = data_matrices["SubMatrix.Id"][0]
profile_name = data_matrices["MeaResult.Name"][0]

print(profile_name)

dm_df = submatrix_to_pandas(conI, dm_id)
if "Time" in dm_df.columns:
    dm_df.set_index("Time", inplace=True)
elif "time" in dm_df.columns:
    dm_df.set_index("time", inplace=True)
dm_df.head()

Profile_68

	U_q	Coolant	Stator_winding	U_d	Stator_tooth	Motor_speed	I_d	I_q	Pm	Stator_yoke	Ambient	Torque
Time
0.0	-1.250622	26.948739	26.401552	-0.269267	26.165591	-0.003114	-2.000875	1.097745	28.011248	26.141670	26.280088	2.785112e-07
0.5	-0.510442	26.950800	26.396721	0.110050	26.172235	21.150647	-4.751546	-14.421556	28.002123	26.155253	26.275925	-1.122167e+01
1.0	-0.388580	26.953776	26.408694	0.907830	26.177153	42.000488	-15.815420	-49.228154	27.996198	26.159354	26.280659	-3.799358e+01
1.5	-0.260576	26.962654	26.401838	1.678359	26.181832	60.023822	-25.347121	-77.327501	27.993540	26.173699	26.280193	-5.986944e+01
2.0	0.134215	26.948778	26.396926	2.581367	26.172657	80.657265	-32.076248	-97.315560	28.001602	26.181490	26.277929	-7.541367e+01

… and now plot the distribution ….

target_features = ['Pm', 'Stator_tooth', 'Stator_yoke', 'Stator_winding']

# prepare colors
color_list = plt.cm.tab10(np.linspace(0, 1, 10)[list(range(10))+[0, 1]])
coi = target_features + [c for c in dm_df if c not in target_features]  # columns of interest
feat_clrs = {k: rgb2hex(color_list[i][:3]) for i, k in enumerate(coi)} if color_list is not None else {}

n_cols = 4
n_rows = np.ceil(dm_df.shape[1] / n_cols).astype(int)
fig, axes = plt.subplots(n_rows, n_cols, figsize=(2.8*n_cols, n_rows*3))
for i, (ax, col) in enumerate(zip(axes.flatten(), list(dm_df.columns))):
    sns.histplot(dm_df[col], kde=True, stat="density", ax=ax, color=feat_clrs[col], bins=30)
    if i % n_cols == 0:
        ax.set_ylabel('Frequency')
plt.tight_layout()

Time series gestalt

First we’ll get an overview of the target temperature trends.

dm_df = submatrix_to_pandas(conI, dm_id)
if "Time" in dm_df.columns:
    dm_df.set_index("Time", inplace=True)
elif "time" in dm_df.columns:
    dm_df.set_index("time", inplace=True)

coi = target_features
fig, ax = plt.subplots()
for c in coi:
    lines = ax.plot(dm_df[c], label=c, color=feat_clrs[c])
ax.set_title(profile_name)
ax.set_ylabel('Temp in °C')
ax.set_xlabel('Time in sec')
    
fig.tight_layout()
# _ = ax.legend(ncol=15, loc='lower center', bbox_to_anchor=(.5, 1), bbox_transform=fig.transFigure)

Now we have a look at the input features

coi = [c for c in dm_df if c not in target_features]
max_values_per_col = dm_df.abs().max(axis=0)
fig, ax = plt.subplots()
for c in coi:
    lines = ax.plot(dm_df[c]/max_values_per_col[c], label=c, color=feat_clrs[c])
ax.set_title(profile_name)
ax.set_ylabel('Normalized feature')
ax.set_xlabel('Time in sec')
fig.tight_layout()
#_ = ax.legend(ncol=15, loc='lower center', bbox_to_anchor=(.5, 1), bbox_transform=fig.transFigure)

_ = dm_df[target_features].rolling(200).mean().plot(kind="line", subplots=False, title=profile_name)
_ = dm_df.drop(target_features,axis=1).rolling(200).mean().plot(kind="line", subplots=True, layout=(4,2), title=profile_name)

Operation points

The operation point of a motor is often explained by its location in the motor_speed-torque-plane.

fig, ax = plt.subplots()
ax.plot(dm_df['Motor_speed'], dm_df['Torque'])
ax.set_title(profile_name)
ax.set_ylabel('Torque in Nm')
ax.set_xlabel('Motor speed in rpm')

Text(0.5, 0, 'Motor speed in rpm')

We see that some driving cycles only capture little of the valid operation region, while other profiles do comprehensive random walks over the full operative range.

Note that motors are power rated, and since power is defined as P=motorspeed⋅torque, there are elliptical borders that can’t be exceeded.

Dimensionality reduced visualization

Depict the sessions principal component axes, shifting the color from blue to red as the permanent magnet temperature rises.

Showing the two most significant principal components.

# normalize
dm_df = submatrix_to_pandas(conI, dm_id)

dm_df = dm_df / dm_df.abs().max(axis=0)
transformed = PCA().fit_transform(dm_df.drop(target_features, axis=1))
sess_df = dm_df
_trans = transformed[sess_df.index, :]
plt.scatter(_trans[:, 0], _trans[:, 1], c=dm_df.loc[sess_df.index, 'Pm'].values, cmap=plt.get_cmap('coolwarm'), marker='.', vmin=dm_df['Pm'].min(), vmax=dm_df['Pm'].max())
plt.xlim(-1, 1)
plt.ylim(-1, 1)
plt.tick_params(axis='both', which='both', bottom=False, top=False,
                        labelbottom=False, right=False, left=False,
                        labelleft=False)
plt.tight_layout()

Close session

Don’t forget to close the session to release the connection license. Otherwise the session will be auto closed after 30 minutes of inactivity.

conI.logout()

Acknowledgement

The data used by this notebook is based on the Kaggle Electric Motor Temperature data provided by Paderborn University under the CC BY-SA 4.0-License

License

Copyright © 2024 Peak Solution GmbH

The training material in this repository is licensed under a Creative Commons BY-NC-SA 4.0 license. See LICENSE file for more information.