Add Thermal Neural Network (TNN) SciML example by mmarkows17 · Pull Request #18 · matlab-deep-learning/SciML-and-Physics-Informed-Machine-Learning-Examples

mmarkows17 · 2026-03-10T19:48:41Z

Adds a Thermal Neural Network (TNN) example demonstrating SciML-based virtual temperature sensor modeling.

bwdGitHub · 2026-03-11T17:19:20Z

thermal-neural-network/DiffEqLayer.m

+    % This layer holds a TNNCell instance, and performs prediction over
+    % tbptt_size time steps.
+    % 
+    %   Copyright 2025 The MathWorks, Inc.


We tend to leave the copyright out of the m-help so it doesn't display on help DiffEqLayer

bwdGitHub · 2026-03-11T17:26:16Z

thermal-neural-network/DiffEqLayer.m

+classdef DiffEqLayer <  nnet.layer.Layer & nnet.layer.Formattable
+    % DiffEqLayer   Differential equation layer


I don't really see a differential equation when I look at the predict implementation - I see either something like recurrence or autoregressive behaviour. I guess in the context of this method it can be seen like an ODE solve.

bwdGitHub · 2026-03-11T17:26:31Z

thermal-neural-network/DiffEqLayer.m

+    % DiffEqLayer   Differential equation layer
+    % 
+    % This layer holds a TNNCell instance, and performs prediction over
+    % tbptt_size time steps.


tbptt_size isn't defined, and it looks like we loop over all steps.

bwdGitHub · 2026-03-11T17:30:11Z

thermal-neural-network/DiffEqLayer.m

+            numSteps = size(input, 3); % Assuming input is [features, batch, time]
+
+            % Preallocate the outputs:
+            outputs = dlarray(zeros(size(state,[1 2 3]),'single'),'CBT');


I'd use zeros(sz, like=input) here. Besides taking the precision from the input I think it also helps out the dlarray autodiff system.

bwdGitHub · 2026-03-11T17:30:45Z

thermal-neural-network/generatePlots.m

@@ -0,0 +1,37 @@
+function generatePlots(profileLengths,test_tensor,prediction,scale,targetNames)


Mix of conventions with test_tensor and targetNames

bwdGitHub · 2026-03-11T17:35:11Z

thermal-neural-network/generatePlots.m

+        mseError = mse(groundTruth,currentPred);
+        plot(0:profileLengths(ii)-1,groundTruth, ...
+            Color = "green",LineWidth = 1.5)
+        hold on


Should there be a hold off somewhere too? I think running this function will leave the current figure with a hold on the axes, so any subsequent plot commands by the user would end up on the same axes unless they create a fresh one or turn hold off themselves.

bwdGitHub · 2026-03-11T17:38:42Z

thermal-neural-network/README.md

+-  Stator tooth 
+-  Stator winding 
+
+![image_0.png](README_media/image_0.png)


A descriptive alt text would be better.

bwdGitHub · 2026-03-11T17:41:33Z

thermal-neural-network/README.md

+
+A simplified discrete state equation for each component $i$ of a LPTN model is:
+
+ $$ T_i (k+1)=T_i (k)+t_s \kappa_i (k)\left(\pi_i (k)+\sum_{m\;\in \;components} \gamma_{im} (k)\Delta T_{im} (k)+\sum_{n\;\in \;ancillary} \gamma_{in} (k)\Delta T_{in} (k)\right) $$ 


The stuff below the sums didn't render so well on the GitHub diff.

bwdGitHub · 2026-03-11T17:48:55Z

thermal-neural-network/TNNCell.m

+
+    methods
+
+        function this = TNNCell(inputStruct)


Using a struct as an input doesn't really tell a user anything about what fields that struct has to contain. This would be clearer with required arguments and optional NVPs with an arguments block.

jess-42

Taken a pass over the repo, focusing on the ReadMe.

jess-42 · 2026-03-18T13:26:13Z

thermal-neural-network/README.md

+
+The data set contains various operational measurements from an electric motor, including currents, voltages, torque, speed, and ambient conditions. The TNN model learns to estimate the internal temperatures that would typically require embedded temperature sensors, which are often impractical in production motors.
+
+This example is intended to be an instructional resource demonstrating how to build and train a TNN in MATLAB. For integration and use with Simulink, see the Github repository [here](https://github.com/wkirgsn/thermal-nn/tree/main/matlab). For video content on this topic, check out this recorded MATLAB Expo talk on [Thermal Neural Network for Temperature Modeling in E-Motors](https://www.mathworks.com/videos/thermal-neural-network-for-temperature-modeling-in-e-motors-1762251869161.html) and this [YouTube video](https://youtu.be/ySRjX9GUiCA?si=s7slG-9Ekt6y9wUe). 


Can we spell out where the link goes to? I.e.,

For an example showing how to integrate and use this model with Simulink, see the Thermal neural network for electric motor temperature estimation GitHub repository.

jess-42 · 2026-03-18T13:35:22Z

thermal-neural-network/README.md

+
+The data set contains various operational measurements from an electric motor, including currents, voltages, torque, speed, and ambient conditions. The TNN model learns to estimate the internal temperatures that would typically require embedded temperature sensors, which are often impractical in production motors.
+
+This example is intended to be an instructional resource demonstrating how to build and train a TNN in MATLAB. For integration and use with Simulink, see the Github repository [here](https://github.com/wkirgsn/thermal-nn/tree/main/matlab). For video content on this topic, check out this recorded MATLAB Expo talk on [Thermal Neural Network for Temperature Modeling in E-Motors](https://www.mathworks.com/videos/thermal-neural-network-for-temperature-modeling-in-e-motors-1762251869161.html) and this [YouTube video](https://youtu.be/ySRjX9GUiCA?si=s7slG-9Ekt6y9wUe). 


Similar to other comments, it is good to say where a link is going, rather than just "YouTube video". How about:

For video content on this topic, see the recorded MATLAB Expo talk on Thermal Neural Network for Temperature Modeling in E‑Motors, as well as a related YouTube video.

jess-42 · 2026-03-18T13:38:59Z

thermal-neural-network/README.md

+
+# References
+<a id="M_560c"></a>
+1.  Wilhelm Kirchgässner, Oliver Wallscheid, Joachim Böcker, Thermal neural networks: Lumped\-parameter thermal modeling with state\-space machine learning, Engineering Applications of Artificial Intelligence 117 (2023) 105537. [https://doi.org/10.1016/j.engappai.2022.105537](https://doi.org/10.1016/j.engappai.2022.105537)


Kirchgässner, Wilhelm, et al. “Thermal Neural Networks: Lumped-Parameter Thermal Modeling with State-Space Machine Learning.” Engineering Applications of Artificial Intelligence, vol. 117, Jan. 2023, p. 105537. DOI.org (Crossref), https://doi.org/10.1016/j.engappai.2022.105537.

jess-42 · 2026-03-18T13:40:37Z

thermal-neural-network/README.md

+
+
+
+The data set contains measurements from various sensors on an electric motor across multiple test profiles. Each row represents a measurement at a specific time point, and each column represents a different sensor or calculated value. In addition to the measured data the `profile_id` column identifies different test profile. You can use the `profile_id` value to separate the training and testing data. 


Add comma and plural:
"In addition to the measured data, the profile_id column identifies different test profiles."

jess-42 · 2026-03-18T13:43:49Z

thermal-neural-network/README.md

+end
+```
+
+Normalize all non\-temperature input columns by dividing by the maximum values.


Is "by their maximum values" clearer?

jess-42 · 2026-03-18T15:13:18Z

thermal-neural-network/TNNCell.m

@@ -0,0 +1,142 @@
+classdef TNNCell <  nnet.layer.Layer


It might be nice throughout to give a 1-line about what each function does, similar to prepareSequenceData.m

jess-42 · 2026-03-20T16:41:44Z

thermal-neural-network/README.md

+Run the example by running [`ThermalNeuralNetwork.m`](ThermalNeuralNetwork_code.m).
+
+# **Requirements**
+MATLAB version R2025a is strongly recommended for full functionality. At a minimum, this example requires MATLAB release R2024a or newer as well as Deep Learning Toolbox. 


What do I miss out out by not having the latest version? Does this example also only require MATLAB + DLT?

jess-42 · 2026-03-20T16:42:50Z

thermal-neural-network/README.md

+
+This example is intended to be an instructional resource demonstrating how to build and train a TNN in MATLAB. For integration and use with Simulink, see the Github repository [here](https://github.com/wkirgsn/thermal-nn/tree/main/matlab). For video content on this topic, check out this recorded MATLAB Expo talk on [Thermal Neural Network for Temperature Modeling in E-Motors](https://www.mathworks.com/videos/thermal-neural-network-for-temperature-modeling-in-e-motors-1762251869161.html) and this [YouTube video](https://youtu.be/ySRjX9GUiCA?si=s7slG-9Ekt6y9wUe). 
+
+# **Setup**


Consider changing this to "Get Started" as this is all you need to do to get started with the repo.

jess-42 · 2026-03-20T16:44:50Z

thermal-neural-network/README.md

+<a id="M_560c"></a>
+1.  Wilhelm Kirchgässner, Oliver Wallscheid, Joachim Böcker, Thermal neural networks: Lumped\-parameter thermal modeling with state\-space machine learning, Engineering Applications of Artificial Intelligence 117 (2023) 105537. [https://doi.org/10.1016/j.engappai.2022.105537](https://doi.org/10.1016/j.engappai.2022.105537)
+
+# **Prepare Data for Training**


I'm assuming you have the full example here for discoverability, but it does give a bit of whiplash to suddenly start preparing data after reading the references.

jess-42 · 2026-03-20T16:45:52Z

thermal-neural-network/README_media/figure_0.png

Consider resizing the plot during training, this will make the text easier to read.

Add Thermal Neural Network (TNN) SciML example

6e34595

mmarkows17 requested review from bwdGitHub and conordaly0 March 10, 2026 19:48

bwdGitHub reviewed Mar 11, 2026

View reviewed changes

mmarkows17 requested a review from jess-42 March 13, 2026 16:09

jess-42 reviewed Mar 20, 2026

View reviewed changes

		classdef DiffEqLayer < nnet.layer.Layer & nnet.layer.Formattable
		% DiffEqLayer Differential equation layer

		@@ -0,0 +1,37 @@
		function generatePlots(profileLengths,test_tensor,prediction,scale,targetNames)


		A simplified discrete state equation for each component $i$ of a LPTN model is:

		$$ T_i (k+1)=T_i (k)+t_s \kappa_i (k)\left(\pi_i (k)+\sum_{m\;\in \;components} \gamma_{im} (k)\Delta T_{im} (k)+\sum_{n\;\in \;ancillary} \gamma_{in} (k)\Delta T_{in} (k)\right) $$


		The data set contains various operational measurements from an electric motor, including currents, voltages, torque, speed, and ambient conditions. The TNN model learns to estimate the internal temperatures that would typically require embedded temperature sensors, which are often impractical in production motors.

		This example is intended to be an instructional resource demonstrating how to build and train a TNN in MATLAB. For integration and use with Simulink, see the Github repository [here](https://github.com/wkirgsn/thermal-nn/tree/main/matlab). For video content on this topic, check out this recorded MATLAB Expo talk on [Thermal Neural Network for Temperature Modeling in E-Motors](https://www.mathworks.com/videos/thermal-neural-network-for-temperature-modeling-in-e-motors-1762251869161.html) and this [YouTube video](https://youtu.be/ySRjX9GUiCA?si=s7slG-9Ekt6y9wUe).




		The data set contains measurements from various sensors on an electric motor across multiple test profiles. Each row represents a measurement at a specific time point, and each column represents a different sensor or calculated value. In addition to the measured data the `profile_id` column identifies different test profile. You can use the `profile_id` value to separate the training and testing data.


		This example is intended to be an instructional resource demonstrating how to build and train a TNN in MATLAB. For integration and use with Simulink, see the Github repository [here](https://github.com/wkirgsn/thermal-nn/tree/main/matlab). For video content on this topic, check out this recorded MATLAB Expo talk on [Thermal Neural Network for Temperature Modeling in E-Motors](https://www.mathworks.com/videos/thermal-neural-network-for-temperature-modeling-in-e-motors-1762251869161.html) and this [YouTube video](https://youtu.be/ySRjX9GUiCA?si=s7slG-9Ekt6y9wUe).

		# Setup

Conversation

mmarkows17 commented Mar 10, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jess-42 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants