PMSM Rotor Temperature Forecasting

Table of Contents

• Introduction
• Supported Combinations
• Dataset and Model Details
  • Dataset
  • Model Architecture
  • Input Features
  • Output
• Feature Extraction Configuration
• Steps to Run the Example
• See Also
• Sample Output

Introduction

Permanent Magnet Synchronous Motors (PMSMs) are widely used in electric vehicles, industrial automation, and wind turbines. However, they carry a critical vulnerability: the permanent magnets can suffer irreversible demagnetization if rotor temperatures exceed safe limits (typically above 150C). Placing temperature sensors directly on the rotor is expensive and mechanically fragile.

This project demonstrates an elegant solution using on-device machine learning. A lightweight LSTM model running on the AM26x MCU estimates rotor temperature from signals that are already available - phase currents, voltage components, ambient temperature, and coolant temperature. This enables preemptive thermal protection actions before damage occurs.

Supported Combinations

Parameter	Value
CPU + OS	r5fss0-0 nortos
Toolchain	ti-arm-clang
Board	am263px-lp
Example folder	examples/ai/forecasting_pmsm_rotor_temp/

Dataset and Model Details

Dataset

This example uses a dataset derived from real sensor measurements collected from a PMSM running on a controlled test bench. The original data was recorded by the LEA department at Paderborn University at 2 Hz sampling rate.

Signal	Description
i_a	Phase current magnitude
u_d	Voltage d-component (active)
u_q	Voltage q-component (reactive)
ambient	Ambient temperature
coolant	Coolant temperature
pm	Permanent magnet surface temperature (target)

The model predicts the permanent magnet surface temperature at the next timestep given the past 3 readings of all six signals.

Model Architecture

The model is FCST_LSTM8 - a single-layer LSTM network with a hidden size of 8 and approximately 393 trainable parameters. The LSTM's gating mechanisms learn how current loading, voltage conditions, and cooling state interact over time to determine rotor temperature.

Input Features

The model takes 4D input (N,C,H,W):

• N (1): batch size, restricted to 1
• C (6): channels - i_a, u_d, u_q, ambient, coolant, pm
• H (3): number of historical timesteps (frame size) used for prediction
• W (1): width, restricted to 1 for timeseries applications

Output

The model produces an output tensor of shape (1, 1), representing the predicted permanent magnet surface temperature at the next timestep. This continuous forecasting output can be fed directly into thermal protection or derating logic.

Feature Extraction Configuration

A SimpleWindow transform segments the continuous stream into fixed-length windows of 3 consecutive timesteps across all 6 input channels. No FFT, binning, or log scaling is applied - the LSTM model learns directly from raw sensor values.

Steps to Run the Example

• When using CCS projects to build, import the CCS project for the required combination and build it using the CCS project menu (see Using SDK with CCS Projects).
• When using makefiles to build, note the required combination and build using make command (see Using SDK with Makefiles)
• Launch a CCS debug session and run the executable, see CCS Launch, Load and Run
• The application will forecast rotor temperature and compare against expected values

See Also

AI Examples

Sample Output

PMSM Rotor Temperature Forecasting Example Started ...
All tests have passed!!
Golden vectors matched: 3 not matched: 0