Torque Measurement

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

Predicting rotor torque in motors presents a significant engineering challenge: direct measurement requires costly sensors that add weight, a critical concern for electric vehicles striving for efficiency. While traditional statistical methods for estimating rotor torque exist, they are limited by the need for domain expertise and motor-specific tuning.

This project leverages deep learning to efficiently predict rotor torque in Permanent Magnet Synchronous Motors (PMSM), enabling a substantial reduction in the hardware sensors required for motor instrumentation and control.

Supported Combinations

Parameter	Value
CPU + OS	r5fss0-0 nortos
Toolchain	ti-arm-clang
Board	am261x-lp
Example folder	examples/ai/torque_measurement/

Dataset and Model Details

Dataset

The dataset contains sensor measurements collected from a PMSM running on a controlled test bench. The measurements were collected by the LEA department at Paderborn University at 2 Hz sampling rate.

Column Name	Description
i_d	Current d-component (active)
i_q	Current q-component (reactive)
u_d	Voltage d-component (active)
u_q	Voltage q-component (reactive)
motor_speed	Motor speed
ambient	Ambient temperature
torque	Torque induced by current (target)

The model uses a subset of easily measurable inputs: Ambient temperature, Coolant temperature, Voltage components (u_d, u_q), and Phase current magnitude to predict torque.

Model Architecture

This regression model REGR_10k contains approximately 10,000 parameters with three convolutional layers (each with BatchNorm and ReLU) followed by two linear layers.

Input Features

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

• N (1): batch size which is restricted to 1
• C (10): channels for multiple parameters
• H (128): samples of timeseries data
• W (1): width of samples is restricted to 1 for timeseries applications

Output

This model produces a 1D output representing the continuous value of torque.

Feature Extraction Configuration

For this torque measurement system, no FFT or frequency-domain feature extraction is applied. The model uses normalization to convert floating point input into fixed point for efficient inference.

Configuration in user_input_config.h:

• SKIP_NORMALIZE: Converts floating point input into fixed point input for the AI model

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 predict torque values and compare against expected outputs

See Also

AI Examples

Sample Output

Torque Measurement Example Started ...
Feature extraction mismatches 0
All tests have passed!!
Golden vectors matched: 3 not matched: 0