Introduction

Vibration analysis is a key diagnostic tool for identifying bearing faults in machinery. When bearings deteriorate, their smooth motion becomes erratic, causing increased vibration levels. While basic time-domain measurements provide limited information, frequency analysis can pinpoint which specific component is failing, allowing engineers to anticipate critical failures and schedule repairs before catastrophic breakdowns occur.

This project demonstrates implementation of an AI-based motor fault detection system on AM26x microcontrollers. It showcases how to deploy machine learning models for real-time mechanical motor fault classification in embedded systems, helping prevent hazards through early detection of motor faults.

Supported Combinations

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

Dataset and Model Details

Dataset

TI has created a specialized motor fault dataset containing vibrational measurements. The dataset is divided into six classes representing different bearing conditions.

Parameter	Value
Sensor	3-axis accelerometer (X, Y, Z axes)
Sampling Rates	10 Hz, 20 Hz, 30 Hz, 40 Hz (variable)
Channels	3 (Vibration X, Y, Z axes)
Samples per File	Variable: 24,000 - 144,000 samples
Total Files	216 files (36 files in each of 6 classes)

Output Classes:

Class 0: Bearing Normal
Class 1: Bearing Contaminated
Class 2: Bearing Erosion
Class 3: Bearing Flaking
Class 4: Bearing No Lubrication
Class 5: Bearing Localized Fault

Model Architecture

This lightweight classification model contains approximately 1,000 parameters and follows a streamlined architecture consisting of four convolutional layers (each enhanced with BatchNorm and ReLU activation functions) followed by a single linear layer.

Input Features

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

N (1): batch size which is restricted to 1
C (3): channels which is 3 for 3-axis vibration data
H (128): samples of timeseries vibrations which is 128 in this example
W (1): width of samples is restricted to 1 for timeseries applications

Output

This model produces a 1D output representing the six possible classes. The position of the highest value in this output indicates the classification result.

Feature Extraction Configuration

Feature extraction transforms raw vibration data into meaningful inputs for the AI model. The feature extraction pipeline uses FFT to identify frequencies, followed by binning and logarithmic scaling.

Configuration flags in user_input_config.h:

FE_FFT: Performs Fast Fourier Transform, converting time-domain signal into frequency components
FE_DC_REM: Removes the DC component from Fourier transform
FE_BIN: Groups frequencies into bins (size 8), reducing dimensionality
FE_BIN_NORMALIZE: Normalizes the bin values
FE_LOG: Applies logarithmic scaling to compress dynamic range
FE_CONCAT: Combines outputs from 8 frames for temporal context

The pipeline takes a 256-sample frame, computes FFT (129 outputs), removes DC (128), bins to 16 features, and concatenates 8 frames for 128 total features per channel.

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 run inference on test vectors and classify bearing condition

Sample Output

Motor Fault Detection Example Started ...
Feature extraction mismatches 0
All tests have passed!!
Golden vectors matched: 6 not matched: 0

Table of Contents