Motor bearing fault classification and detection for AC drives with single MCU and 98% plus accuracy
Protect AC motor drive designs with high-accuracy AI based motor fault classification.
Application overview
Spotting problems early without needing expensive external monitoring systems or shutting the system down makes the system safe without interrupting its operation.?Real-time motor control & automation microcontrollers (MCUs) are equipped with our TinyEngine? NPU which allows to execute the task right where the motor is. Rather than using traditional FFT methods that need manual feature engineering, edge AI is leveraged to learn complex vibration patterns through data based training and adapts to different motor types.
Starting evaluation
Data collection
LAUNCHXL-F28P55X?LaunchPad? development kit is used to perform the task. A vibration sensor attached to the motor is connected to the SPI port of the evaluation board to provide the vibration sensor data. This data is processed in realtime and monitored by edge AI model for detection and classification of potential motor bearing fault. The capture & display data tab in CCStudio? EdgeAI Studio is used to command the LaunchPad and complete the data acquision. The acquired vibration data is used for model trainning.
Data quality assessment
Acquired vibration data can be displayed and plotted in EdgeAI Studio in both time and frequency domain to ensure vibration data sets are separable in features.
Build and train your model
EdgeAI Studio allows model training, assessment and deployment to be completed within a intuitive GUI user interface.
- Explore and train multiple model choices through an easy-to-use GUI based workflow
- Start fast with the motor bearing fault example project, featuring a preloaded dataset
- Use your own data with integrated capture and hosting tools
Model Flexibility to support your application needs based on the memory constraints and outlier detection requirements.
Deploying your model
CCStudio Edge AI Studio provides a start to finish comprehensive workflow for deploying trained models directly to embedded targets.
For developers seeking deeper customization and control, the C2000WARE-DIGITALPOWER-SDK?offers a comprehensive framework for building and integrating edge AI functionality into your own embedded applications.
Choosing the Right device for you
TI's?c28 DSPs and Arm? Cortex?-M33 based MCUs deliver scalable performance for executing and accelerating arc fault detection models, along with key SoC features critical to your application.
Data below is taken into consideration 4K Hz sampling frequency on 256 samples
?
| Product number | Processing core | NPU available | Clock frequency (MHz) | Arc detection benchmark metrics | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Model execution time (ms) | Flash (kB) | SRAM (kB) | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| TMS320F28P550x | C28x DSP core | Yes | 150 | 0.17 | 1.8 | 1.1 | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| TMS320F2800137 | C28x DSP core | No | 120 | 0.42 | 1.2 | 1 | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| TMS320F28P650 | C28x DSP core | Yes | 200 | 0.25 | 1.2 | 1 | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| AM13E23019 | Cortex-M33 core | Yes | 200 | ~0.13* | 3 | 0.9 | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| *AM13E23019 with TinyEngineTM NPU numbers are preliminary | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
All the hardware, software and resources you’ll need to get started
Hardware
LAUNCHXL-F28P55X
Evaluation and develpoment board for TI C2000 TMS320F28P550x series MCUs that in conjuction with plug-on motor drive BoosterPack, can be used to perform the tasks of data collection and realtime inferencing.
Software & development tools
CCStudio? Edge AI Studio
A fully integrated no-code solution for acquiring sensor data, training and compilation of motor bearing fault classification and detection models, deploying onto our MCUs.
Dynamic tensorlab
A command line interface for motor advanced users, who want to develop their own model.
Supporting resources
?
This document provides a guide on developing a Convolutional Neural Network (CNN) based motor fault classification with C2000 devices
Industrial automation | Motor diagnostics & monitoring
Reduce noise, improve efficiency, increase reliability of HVAC system by detecting fan-blade imbalance early with single digit of gram of detection sensitivity, with a single MCU with NPU, in addition to motor control