AI-models for Condition Monitoring on Texas Instruments MCUs

Unexpected failure in motors and other kinds of machines can be very costly. Being able to detect when problems arise or when a machine is about to break down allows operators to plan ahead and ensure that the machine only stops at a non-critical time. Machine learning can help in this regard by providing additional, crucial information that allows operators to better understand if the machine is experiencing some issues and/or plan when maintenance needs to be performed.

Utilising machine learning for condition monitoring and fault/anomaly detection was previously thought to only be feasible with the help of cloud computing. Imagimob, together with Texas Instruments, has shown that this can be achieved also on the Edge: Imagimob machine learning models are able to run locally on a MCU board, like the TI Sitara AM243x or AM64x, placed in the vicinity of the machine.

This means that the sensor signals from the machine are gathered and then processed directly on the board and the AI-model running on it gives real-time feedback to the operator. Such a setup can be even hooked up to a PLC system to automate, for instance, critical functionalities.

Imagimob and Texas Instruments have jointly developed a demo, in which you will be able to run an Imagimob AI-model on the Texas Instruments Sitara Microcontroller AM243x Evaluation Module to predict failures due to insulation deterioration in an electric drive motor.
 
By clicking the link(s) below, you can download the files needed to replicate the demo using your TI board:

Download Firmware

Download Imagimob UI

Download AI-Models

Once the download is complete, you will get

  • The full firmware project, containing the AI-model, to be deployed on your TI AM243x board
  • The imagimob_ui folder with data and Python scripts to be used to run the demo and the Imagimob UI used to visualize the results in real-time in your browser
  • The models folder with the Keras model trained with Imagimob AI, the Python code for the pre-processor and the corresponding C (edge) model which is already embedded in the firmware

To set up the demo, follow the instructions in the README file that you find in the downloaded Imagimob UI folder.

Demo Project Description

Input Data

The input data consist of 3-phase currents and voltages sampled at 4 kHz obtained via the accelerated ageing process of an electric drive motor defined by the IEEE standard 117. Special thanks to KTH, Royal Institute of Technology (Sweden), for performing the accelerated ageing and the hardware setup for data collection.

Demo Setup and Functioning

To run the demo you need a PC with the required software installed and a TI Sitara AM243x board, running the provided firmware. The TI board is connected to the PC via 2 USB cables. 

The PC simulates here the electric motor or, in general, any other kind of (industrial) machine. The 3-phase currents and voltages data are saved on the PC and sampled to the TI AM243x via UART. This input data is then fed to the Imagimob AI model running on the board. The model processes the input data and generates an output which is sent from the board to the PC where it is visualized in a browser in real-time together with the corresponding input data. Such an output is a value between 0 and 1 and it represents the condition of the machine or its age, which ranges from 0 to 12 years in this example.

The demo completes in about 30-60 minutes.

Imagimob AI-model Properties

MODEL TYPE: Regression neural network model
MODEL MEMORY USAGE: 3.5 kB (FLASH) and 6.6 kB (RAM)
MODEL OUTPUT: AI model predictions related to machine age and machine damage probability

Imagimob AI platform

Sign up for the Free Forever Plan and download the Imagimob AI platform

Contact Imagimobs sales.