A Prototype for the State Analysis of an Electric Motor
Development of a prototype of a predictive maintenance system that analyzes the state of an electric motor
The partner is Toradex, the Switzerland-based provider of embedded computing solutions, especially Arm®-based system on modules and customized single-board computers. The company has offices in the USA, Vietnam, China, India, Japan and Brazil.
The operation and productivity of industrial equipment directly depends on the motor performance. That is why a solution that allows maintenance engineers to continuously monitor their production assets and be alert to any aberrations, from slight malfunctions to outage, helps prevent failures.
SaM Solutions aimed to create a predictive maintenance system that assesses the behavior of a motor and, based on that, determines the status and alerts about aberrations, if there are any.
Our team has created the prototype of a system that relies its predictive capability on motor vibration frequency and force, as an abnormal vibration level signals that the motor has probably failed. As a hardware platform for the prototype, they have used Toradex’s computer-on-module Colibri i.MX6-ULL, as it has all the capabilities required for predictive maintenance processes.
The system relies the detection of abnormal behavior on AI methods, while data analytics is based on a pre-trained model. The team has applied TensorFlow/Keras-based algorithms to train and build the model after collecting the data specific to each of the motor performance states. To provide flexibility, the system is based on two cloud solutions: it was trained on Azure, while AWS Greengrass has been used for algorithm deployment.
The process of the motor state analysis includes the following steps:
- The system acquires performance parameters from the motor using MPU-6050 data that is read continuously and transmits it to an AWS Lambda-based algorithm that is written in Python.
- The data is prepared in the way the model can accept it as input characteristics. Then the system applies machine learning algorithms to identify the motor state based on Amazon Greengrass and Amazon Lambda function. An ‘off’ state when it is stopped, a ‘normal’ state when it is on and behaves normally, and an abnormal state that indicates malfunctions such as undervoltage, overvoltage or motor failure.
- The board reports the motor status via a web interface or using various communication protocols.
Microsoft Azure, Amazon Web Services, Microsoft IoT Edge, Amazon Greengrass/Lambda, MQTT, SMTP, TensorFlow/Keras frameworks, Python, Jupiter Tools: Toradex’s NXP i.MX 6ULL Computer on Module - Colibri iMX6ULL, InvenSense’s MPU6050 Gyroscope-accelerometer
SaM Solutions’ team has developed an efficient prototype of a predictive maintenance system that features a probability of prediction results of up to 99%. This system allows companies to avoid down-time-related costs and improve the productivity of their assets.