Embedded machine learning for fault detection in conveyor systems using multi-sensor data and discrete wavelet transform
This study presents a fault detection model utilizing the Seeed Studio XIAO nRF52840 Sense microcontroller development board to identify and diagnose faults in the conveyor belt system of the MPS-PA Bottling Learning System, a component of the Festo Didactic System. The microcontroller is equipped...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Publishing House for Science and Technology
2025-07-01
|
| Series: | Vietnam Journal of Mechanics |
| Subjects: | |
| Online Access: | https://vjs.ac.vn/vjmech/article/view/22277 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This study presents a fault detection model utilizing the Seeed Studio XIAO nRF52840 Sense microcontroller development board to identify and diagnose faults in the conveyor belt system of the MPS-PA Bottling Learning System, a component of the Festo Didactic System. The microcontroller is equipped with a 6-axis Inertial Measurement Unit (IMU) and a Pulse Density Modulation (PDM) microphone, enabling it to monitor vibrations and sounds generated during conveyor belt operation. The collected signals are processed using the Discrete Wavelet Transform (DWT) to extract relevant features, which are then used to train an embedded machine learning model designed to detect faults such as bottle obstructions and falls. The conveyor belt transports empty bottles to a dispenser, but a 90-degree turn in the path frequently causes disruptions, resulting in bottle rotation or falls. The IMU captures vibration data, while the PDM microphone records audio signals during these events. The processed DWT features are utilized to train the fault detection model. The model is developed using the TensorFlow Lite framework, incorporating batch normalization to stabilize and accelerate the learning process. Once deployed, the system predicts faults and sends Bluetooth alerts to a host PC when an issue is detected, allowing the process to be halted to prevent further damage. The proposed fault detection model demonstrates promising results, achieving up to 94% accuracy. Additionally, the system's low power consumption, supported by a 200 mAh LiPo rechargeable battery, enhances energy efficiency, contributing to more sustainable manufacturing operations.
|
|---|---|
| ISSN: | 0866-7136 2815-5882 |