- 90 views
- 170 downloads
Multi-Sensor Data Fusion and Reconfigurable Measurement System: A Machine Learning Approach
-
- Author / Creator
- Soriano Morales, Mario A
-
The fast development of new technologies related to sensor solutions, cyber-physical-systems, cloud computing, the Internet of Things (IoT) and their applications in the industry has led to a new modern era where the industry itself has faced a new industrial revolution called Industry 4.0. With the help of machine learning techniques, sensory solutions and the application of IoT, Industry 4.0 has been able to achieve fully autonomous and intelligent processes that can communicate with each other and could be located hundreds of miles away. As a consequence, in the presented work, an implementation of the concept mentioned earlier is acquired to create an intelligent reconfigurable measurement system technology that takes multiple outputs from different sensors (pressure sensor, accelerometer, temperature, and light absorption) and performed the data analysis and data acquisition. The methodology used is an advanced analytics framework of machine learning as an end-to-end model with a combination of nonlinear multi-layers for structuring the multi-sensor fusion, this framework uses a deep learning approach, which is an end-to-end learning structure that takes the outputs of the multi-sensor network and performs classification, data linearization and calibration for the different sensors. The multi-sensor data fusion is performed using a centralized architecture (microcontroller and PC), taking an IoT implementation for data transfer. The data alignment and data associations are performed within a desktop PC using a microcontroller as a communication node. Then, a convolutional neural network is used for classifying the data and then pass it to a deep fully connected neural network for its linearization and calibration. The validation of the methodology is performed using 150, 000 data points as reference for the calibration and linearization processes as well as the classification of the data coming from the multi-sensor system. A user-to-system communication framework is designed to perform the multi-sensor fusion and also to enable the user control of the processes. With the communication framework mentioned above, an easy-to-use device has been designed and developed to help to understand the structure of sensor fusion using deep learning as a contribution to the academic learning community.
The contributions of the presented work lie in the usage of a deep learning framework for multi-sensor fusion with a centralized low-cost architecture. The main focus is to create a low-cost solution for sensor fusion that relies on the application of an Internet of Things (IoT) and machine learning data structures; this will help to prove how using machine learning methods can contribute to the construction of such measurement system. It is concluded that a multi-sensor fusion approach using deep learning as a framework model gives excellent results compared to benchmark methods for the integration of different sensors, accomplishing at the same time the linearization and calibration of the outputs coming from these sensors. -
- Subjects / Keywords
-
- Graduation date
- Spring 2020
-
- Type of Item
- Thesis
-
- Degree
- Master of Science
-
- License
- Permission is hereby granted to the University of Alberta Libraries to reproduce single copies of this thesis and to lend or sell such copies for private, scholarly or scientific research purposes only. Where the thesis is converted to, or otherwise made available in digital form, the University of Alberta will advise potential users of the thesis of these terms. The author reserves all other publication and other rights in association with the copyright in the thesis and, except as herein before provided, neither the thesis nor any substantial portion thereof may be printed or otherwise reproduced in any material form whatsoever without the author's prior written permission.