Robotic additive manufacturing (AM) is an advanced manufacturing technology based on robotics and traditional methods such as arc welding with metal wire. Within the industrial internet of things (IIoT) architecture, digital twins are used to optimize and make predictions in manufacturing processes. A digital twin captures the physical process' states and is considered a digital copy/replica of the process. For the development of an AM digital twin, process data is needed. Moreover, to ensure qualitative and consistent AM parts, real-time monitoring of the process with sensor is recommended. In this project, AM experiments are conducted, and the process state is captured with a hyperspectral imaging camera sensor (Part A); next, the hyperspectral data is analyzed and mathematical models are created to predict the AM quality (Part B).
Part A: In this project a hyperspectral imaging camera (manufacturer: Specim) is used as a monitoring sensor. Different AM samples (both good and bad quality) shall be produced along with their corresponding sensor dataset. The dataset obtained with the hyperspectral sensor is considered a digital twin of the physical AM structure.
The goal of Part A is to produce AM structures together with their corresponding digital twin (i.e., recorded hyperspectral sensor data).
Part B: The hyperspectral imaging sensor data produced in Part A shall be analyzed with statistical methods, deterministic methods, or machine learning algorithms. The hyperspectral sensor data must also be correlated to the physical AM samples (both good and bad quality). The goal of Part B is first to understand the sensor data output of the hyperspectral camera and then to obtain a model which can predict the AM structure quality (good or bad quality) based on sensor data.
Different Matlab toolboxes will be used for data processing, e.g., image processing, computer vision, system identification and machine learning, etc. The project aims to understand
- the output data of the hyperspectral camera
- the wire-arc additive manufacturing process behavior by analyzing the hyperspectral camera sensor signals
All signal processing and analysis is done with a SpeedGoat real-time target machine which runs Matlab Simulink as a programming language. Data from the hyperspectral sensor is processed by computation algorithms and simulations running on the real-time target machine.
Part A and Part B can be carried out separately by different students from different study programs. The project has a strong relation to the course 4MT321 Sustainable Production; metrics related to sustainable issues will also be included in the project outcome. Project outcomes, such as demos, programming codes and design concepts will be used for the development of course material and labs in the course 4MT321 Sustainable Production. The thesis project is carried out in the Smart Industry Group – SIG Lab which is in the EPIC building.
- Sven Nordebo, Dept. of Physics and Electrical Engineering ([email protected])
- Jetro Kenneth Pocorni (Contact person) and Osama Ziada, Dept. of Mechanical Engineering ([email protected], [email protected])