Data-driven prediction modeling for part attributes and process monitoring in additive manufacturing
- WCU Author/Contributor (non-WCU co-authors, if there are any, appear on document)
- Jayanta Bhusan Deb (Creator)
- Institution
- Western Carolina University (WCU )
- Web Site: http://library.wcu.edu/
- Advisor
- Nazmul Ahsan
Abstract: The first study aimed to use artificial neural networks (ANN) to predict how process parameters would affect the part attributes in an extrusion-based additive manufacturing (AM) process. The study involved parts fabrication using an orthogonal array experimental design with five process parameters at three levels: building orientation, print speed, extrusion temperature, deposition direction, and layer thickness. The fabricated parts were measured for dimensional accuracy, surface roughness, and tensile strength. These attributes were then used to train, validate, and test multilayer ANN models. Three of the four ANN models were for predicting each of the three-part qualities separately, while the fourth was for combining all three attributes. Regarding RMSE and correlation coefficient, the findings showed that the individual part attribute ANN models outperformed the model for combining three attributes. To determine which parameters had a higher impact on the individual part qualities, comparisons between the individual part attributes with respect to the process parameters were made. The trained ANN models can forecast and optimize the part properties in extrusion-based AM processes. The second research developed a new method of collecting time series data for process monitoring in a Fused Filament Fabrication (FFF) system using wireless sensors to predict the machine bed angular velocity of FFF using the Vanilla Long Short-Term Memory (VLSTM) network. With two levels, the printer speed and deposition direction of the nozzle head were used in this study following a full factorial experimental design to investigate their effects on machine vibration during printing. Time series machine bed angular velocity data were collected and used to train and test the proposed VLSTM network. Adam optimizer and VLSTM networks with four cells generated the best training accuracy after 100 epochs. One developed VLSTM model was used to train and test the network by inserting four-time series machine bed angular velocity data. Then four-time series simulation results were investigated to analyze the outputs of our developed and trained model. Simulation and experimental results were analyzed using root mean square error (RMSE). Practical data analysis concluded that the deposition direction of the nozzle head and printer speed both significantly affected the angular velocity of the printer bed. The developed VLSTM model can be used to predict the FFF printer bed angular velocity having different unexplored printer speeds and deposition directions, which will eventually help predict the quality of the printed parts through machine vibration analysis.
Data-driven prediction modeling for part attributes and process monitoring in additive manufacturing
PDF (Portable Document Format)
1664 KB
Created on 6/1/2023
Views: 116
Additional Information
- Publication
- Thesis
- Language: English
- Date: 2023
- Subjects
- Additive manufacturing
- Industrial engineering
- Mechanical engineering
- Neural networks (Computer science)