Gespeichert in:
| 1. Verfasser: | |
|---|---|
| Format: | Recurso digital |
| Sprache: | |
| Veröffentlicht: |
Zenodo
2026
|
| Online-Zugang: | https://doi.org/10.5281/zenodo.19062686 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Inhaltsangabe:
- <p>The thesis work was implemented as a part of a Business Finland co-creation project <strong>"Data Driven Design for Sustainability (3DSUS)" Project Code:</strong><strong>6292/31/2023. </strong></p> <p>This repository contains a hardware configuration file with PLC ladder logic, python script collecting data from the PLC and three Python-based Jupyter Notebooks developed as part of the data logging and analysis framework for a Powder Bed Fusion (PBF) additive manufacturing thesis.</p> <p>The hardware configuration file with PLC ladder logic <strong>Datalogger_withEnergy_Power.smc2 </strong> contains a project which contains all the hardware-software configuration of the sensor modules, PLC and the Ladder logic implemented for collecting the data from the PLC. The python script <strong>pylogix_main.py</strong> collects the data into a csv file by communicating with PLC's IP address in the same subnet. This data is then stored into a local database.</p> <p>The first notebook, <strong>A-PBF_Build_Comparison_v2.ipynb</strong>, implements phase-wise segmentation of two complete PBF build runs by extracting and visualising gas sensor (ArgonFlowRAW, LowFlowRAW, HighFlowRAW) and laser power (Power_W) data from a SQLite database. It automatically detects the post-processing phase using a rolling-median threshold algorithm applied to the HighFlowRAW signal, and generates interactive Plotly figures for the purge, exposure, and post-processing phases of each run.</p> <p>The second notebook, <strong>A-LayerSegmentation.ipynb</strong>, performs layer-wise segmentation by reconstructing and synchronising timestamps from the machine's part statistics export with the sensor data log, merging them via an inner join on timestamp, and producing dual-axis visualisations that map laser power fluctuations to individual layer indices using vertical band overlays and peak/valley detection.</p> <p>The third notebook, <strong>A-ML.ipynb</strong>, implements a machine learning pipeline to convert raw ADC sensor readings into calibrated gas flow values in litres per minute. A Linear Regression model is trained independently for each of the three flow sensors using laboratory calibration reference points, evaluated on held-out test data using RMSE, MAE, R², and a ±5% tolerance-based classification metric, and subsequently applied to the full operational dataset to produce corrected flow values. The corrected flows are then used to compute argon and shielding gas consumption via trapezoidal numerical integration and energy consumption from cumulative meter readings, enabling quantitative comparison of resource usage across build phases and runs.</p>