In-situ surface porosity prediction in DED process using explainable multimodal sensor fusion

This repository maintains data associated with the models used in "In-situ surface porosity prediction in DED (directed energy deposition) printed SS316L parts using multimodal sensor fusion".

The repository contains 4 documents.

1. Surfels.xlsx - Contains the ground truth data (area percentage of porosity, as obtained from ImageJ software) for all surface elements ("surfels"). Surfels with percentage area of pores < 1% are classified as non-porous and those with area percentages > 2% are classified as porous.

2. Tensor_data_72_surfels.zip - Contains the spectrogram data from Accelerometer and Acoustic Emission signals pertaining to printing and milling tracks for all 72 surface elements in a zip file.

3. CNN Model Architecture and Predictions.ipynb - The CNN architectures for the various models developed including k-fold validation.

4. LIME_Explanations_Porosity_Predictions.ipynb - Python code for explaining the CNN model predictions using LIME.

本仓库维护与《基于多模态传感器融合的DED（定向能量沉积，directed energy deposition）打印SS316L构件原位表面孔隙率预测》一文所用模型相关的数据。
本仓库共包含4份文档：
1. Surfels.xlsx：包含所有表面面元（surfels）的真值数据（孔隙面积占比，通过ImageJ软件获取）。其中孔隙面积占比低于1%的表面面元被归类为无孔隙，占比高于2%的则被归类为有孔隙。
2. Tensor_data_72_surfels.zip：包含全部72个表面面元对应的打印与铣削轨迹的加速度计（Accelerometer）、声发射（Acoustic Emission）信号语谱图数据，以压缩包形式存储。
3. CNN Model Architecture and Predictions.ipynb：包含所开发的各类模型的卷积神经网络（Convolutional Neural Network, CNN）架构及k折验证相关内容的Jupyter Notebook。
4. LIME_Explanations_Porosity_Predictions.ipynb：用于借助LIME（局部可解释模型不可知论解释，Local Interpretable Model-agnostic Explanations）解释卷积神经网络模型孔隙率预测结果的Python代码Jupyter Notebook。