README for ECM Prediction and Interpretability Project

Real-time web application: https://mingwucn.github.io/SmartECM/

This project contains the implementation of machine learning models for real-time prediction and optimization in Electrochemical Machining (ECM). The project emphasizes the interpretability of these models through Explainable AI (XAI) techniques, including SHapley Additive exPlanations (SHAP), Gradient-weighted Class Activation Mapping (Grad-CAM), and a custom linear regression explainer. The codebase is organized to facilitate the training, evaluation, and interpretation of models used to predict cavity profiles based on processing parameters and in-process data.

To run the code in this project, the following environment and dependencies are required:

• Operating System: Linux/Windows
• CUDA Version: 9.0
• TensorFlow: 2.0
• NumPy: Latest stable version

1. Install CUDA 9.0:

   • Follow the official CUDA installation guide for your operating system: https://developer.nvidia.com/cuda-90-download-archive

2. Install TensorFlow 2.0:

   • With GPU support:

     pip install tensorflow-gpu==2.0.0

   • Without GPU support:

     pip install tensorflow==2.0.0

3. Install NumPy:

   pip install numpy
   

import tensorflow as tf
import numpy as np

print("TensorFlow version:", tf.__version__)
print("NumPy version:", np.__version__)
print("Is GPU available:", tf.test.is_gpu_available())

The project is organized into the following directories:

/algorithms: Contains the implementations of the interpretability algorithms.

/grad_cam.py: A self-implemented version of the Grad-CAM algorithm for visualizing the focus of convolutional neural networks. /shap_explainer.py: SHAP implementation for providing global explanations of machine learning models. /linear_regression_explainer.py: Custom explainer for interpreting linear regression models. /models: Includes the ML models used in the study.

/logistic_regression.py: Implementation of the logistic regression model. /neural_network.py: Implementation of the neural network model. /cnn.py: Implementation of the convolutional neural network (CNN) model. /data: Placeholder for datasets.

License This project is licensed under the MIT License. See the LICENSE file for more details.