Wafer Fault Detection using MlOps Integration

Overview

Wafer Fault Detection using MlOps Integration

This is an end to end machine learning project with MlOps integration for predicting the quality of wafer sensors.

Demo

  • Link

Table of Contents

  • Problem Statement
  • How to run the application
  • Technologies used
  • Proposed Solution and Architecture
  • WorkFlow of project
  • Technologies used

Problem Statement

Improper maintenance on a machine or system impacts to worsen mean time between failure (MTBF). Manual diagnostic procedures tend to extended downtime at the system breakdown. Machine learning techniques based on the internet of things (IoT) sensor data were used to make predictive maintenance to determine whether the sensor needs to be replaced or not.

How to implement the project

  • Create a conda environment
conda create -n waferops python=3.6.9
  • Activate the environment
conda activate wafer-ops
  • Install the requirements.txt file
pip install -r requirements.txt

Before running the project atleast in local environment (personal pc or laptop) run this command in new terminal, basically run the mlflow server.

mlflow server --backend-store-uri sqlite:///mlflow.db --default-artifact-root artifacts --host 0.0.0.0 -p 5000

After running the mlflow server in new terminal, open another terminal and run the following command, since we are using fastapi. The command to run the application will change a bit

uvicorn main:app --reload

WorkFlow of the Project

To solve the problem statement we have proposed a customized machine learning approach.

WorkFlow of Project

In the first place, whenever we start a machine learning project, we need to sign a data sharing agreement with the client, where sign off some of the parameters like,

  • Format of data - like csv format or json format,etc
  • Number of Columns
  • Length of date stamp in the file
  • Length of time stamp in the file
  • DataType of each sensor - like float,int,string

The client will send multiple set of files in batches at a given location. In our case, the data which will be given to us, will consist of wafer names and 590 columns of different sensor values for each wafer. The last column will have Good/Bad value for each wafer as per the data sharing agreement

  • +1 indicates bad wafer
  • -1 indicates good wafer

These data can be found in the schema training json file.More details are present in LLD documentation of project.

Technical Aspects of the Project

As discussed, the client will send multiple set of files in batches at a given location. After signing the data sharing agreement, we create the master data management which is nothing but the schema training json file and schema prediction json (this is be used for prediction data). We have divided the project into multiple modules, for high level understanding some of them are

Training Validation

In this module,we will trigger the training validation pipeline,which will be responsible for training validation. In the training validation pipeline,we are internally triggering some of the pipelines, some of the internal function are

  • Training raw data validation - This function is responsible for validating the raw data based on schema training json file, and we have manually created a regex pattern for validating the filename of the data. We are even validating length of date time stamp, length of time stamp of the data. If some of the data does not match the criteria of the master data management, if move that files to bad folder and will not be used for training or prediction purposes.

  • Data Transformation - Previously, we have created both good and bad directory for storing the data based on the master data management. Now for the data transformation we are only performing the data transformation on good data folder. In the data transformation, we replace the missing values with the nan values.

  • DataBase Operation - Now that we have validated the data and transformed the data which is suitable for the further training purposes. In database operation we are using SQL-Lite. From the good folder we are inserting the data into a database. After the insertion of the data is done we are deleting the good data folder and move the bad folder to archived folder. Next inserting the good database, we are extracting the data from the database and converting into csv format.

Training Model

In the previous pipeline,after the database operation, we have exported the good data from database to csv format. In the training model pipeline, we are first fetching the data from the exported csv file.

Next comes the preprocessing of the data, where we are performing some of the preprocessing functions such as remove columns, separate label feature, imputing the missing the values if present. Dropping the columns with zero standard deviation.

As mentioned we are trying to solve the problem by using customized machine learning approach.We need to create clusters of data which represents the variation of data. Clustering of the data is based on K-Means clustering algorithm.

For every cluster which has been created two machine learning models are being trained which are RandomForest and XGBoost models with GridSearchCV as the hyperparameter tuning technique. The metrics which are monitoring are accuracy and roc auc score as the metric.

After training all the models, we are saving them to trained models folders.

Now that the models are saved into the trained models folder, here the mlops part comes into picture, where in for every cluster we are logging the parameters, metrics and models to mlflow server. On successful completion of training of all the models and logging them to mlflow, next pipeline will be triggered which is load production model pipeline.

Since all the trained models, will have different metrics and parameters, which can productionize them based on metrics. For this project we have trained 6 models and we will productionize 3 models along with KMeans model for the prediction service.

Here is glimpse of the mlflow server showing stages of the models (Staging or Production based on metrics)

mlflow server image

Prediction pipeline

The prediction pipeline will be triggered following prediction validation and prediction from the model. In this prediction pipeline, the same validation steps like validating file name and so on. The prediction pipeline, and the preprocessing of prediction data. For the prediction, we will load the trained kmeans model and then predict the number of clusters, and for every cluster, model will be loaded and the prediction will be done. The predictions will saved to predictions.csv file and then prediction is completed.

Technologies Used

  • Python
  • Sklearn
  • FastAPI
  • Machine Learning
  • Numpy
  • Pandas
  • MlFlow
  • SQL-Lite

Algorithms Used

  • Random Forest
  • XGBoost

Metrics

  • Accuracy
  • ROC AUC score

Cloud Deployment

  • AWS
Owner
Sethu Sai Medamallela
Aspiring Machine Learning Engineer
Sethu Sai Medamallela
Capstone-Project-2 - A game program written in the Python language

Capstone-Project-2 My Pygame Game Information: Description This Pygame project i

Nhlakanipho Khulekani Hlophe 1 Jan 04, 2022
Enhancing Knowledge Tracing via Adversarial Training

Enhancing Knowledge Tracing via Adversarial Training This repository contains source code for the paper "Enhancing Knowledge Tracing via Adversarial T

Xiaopeng Guo 14 Oct 24, 2022
Source-to-Source Debuggable Derivatives in Pure Python

Tangent Tangent is a new, free, and open-source Python library for automatic differentiation. Existing libraries implement automatic differentiation b

Google 2.2k Jan 01, 2023
🍀 Pytorch implementation of various Attention Mechanisms, MLP, Re-parameter, Convolution, which is helpful to further understand papers.⭐⭐⭐

🍀 Pytorch implementation of various Attention Mechanisms, MLP, Re-parameter, Convolution, which is helpful to further understand papers.⭐⭐⭐

xmu-xiaoma66 7.7k Jan 05, 2023
Scenic: A Jax Library for Computer Vision and Beyond

Scenic Scenic is a codebase with a focus on research around attention-based models for computer vision. Scenic has been successfully used to develop c

Google Research 1.6k Dec 27, 2022
TorchDistiller - a collection of the open source pytorch code for knowledge distillation, especially for the perception tasks, including semantic segmentation, depth estimation, object detection and instance segmentation.

This project is a collection of the open source pytorch code for knowledge distillation, especially for the perception tasks, including semantic segmentation, depth estimation, object detection and i

yifan liu 147 Dec 03, 2022
Mmdet benchmark with python

mmdet_benchmark 本项目是为了研究 mmdet 推断性能瓶颈,并且对其进行优化。 配置与环境 机器配置 CPU:Intel(R) Core(TM) i9-10900K CPU @ 3.70GHz GPU:NVIDIA GeForce RTX 3080 10GB 内存:64G 硬盘:1T

杨培文 (Yang Peiwen) 24 May 21, 2022
Official PyTorch Implementation of paper EAN: Event Adaptive Network for Efficient Action Recognition

Official PyTorch Implementation of paper EAN: Event Adaptive Network for Efficient Action Recognition

TianYuan 27 Nov 07, 2022
Official code for "Decoupling Zero-Shot Semantic Segmentation"

Decoupling Zero-Shot Semantic Segmentation This is the official code for the arxiv. ZegFormer is the first framework that decouple the zero-shot seman

Jian Ding 108 Dec 30, 2022
PyTorch implementation of SimSiam: Exploring Simple Siamese Representation Learning

SimSiam: Exploring Simple Siamese Representation Learning This is a PyTorch implementation of the SimSiam paper: @Article{chen2020simsiam, author =

Facebook Research 834 Dec 30, 2022
Film review classification

Film review classification Решение задачи классификации отзывов на фильмы на положительные и отрицательные с помощью рекуррентных нейронных сетей 1. З

Nikita Dukin 3 Jan 21, 2022
Activating More Pixels in Image Super-Resolution Transformer

HAT [Paper Link] Activating More Pixels in Image Super-Resolution Transformer Xiangyu Chen, Xintao Wang, Jiantao Zhou and Chao Dong BibTeX @article{ch

XyChen 270 Dec 27, 2022
An offline deep reinforcement learning library

d3rlpy: An offline deep reinforcement learning library d3rlpy is an offline deep reinforcement learning library for practitioners and researchers. imp

Takuma Seno 817 Jan 02, 2023
The personal repository of the work: *DanceNet3D: Music Based Dance Generation with Parametric Motion Transformer*.

DanceNet3D The personal repository of the work: DanceNet3D: Music Based Dance Generation with Parametric Motion Transformer. Dataset and Results Pleas

南嘉Nanga 36 Dec 21, 2022
A universal memory dumper using Frida

Fridump Fridump (v0.1) is an open source memory dumping tool, primarily aimed to penetration testers and developers. Fridump is using the Frida framew

551 Jan 07, 2023
TumorInsight is a Brain Tumor Detection and Classification model built using RESNET50 architecture.

A Brain Tumor Detection and Classification Model built using RESNET50 architecture. The model is also deployed as a web application using Flask framework.

Pranav Khurana 0 Aug 17, 2021
Saliency - Framework-agnostic implementation for state-of-the-art saliency methods (XRAI, BlurIG, SmoothGrad, and more).

Saliency Methods 🔴 Now framework-agnostic! (Example core notebook) 🔴 🔗 For further explanation of the methods and more examples of the resulting ma

PAIR code 849 Dec 27, 2022
Doing the asl sign language classification on static images using graph neural networks.

SignLangGNN When GNNs 💜 MediaPipe. This is a starter project where I tried to implement some traditional image classification problem i.e. the ASL si

10 Nov 09, 2022
1st-in-MICCAI2020-CPM - Combined Radiology and Pathology Classification

Combined Radiology and Pathology Classification MICCAI 2020 Combined Radiology a

22 Dec 08, 2022
Computational modelling of ray propagation through optical elements using the principles of geometric optics (Ray Tracer)

Computational modelling of ray propagation through optical elements using the principles of geometric optics (Ray Tracer) Introduction By applying the

Son Gyo Jung 1 Jul 09, 2022