Control Charts and Machine Learning for Anomaly Detection in Manufacturing

✍ Scribed by Kim Phuc Tran

• Publisher: Springer
• Year: 2021
• Tongue: English
• Leaves: 270
• Series: Springer Series in Reliability Engineering
• Edition: 1
• Category: Library

✦ Synopsis

This book introduces the latest research on advanced control charts and new machine learning approaches to detect abnormalities in the smart manufacturing process. By approaching anomaly detection using both statistics and machine learning, the book promotes interdisciplinary cooperation between the research communities, to jointly develop new anomaly detection approaches that are more suitable for the 4.0 Industrial Revolution.

The book provides ready-to-use algorithms and parameter sheets, enabling readers to design advanced control charts and machine learning-based approaches for anomaly detection in manufacturing. Case studies are introduced in each chapter to help practitioners easily apply these tools to real-world manufacturing processes.

The book is of interest to researchers, industrial experts, and postgraduate students in the fields of industrial engineering, automation, statistical learning, and manufacturing industries.

✦ Table of Contents

Contents
Introduction to Control Charts and Machine Learning for Anomaly Detection in Manufacturing
1 Scope of the Research Domain
2 Main Features of This Book
3 Structure of the Book
4 Conclusion
References
Application of Machine Learning in Statistical Process Control Charts: A Survey and Perspective
1 Introduction
2 Machine Learning Techniques Based Control Charts for Process Monitoring
2.1 Kernel-Based Learning Methods
2.2 Dimensionality Reduction
2.3 Neural Network and Deep Learning
3 Machine Learning Techniques Based Control Chart Pattern Recognition
3.1 Regression Tree and Decision Tree Based CCPR
3.2 Neural Network and Deep Learning Based CCPR
3.3 Support Vector Machines Based CCPR
4 Interpreting Out-of-Control Signals Using Machine Learning
5 Difficulties and Challenges for Application of Machine Learning in Statistical Process Control Charts
5.1 Non-stationary Processes
5.2 Big Data Analysis
5.3 Monitoring Image Data
6 Perspectives for Application of Machine Learning in Statistical Process Control Charts in Smart Manufacturing
6.1 Auto-correlated Processes and Non-stationary Processes
6.2 Big Data Analysis
6.3 Real Word Implementation and Hyperparameters Optimization
6.4 Integration of SVM and NN Techniques
6.5 ML Algorithm in the Presence of Drift
6.6 Data Fusion and Feature Fusion
6.7 Control Chart for Complex Data Types
6.8 Monitoring Image Data
7 A Case Study: Monitoring and Early Fault Detection in Bearing
8 Conclusion
References
Control Charts for Monitoring Time-Between-Events-and-Amplitude Data
1 Introduction
2 TBEA Charts for Independent Times and Amplitudes
2.1 Model
2.2 Statistics to Be Monitored
2.3 Control Limit
2.4 Time to Signal Properties
2.5 Comparative Studies
2.6 Illustrative Example
3 TBEA Charts for Dependent Times and Amplitudes
3.1 Motivation
3.2 Model
3.3 Comparative Studies
3.4 Illustrative Example
4 A Distribution-Free TBEA Chart
4.1 Motivation
4.2 Model
4.3 Comparative Studies
4.4 Illustrative Example
5 Conclusions
References
Monitoring a BAR(1) Process with EWMA and DEWMA Control Charts
1 Introduction
2 The Binomial AR(1) Model
3 Methods
3.1 EWMA Control Charts
3.2 DEWMA Control Charts
3.3 Performance Measures
4 Numerical Analysis
5 A Real-Data Example
6 Conclusion and Future Work
References
On Approaches for Monitoring Categorical Event Series
1 Introduction
2 Monitoring Time Series of Event Counts
2.1 Analysis and Modeling of Count Time Series
2.2 Control Charts for Count Time Series
3 Monitoring Categorical Event Series
3.1 Analysis and Modeling of Categorical Time Series
3.2 Control Charts for Categorical Time Series
4 Machine Learning Approaches for Event Sequence Monitoring
5 Real Applications: Sample-Based Monitoring of Categorical Event Series
5.1 A Nominal Event Sequence on Paint Defects
5.2 An Ordinal Event Sequence Regarding Flash on Toothbrush Heads
6 Conclusions
References
Machine Learning Control Charts for Monitoring Serially Correlated Data
1 Introduction
2 Improve Some Machine Learning Control Charts for Monitoring Serially Correlated Data
2.1 Description of Some Representative Machine Learning Control Charts
2.2 Sequential Data De-Correlation
2.3 Machine Learning Control Charts for Monitoring Serially Correlated Data
3 Simulation Studies
4 A Real-Data Application
5 Concluding Remarks
References
A Review of Tree-Based Approaches for Anomaly Detection
1 Introduction
2 Taxonomy and Approaches to Anomaly Detection
2.1 Formal Definition of Anomaly
2.2 Static and Dynamic Problems
2.3 Classes of Algorithms
2.4 Tree-Based Methods
2.5 Structure of the Chapter and Contribution
3 Isolation Tree Based Approaches
3.1 Static Learning
3.2 Dynamic Datasets
3.3 Distributed Approaches
3.4 Interpretability and Feature Selection
4 Experimental Comparison
4.1 Methods Comparison and Available Implementations
4.2 Industrial Case Studies
5 Conclusion and Future Work
References
Joint Use of Skip Connections and Synthetic Corruption for Anomaly Detection with Autoencoders
1 Introduction
2 Related Work
3 Our Anomaly Detection Framework
3.1 Model Configuration
3.2 Corruption Model
3.3 Anomaly Detection Strategies
4 Anomaly Detection on the MVTec AD Dataset
4.1 AESc + Stain: Qualitative and Quantitative Analysis
4.2 Residual- Vs. Uncertainty-Based Detection Strategies
4.3 Comparative Study of Corruption Models
5 Internal Representation of Defective Images
5.1 Formulating Anomaly Detection as an Out-of-Distribution Sample Detection Problem
5.2 Out-of-Distribution Ratio for Clean and Defective Samples
5.3 Analyzing Out-of-Distribution Activations at the Tensor Component Level
5.4 Using the Out-of-Distribution Ratio as a Measure Quantifying the Level of Abnormality
5.5 Related Works and Perspectives
6 Conclusion
References
A Comparative Study of L1 and L2 Norms in Support Vector Data Descriptions
1 Introduction
2 L1 Norm Support Vector Data Description (L1 SVDD)
3 L2 Norm Support Vector Data Description (L2 SVDD)
4 Simulation Study
4.1 Simulation Results
5 Sequential Minimum Optimization (SMO)
5.1 SMO for L1 SVDD
5.2 SMO for L2 SVDD
5.3 Performance Study
6 Stochastic Sub-gradient Descent Solutions
6.1 L1 SVDD
6.2 L2 SVDD
6.3 Performance Evaluation
7 Case Study
7.1 L1 SVDD
7.2 L2 SVDD
8 Conclusion
References
Feature Engineering and Health Indicator Construction for Fault Detection and Diagnostic
1 Condition Monitoring Data Acquisition for Fault Detection and Diagnostic
1.1 Choice of Condition Monitoring Parameters
1.2 Data Acquisition and Preprocessing
2 Signal Processing Techniques for Feature Extraction
2.1 Features in Time Domain
2.2 Features in Frequency Domain
2.3 Features in Time-Frequency Domain
3 Feature Selection
3.1 Common Criteria in Literature for Feature Selection
3.2 A Proposed Metric for Evaluating Feature Performance
3.3 Feature Selection Techniques
3.4 A Proposed Algorithm for Feature Ranking
4 Health Indicator Construction
4.1 Taxonomy of Existing Methods
4.2 Automatic Health Indicator Construction Method
4.3 Applications of the Automatic Health Indicator Construction Method
5 Conclusions
References