IoT Streams for Data-Driven Predictive Maintenance and IoT, Edge, and Mobile for Embedded Machine Learning (Communications in Computer and Information Science)

✍ Scribed by Joao Gama (editor), Sepideh Pashami (editor), Albert Bifet (editor), Moamar Sayed-Mouchawe (editor), Holger Fröning (editor), Franz Pernkopf (editor), Gregor Schiele (editor), Michaela Blott (editor)

• Publisher: Springer
• Year: 2021
• Tongue: English
• Leaves: 317
• Category: Library

✦ Synopsis

This book constitutes selected papers from the Second International Workshop on IoT Streams for Data-Driven Predictive Maintenance, IoT Streams 2020, and First International Workshop on IoT, Edge, and Mobile for Embedded Machine Learning, ITEM 2020, co-located with ECML/PKDD 2020 and held in September 2020. Due to the COVID-19 pandemic the workshops were held online.
The 21 full papers and 3 short papers presented in this volume were thoroughly reviewed and selected from 35 submissions and are organized according to the workshops and their topics: IoT Streams 2020: Stream Learning; Feature Learning; ITEM 2020: Unsupervised Machine Learning; Hardware; Methods; Quantization.

✦ Table of Contents

IoT Streams 2020 Preface
IoT Streams 2020 Organization
ITEM 2020 Preface
ITEM 2020 Organization
Contents
IoT Streams 2020: Stream Learning
Self Hyper-parameter Tuning for Stream Classification Algorithms
1 Introduction
2 Related Work
3 Self Parameter Tuning Method
3.1 Nelder-Mead Optimization Algorithm
3.2 Dynamic Sample Size
3.3 Stream-Based Implementation
4 Experimental Evaluation
5 Conclusion
References
Challenges of Stream Learning for Predictive Maintenance in the Railway Sector
1 Introduction
2 An Overview of Predictive Maintenance
2.1 Knowledge-Based Approach
2.2 Data-Driven Approach
3 An Overview of Stream Learning
3.1 Algorithms
3.2 Concept Drifts
4 Application in the Railway Sector
4.1 The Need of Maintenance for the Railway
4.2 Predictive Maintenance for the Railway
4.3 Benefits of Stream Learning in Railway Maintenance
5 Conclusion
References
CycleFootprint: A Fully Automated Method for Extracting Operation Cycles from Historical Raw Data of Multiple Sensors
1 Introduction
2 Problem Definition
3 Related Work
4 Proposed Solution
5 Algorithm CycleFootprint
5.1 Transformation of Signal to State Sequences
5.2 Mining Footprints
6 Experimental Evaluation
6.1 Dataset
6.2 Configuration
6.3 Results
7 Conclusion
References
Valve Health Identification Using Sensors and Machine Learning Methods
1 Introduction
2 Related Work
3 Data
3.1 Dataset Description
3.2 Time and Frequency Domain Features
3.3 Signal Transformations
4 Classifying Valve States
5 Detecting Anomalous Valve Behaviour
5.1 Effectiveness of Distance Metrics
5.2 Anomaly Detection Using Distance Metrics
6 Conclusion and Future Directions
References
Failure Detection of an Air Production Unit in Operational Context
1 Introduction
2 Problem Definition
3 Related Work
3.1 Predictive Maintenance
3.2 Anomaly Detection
4 Our Proposal
5 Experimental Evaluation
6 Conclusions
References
IoT Streams 2020: Feature Learning
Enhancing Siamese Neural Networks Through Expert Knowledge for Predictive Maintenance
1 Introduction
2 Foundations and Related Work
2.1 Distance-Based Time Series Classification
2.2 Feature-Based Time Series Representation
2.3 Siamese Neural Networks for Time Series Similarity
2.4 Related Work
3 Infusing Expert Knowledge About Attribute Relevance
3.1 Infusing Expert Knowledge at the Input Level
3.2 Infusing Expert Knowledge with 2D CNNs
4 Evaluation
4.1 Fischertechnik Model Factory Data Set
4.2 Approaches for Measuring Time Series Similarity
4.3 Experimental Setup and Training Procedure
4.4 Predictive Maintenance Related Quality Measures
4.5 Results
5 Conclusion and Future Work
A Dataset
References
Explainable Process Monitoring Based on Class Activation Map: Garbage In, Garbage Out
1 Introduction
2 Related Work
3 Control Chart in Process Monitoring
3.1 Multivariate Control Charts
3.2 Latent Variable Based Multivariate Control Charts
4 Proposed Method
5 Experiment and Performance Analysis
6 Conclusion and Future Work
References
AutoML for Predictive Maintenance: One Tool to RUL Them All
1 Introduction
2 Remaining Useful Lifetime Estimation
3 Related Work
4 Combined Feature Engineering and Model Selection
4.1 Extending ML-Plan for RUL Estimation
5 Experiments
5.1 Experimental Setup
5.2 Results
6 Conclusion
References
Forklift Truck Activity Recognition from CAN Data
1 Introduction
2 Definition of the Task
3 The Data
4 Methods
5 Results
5.1 Baseline Classifier
5.2 Classifiers Trained from Unlabeled Data
6 Conclusions
References
Embeddings Based Parallel Stacked Autoencoder Approach for Dimensionality Reduction and Predictive Maintenance of Vehicles
1 Introduction
2 Data Representation and Pre-processing
3 Hypothesis
4 Proposed Method
5 Experimental Evaluation and Results
5.1 Embeddings Evaluations
5.2 Parallel Stacked Autoencoder Regression
5.3 Combining Autoencoder Regression
6 Conclusion
References
IoT Streams 2020: Unsupervised Machine Learning
Unsupervised Machine Learning Methods to Estimate a Health Indicator for Condition Monitoring Using Acoustic and Vibration Signals: A Comparison Based on a Toy Data Set from a Coffee Vending Machine
1 Introduction
2 Toy Data Set for Condition Monitoring of Assets
2.1 The Coffee Vending Machine
2.2 Feature Extraction
2.3 Modeling Algorithms
2.4 Sensor Data Fusion
3 Experimental Setup
3.1 Experimental Settings
3.2 Autoencoder Architectures
4 Results and Discussion
4.1 Model Comparison
4.2 Level/Trend Analysis
4.3 Causality
5 Conclusion
References
Unsupervised Anomaly Detection for Communication Networks: An Autoencoder Approach
1 Introduction
2 Use Case and Requirements
2.1 Use Case
2.2 Anomaly Types
2.3 Requirements
2.4 AD Techniques
3 Related Work
4 Autoencoders for Unsupervised Anomaly Detection
5 Evaluation
5.1 AE Architecture Evaluation
5.2 Skyline Use Case Evaluation
6 Conclusion
References
Interactive Anomaly Detection Based on Clustering and Online Mirror Descent
1 Introduction
2 Related Work
3 Proposed Method
3.1 Preliminaries and Goals
3.2 OMD-Clustering
4 Evaluation
5 Conclusion and Future Work
References
ITEM 2020: Hardware
hxtorch: PyTorch for BrainScaleS-2
1 Introduction
2 Methods and Tools
3 Implementation
3.1 PyTorch Integration
3.2 Graph Representation and Just-in-time Execution
3.3 Partitioning
3.4 Parallel Execution of Convolutional Layers
3.5 Handling Hardware Setup, Initialization and Parameters
4 Results
4.1 Performance Evaluation
4.2 Application Example: Human Activity Recognition
5 Discussion and Outlook
References
Inference with Artificial Neural Networks on Analog Neuromorphic Hardware
1 Introduction
2 Methods
2.1 Structure of a Multiply-Accumulate Operation
2.2 Calibration
2.3 Training with Hardware in the Loop
3 Results
3.1 Characterization
3.2 MNIST Benchmark
4 Discussion
5 Contributions
References
Search Space Complexity of Iteration Domain Based Instruction Embedding for Deep Learning Accelerators
1 Introduction
2 Related Work
3 Tensor Compute Graphs
3.1 Semantics of Tensor Compute Trees
3.2 Instruction Selection
4 Kernel Transformations
5 Search Space Exploration
5.1 Search Space
5.2 Solution Space
5.3 Transformation Impact
5.4 Memory Transformations
6 Conclusion
A Tensor Compute Graphs
A.1 Access Functions
A.2 Data Dependencies
B Instruction Selection
B.1 Instruction Representation
B.2 Preprocessing
B.3 Selection Algorithm
References
On the Difficulty of Designing Processor Arrays for Deep Neural Networks
1 Introduction
2 Related Work
3 Emulator Design
4 Evaluation of Network Architectures
4.1 Case Study: ResNet-152
4.2 Impact of DNN Architecture Development on Systolic Array Performance
5 Robust/Optimal Processor Architecture Configuration
6 Conclusion
References
ITEM 2020: Methods
When Size Matters: Markov Blanket with Limited Bit Depth Conditional Mutual Information
1 Introduction
2 Background
2.1 Conditional Mutual Information
2.2 Markov Blanket Discovery Algorithm
3 Limited Bit Depth Conditional Mutual Information
3.1 Empirical Study
3.2 IAMB with Limited Bit Depth CMI: Approximation of the P-Value
4 Experimental Results: Application in Markov Blanket
4.1 Quality of the Selected Features
4.2 Classification Accuracy
5 Conclusions
References
Time to Learn: Temporal Accelerators as an Embedded Deep Neural Network Platform
1 Introduction
2 Related Work
3 Temporal Accelerators
3.1 Splitting an Accelerator
3.2 Execution and Intermediate Results
3.3 Overhead of Executing a Temporal Accelerator
4 CNN as a Temporal Accelerator
5 Evaluation
5.1 Reconfiguration Overhead
5.2 CNN Temporal Accelerator vs. Single Configuration CNN
5.3 Device Price Reduction
6 Conclusion and Outlook
References
ML Training on a Tiny Microcontroller for a Self-adaptive Neural Network-Based DC Motor Speed Controller
1 Introduction
2 Related Work
3 Background on Direct Current (DC) Motors
4 Self-adaptive DC Motor Control Scheme
4.1 NN-based Direct Inverse Control (DIC)
4.2 Training Strategy for Tiny Microcontrollers
4.3 Self-adaptation During Operation
5 Experiments
5.1 Simulation-Based NN Architecture Search and Training Setup
5.2 Implementation on a Cortex-M0 Microcontroller
6 Conclusion
References
ITEM 2020: Quantization
Dynamic Complexity Tuning for Hardware-Aware Probabilistic Circuits
1 Introduction
2 Background
2.1 Probabilistic Sentential Decision Diagrams
2.2 Cost Versus Accuracy Pareto Trade-Off Extraction
3 Model-Complexity Switching Strategy
4 Experiments
4.1 Ensemble Set Construction
4.2 Comparison of Static Ensembles and Switching Strategy
5 Related Work
6 Discussion
References
Leveraging Automated Mixed-Low-Precision Quantization for Tiny Edge Microcontrollers
1 Introduction
2 Related Work
3 Automated Mixed-Precision Quantization for MCU
3.1 MCU-Aware Optimization Objectives
3.2 Automated Precision Tuning
3.3 HAQ for MCU Deployments
4 Experimental Results
4.1 Experimental Setup
4.2 Automated Search for Weight-Only Quantization Policies
4.3 Automated Search for Weight and Activation Quantization Policies
5 Conclusion
References
Author Index