Improving Quality of Service

in Industry 4.0 using

Federated Learning
Presented by - Patel Kathan Prashant (MHC2023004)

Guided by- Prof. O.P. Vyas

Table of contents
01 02 03

Introduction Literature Review Analysis

01

Introduction
Federated Learning
Here’s the single-sentence explanation:

● Central machine learning: move the data to the computation

● Federated (machine) learning: move the computation to the data

1. You choose a model that is either pre-trained on the central server or isn’t trained at all.
2. The next step would be the distribution of the initial model to the clients (devices or local
servers).
3. Each client keeps training it on-site using its own local data. The important part is that
this training data can be confidential, including personal photos, emails, health metrics,
or chat logs.
4. When locally trained, the updated models are sent back to the central server via
encrypted communication channels. Server doesn’t get any real data, but only trained
parameters of the model (e.g., weights in neural networks). The updates from all clients
are averaged and aggregated into a single shared model, improving its accuracy.
5. Finally, this model is sent back to all devices and servers.

Eg. Google Voice Assistant

Industry 4.0
● Industry 4.0 refers to a new phase in the Industrial Revolution that focuses heavily on
interconnectivity, automation, machine learning, and real-time data

● Industry 4.0 brings three foundational types of disruptive technologies to form the new
future

○ connectivity, data, and computational power: cloud technology, the Internet,

sensors
○ analytics and intelligence: advanced analytics, machine learning, artificial
intelligence
○ human–machine interaction: virtual reality (VR) and augmented reality (AR),
robotics and automation, autonomous guided vehicles
02
Literature
Review
Federated Learning for Industrial Internet of Things in
Future Industries [1]
Authors: Dinh C. Nguyen, Ming Ding, Pubudu N. Pathirana, Aruna Seneviratne, Jun Li,
Dusit Niyato, and H. Vincent Poor

Publication Year: 2021

Research Purpose: The primary goal is to explore how FL can address the limitations of
traditional centralized AI approaches in industrial systems by enabling distributed,
privacy-preserving, and scalable machine learning models at the network edge.The
study seeks to identify key challenges and highlight the significance of FL in IIOT services
● Integration of FL and IIOT Network

○ System Initialization and Device Selection

○ Distributed Local Training and Updates
○ Model Aggregation and Download
○ Iterated Training

● Key Benefits of FL Integration in IIOT

○ Data Privacy Enhancement

○ Low-latency Network Communication
○ Improved Learning Quality
Figure 1: FL-IIOT Architecture [1]
Where FL can be useful in Smart Manufacturing?

● Predictive Maintenance of different equipments

● Supply Chain Optimization
● Quality of Control
● Anomaly Detection
Deep Anomaly Detection for Time-Series Data in
Industrial IoT [2]

Authors: Yi Liu, Sahil Garg, Jiangtian Nie, Yang Zhang Zehui Xiong,Jiawen Kang and M.
Shamim Hossain

Publication Year: 2021

Research Purpose: To develop a cutting edge anomaly detection framework that

combines FL, DL models and attention mechanisms to address the challenges of edge
device failures, user privacy and anomaly detection accuracy in IIOT
Anomaly Detection Techniques: Highlights the importance of anomaly detection in IIoT.
Traditional anomaly detection methods such as machine learning algorithms, and deep
learning models have been widely used. However, these methods may face challenges in
handling complex time-series data and ensuring privacy.

Challenges in Anomaly Detection for Industrial IoT: Discusses the challenges faced in
anomaly detection for industrial IoT applications, such as data heterogeneity, data
privacy concerns, and scalability issues.

Federated Learning: It is introduced as a collaborative machine learning approach that

enables edge devices to train a shared model without sharing their raw data. This
decentralized training process helps in preserving data privacy and security while
improving the overall model performance
Communication-Efficient Federated Learning: By minimizing the communication
overhead and optimizing model aggregation strategies, federated learning can be
effectively applied in resource-constrained IoT environments.

Research Gap: The need for more robust anomaly detection models tailored for
industrial IoT settings, scalable federated learning algorithms, and privacy-preserving
mechanisms.
Communication-Efficient Federated Learning for
Anomaly Detection in Industrial Internet of Things [3]
Authors: Yi Liu, Neeraj Kumar, Zehui Xiong, Wei Yang Bryan Lim, Jiawen Kang, Dusit
Niyato

Publication Year: 2020

Research Purpose: The Primary goal is to develop an efficient anomaly detection model
using Federated Learning that can accurately and timely detect anomalies in IIOT.
Additionallyit also aims to improve communication efficiency through gradient
compression mechanisms.
Federated Learning: It is introduced as a collaborative machine learning approach that
enables edge devices to train a shared model without sharing their raw data. This
decentralized training process helps in preserving data privacy and security while
improving the overall model performance

CNN-LSTM Model: The proposed framework leverages a CNN-LSTM model for anomaly
detection, combining(CNN to extract features from time-series data and LSTM modules
for time-series prediction. This model aims to accurately and timely detect anomalies in
industrial IoT devices

Gradient Compression: To enhance communication efficiency, a gradient compression

scheme based on Top-k selection is proposed to reduce the number of gradients
uploaded by edge devices. This mechanism significantly reduces communication
overhead while maintaining the accuracy of anomaly detection
Toward Accurate Anomaly Detection in Industrial
Internet of Things Using Hierarchical Federated Learning
[4]

Authors: Xiaoding Wang , Sahil Garg,Hui Lin , Jia Hu , Georges Kaddoum Md. Jalil Piran,
and M. Shamim Hossain

Publication Year: 2022

Research Purpose: To propose a reliable anomaly detection strategy using federated

learning techniques. By leveraging deep reinforcement learning algorithms and
incorporating measures for privacy preservation, the study aims to develop a universal
anomaly detection model that can effectively detect anomalies in various IIoT scenarios.
Challenges in Anomaly Detection: Identifies two main challenges in anomaly detection -
privacy preservation and improving detection accuracy with abnormal devices

Deep Reinforcement Learning (DRL) and Federated Learning (FL): Discusses the use of
DRL-based solutions for anomaly detection. It also introduces federated learning as a
decentralized training approach that collaboratively optimizes model parameters
without sharing raw data.

Research Gap: Need to improve anomaly detection accuracy, especially when dealing
with abnormal devices. Many existing Anomaly detection approaches require access to
local datasets which raises issue of privacy leakage
Failure Prediction in Production Line Based on
Federated Learning [5]

Authors: Ning Ge,GuanghaoLi,Li Zhang Yi Liu

Publication Year: 2022

Research Purpose: Investigate the effectiveness of federated learning (FL) in

comparison to centralized learning (CL) for failure prediction in production lines.
Experiment Process
● Evaluation Metrics such as Precision, F1 Score, Area under the Curve are chosen to
assess the FL and CL models.
● Bosch Dataset is used which has measurements and sensor data from various stages of
the production process
● FL models (FedSVM and FedRF) and CL models (SVM and RF) are trained on the training
data using the specified algorithms.
● The trained FL and CL models are evaluated on the testing data using the selected
evaluation metrics
This table presents the accuracy results for the Federated Support Vector Machine (FedSVM),
Support Vector Machine (SVM), Federated Random Forest (FedRF), and Random Forest (RF)
models in the context of failure prediction in production lines

Model Accuracy

FedSVM 85.2

SVM 82.6

FedRF 87.4

RF 84.9
03

Analysis
So why we are choosing FL over CL?
● Data Privacy and Security
● Bandwidth & Latency
● Resource Efficiency
● Scalability
● Resilience to Network Failure
Technical Challenges
● Limited bandwidth and communication latency
● Integration with existing infrastructure
● Scalability to large-scale deployments
● Non-IID data distribution
References
[1] Y. Liu et al., "Deep Anomaly Detection for Time-Series Data in Industrial IoT: A
Communication-Efficient On-Device Federated Learning Approach," in IEEE Internet of Things
Journal, vol. 8, no. 8, pp. 6348-6358, 15 April15, 2021, doi: 10.1109/JIOT.2020.3011726.

[2] D. C. Nguyen et al., "Federated Learning for Industrial Internet of Things in Future
Industries," in IEEE Wireless Communications, vol. 28, no. 6, pp. 192-199, December 2021, doi:
10.1109/MWC.001.2100102.

[3] Y. Liu, N. Kumar, Z. Xiong, W. Y. B. Lim, J. Kang and D. Niyato, "Communication-Efficient

Federated Learning for Anomaly Detection in Industrial Internet of Things," GLOBECOM 2020 -
2020 IEEE Global Communications Conference, Taipei, Taiwan, 2020, pp. 1-6, doi:
10.1109/GLOBECOM42002.2020.9348249
[4] X. Wang et al., "Toward Accurate Anomaly Detection in Industrial Internet of Things Using
Hierarchical Federated Learning," in IEEE Internet of Things Journal, vol. 9, no. 10, pp.
7110-7119, 15 May15, 2022, doi: 10.1109/JIOT.2021.3074382.

[5] Ning Ge & Guanghao Li & Li Zhang & Yi Liu, 2022. "Failure prediction in production line
based on federated learning: an empirical study," Journal of Intelligent Manufacturing,
Springer, vol. 33(8), pages 2277-2294, December.