Fault Diagnosis of Tenessee Eastman Process with Detection Quality Using IMVOA with Hybrid DL Technique in IIOT

Abstract

This work presents a deep learning (DL) based approach to defect identification in continuous systems. The Tennessee Eastman Procedure was chosen as a use case for this study because it involves a dispersed network of sensors throughout a manufacturing facility. To improve detection quality for fault diagnosis, a hybrid DL approach is presented to choose the most representative sensors. Using an enhanced version of the Multi-Verse Optimization technique, the learning velocity of a Recurrent Neural Network is maximised (IMVOA). An exhaustive design space of solutions varying in sensing and detected quality has been made available by the suggested technique. It is an alternative to the standard Industry 4.0 setup, in which a plethora of decentralised sensor networks report their findings to a single cloud repository. In contrast, the suggested method takes a decentralised approach, which means that processing may occur closer to the sensors that create the data, or even at the advantage of the Internet of Things. Bandwidth, privacy, and latency issues, which often plague centralised methods, are circumvented here. Experimental results demonstrate that the suggested technique delivers fault finding solutions for the Tennessee Eastman Procedure at state-of-the-art detection quality levels. Solution times are 35 times faster than the application with the best detection quality, while feature counts are reduced by an average of 1.99 times. It is important to note that the framework’s scalability creates a design space from which the best possible implementation may be selected based on the requirements of the application.

Full text

Vol.:(0123456789)
SN Computer Science (2023) 4:458
https://doi.org/10.1007/s42979-023-01851-9
SN Computer Science
ORIGINAL RESEARCH
Fault Diagnosis ofTenessee Eastman Process withDetection Quality
Using IMVOA withHybrid DL Technique inIIOT
CuddapahAnitha1· T.RajeshKumar2· R.Balamanigandan2· R.Mahaveerakannan2
Received: 19 February 2023 / Accepted: 20 April 2023
© The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd 2023
Abstract
This work presents a deep learning (DL) based approach to defect identification in continuous systems. The Tennessee
Eastman Procedure was chosen as a use case for this study because it involves a dispersed network of sensors throughout
a manufacturing facility. To improve detection quality for fault diagnosis, a hybrid DL approach is presented to choose the
most representative sensors. Using an enhanced version of the Multi-Verse Optimization technique, the learning velocity of a
Recurrent Neural Network is maximised (IMVOA). An exhaustive design space of solutions varying in sensing and detected
quality has been made available by the suggested technique. It is an alternative to the standard Industry 4.0 setup, in which a
plethora of decentralised sensor networks report their findings to a single cloud repository. In contrast, the suggested method
takes a decentralised approach, which means that processing may occur closer to the sensors that create the data, or even at
the advantage of the Internet of Things. Bandwidth, privacy, and latency issues, which often plague centralised methods,
are circumvented here. Experimental results demonstrate that the suggested technique delivers fault finding solutions for
the Tennessee Eastman Procedure at state-of-the-art detection quality levels. Solution times are 35 times faster than the
application with the best detection quality, while feature counts are reduced by an average of 1.99 times. It is important to
note that the framework’s scalability creates a design space from which the best possible implementation may be selected
based on the requirements of the application.
Keywords Tennessee Eastman process· Fault diagnosis· Deep learning· Improved multi-verse optimization algorithm·
Recurrent Neural Network
Introduction
Due to the significant influence that failures of industrial
schemes, fault analysis is the primary focuses of process
nursing [1, 2], particularly in the context of Industry 4.0.
There is potential for using ML-based classifiers to improve
the fault diagnostic system’s detection accuracy. They could
be educated to detect faults and understand their causes by
analysing data collected from a distributed sensor network.
[3]. To clarify, ML methods typically include two phases—
training and inference—with training often requiring greater
processing power.
Allocating all compute in the cloud field may not be in
line with the needs associated with the real-time concepts
in Industry 4.0. This means that we need to look at other
ways of spreading them out throughout the lower levels of
industrial IoT networks. One common strategy [7] involves
moving the inference step closer to the network’s edge
while keeping the training stage on the cloud. In addition
This article is part of the topical collection “Industrial IoT and
Cyber-Physical Systems” guest edited by Arun K Somani, Seeram
Ramakrishnan, Anil Chaudhary and Mehul Mahrishi.
* R. Mahaveerakannan
mahaveerakannanr.sse@saveetha.com
Cuddapah Anitha
dranithacuddapah17@gmail.com;
anithacuddapah@vidyanikethan.edu
T. Rajesh Kumar
t.rajesh61074@gmail.com
R. Balamanigandan
balamanigandanr.sse@saveetha.com
1 Computer Science andEngineering, School ofComputing,
Mohan Babu University, Erstwhile Sree Vidyanikethan
Engineering College, Tirupati, AndhraPradesh, India
2 Department ofComputer Science andEngineering,
Saveetha School ofEngineering, SIMATS, Chennai,
TamilNadu602105, India
Content courtesy of Springer Nature, terms of use apply. Rights reserved.