A new online approach for classification of pumps vibration patterns based on intelligent IoT system

•A new approach for classification of pumps vibration patterns using an Intelligent IoT Systems.•In order to identify a normal stage of cavitation, we use vibration signal as an image.•Combinations with feature extractors and classifiers for detect incipient cavitation.•The results showed that our a...

Full description

Saved in:
Bibliographic Details
Published in:Measurement : journal of the International Measurement Confederation 2020-02, Vol.151, p.107138, Article 107138
Main Authors: Hu, Qinhua, Ohata, Elene F., Silva, Francisco H.S., Ramalho, Geraldo L.B., Han, Tao, Rebouças Filho, Pedro P.
Format: Article
Language:English
Subjects:
Cavitation
Condition monitoring
Downtime
Fault detection
Feature extraction
Industrial plants
Intelligent systems
Internet of Things
Machine learning
Machinery condition monitoring
Microelectromechanical systems
Multilayer perceptrons
Predictive maintenance
Pumps
Support vector machines
Vibration analysis
Vibration monitoring
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:•A new approach for classification of pumps vibration patterns using an Intelligent IoT Systems.•In order to identify a normal stage of cavitation, we use vibration signal as an image.•Combinations with feature extractors and classifiers for detect incipient cavitation.•The results showed that our approach is reliable and efficient to detect cavitation in pumps. Machine condition monitoring is a primordial field of study. It allows to avoid downtime in industrial plants, avoiding financial and time losses. In this article, we use an IoT framework to classify the pump’s vibration signal, in order to identify a normal stage of operation, an incipient cavitation stage and a severe cavitation stage. Our approach uses the vibration signal, which is collected with a MEMS sensor, as an image. The feature extractors used in this study: Hu’s Moments, Gray Level Co-occurrence Matrix, Local Binary Patterns, DenseNet169, ResNet50, VGG19 and MobileNet. The classifiers used in this paper were: Gaussian Naive Bayes, Support Vector Machines, Random Forest, Multilayer Perceptron and k-Nearest Neighbors (kNN). The results showed that Hu’s Moments combined with kNN achieved the best accuracy (99.47%) with a score time of 17 ms. Thus, our approach is reliable and efficient to detect cavitation in pumps.
ISSN:0263-2241
1873-412X
DOI:10.1016/j.measurement.2019.107138