The group of distributed signal processing in Machine Condition Monitoring using Wireless Sensor Networks

Introduction

Wireless sensor networks have a broad range of application. For instance in mechanical engineering, civil engineering, aviation, military, agriculture, traffic control, home automation, medical care or environmental monitoring and natural disaster prediction.
Our group is concerned in employing distributed wireless sensor network in Machine Condition Monitoring (MCM). This approach may bring many advantages compare to a classical wired centralized system:

  • easy installation to operating machine
  • distributed information processing – robustness, self-adaptation
  • easy adaptation for new tasks

WSN is created from nodes, where each node is an embedded system composed of simple microprocessor, sensors, RF transceiver and power source. Thanks to modern technologies as MEMS the whole system can be reduced to millimeters size.

WSN embedded unit Mesh routing Sheme of an IRIS node
WSN embedded unit Mesh routing algorithm Scheme of an IRIS node

Developed system

Currently we are working on an autonomous system based on the Crossbow IRIS network (programmed under TinyOS) for MCM. Nodes are equipped with accelerometers which sense vibration of a rotary machine. In the each node are evaluated features suitable to describe faulty behaviour of a machine and then are used as inputs for the one-class classifier. Then the nodes share acquired information about the machine state. Thanks to the one-class classifier, the system is taught on a health state of a machine and can detect faulty behaviour without the need to give the classifier faulty data (break down a machine). This approach may bring better robustness, self-adaptation, easier deployment and lower price compared to the classical wired MCM systems.

 

Operation of the proposed system is split into the three phases:

  1. In the first phase nodes of WSN are used just for acquisition of raw samples, which are downloaded via gateway to a powerful computational unit (i.e. personal computer).
  2. The computer process all measured samples, selects and extracts suitable features from the signals and trains the classifier. When the classifier is finished, its mathematical model is programmed (coded) in nesC language for TinyOS embedded operating system. Created code is uploaded via the over-the-air-programming to each node placed on a machine.
  3. At each node independently runs the programmed software, which ensures the measurement of samples, feature extraction and classification. The node informs the gateway, if a failure is detected.

MCM sysstem description

 

These three steps can be repeated anytime to adapt system for new circumstances. In the first phase the data flow from nodes to the gateway is very intensive, for that reason data can not be collected from the all nodes in the same time, but are collected from the groups of sensors. However in the third phase the nodes send evaluated simple information and thanks that the network is not overloaded.


Equipment

Our group currently works with following WSN systems

Highlighted instruments

  • Brüel & Kjær PULSE analyzer
  • Agilent 34972A LXI Data acquisition unit
  • Agilent 34410 Multimeter
  • Agilent 33220A Function generator
  • Tektronix TDS2004B oscilloscope

People

Radislav Smid – associate professor
The head of the group – supervisor of involved doctoral students.

Jan Neuzil – doctoral student
Interested in selecting suitable features for MCM, one-class classification and TinyOS programming. Creates MCM system based on IRIS nodes.

Ondrej Kreibich 
Deals with data fusion in WSN. Creates the rotary machine test equipment. MCM expert.

Grants – programmes

  • Synergy – Mobile Sensor Systems and Networks. Grant Agency of Czech Republic project 102/09/H081
  • Czech Technical University students grant SGS10/207/OHK3/2T/13