design and deployment of industrial sensor networks: experiences from a semiconductor plant and the...

Design and Deployment of Industrial Sensor Networks:Experiences from a Semiconductor Plant and the North Sea

Lakshman Krishnamurthy†, Robert Adler†, Phil Buonadonna‡, Jasmeet Chhabra†,

Mick Flanigan†, Nandakishore Kushalnagar†, Lama Nachman†, Mark Yarvis† 1

Sensys 2005

Presenter SY

Outline Introduction Background and Motivation Application Analysis Pre-planning and Requirements Hardware Architecture and Comparison Network Architecture and Comparison Mica2 and Inter Mote Performance Comparison Conclusions Comments Discussion – what we need.

Introduction

Predictive Maintenance (PdM) – monitor and assess the health status of a piece of equipment that is in service.

They have chosen one form of PdM – Vibration analysis.

Deployed on two environments. Semiconductor fabrication plant Oil tanker

Introduction

Semiconductor fabrication plant – Central utility support building (CUB) Produce pure water, handle gases, and

process waste water. Two platforms in same deployment. Mica2 and

Intel Mote Oil tanker

Roughest environments for industrial sensor networks.

Background and Motivation

PdM technologies: Vibration Analysis: Time domain and frequenc

y domain waveform analysis. Oil Analysis: wear particles, viscosity, acidity,

and raw elements. Infrared Thermography: detect abnormal heat

sources and compare to baseline data. Ultrasonic Detection: detect wall thickness, cor

rosion and blistering, erosion, flow dynamics, and wear patterns.

Background and Motivation

Why predictive maintenance (PdM) Reduce cost: equipment failures, repair, replac

ement Change business model: from calendar-based

maintenance to indicator-driven maintenance Quantify the quality of a new system within the

warranty period. Meeting factory uptime and reliability requirem

ents

Background and Motivation

Background and Motivation

An industry cross section shows: Online system in the market is less than 10% 20% use manual data collection. Finding a solution to address this market and t

ap into the remaining 70% may represent a killer application for wireless sensor netwroks.

Application Analysis

Vibration analysis. Wilconxon model 786A sensors:

Calibrated 100mV/G with 5% calibration sensitivity.

80g’s peak. Sampling rate of 19.2KHz.

Application Analysis

Outline Introduction Background and Motivation Application Analysis Pre-planning and Requirements Hardware Architecture and Comparison Network Architecture and Comparison Mica2 and Inter Mote Performance Comparison Conclusions Comments Discussion – what we need.

Pre-planning and Requirements

Site survey: an industrial site-survey addresses the following issues: RF coverage – identify shadows. RF interference – 802.11 AP, motor

controllers, solid state switchgear. Mechanics – Practical matters of where and

how to mount sensors, sensor nodes, and gateways.

Pre-planning and Requirements

Site Survey Experiences -- Setting

Sensor nodes placed close to sensing points. Gateway were placed based near available

power outlets and wired network connectivity.

Site Survey Experiences – Oil tanker

Site Survey Experiences – Procedures

Sensor nodes executed an end-to-end test. Statistics were collected: packet loss and

packet hop count. For the gateways, a simple data copy was

used to evaluate 802.11b connectivity. Passive RF spectrum analysis was performed

with the sensor network disabled.

Site Survey Experiences -- CUB

Overall good connectivity between potential gateway locations and sensing points.

916MHz Mica2 exhibited some RF shadows at certain points.

Moving to alternate points alleviated the shadowing effect.

Intel Motes has no shadow…………… 802.11b connectivity between Stargate gatew

ays was initially excellent, be it changed over the course of days.

Site Survey Experiences – Oil Tanker

Excellent sensor node and gateway connectivity.

433MHz Mica2 were not able to communicate through a non-watertight bulkhead/doorway.

Intel Motes again …… goods. 802.11b connectivity was excellent Shipboard spectrum analysis test showed no

adverse interference conflicts for the motes, the gateways, or the ship.

Site Survey Experiences -- conclusion

Fewer gateway nodes were required to achieve adequate coverage than initially anticipated.

Coverage of outdoor nodes from an indoor gateway was shown to be feasible.

Higher RF frequency was preferred in the industrial environment.

Requirements

Fault tolerance and reliability Long-lived battery powered operation Maintainable Seamless integration into existing applicatoin Security

Hardware Architecture and comparison

Hardware Architecture and comparison

Hardware Architecture and comparison – summaries Multiple sensors connected to one sensor

board. Intel motes is more powerful but more

complex. Intel Mote’s fast UART (up to 960 kb/s) was

more than adequate to support required sampling rate of 16 bit data at 19.2 kHz.

Network Architecture – Mica2

Three tier architecture. Tier 1: sensor nodes form clusters. Tier 2: Stargate gateway, clusters head. Form

a 802.11b backbone. Tier 3: Bridge Stargate and server.

Mica2 Power Management and transfer Protocol Centralized wake/sleep protocol. Cluster-head (gateway) schedules the wake and slee

p period. When cluster awake, it initiates single-destination-DS

DV routing to find a path to cluster-head. Cluster-head schedules the capture/transfer for every

sensor nodes. When data has been transferred to the gateway, it is t

ime-stamped. Gateway periodically copies data files to the root Star

gate.

Bulk Transfer Protocol

When a sensor node is scheduled to transmit, it sends a connection request to the gateway.

This request contains a set of connection parameters: fragment size, data size, and transfer rate.

Intel Mote (bluetooth) has larger sliding window, larger fragment size, higher transfer rate.

Intel Mote – The Different

Scatternet formation algorithm: create a tree topology with a predefined root node.

Implemented a network low power mode.

Fault Tolerance

Multiple watchdog timers were used to recover from any non operational state.

Storage of the core network states in cluster heads.

Intentional re-initialization of sensor nodes after each collection cycle.

Non-volatile storage of critical state at the cluster head after every collection.

Outline Introduction Background and Motivation Application Analysis Pre-planning and Requirements Hardware Architecture and Comparison Network Architecture and Comparison Mica2 and Inter Mote Performance Comparison Conclusions Comments Discussion – what we need.

Performance Across Cluster

Power Consumption

Power Consumption

Conclusion

More capabilities in the sensing platform enabled a simpler and more effective overall system design.

Sparse clusters of sensors in industrial environment.

Multi-hop is uncommon. Tiered architectures is more suitable.

Comments

Ya, more powerful node is more suitable for high data rate applications, we all know.

But the system is more complex. Mica2 is an old design with quite low ability in

terms of microcontroller.

Discussions – what do we need for our own deployment. We need a plan before deployment. Site-survey Consider the maintainability Battery power or permanent power. Fault tolerance – watch dog.