A Data Driven Method for Model Based Diagnostics and Prognostics

Michael D. Bryant
Submission Type: 
Full Paper
AttachmentSizeTimestamp
phmc_14_015.pdf3.21 MBSeptember 23, 2014 - 4:20pm

Machines focus power to do work. Implicit in every machine design is an organization to guide power through the machine. We view a machine as a channel through which power must flow. Each component along the channel performs a function to organize power flow. Broken or degraded components disrupt the power flows through the system, which alters system states. A model based diagnostics approach, based on this view of a machine, involves:
1) Construct detailed physics based models of the machine, with direct physical correspondence between elements in the model, and components and faults in the machine.
2) With sensors, measure states off the real machine under service loads.
3) With the model, simulate machine operation under service loads up to and including the sensor outputs,
4) Compare simulated sensor outputs to real sensor outputs.
5) Adjust (tune) the model’s parameters, until simulated sensor readings mimic real sensor readings. The tuned model now emulates current machine behavior, and the parameter values reflect the machine’s components’ condition.
6) Compare the “tuned” model to an “ideal” model of the machine in perfect health (without faults). This “ideal”, provided by the machine designer’s intent, provides a reference for health assessment.
7) Detect and locate faults by tracking changes in the numerical values of parameters of the “tuned” model.
8) Assess machine functional condition by applying information theory theorems to the machine viewed as a communications channel.
9) With the history of the model’s parameters available from periodic tunings, extrapolate the parameter values forward in time, to predict machine future condition, and assess future functional condition via the aforementioned methods.

This diagnostics method will be demonstrated on various machines, including motors, pumps, and gears.

Publication Year: 
2014
Publication Volume: 
5
Publication Control Number: 
015
Submission Keywords: 
model based diagnostics
Data Driven
condition assessment
Submission Topic Areas: 
Model-based methods for fault detection, diagnostics, and prognosis
Submitted by: 
  
 
 
 

follow us

PHM Society on Facebook Follow PHM Society on Twitter PHM Society on LinkedIn PHM Society RSS News Feed