System Monitoring, Diagnosis, and
Prognosis
What is the status of your equipment? When does it need maintenance service? Is the performance of your production/service
system optimal? Where is the
bottleneck? How to improve your system
productivity? Our work on system
monitoring, diagnosis, and prognosis will help answer these questions. The focus is two-fold: (1) condition-based
maintenance (CBM) to maximize equipment uptime, and (2) diagnostic metrics to
detect bottlenecks and improve system productivity.
Condition-based
Maintenance
The idea of CBM is to
monitor equipment using various sensors to enable real-time diagnosis of
impending failures and prognosis of equipment health. An intelligent equipment health monitoring
system architecture has been developed, as shown in Figure 1. It
consists of four phases; namely, data acquisition, feature extraction, model
generation, and model deployment. Our
focus is on feature extraction and model generation. Feature extraction refers to the processing of
raw sensor data into useful features for model building. It includes noise filtering, missing value
imputation, and feature extraction from time-series. Although many features can be extracted, not
all of them are needed for diagnosis/prognosis.
Therefore, a dimensionality reduction procedure is used to select
appropriate features for model generation.
Model generation follows the
knowledge-based modeling approach. An
advanced model tuning technique, Adaptive Mamdani
Fuzzy Model (AMFM), has been developed.
The resultant diagnosis/prognosis model is accurate, robust, and highly
interpretable, as demonstrated in a paper entitled “A Comparison of
Computational Intelligence and Statistical Methods in Condition Monitoring for
Hard Turning” by R. Kothamasu,
S. H. Huang, and W. H. VerDuin (International Journal of Production Research, Vol. 43, No.
3, 2005, pp. 597-610). We have successfully applied the developed
CBM technology to a number of real-world cases as summarized in Table 1.
Figure 1. Intelligent equipment health
monitoring system architecture.
Table 1. Summary of successful real-world
CBM applications.
|
Application |
Monitored
Parameters |
Result |
Collaborator |
|
Aircraft Engine Fault
Diagnosis |
Temperate, fuel flow, shaft
speed, shaft vibration |
Achieved 97% diagnosis
accuracy (two failure modes) with sparse data, a 15% improvement over human
experts |
Rolls-Royce Corporation |
|
Bearing Fault Diagnosis |
Acoustic and vibration
signals |
Achieved 100% diagnosis
accuracy with 2 features |
NIST |
|
Tool Wear Monitoring |
Force in 3 axes, feed,
speed |
Achieved 96% and above
predictive accuracy in 85 samples |
TechSolve, Inc. |
Diagnostic
Productivity Metrics
While CBM can help maximize the uptime of individual
equipment, diagnostic productivity metrics aim to detect system bottleneck and
improve the productivity of an entire production/service system. This goal can be achieved through system
monitoring and the use of system level metrics that capture process
flow/station interconnectivity information and individual station productivity
measure, as illustrated in Figure 2. A
system is decomposed into a number of common configurations including series,
parallel, assembly, and expansion. For
each configuration, diagnostic metrics based on the concept of overall
equipment effectiveness (OEE) are developed to enable bottleneck detection and
improvement opportunity identification.
Figure 2. Diagnostic metrics for productivity improvement.
Technical Contact
Dr. Samuel H. Huang,
Associate Professor and Director
Intelligent Systems
Laboratory
Department of Mechanical,
Industrial and Nuclear Engineering
Phone:
(513)556-1154
Fax: (513)556-3390
E-mail: sam.huang@uc.edu