Maximal information-based nonparametric exploration for condition monitoring data

Yang Hu, Thomas Palmé, and Olga Fink
Submission Type: 
Full Paper
phmec_16_040.pdf1.02 MBJune 27, 2016 - 11:57pm

The system condition of valuable assets such as power plants is often monitored with thousands of sensors. A full evaluation of all sensors is normally not done. Most of the important failures are captured by established algorithms that use a selection of parameters and compare this to defined limits or references.
Due to the availability of massive amounts of data and many different feature extraction techniques, the application of feature learning within fault detection and subsequent prognostics have been increasing. They provide powerful results. However, in many cases, they are not able to isolate the signal or set of signals that caused a change in the system condition.
Therefore, approaches are required to isolate the signals with a change in their behavior after a fault is detected and to provide this information to diagnostics and maintenance engineers to further evaluate the system state.
In this paper, we propose the application of Maximal Information-based Nonparametric Exploration (MINE) statistics for fault isolation and detection in condition monitoring data.
The MINE statistics provide normalized scores for the strength of the relationship, the departure from monotonicity, the closeness to being a function and the complexity. These characteristics make the MINE statistics a good tool for monitoring the pair-wise relationships in the condition monitoring signals and detect changes in the relationship over time.
The application of MINE statistics in the context of condition monitoring is demonstrated on an artificial case study. The focus of the case study is particularly on two of the MINE indicators: the Maximal information coefficient (MIC) and the Maximum Asymmetry Score (MAS).
MINE statistics prove to be particularly useful when the change of system condition is reflected in the relationship between two signals, which is usually difficult to be captured by other metrics.

Publication Year: 
Publication Volume: 
Publication Control Number: 
Page Count: 
Submission Keywords: 
Fault Detection; Condition Monitoring; Maximal information;
Submission Topic Areas: 
CBM and informed logistics
Data-driven methods for fault detection, diagnosis, and prognosis
Submitted by: 

follow us

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