際際滷

際際滷Share a Scribd company logo

Condition Monitoring of Variable Speed Machinery

Download as PPTX, PDF
Jordan McBain
Jordan McBain

This document proposes methods for novelty detection in variable speed machinery. It summarizes that: 1) Monitoring machinery is limited by changes in speed and load, termed "nuisance parameters". 2) A novel method called "multi-modal novelty detection" is proposed to employ intuition from "statistical parameterization" without its limitations by adding modal parameters like speed to feature vectors. 3) An experimental methodology is described involving sensors, data acquisition, and segmentation to generate feature vectors from variable speed tests with simulated faults, and results show standard techniques fail without speed adaptation while multi-modal novelty detection compares well.

1 of 14
Downloaded 46 times
Advancing the Online Monitoring of Variable Speed Machinery Jordan McBain, P.Eng. mcbainjj@gmail.com Sudbury, Ontario
Condition Monitoring of Variable Speed Machinery
Introduction Monitoring of machinery largely limited to constant conditions Changes in speed and load termed nuisance parameters Variable speed/load machinery ubiquitous Ref: Stack Resonances/vibration power
Novelty Detection Limited data characterizing normal state Little or no data for abnormal states Compute feature vectors of vibration (e.g. AR model) Methods SVDD and Statistical Boundaries
Statistical Parameterization Vibration strongly tied to temp (speed) Advanced by Keith Worden (Structural health monitoring) Segment feature vectors into small groups of modal value Compute statistics for each group (bin) Trend with regression or interpolation Suffers from Double curse of dimensionality Describe healthy state for all segments of modal parameter Gaussian distribution Good heuristic
Multi-Modal Novelty Detection Employ intuition from Statistical Parameterization Dont flatten data into bins Add modal parameter (speed) to feature vector Use any novelty detection technique One parameter only Gaussian Distribution eliminated Curse of dimensionality Dependent on underlying novelty detection technique
Experimental Methodology
Experimental Methodology Sensors 2500 ppr Tach 4 accel (10 kHz) AE Hall effect sensors Inline torque meter Variable Speed/Fixed Load (10 Nm) DAQ and Control NI FPGA and Accel Card Vibration data Segmentation: 30 shaft rotations, 70% overlap, Gaussian window Feature vectors: Auto-Regressive (AR) Models and Statistics Training: 20% of data for training, 80% for validation Faults Gears (96:32 and 80:48): missing tooth, root crack, chipped pinion Bearings: rough ball, outer race, inner race, chopped ball
Classification Results No speed adaptation (SVDD)
Classification Results Statistical Parameterization
Classification Results
Conclusions No speed adaptation = poor results Statistical Parameterization Good results Double Curse of Dimensionality Gaussian Distribution Multi-Modal Novelty Detection Comparable Results More to come
Future Work Novelty Detection Augmented for Fault Detection with Variable Speed Machinery (MSSP) Multi-Modal Novelty Detection for Variable Load and Speed Machinery Other multi-modal novelty detection techniques No modal sensors
References [1] J McBain, M Timusk. Fault detection in variable speed machinery: Statistical parameterization, Journal of Sound and Vibration 327 (2009) 623-646. [2] K Worden, H Sohn, CR Farrar. Novelty detection in a changing environment: Regression and interpolation approaches, J.Sound Vibrat. 258 (2002) 741-761. [3] JR Stack, TG Habetler, RG Harley. Effects of machine speed on the development and detection of rolling element bearing faults, IEEE Power Electronics Letters. 1 (2003) 19-21.

More Related Content

Condition Monitoring of Variable Speed Machinery

  • 1. Advancing the Online Monitoring of Variable Speed Machinery Jordan McBain, P.Eng. mcbainjj@gmail.com Sudbury, Ontario
  • 3. Introduction Monitoring of machinery largely limited to constant conditions Changes in speed and load termed nuisance parameters Variable speed/load machinery ubiquitous Ref: Stack Resonances/vibration power
  • 4. Novelty Detection Limited data characterizing normal state Little or no data for abnormal states Compute feature vectors of vibration (e.g. AR model) Methods SVDD and Statistical Boundaries
  • 5. Statistical Parameterization Vibration strongly tied to temp (speed) Advanced by Keith Worden (Structural health monitoring) Segment feature vectors into small groups of modal value Compute statistics for each group (bin) Trend with regression or interpolation Suffers from Double curse of dimensionality Describe healthy state for all segments of modal parameter Gaussian distribution Good heuristic
  • 6. Multi-Modal Novelty Detection Employ intuition from Statistical Parameterization Dont flatten data into bins Add modal parameter (speed) to feature vector Use any novelty detection technique One parameter only Gaussian Distribution eliminated Curse of dimensionality Dependent on underlying novelty detection technique
  • 8. Experimental Methodology Sensors 2500 ppr Tach 4 accel (10 kHz) AE Hall effect sensors Inline torque meter Variable Speed/Fixed Load (10 Nm) DAQ and Control NI FPGA and Accel Card Vibration data Segmentation: 30 shaft rotations, 70% overlap, Gaussian window Feature vectors: Auto-Regressive (AR) Models and Statistics Training: 20% of data for training, 80% for validation Faults Gears (96:32 and 80:48): missing tooth, root crack, chipped pinion Bearings: rough ball, outer race, inner race, chopped ball
  • 9. Classification Results No speed adaptation (SVDD)
  • 12. Conclusions No speed adaptation = poor results Statistical Parameterization Good results Double Curse of Dimensionality Gaussian Distribution Multi-Modal Novelty Detection Comparable Results More to come
  • 13. Future Work Novelty Detection Augmented for Fault Detection with Variable Speed Machinery (MSSP) Multi-Modal Novelty Detection for Variable Load and Speed Machinery Other multi-modal novelty detection techniques No modal sensors
  • 14. References [1] J McBain, M Timusk. Fault detection in variable speed machinery: Statistical parameterization, Journal of Sound and Vibration 327 (2009) 623-646. [2] K Worden, H Sohn, CR Farrar. Novelty detection in a changing environment: Regression and interpolation approaches, J.Sound Vibrat. 258 (2002) 741-761. [3] JR Stack, TG Habetler, RG Harley. Effects of machine speed on the development and detection of rolling element bearing faults, IEEE Power Electronics Letters. 1 (2003) 19-21.