1 / 13

1. Introduction 2. Fuzzy inference systems and fuzzy modelling

FAULT DIAGNOSIS OF THE DAMADICS BENCHMARK ACTUATOR USING NEURO-FUZZY SYSTEMS WITH LOCAL RECURRENT STRUCTURE Letitia Mirea*, Ron J. Patton** * ”Gh.Asachi” Technical University of Iaşi, Dept. of Automatic Control ** University of Hull, Dept. of Engineering. 1. Introduction

gino
Download Presentation

1. Introduction 2. Fuzzy inference systems and fuzzy modelling

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. FAULT DIAGNOSIS OF THE DAMADICS BENCHMARK ACTUATOR USING NEURO-FUZZY SYSTEMS WITH LOCAL RECURRENT STRUCTURELetitia Mirea*, Ron J. Patton*** ”Gh.Asachi” Technical University of Iaşi, Dept. of Automatic Control** University of Hull, Dept. of Engineering 1. Introduction 2. Fuzzy inference systems and fuzzy modelling 3. Adaptive neuro-fuzzy systems with local recurrent structure 4. Neuro-fuzzy design of an FDI system 4.1 Residual generation 4.2 Residual evaluation 5. Application

  2. 2. Fuzzy inference systems and fuzzy modelling  Fuzzy Inference System (FIS): - rule base - data base - reasoning mechanism  Most applied FIS for system modelling  Sugeno fuzzy system Rule i: if x1 is A1 and x2 is A2and ... and xn is Anthen  Consequents of each fuzzy rule  local model  Antecedents of each fuzzy rule  define region in input space where local model applies  Sugeno model can be implemented as special type of neural network  Adaptive Neuro - Fuzzy System (ANFS)

  3. 2. Fuzzy inference systems and fuzzy modelling  ANFS combines: - capability to handle uncertain & imprecise information (from fuzzy systems) - ability to learn from examples (from neural networks)  Identification of dynamic systems  models with adequate memory   ANFS should be provided with dynamic elements: - ANFS, external dynamic elements (ext. cascades of linear filters) - ANFS, internal dynamic elements (recurrent connections, internal filters)  ANFS with Local Recurrent Structure (ANFS-LRS)

  4. 3. Adaptive NF systems with local recurrent structure  local model described by:

  5. 3. Adaptive NF systems with local recurrent structure  Layer 1: adaptive, membership functions  Layer 2: computes the firing strengths of fuzzy rules  Layer 3: computes normalised firing strengths of the fuzzy rules  Layer 4: adaptive, outputs of local models  Layer 5: computes the overall output of the ANFS-LRS

  6. 3. Adaptive NF systems with local recurrent structure  identification of MIMO system ANFS-LRS model for each output of process:  ANFS-LRS learning: - number of fuzzy rules and initial values of premise parameters  fuzzy clustering algorithm (Chiu, 1994) - ANFS-LRS parameters  gradient method: N - number of the training data yP,i - the i-th output of the process  - learning rate

  7. 4. Neuro-fuzzy design of an FDI system4.1 Residual generation  FDI system: residual generation and residual evaluation  Residual generation: - an ANFS-LRS model for each system output is identified (MISO model) - MISO models  neuro-fuzzy observer scheme - generated symptoms (current state)  residuals  Neuro-Fuzzy Simplified Observer Scheme (NF-SOS): - MIMO process with I inputs uP,i[k], i=1,...,Iand O outputs yP,j[k], j=1,...,O - NF-ARX models: normal behaviour of the process - residuals  one-step ahead prediction error

  8. 4. Neuro-fuzzy design of an FDI system4.2 Residual evaluation  Residual evaluation pattern classification using neural networks - pattern classifier  static Multi-Layer Perceptron - decision logic  Euclidean distance

  9. 5. Application  Investigated process: actuator from the steam boiler used to control the water level in the 4th boiler station (Lublin sugar factory, Poland)  Real data corresponding to the normal behaviour of the process have been used to: - design the NF-SOS scheme using ANF-LRS - generate faulty data using the DAMADICS benchmark Considered faults: F1: Valve clogging F2: Valve plug or valve seat sedimentation F3: Servo-motor’s diaphragm perforation F4: Electro-pneumatic transducer fault F5: Rod displacement sensor fault F6: Positioner feedback fault F7: Fully or partly opened bypass valve F8: Flow rate sensor fault

  10. 5. Application Methodology: Data filtering: low-pass Butterworth filters  noise reduction and data decimation Selection of used data:  training data set: 360 out of 3600 measurements – NORMAL behaviour  testing data sets: - data set 1: 3600 measurements (another hour, same day) - data set 2: 3600 measurements (previous day) - data set with faults Residual generation:  system identification using ANFS-LRS  neuro-fuzzy simplified observer scheme Residual evaluation:  static neural classifier (MLP/ BP)  decision mechanism based on the Euclidean distance

  11. 5. Application inputs: u1 – level controller output u2 – valve input water pressure u3 – valve output water pressure u4 – temperature of the water outputs: y1 – servo-motor rod displacement y2 – water flow to steam boiler inlet Testing data set 1 Testing data set 2  Electro-pneumatic actuator:system identification using ANFS-LRS

  12. 5. Application  data with faults  example for fault F3:  Residuals generated with NF-SOS based on ANFS-LRS corresponding to: - the normal behaviour - the considered faulty behaviours were evaluated using a neural classifier  Multi-Layer Perceptron  Obtained recognition rate: 93.67%

  13. Conclusions  The present paper investigates the development of a neuro-fuzzy system with local recurrent structure and its application to fault diagnosis of an electro-pneumatic actuator (DAMADICS benchmark).  The advantages of using such a neuro-fuzzy system are: - it is abble to process uncertain information; - automatic extraction of the rule-base; - it is able to learn from examples; - it has a reduced input data space because of its locally recurrent structure.  The obtained experimental results by using the suggested neuro-fuzzy system reveal its good performances of approximation and generalisation.  Its application to fault diagnosis of an industrial process leads to good results reflected in a recognition rate greater than 90%.

More Related