Fault Detection and Diagnosis using Adaptive Neural Networks

1
Fault Detection and Diagnosis using Adaptive
Neural Networks

• Tan Shing Chiang, Prof. M.V.C. Rao
• FIST, FET
• MMU

2
Artificial Neural Networks

• Artificial neural networks are inspired by
  biological neural networks (BNN), which are
  inherently nonlinear, highly parallel, robust and
  fault tolerant.
• Consists of large number of highly interconnected
  processing elements (neurons) working together

3
Inspiration from Biological Neural Networks

• A neuron inputs-output
• output can be excited or not excited
• incoming signals from other neurons determine if
  the neuron shall excite ("fire")
• Output subject to attenuation in the synapses,
  which are junction parts of the neuron

4
Mathematical Representation
5
Learning

• Learned from examples / training data
• Strength of connection between the neurons is
  stored as a weight-value for the specific
  connection
• Learning the solution to a problem changing the
  connection weights

6
Learning

• Continuous process of
• Evaluate outcome
• Update weights
• Receive new inputs
• ANN evolving causes stable state of the weights,
  but neurons continue working network has
  learned dealing with the problem

7
Competitive learning

• Example Adaptive Resonance Theory
• network / Kohonen network
• Winner takes all
• only update weights of winning neuron

8
Applications

• Prediction learning from past experience
• Pick the best stocks in the market
• Predict weather
• Identify people with cancer risk
• Classification
• Fault detection and diagnosis
• Image processing
• Risk assessment

9
Aims of the Project

• To develop neural-network-based autonomous
  learning systems that can acquire knowledge
  incrementally in real-time, with as little
  supervision as possible
• To deploy effective strategies for practical
  application of such systems for fault detection
  and diagnosis.

10
Project Schedule

Current status