based on Co-evolutionary Learning in Spiking Neural Network (SNN)

(1)

A Traction Control System

based on Co-evolutionary Learning in Spiking Neural Network (SNN)

Speaker:

Javier Pérez Fernández Authors:

Javier Pérez, Juan A. Cabrera & Juan J. Castillo Department of Mechanical Engineering

University of Malaga, Malaga, Spain

(2)

Outline

1. Introduction

2. Motorcycle Model

3. Control Spiking Neural Network 4. Co-evolution Learning

5. Results

6. Conclusion

(3)

Introduction

TCS Operation Adhesion characteristic curves

Wheel slip Wet Asphalt

Fz

Fx= μ Fz T_motor

(4)

Dry AsphaltDry AsphaltICE ICE

Introduction

TCS Operation

(Road change high-low)

Adhesion characteristic curves

Wheel slip Spiking Neural Network

(5)

Motorcycle Model

1. Introduction

2. Motorcycle Model

5. Results

6. Conclusion

(6)

Motorcycle Model

Motor Model

(7)

Control Spiking Neural Network

1. Introduction

2. Motorcycle Model

3. Control Spiking Neural Network 4. Co-evolution Learning

5. Results

6. Conclusion

(8)

Control Spiking Neural Network

Izhikevich Neuron Model

Izhikevich Neuron (a,b,c,d)

Synapse Model

I(t) v(t)

Izhikevich Neuron (a,b,c,d)

(9)

Control Spiking Neural Network

Neural Network Operation

Input layer Output layer

N Z P

OUTN OUTP

N Z P

OUTN OUTP

(10)

Control Spiking Neural Network

Neural Network Structure

(11)

Control Spiking Neural Network

Neural Network Scheme

(12)

Co-evolution Learning

1. Introduction

2. Motorcycle Model

3. Control Spiking Neural Network 4. Co-evolution Learning

5. Results

6. Conclusion

(13)

Co-evolution Learning

Prey

Predator

(14)

Co-evolution Learning

Training Method (Genetic Algorithm)

Control #0

Control #1

Control #2

Test #0

Test #1

Test #2

Training #0 Training #1 Training #2

(15)

Co-evolution Learning

Training #0

Training #1

Training #2

Best Controller (Genetic Algorithm)

(16)

Co-evolution Learning

Training #0

Training #1 Best Controller

Best Controller (Co-evolutionary Algorithm)

(17)

Co-evolution Learning

(18)

Results

1. Introduction

2. Motorcycle Model

5. Results

6. Conclusion

(19)

Results

(20)

Results

(21)

Conclusion

1. Introduction

2. Motorcycle Model

5. Results

6. Conclusion

(22)

Conclusion

• A controller for motorcycle T.C.S. is presented. The method is based on the use of Spiking Neural Networks and is optimized to road variations.

• The proposed controller has the best overall performance not only in constant adherence conditions but also when adherence changes during the test.

(23)

Conclusion

In future works, tests will be carried out on real roads with an electric bike.

(24)

Conclusion

NO Control

Fuzzy Control

(25)

A Traction Control System