Artificial Neural Networks with Machine Learning Design for a Polyphasic Encoder.

Area of Science:

• Optoelectronics
• Artificial Intelligence
• Machine Learning

Background:

• Managing complex data requires advanced processing techniques.
• Optical encoders are crucial for measuring angular position.
• Artificial neural networks (ANNs) offer powerful data interpretation capabilities.

Purpose of the Study:

• To design and implement a backpropagated multilayer artificial neural network (ANN) for processing optical encoder data.
• To develop a machine learning technique for training ANNs to predict angular position.
• To enhance the accuracy and range of angular measurements for industrial sensors.

Main Methods:

• Utilized backpropagated multilayer ANNs with vector input, hidden layers, and an output node.
• Employed light's phase-shifting arrangements from an optical encoder as input.
• Developed a training methodology to ensure accuracy across 0–360° measurement ranges.
• Evaluated performance using total error as the primary metric in a simulation environment.

Main Results:

• The neuro-encoder demonstrated remarkable accuracy in predicting angular position.
• The proposed methodology successfully extended high performance to larger angular intervals (0–360°).
• Simulations confirmed the system's effectiveness, with total error as the key performance indicator.

Conclusions:

• Backpropagated ANNs are effective for interpreting optical encoder data for precise angular sensing.
• The developed neuro-encoder system offers a viable solution for high-precision angular measurement.
• This technology can significantly improve the performance of resolvers and other polyphasic sensors in challenging industrial settings.