Sensorless Estimation Based on Neural Networks Trained with the Dynamic Response Points.

Area of Science:

• Control Engineering
• Artificial Intelligence
• System Dynamics

Background:

• Accurate prediction of system dynamic responses is crucial for control and monitoring.
• Traditional methods often require physical sensors and can be computationally intensive.
• Developing sensor-less estimation techniques offers significant advantages in remote monitoring and complex systems.

Purpose of the Study:

• To develop and validate a neuronal dynamic response prediction system for remote, sensor-less estimation of stable system behavior.
• To leverage Neural Networks for real-time dynamic response prediction.
• To compare the performance of the proposed system against traditional linear regression techniques.

Main Methods:

• Utilized a set of Neural Networks trained on a large dataset (1,500,000 data points).
• Extracted six basic characteristics of dynamic response to derive an equivalent Transfer Function.
• Trained the network to map dynamic response characteristics and Transfer Functions to system behavior.

Main Results:

• Achieved an average Mean Squared Error (MSE) of 2% for the Neural Network system.
• Demonstrated precise estimation with less than 1% error for simulated systems.
• Showcased high performance with real-world signals, effectively handling typical acquisition noise.

Conclusions:

• The proposed Neural Network system provides an effective and accurate method for sensor-less dynamic response prediction.
• Real-time estimation capabilities offer significant advantages over offline, sensor-dependent methods.
• The system demonstrates robust performance in both simulated and practical applications, including noisy real-world data.