.png){kind=icon}
.png){kind=icon}
Return
Research on adaptive weighting method for position-sensitive detector error optimization based on genetic algorithm-backpropagation
.png){kind=icon}
PRE
AI
DOI：10.1117/1.OE.64.12.127101.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Position-sensitive detector (PSD), a sensor device for detecting the position of the center of gravity of a spot, is widely used in industrial manufacturing, aerospace, and other fields. However, due to the physical properties of its devices and fabrication process, there are nonlinear errors, which bring challenges to the improvement of testing accuracy. To address this problem, first, a sample preprocessing method is established, which integrates genetic algorithm (GA)-based adaptive weighting and oversampling. Then, the Levenberg-Marquardt backpropagation (LM-BP) neural network is applied to correct the PSD's nonlinear error. This method combines the GA and the LM-BP network. The GA first utilizes its strong global search capability to determine the optimal weights and thresholds for the LM-BP network. Then, the LM-BP network leverages its ability to approximate any nonlinear function to fit the sample data to the theoretical values. A PSD nonlinear calibration test platform was set up in the laboratory. Experimentally, the optimized model achieved an absolute mean error of 5.41 mu m across the entire photosensitive surface and 5.7 mu m in the peripheral regions. Compared with the traditional GA-LM-BP model's results of 7.98 and 11.8 mu m, these improvements represent increases of 32.2% and 51.7%, respectively. Simulation results show that the method overcomes the dependence of the LM-BP network on the initial weight threshold and the low fitting accuracy in the large error region of the PSD edge. It significantly enhances the measurement precision and detection area of PSD sensors, thereby improving the accuracy of the spatial three-dimensional coordinate measurement system.
Keywords:
position-sensitive detector
nonlinear
adaptive weighting
genetic algorithm-Levenberg-Marquardt backpropagation