We have been developing a measurement system for microstructures using a small-diameter optical fiber probe. The contact detection method using a threshold value may cause a large error due to the time lag between the real contact time and the time detected by the measurement system. In this paper, we investigated a method for detecting the starting point of contact using an AR model and LSTM, a type of deep learning when the stylus tip contacts the measured surface. The results of measurement experiments using an inclined surface showed that the detection accuracy using AR model decreased as the inclination angle increased, but the LSTM was able to detect the contact point with high accuracy even when the inclination angle increased. It was also confirmed that the method using LSTM can detect contact with high accuracy without being affected by drift over time and various noises (other than random noise).
All Science Journal Classification (ASJC) codes
- Electrical and Electronic Engineering
- Industrial and Manufacturing Engineering
- Mechanics of Materials
- Electronic, Optical and Magnetic Materials