Computational Neuroscience Applied in Surface Roughness Fiber Optic Sensor

Abstract Computational neuroscience has been widely used in fiber optic sensor signal output. This paper introduces a method for processing the Surface Roughness Fiber Optic Sensor output signals with a radial basis function neural network. The output signal of the sensor and the laser intensity signal as the light source are added to the input of the RBF neural network at the same time, and with the ability of the RBF neural network to approach the non-linear function with arbitrary precision, to achieve the nonlinear compensation of the sensor and reduction of the effect of changes in laser output light intensity at the same time. The Surface Roughness Fiber Optic Sensor adopting this method has low requirements on the stability of the output power of laser, featuring large measuring range, high accuracy, good repeatability, measuring of special surfaces such as minor area, and the bottom surface of holed etc. The measurements were given and various factors that affect the measurement were analyzed and discussed.


Introduction
Surface roughness is one of the most important parameters for monitoring the machining process and workpiece quality.
Although the current profiler can accurately detect surface roughness, it is not suitable for on-line measurement and control in automatic manufacturing, and destructive contact measurement is often not allowed for superfinishing [1,2]. For this reason, people have been trying to develop non-contact photoelectric methods and devices to measure surface roughness. Although the Intensity Modulated Fiber Optic Sensor (IM-FOS) has the advantages of simple structure and low cost, the drift of the light intensity of light source and some interference light have a great influence on the signal stability of the sensor.
If this effect can be effectively eliminated and the measurement signal is sufficiently stable, the application prospect of this kind of sensor will be broad. The Intensity Modulated Fiber Optic Sensor (IM-FOS) will measure the surface roughness based on the principle of scattering, so the measurement is not only related to the processing quality of the workpiece, but also directly related to the cleanliness degree of workpiece surface [3]. Therefore, to measure the surface roughness by the electro-optical method, the cleaning of the workpiece surface must be standard, which is essential to ensure the repeatability of the measurement.

Working principle of optical fiber displacement sensor
In When the measured surface is gradually away from the fiber optic probe, the area for the send fiber illuminating the measured surface becomes larger [4,5].  operations, identification or process control [7].
Artificial neural networks must first learn with certain learning guidelines before they can work. For example, the artificial neural network used for displacement is shown in Figure 4.

System design
The surface roughness of parts is random. The main reason is that different processing methods lead to different regular geometric contours.

The training of neural network
The surface roughness of the parts is random; the main reason is that different machining methods lead to different inherent geometric   which is amplified and converted into a digital signal Uri. As another output signal of RBF neural network, it is found from the analysis that the signal is not only related to the surface roughness of the measured object, but also reflects the output intensity IO of the laser. The signal can be expressed as: If the laser output power for the measurement is not stable, then IO in Equation   5 will change, and will also change accordingly.  The difference between U and the expected numberRa is used to adjust the weight W of the network, so that U constantly approaches Ra, Where X is a given small positive number.

RFB Artificial neural network sample training model
In Table 1, Ra is the calibrated surface roughness value, p represents the fluctuation of the laser output power, (the presence of p will cause change to the light intensity IO), the output of sensor in front of the neural network processor, U is the output of the RBF neural network. It can be seen from Table 1 that there is a nonlinear relationship between Uri and Ra, moreover, with change to the laser output  power, Uri will also change and cause error.

Conclusions
In this paper, with the ability of the neural network to approach non-linear functions with arbitrary precision, and the advantages of fast training speed, the output signal of the surface roughness fiber optic sensor and the laser output power signal as the light source are added to the input of the neural network at the same time, to achieve nonlinear compensation of the sensor and reduces the influence caused by the change to the laser output power.