A Neural-Fuzzy System for Predicting the Areal Surface Metrology Parameters

Ronak Sharma, Mahdi Mahfouf, Olusayo Obajemu

Abstract

With the increasing demand for faster manufacturing, Industry 4.0 has now only started to contribute towards streamlining the manufacturing processes. Despite the availability of high dimensional manufacturing data, a significant amount of time is still spent on testing the end products. Therefore, with a drive to substitute these inspection processes with a “digital twin”, this paper presents a framework for predicting the optimal surface metrology parameters such as force and vibration, required to achieve the desired surface roughness of an end product. Firstly, an Adaptive Neuro-Fuzzy Inference System (ANFIS) was designed to predict the surface roughness using vibration, force and temperature. A low RMSE of 0.07 was obtained between the predicted and desired surface roughness. This model was then reverse engineered to predict the optimal surface conditions (force, vibration and temperature) required to achieve the desired surface roughness. For this, optimisation was applied to minimise the error between the target and predicted surface roughness. This framework will help manufacturing industries to discard frequent in-depth product inspection processes in favour of this “digital twin” due to the possibility of achieving right-first-time production.

Download


Paper Citation