Abstract
CNC milling has long been one of the most widely utilised production methods for various tasks ranging from tiny integrated circuits to large-scale mining machine gearboxes. It is a well-known machining method that provides tight tolerances and repeatability. End-milling is an old cutting procedure that uses rotating cutter tools operating at relatively high speed and a combination of tool and work-piece movement to remove materials from the main component. While gradually expelling a predetermined amount of materials from a main block, the CNC end-mill tool operates to create a suitable work-piece surface roughness (SR) to meet a particular application requirement. Therefore, the surface appearance it generates is a good indicator of a product's quality. Surface finishing forecasting is critical to prevent rejects, reprocessing of faulty works, and schedule constraints as well as to achieve production trust by producing dependable products, lowering manufacturing costs and being competitive. It largely relies on visual inspection to ensure a product is defect-free. Surface roughness on a machined object can be a flaw that detracts from a product’s appearance. CNC machine setting is a set of parameters that can be tweaked to alter the texture of the surface. However, the relationships between machine’s influencing parameters such as the spindle speed, federate, tool shape, vibration, surface roughness…are non-linear, there is no correlation between these variables. Therefore, the choice of these parameters within computing systems to produce the appropriate roughness of an item and to design a model for a specific outcome prediction, remains a concern. Trial-and-error as well as analytical methods have been employed so far, using machining parameters to predict surface roughness. However, these methods have shown some flaws and limitations. They have been found less reliable, short terms and time consuming. In the present study, artificial neural network (ANN) and adaptive network-based fuzzy inference system (ANFIS) were used to forecast and monitor surface roughness of aluminium Al6061 blocks machined using CNC end-milling. In addition, these computing methods have been coupled with genetic algorithm (GA) and particle swarm optimisation (PSO) to hybrid algorithms and to determine which of the hybrids algorithms (PSO or GA) best optimises ANFIS and ANN for the prediction of Al6061 SR. The methods involved varying four machine input parameters – spindle speed of rotation, feed rate, radial and axial depths – while performing a parametric within all the systems, namely ANN, ANFIS, ANN-GA, ANN-PSO, ANFIS-GA and
vi
ANFIS-PSO. These models have been compared and the most robust model has been selected for the prediction of the SR.
Results have been expressed using the regression values (R2) for both training and testing to measure the prediction performance. The outcome revealed that the hyperparameters of each machine learning (ML) configuration substantially impact the prediction performance of the suggested models. The hybrid models performed better than their counterparts stand-alone with ANFIS-GA yielding the best results with R2 of 0.9885 for training and R2 of 0.8478 for testing. This study sheds light on developing appropriate prediction models and the potential of soft computing techniques to forecast surface roughness of aluminium Al6061 blocks on end-milling CNC machines. While analysing the impact of models hyperparameters combination on the prediction ability, this study provides insight into developing suitable prediction models and the potential of soft computing techniques to predict surface roughness of the aluminium Al6061 blocks on CNC machines.