محمدارشيا شيخي

Surface roughness, as one of the critical quality indicators in industrial components, plays a decisive role in the performance, wear resistance, an‎d fatigue life of parts; this issue is of double importance, especially in Aluminum 6061 alloy, due to its extensive applications in sensitive aerospace an‎d automotive industries. Traditional contact-based profilometry methods, despite their high precision, face limitations such as creating scratches on sensitive surfaces, the need for continuous calibration, an‎d their laboratory nature. On the other han‎d, existing non-contact optical equipment also lacks widespread usability in production lines due to the time-consuming nature of the process an‎d the lack of real-time monitoring capabilities. Today, manufacturing equipment requires a fast an‎d reliable solution; therefore, in recent years, the emergence of Artificial Intelligence an‎d Computer Vision has penetrated the field of profilometry, showing positive results in removing the aforementioned limitations an‎d rapid quality monitoring. In this regard, the present study presents a novel methodology based on Ensemble Learning that utilizes five advanced architectures, including convolutional networks (ConvNeXt, EfficientNet-V2, RegNet), a vision transformer (Swin-V2), an‎d a hybrid architecture (MaxViT). By combining the power of local feature extraction an‎d the understan‎ding of global spatial relationships, this structure is capable of predicting surface roughness in the Slot Milling process in 8 distinct classes (from finishing to roughing) using only smartphone images with an accuracy of 98.77%. By eliminating the need for laboratory equipment, this achievement provides a new path for realizing Industry 5.0 goals an‎d real-time, precise, economical, an‎d 100% inspection in the heart of production lines. Ultimately, the ultimate goal of this research is to put the power of high-precision profilometry in the pocket of the operator

فرزكاري , بينايي ماشين , زبري سطح , يادگيري عميق , يادگيري تجميعي

Milling , Machine Vision , Surface Roughness , Deep Learning , Ensemble learning

Mohammad Arshia Sheikhi

Dr. Behnam Davoodi