[BKYST] Using Deep Learning Models in Prediction of Surface Roughness

Abstract

In contemporary industrial operations, machining precision has undergone significant advancements, placing surface roughness measurement at the forefront of quality assessment. However, current methods for measuring surface roughness are not only time-consuming but also labor-intensive, often requiring additional effort for equipment preparation and fixture setup. Moreover, the current practice often involves direct measurement on the sample, necessitating the removal of the sample from the machining equipment, which can introduce setup errors and potentially compromise the original machining standards. Recognizing these challenges, our project aimed to streamline the post-machining inspection process by evaluating surface roughness through imaging of the machined surface. This article explores the application of deep learning models, facilitated by MATLAB software, to diagnose surface roughness in machining. Additionally, we propose a comprehensive method for sampling measurement, including procedures and conditions, to guide further project development based on collected samples or by incorporating any missing conditions to enhance the project in the future. Leveraging non-contact measurement methods ensures precise surface details regarding roughness and glossiness, making them suitable for processing. This advancement represents a significant step forward in testing and measurement, with wide-ranging applications in mechanical machining aimed at boosting productivity, reducing costs, and machining time, ultimately optimizing profits and yielding substantial economic benefits in the long run. The results obtained from our study exhibit promising signals and a high level of feasibility in diagnosing and verifying surface quality in machining. The number of measured samples will be synthesized, supplemented, and provided to relevant parties to significantly increase the sample size, thereby enhancing the accuracy of the AI model and accelerating prediction capabilities through vast data. To bolster the reliability of these findings, it is imperative to augment the model with additional data, maximizing its effectiveness and applicability in real-world scenarios.