Data-driven prediction of surface roughness using advanced signal processing methods

In machining processes, the surface quality of the processed parts directly affects product performance, production time, and cost. Especially in precise operations such as turning, cutting parameters and machining environment have a direct impact on the formation of surface quality. In this study, cutting force and vibration data were collected under 16 different conditions using SAE-AISI 5140 steel. Force data were obtained via a dynamometer, while vibration data were recorded in real time using an MPU6050 MEMS sensor. The obtained data were segmented, and time, frequency, and time-frequency analyses were applied to each segment. Each segment was treated as a signal, and a total of 63 features were extracted. Additionally, surface roughness was measured for each segment. In the first phase of the study, these features were used to classify cutting parameters and the machining environment; model performances were compared. The classification results showed that the sensor data could identify the machining conditions with high accuracy. In the second phase, the same features were used in regression models to predict surface roughness. MATLAB software was used for training both classification and regression models. This two-stage approach allows the system to contextually identify the machining conditions using sensor data and predict the resulting surface quality under those conditions. The developed method was implemented in an offline environment but is adaptable for integration into real-time monitoring and control systems. In this respect, the study makes a scientific contribution to data-driven production monitoring, intelligent decision support systems, and predictive quality control applications.