A Convolutional Neural Network for CNC Milling Machines Processes Classification

In recent years, manufacturing processes have undergone many technological innovations. Such technological leverages are accelerated even more with the advent of Industry 4.0, resulting in embedding advanced techniques such as Artificial Intelligence (AI) and Machine Learning (ML) in almost every industrial system. AI and ML play a key role in the field of Fault Diagnosis and Prognosis (FDP), due to their capacity to predict failures in the system or to locate faults, reducing both the time needed for maintenance and its costs. The latter becomes more crucial in the field of Computer Numerical Control (CNC) machines as one of the most robust pillars of the production chain where achieving a reliable FDP becomes a challenging task. However, the variety of tool types and operations makes CNC machines the perfect training field for ML, ensuring a large amount of different data to learn. In this paper, a data-driven model, more specifically a Convolutional Neural Network (CNN), is developed to assert if the quality of the process remains acceptable or not, and so if the machine has to be maintained based on a vibration signal captured by a triaxial accelerometer installed on 4-axis horizontal Brownfield CNC milling machines. In addition, data are collected over a period of two years, we developed a second CNN model to classify processes based on the month and year, highlighting the presence of information regarding the aging of the machine during the whole period of evaluation. As the result shows, both time and frequency domain-based features reach an accuracy of about 100% for the first classification problem. Considering different machines and various working processes in addition to different time windows from February 2019 to August 2021, time domain-based features correctly predict more than 85% of all processes containing information about the aging of the machine.