Minimum sufficient signal condition of identifying process incidence in stacked drilling through deep learning

The determination of minimum sufficient condition of machining signals in representing events is vital to not only the procedure of signal acquisition, transfer, and storage, but also the design, training and deployment of deep learning and its integrated system. This paper proposed a deep learning-based approach to accurately identify the crucial process incidence in drilling of hybrid stacks. The influence of three sampling elements: duration, frequency and phase on both signal fidelity and the performance of deep learning is investigated. Based on the property of translation-invariance of convolutional neural network (CNN) and statistical probability, a theory establishing continuous classification equivalent accuracy is proposed and proved through experimentation, which helps strike a balance between immediacy and accuracy in deep learning application. Both minimum sufficient duration (MSD) and frequency (MSF) are found to have a close relation with harmonics driven by machining operation and are determined by spindle rate in the drilling process. Through investigation into phase, the condition and property for minimum sufficient unit are examined. The work establishes the machine signal’s feasibility boundary for deep learning models, supporting a safe, compact, and lossless application of deep learning in industry especially where real-time and low latency is valued.