Hybrid heterogeneous prognosis of drill-bit lives through model-based spindle power analysis and direct tool inspection

Abstract: In the context of Industry 5.0, manufacturing systems are driven by human-centered production processes, assigning high-level supervisory tasks to operators. This necessitates that machines can perform low-level decision-making actions. This paper presents a novel hybrid heterogeneous prognosis algorithm designed to autonomously inspect the cutting edges of drill-bits and to forecast their Remaining Useful Life along with the associated probability density function. The algorithm leverages specific force coefficients from spindle power and feed axis current measurements, as features correlated with tool wear, to detect tool brittle failures. Additionally, flank wear is automatically measured through a specifically conceived image processing algorithm, using thresholding, convolutional filters, and edge detection techniques. Direct tool wear measurements are analyzed by a hybrid prognosis algorithm, fusing particle filter and multi-layer perceptron, to predict drill-bits’ remaining useful lives. The proposed solution offers several advantages. It reduces the need for extensive experimental run-to-failure tests typically required for training standard machine learning algorithms. Instead, it allows for real-time adaptation, even in scenarios involving untested and varying cutting process conditions. Furthermore, it utilizes both indirect wear observations during cutting operations and direct wear observations during setup times (e.g. tool changes, workpiece changes), without interrupting the ongoing process. Exponent of Kronenberg’s models for specific force coefficients was found to be sensitive to tool wear. Prognosis could correctly predict the 67% of end-of-lives with an average prognosis horizon of 30%.