Automatic Diagnostics of Drill Condition by Several Diagnostic Signs

Abstract

A key problem in mechanical engineering is the prediction of tool wear in real time. Traditional methods rely on conservative tool replacement, which leads to premature tool replacement or excessive tool wear, as well as increased production downtime. Manual assessment of drill condition is time-consuming. To reduce downtime, it is necessary to perform the process of diagnosing the drill condition in production conditions automatically, with high accuracy. Automatic tool condition monitoring systems have shown great potential for detecting and predicting tool wear, reducing the risk of tool breakage and optimizing tool change intervals, but the prediction accuracy is not always sufficient. This study presents an automatic system for predicting tool life in real time, which helps to solve this limitation by integrating several control modules. The prediction system includes machine-mounted modules for direct control of feed force and cutting power consumption signals, a signal integration module by connecting them for real-time data management, and a data processing module for predicting tool wear. In the last module, based on these signals, an integral indicator of drill wear is generated, the degree of approximation of which to the permissible wear value determines the condition of the drill as “fit for work” or “unfit for work”. Obviously, an approach of model aggregation in machine learning is proposed here, when several models are combined to obtain better results than any individual model gives in order to increase the accuracy, stability and generalization ability of the diagnostic system. The paper uses an aggregated forecasting model based on the use of the axial cutting force signal and the power consumed by the spindle signal to control tool wear.