The capacity of statistical features extracted from multiple signals to predict tool wear in the drilling process

Industrial processes are being developed under a new scenario based on the digitalisation of manufacturing processes. Through this, it is intended to improve the management of resources, decision-making, production costs and production times. Tool control monitoring systems (TCMS) play an important role in the achievement of these objectives. Therefore, it is necessary to develop light and scalable TCMS that can provide information about the tool status using the signals provided by the machine. Due to the lack of this type of systems in industrial environments, this work has two main objectives. First, the predictive capacity of statistical features in the time domain of internal and external signals for the prediction of tool wear in drilling processes was analysed. To this end, a methodology based on automatic learning algorithms was developed. Secondly, once the most sensitive signals to tool wear were identified, algorithms with signals of a certain tool geometry were trained and a model was obtained. Then, the model was tested using signals from two different tool geometries. The experiments were carried out on a vertical milling machine on a steel with composition 35CrMo4LowS under pre-established cutting conditions. The results show that the most sensitive signals to monitor the tool wear in the time domain are the feed force (external) and the z-axis motor torque (internal). The models created for the fulfilment of the second objective show a great capacity of prediction even when dealing with tools with different geometrical characteristics.