This paper presents a tool wear monitoring strategy based on a large number of signal features in the rough turning of Inconel 625. Signal features (SFs) were extracted from time domain signals as well as from frequency domain transforms and their wavelet coefficients (time-frequency domain). All of them were automatically evaluated regarding their relevancy for tool wear monitoring based on a determination coefficient between the feature and its low-pass-filtered course as well as the repeatability. The selected SFs were used for tool wear estimation. The accuracy of this estimation was then used to evaluate the sensor and signal usability.

The paper presents a comparison of efficiency of tool wear monitoring strategies based on one signal feature, on a single neural network with several input signals, and on a hierarchical algorithm and a large number of signal features. In the first stage of the hierarchical algorithms, the tool wear was estimated separately for each signal feature. This stage was carried out using either simple neural networks or polynomial approximation. In the second stage, the results obtained in the first one, were integrated into the final tool wear evaluation. The integration was carried out by the use of either a neural network or averaging. The paper shows a considerable advantage of the hierarchical models over conventional industrial solutions (single signal feature) and typical laboratory solutions (single, large neural network).

No abstract

No abstract

No abstract