Stochastic tool wear assessment in milling difficult to machine alloys

In the machining industry, maximising profit is intuitively a primary goal; therefore continuously increasing machining process uptime and consequently productivity and efficiency is crucial. Tool wear plays an important factor in both machining uptime and quality, and since tool failure is related to the surface quality and the dimensional accuracy of the end product, it is essential to quantify and predict this phenomenon with the best possible certainty. One of the most common ways of tool wear prediction is through the use of low cost spindle current sensing technology which is used to measure spindle power consumption in CNC machines and relate power increase to tool wear. In this work, two methods of stochastic filtering (i.e. Kalman and particle filter) were used in predicting tool flank wear in machining difficult-to-machine materials through spindle power consumption measurements. Results show a maximum of 15% average error in estimation, which indicates the good potential of using stochastic filtering techniques in estimating tool flank wear. In addition, the particle filter was used for online estimation of a spindle power model parameter with uniform and Gaussian mixture models as the initial probability density functions, and the evolution of this parameter to the true posterior density function over time was investigated.