¡¡
Project Description
This study was a collaborative project undertaken by Icams lab in University of Cincinnati and TechSolve Inc., and it investigates the feasibility of monitoring tool wear and surface finish based on force signal by conducting a set of experiments in turning operations.
Introduction
In current metal-cutting manufacturing practice, a tool will be used for only a fixed amount of time in accordance with some tool manufacturer recommendations or past experiences. This tool change policy has two prominent drawbacks, at one end, a worn tool without being exchanged in time will produce out-of-specification parts or even cause catastrophic tool breakage, and at the other tools being thrown away prematurely over time will incur a huge waste of manufacturing resource.
Developing a real-time tool wear monitoring system has been an intensive research topic for almost two decades and two main monitoring schemes exist. Direct monitoring is attempting to measure tool wear directly using some optical instrument such as video camera, which requires cutting operations to be interrupted periodically. Indirect monitoring is dealing with indirect signals that are believed to be influenced by or concomitant with tool wear progression. Among those indirect signals, force, vibration and acoustic emission are the most widely applied three, and they share some common characteristics¡ªcapable of being monitored in real time, easy to implement in hardware, and causing no interruption to cutting operation. If the underlying dependency between tool wear and these monitored indirect signals does exist and could be captured and interpreted, a tool wear monitoring system based on indirect methods seems to have great economic potentials.
A great variety of indirect monitoring techniques have been reported and most of publications claimed satisfactory performances. Part I of this report introduces general tool monitoring approaches taken by researchers based on a literature review conducted by Sick (2002). As suggested in Sick (2002) ¡°despite of more than a decade of intensive scientific research, the development of tool wear monitoring systems is an on-going attempt¡±, his review is however unable to answer why this is the case, which might be attributed to the lack of a common ground among reported methods for any meaningful evaluation and comparison to be made.
The goal of this project is to construct such a common ground by conducting a set of turning operations and investigate the feasibility of monitoring tool wear and surface finish based on forces signal. In Part II of this report, great details of experiments conducted in TechSolve are provided, including the experimental setup, hardware specifications and tool wear and surface finish measurements.
In Part III, data obtained from experiments are processed and analyzed in time domain and frequency domain. In time domain, forces signal appears to be nicely curvi-linear to both uniform and maximal tool wear and shifts systematically across different cutting conditions. In frequency domain, it is found low frequency band (< 20Hz) responds to tool wear progression in a much more sensitive way than other spectrum in terms of power variations, however, the manner of responses can not be clearly identified. The behavior of surface finish seems rather difficult to predict.
In Part VI, two modeling techniques¡ªmulti-layer neural networks and multiple linear regressions¡ªare tentatively applied to represent the dependency between tool wear, surface finish and forces signal. It is shown that force signal is adequate in monitoring tool wear progression based on the data obtained from the experiment and both neural network and linear regression can satisfactorily capture the underlying dependency. It can not be shown that force frequency spectrum analysis contains extra meaningful information as to tool wear progression. Analysis of force frequency spectrum seems to provide some more insight of surface finish variations than force analysis alone, however, the tremendous amount of data in frequency domain pose a great challenge for any modeling technique to accommodate.
In Part V, attempts are made to answer several key questions regarding some approaches widely taken by current tool wear monitoring researchers. Key findings from this study are summarized together with suggestions of further research.
Part I Literature Review
1.1
Objectives and Measured Signals
1.2
Five Stages in Tool Wear Monitoring
1.3
Evaluation and Comparison
1.4
Reference
Part II Experimental Procedure
2.1 Experimental Setup
2.2 Tool Wear Criteria and Measurement
Part III Data Processing and
Analysis
3.1 Time Domain
3.2 Frequency Domain
Part IV Model Building
4.1 Multi-layer Feed-forward Neural Network
4.2 Multiple Linear Regression Models
4.3 Comparison of Model Performance
Part V Conclusion and Discussion