Multi-sensor data fusion framework for CNC machining monitoring

Highlights

• AE sensors are highly sensitive to sensor position and cutting parameters.

• A multi-sensor data fusion framework for CNC machining monitoring is proposed.

• The framework is able to enhance the periodic component and SNR of the signal.

• We study the robustness of the framework for a wide range of machining parameters.

• With only three sensors it is possible to improve the signal interpretation.

Abstract

Reliable machining monitoring systems are essential for lowering production time and manufacturing costs. Existing expensive monitoring systems focus on prevention/detection of tool malfunctions and provide information for process optimisation by force measurement. An alternative and cost-effective approach is monitoring acoustic emissions (AEs) from machining operations by acting as a robust proxy. The limitations of AEs include high sensitivity to sensor position and cutting parameters. In this paper, a novel multi-sensor data fusion framework is proposed to enable identification of the best sensor locations for monitoring cutting operations, identifying sensors that provide the best signal, and derivation of signals with an enhanced periodic component. Our experimental results reveal that by utilising the framework, and using only three sensors, signal interpretation improves substantially and the monitoring system reliability is enhanced for a wide range of machining parameters. The framework provides a route to overcoming the major limitations of AE based monitoring.

Introduction

In the manufacturing sector there is an ever growing-demand to increase the quality and diversity of products, as well as lowering production time and costs. This is mostly due to intensified global competition, diversified demand and shrinkage of product life cycles [1], [2]. To meet the above demands, manufacturers׳ interests are turning increasingly towards automated machining systems, where there is less dependence on the operator during the production process. The success of an automated machining system depends vastly on a robust and reliable monitoring system for on-line and off-line supervision of key machining processes. This is considered a challenging task due to the following main reasons [3]:

• The geometrical complexity of the components that form the final machined product requires complex tool path strategies along with a variety of machining techniques, as found, for instance, in high-speed milling [4], [5] and machining of sculptured surfaces [6].

• Materials that possess low machinability such as difficult-to-cut nickel based and titanium superalloys can lead to tool failure during machining operations since they require more energy than that of lower strength materials. Some typical wear features that are caused during machining include rapid flank wear and notching [7].

• Sensory signals derived from machining operations that could indicate the presence of machine failure are not always easy to interpret owing to the complexity of the cutting tool, and also due to its geometry and paths.

• The high-cost associated with certain machining components which prevent the occurrence of wastage and/or any additional machining.

The application of intelligent systems to monitor computer numerical control (CNC) machining operations is rapidly increasing in the industry. Several approaches have been proposed that accomplish tool monitoring and some of them have been successfully adapted to industrial applications. An extensive review on sensor-based systems for tool condition monitoring with a special focus on industrial applications can be found in [8]. Despite earlier efforts, and due to the reasons mentioned above, the existing intelligent monitoring systems are still not considered reliable enough to completely replace human supervision. In that, human operators are still essential in the industry to detect the end of tool life and to correct the cutting parameters whenever it is required [9]. Currently, there are three main goals related to machining process monitoring:

• Prevent and detect any machining tool and workpiece malfunctions. This can reduce the number of scrapped components during machining operations and prevent any irreversible damage to the tool and/or final machined product.

• Provision of information that can be utilised towards the machining process optimisation. For instance, in [10] energy consumption readings are utilised to optimise the process planning in CNC machining.

• Contribution to the development of a database towards the determination of an optimal set of cutting (control) parameters for the given machining process.

An approach that is becoming increasingly popular is to analyse the acoustic emissions (AEs) derived from machining cutting operations. Despite their advantages, AE-based systems are not considered to be totally reliable due to (a) their sensitivity to AE generated by sources other than tool and workpiece which can be picked up by the sensor and confuse the signal processing task [11], (b) the requirements of adjusting the signal amplification which is dependent on the process to be monitored [12], (c) the sensitivity of the AE measurements to sensor location and cutting parameters [2], and (d) limitations related to the practical implementation of a microphone in an industrial setting, such as directional consideration, frequency response, and environmental sensitivity [13]. To address the above limitations, this paper proposes a multi-sensor data fusion framework that relies on the information captured by more than one sensor and subsequent processing allows for

• Identification of which sensor provides the best signal representation and best location for monitoring the cutting operation. The identified sensor yields the highest periodic component strength that corresponds to the cutting tool rotation period.

• Derivation of a signal with an enhanced periodic component corresponding to the cutting tool rotation period when compared with individual sensor signals. The derived signal, known as signal estimate, improves the signal representation by further enhancing the signal over the noise that best describes the cutting operation for the given cutting parameters.

To validate the proposed framework a set of three microphones are placed at different locations inside a CNC machining structure and measurements are taken for a wide range of cutting parameters.

The remaining structure of the paper is as follows: an introduction to acoustic emissions with special focus on machining applications is provided in Section 2, two approaches that form the basis for the proposed framework are described in Section 3 while Section 4 presents the framework. The experimental results on CNC machining data are covered in Section 5 and the paper concludes with Section 6.