A cyber physical system for tool condition monitoring using electrical power and a mechanistic model
Introduction
Computer Numeric Control (CNC) has enabled the automation of machining processes, such as milling, turning and drilling, by programming machining operations and parameters such as paths, spindle speed and feed rates. However, CNC machines generally have limited or no knowledge of their working environment, meaning that they cannot detect, adapt or deviate from a cutting program to avoid issues such as chatter, tool wear and breakage, tool deflection or collisions without the intervention a human operator (Cao et al., 2016).
Cutting tools are consumable components of the CNC machine which degrade and wear with use through mechanisms such as abrasion, adhesion, diffusion, fatigue and chemical wear (Altintas, 2000). Tool wear can significantly impact the machining process of a part, reducing the quality of the surface finish and potentially damaging the part, resulting in either additional rework or the need to scrap the part (Pimenov et al., 2018). Research into Tool Condition Monitoring (TCM) systems, where sensing, decision making and control strategies are introduced to detect the onset of tool wear, has therefore attracted a significant amount of interest from the academic community with a number of review articles listed (Cao et al., 2016; Teti et al., 2010; Siddhpura and Paurobally, 2013; Ambhore et al., 2015; Abellan-Nebot and Romero Subirón, 2010; Visariya et al., 2018). However, key challenges remain, including the development of TCMs which can be cost effective to install and maintain and are robust and reliable in the presence of the variability and discontinuities of real-world machining.
Several sensing methods have been evaluated within the literature to determine tool wear indirectly including force (Nouri et al., 2015; Choi et al., 2004), vibration (Rubio and Teti, 2010), spindle power (Shao et al., 2004; Al-Sulaiman et al., 2005; Drouillet et al., 2016), acoustic emission (Patra, 2011), eddy current (Abbass and Al-Habaibeh, 2015) or combinations of these (via sensor fusion) (Chattopadhyay et al., 2006; Cuka and Kim, 2016; Zhang et al., 2016). However, no research has been published focused on investigating the feasibility of monitoring the electrical power consumption of the entire CNC machine as a method of detecting tool wear. Electrical power monitoring has advantages over some of the aforementioned approaches due to its relative low cost (Nouri et al., 2015), non-intrusive installation (Abellan-Nebot and Romero Subirón, 2010) and non-reliance upon open machine architectures to allow access and integration with existing sensor systems (e.g. spindle current sensors) (Cao et al., 2016), which may also cause interoperability and data veracity issues when deployed across different machines.
Developing robust methods of interpreting sensor data to infer tool wear levels causes additional challenges due to the variability and discontinuous nature of cutting processes such as end milling. These variabilities are caused by wide operating and transient cutting parameters, such as spindle speed, feed rate, depth and width of cut, tool type, workpiece geometry and material, discontinuous tool paths and noise from background operations influencing the measured signal (e.g. coolant pump, lighting, computers). Several black-box approaches, such as Artificial Intelligence (AI), have been used to identify tool wear with good accuracy (Chattopadhyay et al., 2006; Zhang et al., 2016; Corne et al., 2017). However, these are usually only evaluated within limited operating conditions; therefore when changes occur to the system (i.e. different machines, tools, workpiece materials) retraining of the algorithms is needed which requires further controlled testing and data collection. These approaches reduce the robustness and scalability of the system and increase the operating cost which will limit the relevance of the research to industry. On the other hand, white box approaches, such as mechanistic models, use scientific first principles to represent the real-world complexities through mathematical relationships, with examples developed to determine tool wear based upon changes in electrical spindle power (Shao et al., 2004) and force measurements (Nouri et al., 2015). Mechanistic models are useful as they can be reused over a number of scenarios by changing the model parameters to match the changes within the real world. However, it is difficult to model and measure all of the real-world complexities that may impact the tool wear measurements. This leads to an inevitable trade-off between accuracy and usability/robustness when developing models in real applications where not all parameters can accurately be predetermined.
Several studies relevant to tool wear condition monitoring are listed in Table 1. Shao et al. (Shao et al., 2004) develop a mechanistic model to predict tool wear using the electrical power of the spindle and known cutting parameters. Whilst experimental results from this study appear to show good correlation with the mechanistic model, only limited validation of the technique is demonstrated over a small range of cutting parameters and tool wear is only presented as new or worn. Drouillet et al. (Drouillet et al., 2016) present an interesting approach using neural networks to update and correct Remaining Useful Life (RUL) prediction of an end milling tool, using historical tool degradation patterns from the electrical spindle power. The approach is used to correct errors in RUL prediction which is caused by stochastic tool wear degradation. However, all predictions are based upon nominal spindle power (i.e. cutting power of a new tool), so at present the approach is unable to accommodate changing cutting parameters. Nouri el al. (Nouri et al., 2015) develop a real-time tool condition monitoring method to track changes within cutting force coefficients which are independent of cutting conditions and correlated with tool wear within a mechanistic model. A Cumulative Sum (CUSUM) control chart is used to determine the transition of the cutting coefficient from gradual wear to the failure region. Wang et al. (Wang et al., 2019) use a combination of cutting force measurement and electrical power sensors to predict tool wear in drilling. A mechanistic model is used to predict the cutting force, whilst idle and auxiliary power demand of the motor is predicted based upon a regression model using empirical data.
This research investigates the use of electrical power monitoring of the entire CNC machine (rather than just the spindle) in combination with a mechanistic model as a low cost and process robust method to detect and determine tool wear in CNC end milling.
There are two novel aspects of this work: (i) a comprehensive assessment of the mechanistic model using only the electrical power under variable conditions (i.e. different spindle speed and feed rates), which has only been assessed under a limited set of experiments by Shao et al. (Shao et al., 2004), and (ii) the proposal of a cyber-physical system to support parameter calculation of the mechanistic model from the digital part program to enable real time implementation of a TCM system.
The structure of the paper is as follows. In section 2 the mechanistic model used to determine cutting power and tool wear is discussed. Section 3 details the test methodology used to assess the mechanistic model under variable cutting conditions. In Section 4 the results of the experiments are presented and discussed, whilst within section 5 an approach to tool conditioning monitoring combining a mechanistic model and cyber-physical system is presented and discussed. Conclusions to the research and future work are presented in Section 6.
Section snippets
Mechanistic model
A number of mechanistic models have been developed to predict cutting forces which are often expressed in terms of their tangential and radial components (Nouri et al., 2015). Electrical power monitoring, unlike force monitoring, does not detect individual axial components. Rather the magnitude of the total force is inferred through characterisation of the electrical power signal. Within this work the suitability of a mechanistic model originally developed by Shao et al. (Shao et al., 2004) is
Experimental setup
Experiments have been designed to investigate if the amount of tool wear can be determined based upon the measured electrical power under multiple cutting parameters using the test procedure shown in Fig. 3.
All cutting experiments were conducted using a Hurco 3-axis VM1 CNC machine. Cutting tools chosen for the experiment were four flute High Speed Steel with 8 % Cobalt (HSS-Co8) end mills with three different diameters (6 mm, 8 mm, 10 mm) to examine the effect of tool wear for different tool
Prediction of tool wear
The first part of the analysis explores the suitability of the electrical power monitoring of the CNC machine and the mechanistic model outlined in Eq. (12) to predict the amount of tool wear. The experiment was conducted in accordance with the methodology shown in Fig. 4, with a total of 8 different levels of wear per tool. The tools showed progressive wear, highlighted in Fig. 7 and Fig. 8, with the 6 mm exceeding the recommended limits of 0.2 mm uniform wear after 140 min of cutting on steel
Challenges
The previous section has demonstrated that the level of wear of end milling tools can be estimated using a mechanistic model and electrical power consumption data. Within this section the concept for a TCM system is proposed combining the mechanistic model with a cyber-physical model of the milling process.
Unlike the straight-line cuts performed in section 4, real world milling is used to cut more complex geometries, where the cutting parameters (i.e. spindle speed, feed rate, width and depth
Conclusion
This research has investigated the use of electrical power monitoring of a CNC machine as a method of detecting tool wear. A mechanistic model is used to predict the level of tool wear based upon cutting parameters and the difference between measured electrical power for new and worn tools. Experiments were conducted to evaluate the suitability of electrical power monitoring and the mechanistic model to detect tool wear and predict cutting power under various conditions (spindle speed, feed
CRediT authorship contribution statement
Paul Goodall: Conceptualization, Methodology, Investigation, Formal analysis, Writing - original draft. Dimitrios Pantazis: Conceptualization, Software. Andrew West: Writing - review & editing, Supervision, Funding acquisition.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
This work was supported by the Engineering and Physical Science Research Council (EPSRC) from funding for the project “AIIM” [grant reference EP/K014137/1].
References (27)
- et al.
Use of electrical power for online monitoring of tool condition
J. Mater. Process. Technol.
(2005) - et al.
Tool condition monitoring system: a review
Mater. Today Proc.
(2015) - et al.
Study of spindle power data with neural network for predicting real-time tool wear/breakage during inconel drilling
Int. J. Ind. Manuf. Syst. Eng.
(2017) - et al.
Tool life predictions in milling using spindle power with the neural network technique
J. Manuf. Process.
(2016) - et al.
Real-time tool wear monitoring in milling using a cutting condition independent method
Int. J. Mach. Tools Manuf.
(2015) - et al.
An automated feature extraction method with application to empirical model development from machining power data
Mech. Syst. Signal Process.
(2019) - et al.
A cutting power model for tool wear monitoring in milling
Int. J. Mach. Tools Manuf.
(2004) - et al.
Advanced monitoring of machining operations
CIRP Ann. Manuf. Technol.
(2010) - et al.
A mechanics based prediction model for tool wear and power consumption in drilling operations and its applications
J. Clean. Prod.
(2019) - et al.
A comparative study of using spindle motor power and eddy current for the detection of tool conditions in milling processes
Proc. 2015 IEEE Int. Conf. Ind. Inf. INDIN 2015
(2015)
A review of machining monitoring systems based on artificial intelligence process models
Int. J. Adv. Manuf. Technol.
Manufacturing Automation; Metal Cutting Mechanics, Machine Tool Vibrations and CNC Design
An update on AC induction motor efficiency
IEEE Trans. Ind. Appl.
Cited by (21)
Predicting electrical power consumption of end milling using a virtual machining energy toolkit (V_MET)
2023, Computers in IndustryIntelligent tool wear monitoring based on multi-channel hybrid information and deep transfer learning
2023, Journal of Manufacturing SystemsMulti-condition wear prediction and assessment of milling cutters based on linear discriminant analysis and ensemble methods
2023, Measurement: Journal of the International Measurement ConfederationA new tool wear condition monitoring method based on deep learning under small samples
2022, Measurement: Journal of the International Measurement ConfederationCitation Excerpt :As a widely used processing method, machining automation is one of the main contents of production process automation. In machining process, cutting tool is the terminal executive part, and contact directly with the workpiece, which leads easily to wear, and then affects the surface quality of workpiece [1–3]. According to [4,5], tool wear and damage are the primary factor causing the failure of machining process.
Event-driven tool condition monitoring methodology considering tool life prediction based on industrial internet
2021, Journal of Manufacturing SystemsCitation Excerpt :Zhou et al. introduced a TCM approach that utilized the SVM model and a transition point identification method (TPIM) to reach more accurate classification results [16]. Paul et al. proposed a cyber-physical system for TCM using electrical power and a mechanistic model to improve cutting parameters [17]. Zhang et al. proposed a TCM and life estimation system based on multiple sensors such as vibration, cutting force, and power data, as well as actual machining parameters [18].
A tool wear prediction and monitoring method based on machining power signals
2023, International Journal of Advanced Manufacturing Technology