Program Name:

MTConnect Challenge

Title of Idea:

A MTConnect Enabled Platform for Machine Tool Component Health Prognosis and Tool Condition Monitoring

Respondent Name:

David Siegel

Organization:

NSF I/UCRC Center for Intelligent Maintenance Systems – University of Cincinnati

Phone Number:

513-290-8163

Email address:

Date of Proposal:

May 31, 2013

Section 1: Abstract

The maintenance cost and losses in productivity with unplanned downtime for machine tool components such as spindle bearings and ball screws could be reduced if one could proactively take action prior to failure. In addition, cutting tools and inserts are expensive to replace when they are still in good condition, but replacing the tools too late can be costly due to scrap and re-work. The proposed health monitoring applicationwill use MTConnect to extract controller data and pattern recognition algorithms to assess the health condition of the spindle and machine tool axes. The health assessment approach is based on running a routine program each shift in which the most recent data patterns are compared to the baseline data patterns. An online tool condition monitoring module is also proposed and uses controller data such as the spindle motor current, with other add on sensors (vibration, acoustic emission) to accurately estimate and predict tool wear. By collecting data using the MTConnect standard, the platform aims to have widespread adoption in factories. In addition, with the added transparency of the machine tool health information, one can take proactive actions before significant downtime or productivity losses occur.

Section II: Proposed Idea

The era of the “fail and fix” maintenance policy for machine tools is over and has been replaced by the “predict and prevent” strategy that is being enabled by the ease of data collection using the MTConnect standard and the maturation of health monitoring algorithms. In order to achieve this widespread factory adoption of the “predict and prevent” approach, an intelligent machining platform is proposed, in which the architecture of such a platform is illustrated in Figure 1. The idea would be to leverage the SkyMars software platform since it already has the existing source code to use the MTConnect protocols to extract the controller data. In addition, various software modules can be added to the SkyMars platform, in which an energy monitoring application has already been developed. The proposed work would aim to enhance this existing platform by developing a machine tool health prognosis module and a tool condition monitoring software module.

For the machine tool health prognosis software application, the initial step would be to program the CNC machine tool to run a routine program once a shift or once a day for the spindle and the other translational or rotational axes. The routine program would be performed when the machine tool is not cutting any work piece, and would simply be done to characterize the spindle and axes health by running them through typical motion and speed profiles. This effectively provides the same working conditions for each data set and thus is a fair way of comparing the measured signals to the baseline state. In terms of the measured signals, the feed axis health condition could be assessed quiet well using only controlled based signals. A previous study for feed axis condition monitoring noted that the servo motor current (torque) signal was the most effective signal for monitoring its health state [1]. In addition, a recent project with an aerospace manufacturing facility in which only controller signals were used, showed that an anomalous feed axis health condition could be observed several days prior to the actual failure [1].

Figure 1: Machine Tool Intelligent Platform (Modules for Health Prognosis, Energy Monitoring, and Tool Wear Prognosis)

The spindle bearing would also be monitored in a routine program, in which the measured signals and data patterns would be compared to a known baseline condition. For monitoring the spindle bearing, improved coverage and more incipient detection of bearing problems would require the use of additional sensing beyond the existing controller signals. An accelerometer located next to the spindle bearing would help augment the relevant controller signals including the spindle speed, spindle motor current, and the spindle bearing temperature. An initial data processing algorithm would extract time, frequency, and envelope features from the vibration signal. A data analysis would then fuse the characteristic parameters from the vibration, current, and other signals into a single health index that is based on comparing the current data pattern to the baseline data pattern. This health index would then be tracked over time with respective thresholds and logic for triggering warning levels and also alarm levels. In addition, a root-cause algorithm would provide more detailed information on the potential failure mode and suggested maintenance action. Lastly, time series based prediction tools could be used to model the component health index values over time and predict the future health state with confidence bounds. More specifics on the type of data processing algorithms for the feed axis and spindle bearing health monitoring are discussed in Section 3.

The tool condition module would use an online monitoring approach and would estimate the tool wear after each cut. For monitoring the tool wear, key controller signals such as the spindle speed, spindle motor current, and the axis position, and axis speed would be extracted using the MTConnect standard. The software module would also include the ability to process add-on sensors, including vibration, current or power sensing, force measurements, or acoustic emission. The data processing steps would include signal segmentation, comparing measured signal characteristics and patterns to the initial set of baseline patterns, and calculating a tool wear health index with a corresponding set of thresholds for warning and alarm levels. The feasibility of this data processing approach was demonstrated in a previous case study, [2] in which there was a very high correlation between the increase in spindle motor current, the tool wear, and the surface roughness of the work piece. Refinement of this data analysis method and developing a graphical user interface and software application are the next steps in commercialization the tool condition monitoring approach.

The measureable benefits of the proposed software modules is theactionable health informationprovided to operators so they can take action and prevent machine failure, reduce downtime and reduce scrap and rework. The health monitoring methods and algorithms proposed for this software application have been refined and developed for machine tool applications for over 10 years and previous studies show that it can detect the early symptoms prior to the component failures [1]. In addition, the use of the MTConnect standard provides a means to extract controller data in an efficient and standardized manner and allows for more widespread reach of the developed software modules.

Section III: Technical Requirements

One aspect of the technical solution is the SkyMars platform, which can be considered as a kind of MTConnect multi-adapter. It provides an already developed platform that allows one to connect many machines at one time in an easy and time efficient manner. The main functionality of the multi-adapter is that it can support several types of controllers including FANUC, Mitsubishi, Heidenhain, Siemens, among others. The platform can also can be configured to include add-on sensors such as vibration, power, and acoustic emission. The SkyMars MTConnect multi-adapter provides a link to each controller through the intranet in the shop floor and provides the data that is used by the health prognosis and tool condition monitoring software applications. Also, the ability to connect with multiple controllers should make a widespread implementation of the health monitoring software much more achievable.

The SkyMars platform can be considered as a data source provider and also a platform for hosting the data processing algorithms. Selecting and customizing the data processing algorithms requires a proper library of tools along with a methodology of how to apply the algorithms. Based on prior experience, a tabular listing of the inputs and algorithms used for each step are provided in Table 1, for both the health prognosis and tool condition monitoring software applications.

Table I: Listing of Algorithms for Health Prognosis Application and Tool Condition Monitoring

Spindle Bearing / Feed Axis / Tool Condition Monitoring
Inputs / Spindle current, speed, vibration signal / Servo-motor current, position, speed, position and acceleration delay / Spindle motor current and / or vibration
Pre-processing / Segmenting vibration and current data by spindle speed / Segmentation by position and speed signal / Segmenting signal during cutting process
Feature Extraction / Envelope bearing features for vibration signal, statistics from motor current / Statistics extracted during the different motion segments / Motor and vibration RMS, frequency band energy features
Health Index / Three different algorithms could be used (PCA T2/SPE [3], SOM-MQE [1], auto-associative neural network [3] )
Diagnosis / Diagnosis recommendation based on most similar historical cases (k-nearest neighbor method)
Prediction / Robust regression curve fitting [4], similarity based prediction [5]

The health index calculations are based on fusing information from multiple sensors, and more details on each specific health index algorithm can be found in [1,3]. The feasibility of these health monitoring algorithms were demonstrated for different health monitoring applications and components. The tool condition monitoring application will also use a similar setof algorithms for the health index, diagnosis and prediction methods. However, the tool condition monitoring application will provide online monitoring results and thus will be configured with different input signals and use customized data pre-processing and feature extraction methods for that application.

Some additional comments on the diagnosis and prediction aspects should be highlighted, in that they are technology differentiators that are not currently available with commercial solutions. The diagnosis algorithm would not be initiated from the start but would learn over time and be deployed after a sufficient number of historical cases has been accumulated. Effectively it would match the historical health values and the variable contribution plots to the historical problem cases and find the nearest match. The historical cases would contain maintenance information on the severity of the anomalous condition, what component was replaced, and how much time it took to perform the maintenance action. Thus, when an anomalous health condition is detected, the operator can be identified of the problem and also presented with similar historical case that were the most similar to the current situation so they have an idea on what maintenance action might be the most suitable. The prediction algorithm would initially use robust regression fitting to extrapolate the health trend and provide an estimate of the failure time [4] with confidence bounds. However, as more historical cases are accumulated, a similarity based prediction method will be initiated; this method is more accurate since it uses pattern matching techniques to match the current health trajectory with historical degradation trends [5].

Section IV: Benefits

The benefits of the proposed software platform and modules for machine health prognosis and tool condition monitoring include the following:

  • Extracting controller data in a standard way using MTConnect multi-adapter (SkyMars).
  • SkyMars adapter provides a feasible way to implement the monitoring method in a widespread manner for different machine tool brands and across different factories.
  • Predict the machine tool component health problem before the failure occurs, providing savings related to maintenance cost and reduced downtime for the end user.
  • Tool wear health information can be used to yield improved part quality, and result in less scrap and rework.
  • Long term health information can also provide insight on the overall factory performance and which equipment is underperforming regarding maintenance cost and downtime.

References

  1. T. Skirtich, A comparative study of prognostic and health assessment methods in sensor rich and senorless environments, Master Thesis, University of Cincinnati, 2012.
  2. A. Kao, J. Lee, S. Yang, Y. Huang, N. Yen, iFactory Cloud Service Platform based on IMS Tools and Servo-lution., World Congress on Engineering Asset Management (WCEAM), 2011, Cincinnati, OH, USA
  3. D. Siegel, C. Ly, A.J. Bayba, K. Tom, J. Lee, Advanced Diesel Engine Health Monitoring Algorithms for Ground Vehicles, Machine Failure Prevention Technology Conference, 2012.
  4. D. Siegel, C. Ly, J. Lee, Methodology and framework for predicting helicopter rolling element bearing failure, IEEE Transactions on Reliability, Volume 61, pp. 846-875, 2012.
  5. T. Wang, J. Yu, J. Lee, A similarity-based prognostics approach for remaining useful life estimation of engineered systems, International Conference on PHM, 2008.