0
Research Papers

# Real-Time Identification of Incipient Surface Morphology Variations in Ultraprecision Machining Process

[+] Author and Article Information

Systems Engineering,
Virginia Polytechnic Institute and State University,
Blacksburg, VA 24061

Satish Bukkapatnam

School of Industrial Engineering
and Management,
Oklahoma State University,
Stillwater, OK 74078
e-mail: satish.t.bukkapatnam@okstate.edu

Omer Beyca

School of Industrial Engineering
and Management,
Oklahoma State University,
Stillwater, OK 74078
e-mail: omer.beyca@okstate.edu

Zhenyu (James) Kong

Systems Engineering,
Virginia Polytechnic Institute and State University,
Blacksburg, VA 24061
e-mail: zkong@vt.edu

Ranga Komanduri

School of Mechanical and
Aerospace Engineering,
Oklahoma State University,
Stillwater, OK 74078
e-mail: ranga.komanduri@okstate.edu

We also include n exogenous inputs ($y$) independent of the primary input x. For example, if x is chosen to represent the feed direction vibration sensor (VX) data and its time lags, then $y$ could represent a combination of data from other sensors, such as vertical direction vibration (VY), feed direction force (FX), AE, etc.

Here, for each RPNN parameter w, we created a separate state vector consisting of the current and its order-P autoregressive terms, i.e., for each RPNN parameter, wk = [wk wk-1… wk-p+1]T, where the superscript []T denotes the transpose operator.

The reason for obtaining this surface roughness value is because we used a tungsten carbide tool instead of a single crystal diamond tool. It is well known that single crystal diamond tool would give a surface finish in the nanometric range. The main objective of this study was to investigate a multiple sensor fusion approach (with RPNN and PF) for early detection of changes in surface characteristics.

Since, we use only one sensor signal, the term $WET×y(t)$ will become zero in Eq. (4)

We have tested several combinations of primary and exogenous inputs to arrive at this conclusion.

1Corresponding author.

Manuscript received December 11, 2011; final manuscript received April 1, 2013; published online January 16, 2014. Assoc. Editor: Eric R. Marsh.

J. Manuf. Sci. Eng 136(2), 021008 (Jan 16, 2014) (11 pages) Paper No: MANU-11-1393; doi: 10.1115/1.4026210 History: Received December 11, 2011; Revised April 01, 2013

## Abstract

Real-time monitoring and control of surface morphology variations in their incipient stages are vital for assuring nanometric range finish in the ultraprecision machining (UPM) process. A real-time monitoring approach, based on predicting and updating the process states from sensor signals (using advanced neural networks (NNs) and Bayesian analysis) is reported for detecting the incipient surface variations in UPM. An ultraprecision diamond turning machine is instrumented with three miniature accelerometers, a three-axis piezoelectric dynamometer, and an acoustic emission (AE) sensor for process monitoring. The machine tool is used for face-turning aluminum 6061 discs to a surface finish (Ra) in the range of 15–25 nm. While the sensor signals (especially the vibration signal in the feed direction) are sensitive to surface variations, the extraneous noise from the environment, machine elements, and sensing system prevents direct use of raw signal patterns for early detection of surface variations. Also, nonlinear and time-varying nature of the process dynamics does not lend conventional statistical process monitoring techniques suitable for characterizing UPM-machined surfaces. Consequently, instead of just monitoring the raw sensor signal patterns, the nonlinear process dynamics wherefrom the signal evolves are more effectively captured using a recurrent predictor neural network (RPNN). The parameters of the RPNN (weights and biases) serve as the surrogates of the process states, which are updated in real-time, based on measured sensor signals using a Bayesian particle filter (PF) technique. We show that the PF-updated RPNN can effectively capture the complex signal evolution patterns. We use a mean-shift statistic, estimated from the PF-estimated surrogate states, to detect surface variation-induced changes in the process dynamics. Experimental investigations show that variations in surface characteristics can be detected within 15 ms of their inception using the present approach, as opposed to 30 ms or higher with the conventional statistical change detection methods tested.

<>

## Figures

Fig. 1

MicroXAM 3D profile of the workpiece shown in Fig. 3 taken near the outer edge showing abrupt changes in the surface finish

Fig. 2

Two views of the UPM experimental setup showing the force, vibration, and AE sensors

Fig. 3

SCD tool used in UPM experiments and its image on a typical surface (Ra ∼ 22 nm) obtained on an Al 6061 sample workpiece machined with the UPM setup. The sample was machined with the SCD tool under the following conditions: spindle speed 2000 RPM, feed rate 15 mm/min, and depth of cut 4 μm.

Fig. 4

The architecture of an RPNN with input node I = 2, hidden node P = 2, and n exogenous inputs

Fig. 5

Overview of the RPNN-PF method applied to precision face-turning (facing) process. (a) Surface obtained during machining of aluminum before application of coolant (portion A) and after application of coolant (portions B and C). (b) Feed direction vibration signal (Vx) observed isochronously during machining of portions A, B, and C. (c) Behavior of one particular PF-updated RPNN weight (W˜4,41) before and after application of coolant. (d) SPRT applied to the raw times series shown in (b) and the RPNN weight shown in (c). (e) Mean-shift clustering applied to the RPNN weight.

Fig. 6

Surface and concurrent signal obtained during facing test at 2000 RPM, 60 mm/min feed rate, and 12 μm depth of cut. Signal down sampled from 10 kHz to 100 Hz.

Fig. 7

Summary of the application of the present approach to signals from the feed direction vibration sensor for the portions where no change is apparent: (a) variation of the SPRT for weights and raw signal statistics. (b) Behavior of the mean-shift statistic as applied to the raw signal. (c) Behavior of the mean-shift statistic as applied to RPNN weights.

Fig. 8

Effect of multiple sensor fusion on change detection delay, comparing performance of the mean-shift and SPRT applied to RPNN weights with (i) feed direction vibration sensor (VX) alone, (ii) with feed direction vibration and force sensor (VX, FX), and (iii) all sensors. The saturation of prediction accuracies with two sensors may be peculiar to the current situation, where the change in surface morphology is prominent.

## Discussions

Some tools below are only available to our subscribers or users with an online account.

### Related Content

Customize your page view by dragging and repositioning the boxes below.

Related Journal Articles
Related Proceedings Articles
Related eBook Content
Topic Collections