Ex Parte Mukherjee et al

Patent Trial and Appeal Board

Dec 17, 2012

11201694 (P.T.A.B. Dec. 17, 2012)

UNITED STATES PATENT AND TRADEMARK OFFICE
____________

BEFORE THE PATENT TRIAL AND APPEAL BOARD
____________

Ex parte JOYDEB MUKHERJEE, VENKATARAMANA B. KINI and
SUNIL K. MENON
____________

Appeal 2010-011182
Application 11/201,694
Technology Center 3600
____________

Before GAY ANN SPAHN, BRETT C. MARTIN, and
RICHARD E. RICE, Administrative Patent Judges.

RICE, Administrative Patent Judge.

DECISION ON APPEAL

STATEMENT OF THE CASE
Joydeb Mukherjee et al. (Appellants) seek our review under 35 U.S.C.
§ 134 of the Examiner’s rejection of claims 1-23. We have jurisdiction
under 35 U.S.C. § 6.
We AFFIRM-IN-PART.

Appeal 2010-011182
Application 11/201,694

2
The Claimed Subject Matter
The claimed subject matter “generally relates to diagnostic
systems, and more specifically relates to fault detection.” Spec. 1,
para. [0002]. Claim 1, reproduced below, is representative of the
subject matter on appeal:
1. A fault detection system for detecting
faults in a turbine engine, the fault detection
system comprising:
a kernel-based Maximum Representation
Discrimination Features (MRDF) feature extractor,
the kernel-based MRDF feature extractor receiving
measured turbine sensor data from the turbine
engine, the kernel-based MRDF feature extractor
using a nonlinear kernel function, the nonlinear
kernel function comprising a nonlinear mapping of
the measured turbine sensor data to a higher
dimensional space, the kernel-based MRDF feature
extractor using the nonlinear kernel function to
extract features indicative of nonlinear correlations
in the sensor data; and
a classifier, the classifier receiving the
extracted features and classifying the extracted
features to determine if a fault occurred in the
turbine engine.

The Rejection
Claims 1-23 stand rejected under 35 U.S.C. § 103(a) as unpatentable
over:
Talukder (Ashit Talukder, et al., “A General Methodology For
Simultaneous Representation And Discrimination Of Multiple
Object Classes,” Carnegie Mellon University, Department of
Electrical and Computer Engineering, Laboratory for Optical
Data Processing, pp. 1-32 (1998)); and

Appeal 2010-011182
Application 11/201,694

3
Goebel (Kai Goebel, et al., “Rapid Detection of Faults for
Safety Critical Aircraft Operation,” 2004 IEEE Aerospace
Conference Proceedings, pp. 3372-3383 (2004)).1

Ans. 2-3.

OPINION
Claims 1, 7, 13 and 19
Appellants argue claims 1, 7, 13 and 19, which are the only
independent claims, as a group. App. Br. 8-11. We select claim 1 as
representative. See 37 C.F.R. § 41.37(c)(1)(vii) (2011).
The Examiner finds that Talukder discloses a binary classifier for
detecting faults, while Goebel teaches “using binary classifiers . . . similar to
the binary classifier of Talukder in a fault detection system used to identify
faults in aircraft operation, including turbine engines.” Ans. 4 (citing
Goebel, Part 4 “DIAGNOSTICS” and p. 1, col. 2, para. 2, l. 7). The
Examiner concludes:
Therefore, it would have been obvious to one skilled in the art
(e.g.[,] a reliability engineer) at the time the invention was
made, to use the binary classifier of Talukder in the fault
detection system of Goebel, for the advantage of developing
automatic fault accom[mo]dation in aircraft as envisioned by
Goebel.

Ans. 4 (citing Goebel, p. 1, col. 2).
We are not persuaded by Appellants’ arguments: that the “maximum
likelihood hypothesis test” used in Goebel’s turbine engine fault detection
system “is . . . unrelated to the claimed MRDF techniques” (App. Br. 9); that

1 We have relied on copies of Talukder and Goebel located in the file of the
instant Application on appeal.
Appeal 2010-011182
Application 11/201,694

4
Talukder merely teaches “the statistical classification of images,”
particularly, “pistachio nut images” (id.); and that “[t]here is no evidence
that one of ordinary skill in the art would have believed that a technique
applied to images of pistachio nuts could be applied to the type of sensor
data that is available from a turbine engine and achieve any type of
predictable results” (id. at 11). Appellants mischaracterize Talukder, which
explicitly teaches classification of signal data as well as image data:
Feature extraction for signal/image representation is an
important issue in data processing. . . .
In this paper, we develop a new feature extraction
method called the maximum representation and discrimination
feature (MRDF) that allows for simultaneous representation of
each class and separation between the different classes. The
measure used in the MRDF allows each c1ass to have multiple
clusters in the image/signal data, while [other mathematical
techniques such as] the MDF, PCA, Fisher and DCCF methods
assume only one cluster per class.

Talukder, pp. 1-2 (emphasis added); see also id. at 3-4.2 Further, as
explained by the Examiner, “[a] binary classifier is a mathematical (usually
statistical) algorithm used by computers to classify a set of data as belonging
to one of two groups.” Ans. 6. Appellants do not dispute the Examiner’s
finding that Talukder and Goebel both teach use of binary classifiers. See
App. Br. 20. We thus agree with the Examiner that, because Talukder and

2 While Talukder discloses a test of the nonlinear MRDF as applied to a
product production problem involving classification of pistachio nuts as
clean or infested based on features extracted from real-time X-ray images,
the purpose of the test was not limited to pistachio nuts or even image data,
but was to show that “higher-order correlation information exists in real data
and that [the] nonlinear MRDF can locate such information.” Talukder,
p. 20.
Appeal 2010-011182
Application 11/201,694

5
Goebel both teach use of binary classifiers, it would have been obvious to
one of ordinary skill in the art to substitute Talukder’s binary classifier for
the similar classifier in Goebel’s fault detection system:
Since both the MRDF/NLEF3 of Talukder and the maximum
likelihood test used by Goebel[] are both [b]inary [c]lassifiers,
it would be obvious to one of skill in the art (anyone who could
understand and implement the paper of Talukder) that the two
algorithms are each different species of the genus [b]inary
[c]lassif[i]ers, and are similar enough to be replac[e]able/
swappable.

Ans. 7. See KSR Int'l Co. v. Teleflex Inc., 550 U.S. 398, 416 (2007) (“when
a patent claims a structure already known in the prior art that is altered by
the mere substitution of one element for another known in the field, the
combination must do more than yield a predictable result”) (citing United
States v. Adams, 383 U.S. 39, 50-51 (1966)).
We also are not persuaded by Appellants’ argument that “the
Examiner has provided . . . no evidence that the substitution would have
provided predictable results.” App. Br. 10. The Examiner’s finding that
Talukder’s classifier will function as Goebel’s classifier to classify a set of
data as belonging to one of two groups is sufficient to show that the
proposed substitution will yield predictable results and supports the
Examiner’s conclusion that the substitution would have been obvious.
Appellants have neither shown insufficient evidence of prima facie
obviousness nor provided evidence of secondary indicia of nonobviousness.4

3 “NLEF” refers to Talukder’s nonlinear eigenfeature extraction technique.
Talukder, p. 1.
4 “On appeal to the Board, an applicant can overcome a rejection [under
Appeal 2010-011182
Application 11/201,694

6
Nor have Appellants adequately explained why the Examiner’s finding that
Talukder and Goebel both teach binary classifiers is not sufficient to show
interchangeability and predictability. We determine that the Examiner’s
conclusion of obviousness is supported by adequate articulated reasoning
with rational underpinning.
Accordingly, we sustain the rejection of claims 1, 7, 13 and 19 under
35 U.S.C. § 103(a) as unpatentable over Goebel and Talukder.
Claims 2, 8, 14 and 20
Each of dependent claims 2, 8, 14 and 20 calls for the kernel-based
MRDF feature extractor of claims 1, 7, 13 and 19, respectively, to be
“developed using historical sensor data from known good and known bad
turbine engines.” With respect to this additional limitation, the Examiner
merely finds that Talukder discloses and discusses “training data with good
and bad discrimination.” Ans. 4 (citing Talukder, p. 18, second para., ll. 7-
9). This finding does not address the “sensor data” and “known good and
known bad turbine engines” aspects of the claim. Because the Examiner has
not provided a sound basis for believing that Talukder discloses a feature
extractor developed using historical sensor data from known good and
known bad turbine engines, we do not sustain the rejection of claims 2, 8, 14
and 20 under 35 U.S.C. § 103(a) as unpatentable over Goebel and Talukder.
Claims 3, 9, 15 and 21
Each of claim 3, 9, 15 and 21 depends from claims 2, 8, 14 and 20,
respectively, and thus includes the subject matter recited in claims 2, 8, 14

§ 103(a)] by showing insufficient evidence of prima facie obviousness or by
rebutting the prima facie case with evidence of secondary indicia of
nonobviousness.” In re Kahn, 441 F.3d 977, 985-86 (Fed. Cir. 2006)
(emphasis omitted).
Appeal 2010-011182
Application 11/201,694

7
and 20, respectively. Accordingly, for the same reasons as discussed supra
with respect to claims 2, 8, 14 and 20, we do not sustain the rejection of
claims 3, 9, 15, and 21 under 35 U.S.C. § 103(a) as unpatentable over
Goebel and Talukder.
Claims 4-6, 10-12, 16-18, 22 and 23
Because separate patentability is not argued for dependent claims 4-6,
10-12, 16-18, 22 and 23 (App. Br. 11-12), we sustain the rejection of these
claims for the reasons discussed supra with respect to independent claims 1,
7, 13 and 19.

DECISION
We affirm the rejection of claims 1, 4-7, 10-13, 16-19, 22 and 23 as
unpatentable over Goebel and Talukder.
We reverse the rejection of claims 2, 3, 8, 9, 14, 15, 20 and 21.
No time period for taking any subsequent action in connection with
this appeal may be extended under 37 C.F.R. § 1.136(a). See 37 C.F.R.
§ 1.136(a)(1)(iv).

AFFIRMED-IN-PART

Klh