Downloaded from http://rsta.royalsocietypublishing.

org/ on September 13, 2016

Phil. Trans. R. Soc. A (2007) 365, 493–514

doi:10.1098/rsta.2006.1931
Published online 12 December 2006

Static and dynamic novelty detection methods

for jet engine health monitoring
B Y P AUL H AYTON 1 , S IMUKAI U TETE 1 , D ENNIS K ING 2 , S TEVE K ING 2 ,
P AUL A NUZIS 2 AND L IONEL T ARASSENKO 1 , *
1
Department of Engineering Science, University of Oxford,
Oxford OX1 3PJ, UK
2
Rolls-Royce Civil Aero-Engines, Derby DE24 8BJ, UK

Novelty detection requires models of normality to be learnt from training data known to
be normal. The first model considered in this paper is a static model trained to detect
novel events associated with changes in the vibration spectra recorded from a jet engine.
We describe how the distribution of energy across the harmonics of a rotating shaft can
be learnt by a support vector machine model of normality. The second model is a
dynamic model partially learnt from data using an expectation–maximization-based
method. This model uses a Kalman filter to fuse performance data in order to
characterize normal engine behaviour. Deviations from normal operation are detected
using the normalized innovations squared from the Kalman filter.
Keywords: novelty detection; health monitoring; support vector machines;
Kalman filtering; expectation–maximization

1. Introduction

The novelty detection paradigm is ideally suited to monitor the health

(or condition) of safety-critical systems such as jet engines, for which abnormal
events occur very rarely. Jet engines are now routinely monitored using vibration
and performance sensors, both during their pass-off tests and during flight. In
any database of vibration and performance data, however, there will be very few
examples of abnormal behaviour. Nevertheless, the engine being monitored
might unexpectedly display an abnormal type of behaviour which has not been
seen before, and it is essential that such an occurrence should be detected as early
as possible. With novelty detection, a model of normality is learnt from training
data known to be normal and abnormalities are subsequently identified in test
data by testing for novelty against the previously learnt model.
There are many examples of model-based fault detection algorithms in the
literature (e.g. Kadirkamanathan et al. 2002; Korbicz & Janczak 2002;
Venkatasubramanian et al. 2003). A fault is a known example of an anomalous
condition for the plant being monitored. A novel event, in contrast, is a previously
unseen example of an anomalous condition. In this paper, we concentrate on
model-based novelty detection using two different types of models.
* Author for correspondence ([email protected]).
One contribution of 15 to a Theme Issue ‘Structural health monitoring’.

493 q 2006 The Royal Society

 Downloaded from http://rsta.royalsocietypublishing.org/ on September 13, 2016

494 P. Hayton et al.

The first distinction to be drawn between the different types of models

available to us relates to whether they are physical or statistical models. A
thermodynamic model is an example of a physical model, whereby the steady-
state aerothermodynamic performance of a jet engine can be simulated by an
assembly of thermodynamic models that are solved numerically subject to
continuity, momentum and energy constraints. This paper is concerned with
statistical models, for which the parameters of the models are learnt from large
numbers of examples of engine data acquired during normal operation (training
data). These statistical models may either be static or dynamic models.
Static models are sometimes divided into statistical and neural network
models (Markou & Singh 2003), but we prefer to view the latter within the
framework of statistical pattern recognition; therefore we do not make such a
distinction. In §2, we describe the training of a static neural network model to
detect novel events arising out of changes in the vibration spectra recorded from
a jet engine. Dynamic models explicitly encode temporal information which may
be essential to identify certain types of novel events. In §4, we investigate a
dynamic model for the same jet engine in which different types of performance
parameters are fused in the learnt model of normality.

2. Static model of normality

(a ) Introduction
Vibration information has traditionally been the main source of information for
identifying abnormal behaviour in a jet engine. Jet engines have a number of
rigorous pass-off tests before they can be delivered to the customer. The main
test is a vibration test over the full range of operating speeds. Vibration gauges
are attached to the casing of the engine (see figure 14 in appendix A for a block
diagram of a typical jet engine) and the speed of each shaft is measured using a
tachometer. The engine on the test bed is slowly accelerated from idle to full
speed and then gradually decelerated back to idle. As the engine accelerates, the
rotation frequency of the two (or three) shafts increases and so does the
frequency of the vibrations caused by the shafts. A tracked order is the amplitude
of the vibration signal in a narrow frequency band centred on a harmonic of the
rotation frequency of a shaft, measured as a function of engine speed. It tracks
the frequency response of the engine to the energy injected by the rotating shaft.
Most of the energy in the vibration spectrum is concentrated in the fundamental
tracked orders and their main harmonics. These, therefore, constitute the
‘vibration signature’ of the jet engine under test. It is very important to detect
departures from the normal or expected values of these tracked orders as they
will indicate an abnormal pattern of vibration.
In our previous work (Nairac et al. 1999), we investigated the vibration spectra
of a two-shaft jet engine, the Rolls-Royce Pegasus. In the available database,
there were vibration spectra recorded from 52 normal engines (the training data)
and from 33 engines with one or more unusual vibration feature (the test data).
The shape of the tracked orders with respect to speed was encoded as a low-
dimensional vector by calculating a weighted average of the vibration amplitude
over six different speed ranges (giving an 18-dimensional vector for three tracked
orders). With so few engines available, the K-means clustering algorithm was

Phil. Trans. R. Soc. A (2007)

 Downloaded from http://rsta.royalsocietypublishing.org/ on September 13, 2016

Static and dynamic novelty detection methods 495

used to construct a very simple model of normality, following component-wise

normalization of the 18-dimensional vectors. The distribution of the training
data in the transformed space was approximated by a small number (KZ4) of
spherical clusters representing the different types of vibration patterns, according
to the ‘smoothness’ of the engine.
The novelty of the vibration signature for a test engine was assessed as the
shortest distance to one of the kernel centres in the clustering model of normality
(each distance being normalized by the width associated with that kernel). If the test
vector was sufficiently far from all cluster centres, then it was clearly in a region of
space with very few training patterns; hence, it was deemed to be novel. The novelty
threshold was chosen so as to accept all training patterns as normal. When
cumulative distributions of novelty scores were plotted both for normal (training)
and test engines, there was little overlap found between the two distributions
(Nairac et al. 1999). A significant shortcoming of the method, however, is its inability
to rank engines according to novelty, since the shortest normalized distance is
evaluated with respect to different cluster centres for different engines.
In the section that follows, we describe the construction of a support vector
machine (SVM) model of normality to characterize the distribution of energy across
the vibration spectra acquired from a three-shaft engine, the Rolls-Royce Trent
500. The SVM paradigm provides a geometrical description of normality in an
appropriate space, a direct indication of the patterns on the boundary of normality
(the support vectors) and, perhaps most importantly, a ranking of ‘abnormality’
according to the distance to the separating hyperplane in feature space.

(b ) Support vector machine for novelty detection

The SVM approach to novelty detection (Hayton et al. 2001) is sometimes
described as the one-class SVM problem (Schölkopf et al. 2000b; Rätsch et al.
2002). The one-class SVM problem is easier than density estimation as it deals
with estimating regions of high probability for the training data. The normal
patterns from which the training set is assembled are known as ‘positive data’
and the abnormal patterns which may occur in the test set represent the
‘negative data’. The model is learnt from the positive data only and this
description is then used to assess whether a test pattern is likely to have been
generated by the same process as the training set (normal or positive data).
Geometrically, this means finding some hyperplane w 2F that separates the
normal training data from the origin at some threshold r. One estimates a
function f (x)Z(w$F(x)) and decides that a pattern x belongs to the normal class
whenever f (x)Rr.
Suppose we are given a set of ‘normal’ data points X Z fx 1 ; .; x [ g. Our goal is
to construct a real-valued function which, given a previously unseen test point x,
characterizes the ‘X-ness’ of the point x, i.e. which takes on large values for
points similar to those in X. The algorithm that we present here will return such
a function, along with a threshold value, such that a prespecified fraction of X
will lead to function values above threshold. In this sense, we are estimating a
region which captures a certain probability mass.
This approach employs two concepts from SVM theory (Vapnik 1995) that are
essential for good generalization performance in high-dimensional spaces: (i)
maximizing a margin and (ii) nonlinearly mapping the data into some feature

Phil. Trans. R. Soc. A (2007)

 Downloaded from http://rsta.royalsocietypublishing.org/ on September 13, 2016

496 P. Hayton et al.

space F endowed with a dot product. The latter need not be the case for the input
domain c which may be a general set. The connection between the input domain
and the feature space is established by a feature map F: c/F, i.e. a map such
that some simple kernel (Boser et al. 1992; Vapnik 1995)
kðx; yÞ Z ðFðxÞ$FðyÞÞ; ð2:1Þ
such as the Gaussian
2
kðx; yÞ Z eKkxKyk =c ; ð2:2Þ
provides a dot product in the image of F. In practice, we need not necessarily
worry about F, as long as a given k satisfies certain positivity conditions
(Vapnik 1995).
As F is a dot product space, we can use tools of linear algebra and geometry to
construct algorithms in F, even if the input domain c is discrete. Below, we
derive our results in F, using the following shorthand notation:
xi Z Fðx i Þ; ð2:3Þ
X Z fx1 ; .; x[ g: ð2:4Þ
Indices i and j are understood to range over 1; .; [ (in compact notation:
i; j 2½[
); similarly, n,p2[t]. Boldface Greek letters denote [-dimensional vectors
whose components are labelled using normal face typeset.
Using an algorithm proposed for the estimation of a distribution’s support
(Schölkopf et al. 2000b), we seek to separate X from the origin with a large
margin hyperplane committing few training errors. Projections on the normal
vector of the hyperplane then characterize the X-ness of test points, and the area
where the decision function takes the value 1 can serve as an approximation of
the support of X.
The decision function is found by minimizing a weighted sum of a support
vector type regularizer and an empirical error term depending on an overall
margin variable r and individual errors xi,
1 1 X
min kwk2 C x Kr; ð2:5Þ
w2F;x2R[ ;r2R 2 n[ i i
subject to ðw$xi ÞR rKxi ; xi R 0: ð2:6Þ
The precise meaning of the parameter n governing the trade-off between the
regularizer and the training error will become clear later. Since non-zero slack
variables xi are penalized in the objective function, we can expect that if w and r
solve this problem, then the decision function
f ðxÞ Z sgnððw$xÞKrÞ; ð2:7Þ
will be positive for many examples xi contained in X, while the support vector
type regularization term kwk will still be small.
We next compute a dual form of this optimization problem. The details of the
calculation, which uses standard techniques of constrained optimization, can be
found in Schölkopf et al. (2000a). We introduce a Lagrangian and set the
derivatives with respect to w equal to zero, yielding in particular
X
wZ ai x i : ð2:8Þ
i

Phil. Trans. R. Soc. A (2007)

 Downloaded from http://rsta.royalsocietypublishing.org/ on September 13, 2016

Static and dynamic novelty detection methods 497

All patterns fx i i 2½[
; ai O 0g are called support vectors. The expansion (2.8)
turns the decision function
P (2.7) into a form which depends only on the dot
products, f ðxÞZ sgnðð i ai xi $xÞKrÞ.
By multiplying out the dot products, we obtain a form that can be written as a
nonlinear decision function on the input domain c in terms of a kernel (2.1) (cf. (2.3)).
A short calculation yields !
X
f ðxÞ Z sgn ai kðx i ; xÞKr :
i
In the argument of the sgn, only the first two terms depend on x; therefore, we
may absorb the next terms in the constant r, which we have not fixed yet. To
compute r in the final form of the decision function
!
X
f ðxÞ Z sgn ai kðx i ; xÞKr ; ð2:9Þ
i
we employ the Karush–Kuhn–Tucker (KKT) conditions of the optimization
problem (e.g. Vapnik 1995). They state that for points xi where 0! ai ! 1=ðn[Þ,
the inequality constraint (2.6) becomes equalities (note that in general,
ai 2½0; 1=ðn[Þ
), and the argument of the sgn in the decision function should
equal 0, i.e. the corresponding xi sits exactly on the hyperplane of separation.
The KKT conditions also imply that only those points xi can have a non-zero
ai for which the first inequality constraint in (2.6) is precisely met; therefore, the
support vectors Xi with aiO0 will often form but a small subset of X.
Substituting (2.8) (the derivative of the Lagrangian by w) and the
corresponding conditions for x and r into the Lagrangian, we can eliminate the
primal variables to get the dual problem. A short calculation shows that it
consists of minimizing the quadratic form
1X

W ðaÞ Z ai aj kðx i ; x j Þ ; ð2:10Þ
2 ij
subject to the constraints X
0% ai % 1=ðn[Þ; ai Z 1: ð2:11Þ
i
This convex quadratic program can be solved with standard quadratic
programming tools. Alternatively, one can employ the sequential minimal
optimization algorithm described in Schölkopf et al. (1999), which was found to
approximately scale quadratically with the training set size.

3. Application of support vector machine to jet engine

vibration signatures

The SVM algorithm for novelty detection is applied to the distribution of vibration
energies across the vibration harmonics for the three shafts of the Rolls-Royce Trent
500. Typically, in a jet engine, the amplitude of vibration is a measure of how well
balanced an engine shaft is and does not necessarily give an indication of the health of
an engine. However, interactions between components of the engine give rise to
harmonics (known as ‘multiple orders’) or sub-harmonics (known as ‘fractional
orders’) of the shaft speed which can indicate a developing fault.

Phil. Trans. R. Soc. A (2007)

 Downloaded from http://rsta.royalsocietypublishing.org/ on September 13, 2016

498 P. Hayton et al.

The approach taken is to model the distribution of energy associated with

each shaft over the harmonics relative to the fundamental vibration amplitude.
A vector representing the vibration energy for a shaft is therefore generated,
0 1
speed
B C
B 0:5s=1s C
B C
B C
vð Z B 1:5s=1s C;
B C
B C
@ 2s=1s A
3s=1s
where speed is the shaft speed as a percentage of the full shaft speed and s
represents the shaft frequency so that 0.5s is the amplitude of vibration at half
the shaft vibration frequency (a half-order). By dividing the vibration
amplitude for each fractional or multiple tracked order by the fundamental
amplitude, we are removing the dependency on absolute vibration amplitude.

(a ) Data extraction
A number of Trent 500 engines were monitored for several months during
the Engine Development Programme and the vibration data were recorded
from each of the three shafts. As vibration spectra are extracted by the system
every 0.2 s (King et al. 2002), this represents a considerable amount of data.
To generate a set of training vectors of manageable size, only vectors which
differ from the previous accepted vector significantly are added to the training
set. When a jet engine maintains a constant condition for long periods of time,
only one training vector is generated but as the engine changes state (for
example, during acceleration), many vectors are generated.
Using the training dataset, the SVM algorithm finds the hyperplane that
separates the normal data from the origin in feature space with the largest
margin. The number of support vectors gives an indication of how well the
algorithm is generalizing (if all data points were support vectors, the algorithm
would have memorized the data). A Gaussian kernel was used with a width
cZ40.0 in equation (2.2). This value was chosen by starting with a small kernel
width (so that the algorithm memorizes the data), increasing the width and
stopping when similar results are obtained both on the training set and another
set of data kept apart for validation. The number of support vectors generated
depends both on the similarity criterion and the number of training patterns.
With of the order of 103 training patterns generated from 54 engine runs (four
different engines), the number of support vectors varied from a minimum of 7
(for the low-pressure (LP) shaft) to a maximum of 36 (for the intermediate-
pressure (IP) shaft).

(b ) Data visualization
To illustrate the effectiveness of the trained SVM model, the data were
visualized in two dimensions, using the Neuroscale algorithm (Tipping & Lowe
1998). The mapping of a dataset onto a two-dimensional space for visualization
purposes is known in the pattern recognition literature as multi-dimensional
scaling. A well-known example of such a dimensionality-reduction mapping is

Phil. Trans. R. Soc. A (2007)

 Downloaded from http://rsta.royalsocietypublishing.org/ on September 13, 2016

Static and dynamic novelty detection methods 499

neuroscale
1.1

0.8

0.4

– 0.1

– 0.2

– 0.6
–1.7 –1.5 –1.4 –1.2 –1.0 – 0.8 –0.7 –0.5 –0.3 –0.2 0.0

Figure 1. Neuroscale visualization plot for the five-dimensional vectors of speed and tracked orders
for the IP shaft.

Sammon’s (1969) mapping, which seeks to find a configuration of image points in

the two-dimensional visualization space, such that the distances dij between
image points are as close as possible to the corresponding distances dij, in the
high-dimensional input space. Since it is not possible to find a configuration for
which dijZdij, Sammon’s mapping uses the following sum-of-squared-error
criterion for assessing the suitability of a configuration with respect to the others:
1 X ðdij K dij Þ2
ES2 Z P :
i!j dij i!j
dij

The Neuroscale algorithm is an extension of Sammon’s mapping, in which the

mapping from the high-dimensional input space to the two-dimensional
visualization space is parameterized using a Radial Basis Function neural
network (Tipping & Lowe 1998).
The data from two engine tests where events of interest occurred were
visualized alongside the training data for two of the engine shaft models. The first
event was a simultaneous increase in two of the vibration modes of the IP shaft of
one of the development engines.
The Neuroscale visualization plot shows the training data as the densely
packed grey circles and the points from the test data sequence as black crosses.
As can be seen from figure 1, the latter indicates a significant excursion away
from the region of normality in the Neuroscale visualization plot.
Figure 2 shows that running the SVM model of normality over the engine test data
for the IP shaft identifies the novel vibration pattern associated with abnormal
increases in amplitude (relative to the others) for the fundamental, 1i, and half-order,
0.5i. With the novelty threshold used (previously determined empirically on past
data), the period of novelty lasts for 4 s, as denoted by the two vertical lines.

Phil. Trans. R. Soc. A (2007)

 Downloaded from http://rsta.royalsocietypublishing.org/ on September 13, 2016

500 P. Hayton et al.

in /s pk
(C: contractual limit)
0.54
0.51 1i
0.48 0.5i
0.45 1.5i
2i
0.42 3i
0.39 C
0.36
0.33
0.30
0.27
0.24
0.21
0.18
0.15
0.12
0.09
0.06
0.03
0.00
01 : 03 : 52.1 01:04:10.3
time
Figure 2. An example vibration pattern identified as novel (between the two vertical lines) by the
SVM model of normality for the IP shaft. The 1i and 0.5i tracked orders show a large increase in
vibration level during the interval marked by the two vertical lines.

0.2 neuroscale

0.1

0.1

–0.0

–0.1

–0.1
–0.0 0.0 0.1 0.1 0.1 0.2 0.2 0.2 0.3 0.3 0.4
Figure 3. Neuroscale visualization plot for the five-dimensional vectors of speed and tracked orders
for the LP shaft.

The second use of the SVM model of normality is for an abnormal pattern of
vibration for the LP shaft as the engine decelerated by 7%, 2 min after suffering a
foreign object damage event. Again, the Neuroscale visualization plot (figure 3)
shows that the test sequence data vectors (black crosses) are some distance away

Phil. Trans. R. Soc. A (2007)

 Downloaded from http://rsta.royalsocietypublishing.org/ on September 13, 2016

Static and dynamic novelty detection methods 501

in/s pk
1.8 (C: contractual limit)
1.7 1l
1.6 0.5l
1.5 1.5l
1.4 2l
1.3 3l
1.2
1.1 C
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
15 : 07 : 11.0 15:07:20.0
time

Figure 4. A second example vibration pattern identified as novel (between the two vertical lines)
by the SVM model of normality for the LP shaft. The 1l and 2l tracked orders show a marked
decrease in vibration level towards the end of the interval marked by the two vertical lines.

from the (normal) training data shown as small grey circles, although the black
crosses are closer to the latter than in the IP shaft example. The short excursion
away from normality causes the SVM model to identify novelty for approximately
1 s, during the period indicated by the two vertical lines of figure 4.

4. Dynamic model

(a ) Introduction
A state-variable model is a dynamic model (usually linear), for which the
interrelation between engine ‘states’, inputs and outputs is described with a set
of (linear) differential equations. State-space descriptions provide a mathemat-
ically rigorous tool for system modelling and residual generation, which may be
used in fault detection (Chen & Patton 1998) or, as in this paper, novelty
detection. Residuals (the difference between the observations and predicted
observations from the model) need to be processed in order to identify the novel
event reliably and robustly. By assembling a set of linear models for a range of
power conditions, a piece-wise linear state-space model can be constructed. In
this paper, we present results from a linear dynamic model used at a given power
condition (greater than 70% of maximum engine power).
The Kalman filter is a recursive linear algorithm for estimating system states
(Gelb 1974). This section describes the use of a Kalman filter as a linear dynamic
model capable of fusing performance data to detect novel events. A further aspect
of this work, in common with §2 on static models, is the emphasis on learning from

Phil. Trans. R. Soc. A (2007)

 Downloaded from http://rsta.royalsocietypublishing.org/ on September 13, 2016

502 P. Hayton et al.

data: some of the parameters of the linear dynamic model are learnt, using an
expectation–maximization (EM)-based method described in Ghahramani &
Hinton (1996), during a prior training phase with normal data only.

(b ) Model description
It is assumed that the system to be monitored can be described in its fault-free
condition by a linear, discrete-time, dynamic model described by equations (4.1)
and (4.2). Note that the variables which are to be related in the model must be
ones for which a linear relationship can be extracted.
The state equation of the system is given by
xðk C 1Þ Z AðkÞxðkÞ C wðkÞ; ð4:1Þ
where x(k) is the state at time-step k; A(k) is the process model at time-step k;
and w(k) is a zero-mean Gaussian noise process, with covariance matrix Q.
x(k) is a hidden state—there is no direct access to it. However, observations
y(k) are made, which are assumed to be described by a measurement equation
relating the observations to the state.
The measurement equation is
yðkÞ Z CðkÞxðkÞ C vðkÞ; ð4:2Þ
where y(k) is the observation vector; C(k) is the observation model at time-step
k; and v(k) is a zero-mean Gaussian noise process with covariance matrix R.
The model matrices A, C, Q and R are set during a training phase under
normal operating conditions (see below). During monitoring, excursions away
from the trained model can be identified using the Kalman filter innovations,
^
nðkÞ Z yðkÞKCðkÞxðkjk K1Þ; ð4:3Þ
^
where xðkjk K1Þ is the estimate of the state at time-step k, given the knowledge
at time-step k K1. The Kalman filter runs through a cycle of prediction and
correction. The filter’s estimate of state and the error covariance are updated
when a measurement is obtained.
The Kalman filter prediction
^
xðkjk K1Þ Z AðkÞxðk^ K1jk K1Þ; ð4:4Þ
has associated uncertainty, given by the state prediction covariance
Pðkjk K1Þ Z AðkÞPðk K1jk K1ÞAT ðkÞ C QðkÞ: ð4:5Þ
The Kalman gain is given by
KðkÞ Z Pðkjk K1ÞC T ðkÞS K1 ðkÞ; ð4:6Þ
where S(k) is the innovation covariance.
The innovation covariance S(k) is
SðkÞ Z CðkÞPðkjk K1ÞC T ðkÞ C RðkÞ: ð4:7Þ
The state estimate is
^
xðkjkÞ ^
Z xðkjk K1Þ C KðkÞnðkÞ: ð4:8Þ
The associated state covariance P(k) is given by
PðkÞ Z ðI KKðkÞCðkÞÞPðkjk K1Þ: ð4:9Þ

Phil. Trans. R. Soc. A (2007)

 Downloaded from http://rsta.royalsocietypublishing.org/ on September 13, 2016

Static and dynamic novelty detection methods 503

5. Novelty detection using dynamic model

In the sections that follow, we present an approach for learning two of the four
model matrices A, C, Q and R from data. The models are speed-based
performance models of the engine, in which the relationships are learnt from
training sets of normal data. Changes in test data are then detected by
monitoring the normalized innovations squared (NIS) from the Kalman filter,
where NIS is defined as follows:
NISðkÞ Z nT ðkÞS K1 ðkÞnðkÞ: ð5:1Þ
The innovations should be zero mean and white, with covariance consistent
with that calculated by the filter (Bar-Shalom & Li 1993).
Whenever there is a departure from the normal behaviour captured in the
learnt model, there is a rise in NIS. This can be caused by gradual component
deterioration or by unexpected events which affect the relationship between the
performance parameters, or observations y, measured on the engine. To illustrate
this, we consider one such event, which occurred in a development Trent 500
engine on a test bed. A problem with the radial driveshaft caused the three shaft
speeds to undergo a sudden change from normal behaviour. In the main event
region, the speeds diverge for approximately 10 s: the high-pressure (HP) shaft
speed increases, the LP shaft speed remains approximately constant and the IP
shaft speed decreases. The engine is then shut down.

(a ) Expectation–maximization-based learning method

The learning of the model of normality uses the EM algorithm within the
framework described by Ghahramani & Hinton (1996). For a sequence of
observation vectors y and state vectors x, the joint log probability, given a
Gaussian initial state density, is shown to be expressible as a sum of quadratic
terms. The EM algorithm consists of successive expectation and maximization
steps. In the expectation stage, the expected log likelihood is computed. The
maximization step involves estimating the system parameters by taking partial
derivatives of the expected log likelihood and equating them to zero. This method
is further described in Roweis & Ghahramani (1999), which also presents
algorithms for determining the parameters. Software implementing these
algorithms can be found in a well-known Kalman filtering toolbox (Murphy
1998–2002). In the sections that follow, we show how these algorithms can be used
to learn performance-based models for the Trent 500 from normal data. Learning
of the model matrices from data does not give a unique state-space representation
and so the learning process has to be guided by imposing strong prior constraints.

(b ) Training and test data

Training data sequences are taken from 3 days of engine running: from runs
several days before the event and from running on the event day, but prior to the
event. The test data sequences are taken from the event day—from a later
sequence than any of the training data, but prior to the main event, and from the
main event itself. The data are subsampled, so that one in every five points is
extracted. This corresponds to a data point per second. The subsampled data are
then averaged over a five-point sliding window.

Phil. Trans. R. Soc. A (2007)

 Downloaded from http://rsta.royalsocietypublishing.org/ on September 13, 2016

504 P. Hayton et al.

In the engine datasets, there are long periods of running at constant speed.
Accelerations and decelerations, by contrast, tend to cover a relatively small number
of data points. The data points are required to be sequential as their time history is
important. The datasets used for training models are balanced so that the different
types of operating regime are given as equal a weighting as possible. In this paper, we
also concentrate on models of normality for smooth acceleration and deceleration.
A separate model is required for sudden acceleration and deceleration, during which
the rates of change of speed are much greater than presented here.
The training data are shown in the eight sequences of figure 5. In both training
and testing, the shaft speed data used are in the region where the LP speed is
between 70 and 90% of maximum LP speed. Within this speed sub-range, the
assumption of model linearity is taken to be valid. Figure 6 presents the two
sequences of test data; figure 6b covering the main event. Although the three
shafts are separate systems, the shaft speeds should change in tandem. This
condition is violated as a result of the radial driveshaft problem in the event
sequence of figure 6, between tZ30 and 50, when (i) the LP shaft speed does not
change, (ii) the IP shaft speed decreases, and (iii) the HP shaft speed increases.
Initial tests are carried out with a dynamic model of normality based solely on
a three-speed observation vector. In order to make learning possible, we
introduce the constraint that the measurement matrix C should be equal to the
identity matrix. Hence, equation (4.2) becomes
yðkÞ Z xðkÞ C vðkÞ: ð5:2Þ
The observation vector is therefore the state corrupted by noise. The
measurement noise covariance matrix R is assumed to be diagonal and set using
engine measurement uncertainty data sheet values provided by Rolls-Royce,
R Z ½0:1538 0:0 0:0
0:0 0:0989 0:0
0:0 0:0 0:0902
:
The C and R values remain fixed throughout (no learning). The state transition
matrix, A, the process noise covariance, Q, and the initial state covariance, P(0),
are learned using the EM-based algorithm of Ghahramani & Hinton (1996). The
state transition matrix is initialized to I and the process noise covariance to 0.1I.
The initial state is taken from an average over five points for one of the training
sequences, and the initial state covariance is set to I.
The learning curve for the system is shown in figure 7. After 100 EM
iterations, the state transition matrix is found to be
A Z ½0:9661 0:06601 K0:03673
K0:02079 1:041 K0:02265
K0:01780 0:03546 0:9799
:
Thus, learning does not change the state transition matrix, A, significantly as it
remains approximately equal to I. The process noise covariance matrix becomes
Q Z ½0:01878 0:01166 0:01011
0:01166 0:007916 0:006619
0:01011 0:006619 0:005852
:

Phil. Trans. R. Soc. A (2007)

 Downloaded from http://rsta.royalsocietypublishing.org/ on September 13, 2016

Static and dynamic novelty detection methods 505

(a) 100

95

90

85

80

75
n1v
n2v
n3v
70
0 50 100 150 0 50 100 0 50 0 50

(b) 90

88

86 n1v
n2v
n3v
84

82

80

78

76

74

72
0 50 0 100 200 300 0 100 200 300 0 50 100 150

Figure 5. Training data. The figure shows shaft speed data (acceleration and deceleration
manoeuvres) taken from several days prior to the event (a, four plots) and also from the event day
(b, four plots), but from sequences occurring well before the event. n1v, LP shaft speed; n2v, IP shaft
speed; n3v, HP shaft speed (all expressed as a percentage of the maximum speed for that shaft).

Phil. Trans. R. Soc. A (2007)

 Downloaded from http://rsta.royalsocietypublishing.org/ on September 13, 2016

506 P. Hayton et al.

(a) (b)
95
n1v
n2v
n3v

90

85

80

75

70
0 20 40 60 80 0 20 40 60 80

Figure 6. Test data. The figure shows shaft speed data taken from the event day. (a) Test data
recorded before the event. (b) Data acquired just before and at the time of the event (tZ30).

learning curve
200

–200
log likelihood

– 400

–600

– 800

–1000

–1200
0 10 20 30 40 50 60 70 80 90 100
EM iteration number

Figure 7. The log likelihood values over 100 EM iterations, for a system with a three-speed
observation vector, are shown.

Phil. Trans. R. Soc. A (2007)

 Downloaded from http://rsta.royalsocietypublishing.org/ on September 13, 2016

Static and dynamic novelty detection methods 507

The learned initial state uncertainty is

Pð0Þ Z ½20:36 14:77 11:10
14:77 11:71 7:796
11:10 7:796 6:190
:
Once the three-speed model has been learned, it is applied to the training
sequences of figure 5 to check the range of NIS values obtained during the
(normal) changes of shaft speed associated with acceleration and deceleration
manoeuvres.
Figure 8 shows the changes in NIS for the training data sequences taken
from an engine for several days prior to the event and figure 9 shows the
changes in NIS for training data sequences recorded during the morning of the
event day. (Note that figures 8 and 9 show the time of day on the horizontal
axis.) With each sequence, there is a transient rise in NIS when the
acceleration (or deceleration) manoeuvre occurs, but never above a value of
3.0. On the basis of these results, it would be possible to set a threshold of 8.0
for the detection of a novel event (approximately 2.5 times the maximum
value of NIS on the training data).
When the test data of figure 10 are run through the same three-speed model,
figure 10a gives a maximum NIS of 0.6, but the diverging speeds from
tZ17 : 30 : 03 in the event sequence cause the NIS value to rise first to 8, then to
10 at tZ17 : 30 : 33 and eventually to 15, just before the automatic shutdown
occurs. With the novelty threshold set at NISZ8.0, the event would have been
detected at tZ17 : 30 : 16.

(c ) Speeds, pressure and temperature

The model of normality is now extended to five performance parameters by
incorporating the high-pressure compressor (HPC) exit pressure, p30 (p30 is
one of the P3 set of pressure readings; see figure 14 in appendix A), and the
turbine gas temperature, tgt, into the observation vector, in addition to the three
shaft speeds.
As with the three-speed model, C is set to I. R is now a diagonal matrix, with
two extra values for the measurement uncertainties for p30 and tgt,
R Z ½1:1538 0:0 0:0 0:0 0:0
0:0 0:0989 0:0 0:0 0:0
0:0 0:0 0:0902 0:0 0:0
0:0 0:0 0:0 0:7300 0:0
0:0 0:0 0:0 0:0 2:000
:
The state transition matrix, A, and the process noise covariance, Q, are again
learnt, after being set to initial values of I and 0.1I, respectively. The initial state
is taken from an average over five points for one of the training sequences and the
initial state covariance P(0) is set to I. A, Q and P(0) are all learned, as with the
three-speed model.

Phil. Trans. R. Soc. A (2007)

 Downloaded from http://rsta.royalsocietypublishing.org/ on September 13, 2016

508 P. Hayton et al.

(a) (b)
100 100

95 95
n1v n2v n3v

90 90

85 85

80 80

1.4 2.5
1.2
2.0
1.0
0.8 1.5
NIS

0.6 1.0
0.4
0.5
0.2
0 0
19:23:14 19:23:31 19:23:48 19:24:06 19:24:23 19:24:40 19:24:58 19:25:15 19:25:32 19:25:49 19:27:16 19:27:33 19:27:50 19:28:08 19:28:25 19:28:42 19:29:00 19:29:17

data and NIS data and NIS

number of observations =3 state size = 3 number of observations = 3 state size = 3
(c) observations: n1v n2v n3v (d) observations: n1v n2v n3v
95 100
95
90
n1v n2v n3v

90
85 85
80
80
75
75 70

3.0 3.5
2.5 3.0
2.5
2.0
2.0
NIS

1.5
1.5
1.0
1.0
0.5 0.5
0 0
19:35:02 19:35:11 19:35:20 19:35:28 19:35:37 19:35:46 19:35:54 19:36:03 19:36:12 19:36:20 11:42:26 11:42:43 11:43:00 11:43:18 11:43:35 11:43:52 11:44:10
time time

Figure 8. (a–d ) The change in NIS for the training data sequences taken from an engine run several
days prior to the event. This test on training data is included as a check.

After 100 EM iterations, the learned value of A is

AZ ½0:7467 K0:1705 0:4583 0:02863 K0:01574

K0:1537 0:8918 0:2847 0:01771 K0:00982
K0:1232 K0:08763 1:232 0:01448 K0:00819
K2:807 K1:8870 5:110 1:320 K0:1781
K3:558 K2:6820 6:827 0:4233 0:7631
:

The learned process noise covariance is

QZ ½0:00148 0:000414 0:000396 0:01429 K0:00542

0:000411 0:000661 0:000473 0:00846 K0:00147
0:000394 0:000473 0:000600 0:00725 K0:000203
0:0143 0:00846 0:00725 0:2048 K0:0624
K0:00542 K0:00148 K0:000207 K0:0624 0:0669
:

Phil. Trans. R. Soc. A (2007)

 Downloaded from http://rsta.royalsocietypublishing.org/ on September 13, 2016

Static and dynamic novelty detection methods 509

(a) (b)
90

85
n1v n2v n3v

80

75

70

0.7
0.6
0.5
0.4
NIS

0.3
0.2
0.1
0
12:32:41 12:32:50 12:32:59 12:33:07 12:33:16 12:33:24 12:33:33 12:33:42 12:33:50 12:33:59 12:46:05 12:46:48 12:47:31 12:48:14 12:48:58 12:49:41 12:50:24 12:51:07

data and NIS data and NIS

number of observations=3 state size=3 number of observations=3 state size=3
(c) observations: n1v n2v n3v (d) observations: n1v n2v n3v
90 90
88
85 86
n1v n2v n3v

84
80
82
75 80
78
70 76

2.0 1.0

0.8
1.5
0.6
NIS

1.0
0.4
0.5
0.2

0 0
13:27:50 13:28:34 13:29:17 13:30:00 13:30:43 13:31:26 13:45:07 13:45:24 13:45:42 13:45:59 13:46:16 13:46:34 13:46:51 13:47:08
time time

Figure 9. (a–d ) The change in NIS for training data sequences taken from the day of the event, but
well before the event. This test on training data is included as a check.

data and NIS data and NIS

number of observations = 3 state size = 3 number of observations = 3 state size = 3
(a) observations: n1v n2v n3v (b) observations: n1v n2v n3v
90 95
88
90
86
n1v n2v n3v

84 85
82 80
80
75
78
76 70

0.8 15

0.6
10
NIS

0.4
5
0.2

0 0
17:01:06 17:01:15 17:01:24 17:01:32 17:01:41 17:01:49 17:01:58 17:02:07 17:02:15 17:02:24 17:29:28 17:29:37 17:29:46 17:29:54 17:30:03 17:30:12 17:30:20 17:30:29 17:30:37 17:30:46
time time

Figure 10. The changes in NIS for the two test sequences (three-dimensional observation model),
(a) prior to the radial driveshaft event and (b) at the time of the event. Note the two different
scales used for the NIS axis.

Phil. Trans. R. Soc. A (2007)

 Downloaded from http://rsta.royalsocietypublishing.org/ on September 13, 2016

510 P. Hayton et al.

(a) (b)
100
n1v n2v n3v

90

80
86
84
p30

82
80
86
84
tgt

82
80
10
NIS

0
19:23:14 19:23:31 19:23:48 19:24:06 19:24:23 19:24:40 19:24:58 19:25:15 19:25:32 19:25:49 19:27:16 19:27:33 19:27:50 19:28:08 19:28:25 19:28:42 19:29:00 19:29:17

data and NIS data and NIS

(c) number of observations = 5 state size = 5
observations: n1v n2v n3v p30 tgt
(d) number of observations = 5 state size = 5
observations: n1v n2v n3v p30 tgt
100 100
n1v n2v n3v

90 90
80 80
70 70
82 90
80
p30

80
78
76 70
82 90
80
tgt

80
78
76 70
6 15
4 10
NIS

2 5
0 0
19:35:02 19:35:11 19:35:20 19:35:28 19:35:37 19:35:46 19:35:54 19:36:03 19:36:12 19:36:20 11:42:26 11:42:43 11:43:00 11:43:18 11:43:35 11:43:52 11:44:10
time time

Figure 11. (a–d ) The change in NIS for the first four training sequences (several days before the
event) using the five-dimensional model (speeds plus tgt and p30).

The learned initial state covariance is

Pð0Þ Z ½20:75 14:88 10:83 264:1 309:6
14:88 11:77 7:56 198:3 216:4
10:83 7:56 5:79 136:3 165:4
264:1 198:3 136:3 3436 3898
309:6 216:4 165:4 3898 4728
:
As before, once the five-dimensional observation model has been learned, it is
applied to the training sequences of figure 5 to check the range of NIS values on
the training data. Figure 11 shows the three speeds, p30, tgt and NIS for the first
four of the eight training sequences (taken from engine runs which took place
several days before the event).
Figure 12 shows the three speeds, p30, tgt and NIS values for the four training
sequences taken from the event day, but well before the occurrence of the event.
These figures reveal that the largest value of NIS on the training data for the five-
dimensional observation model is just below 15.0, which occurs for the fourth of

Phil. Trans. R. Soc. A (2007)

 Downloaded from http://rsta.royalsocietypublishing.org/ on September 13, 2016

Static and dynamic novelty detection methods 511

(a) (b)
90 90
n1v n2v n3v

80 80

70 70
75 74.5
74.0
p30

74
73.5
73 73.0
75 74.5
74.0
tgt

74
73.5
73 73.0
10 3
2
NIS

5
1
0 0
12:32:41 12:32:50 12:32:59 12:33:07 12:33:16 12:33:24 12:33:33 12:33:42 12:33:50 12:33:59 12:46:05 12:46:48 12:47:31 12:48:14 12:48:58 12:49:41 12:50:24 12:51:07

data and NIS data and NIS

number of observations = 5 state size = 5 number of observations = 5 state size = 5
(c) observations: n1v n2v n3v p30 tgt (d ) observations: n1v n2v n3v p30 tgt
90 90
n1v n2v n3v

85
80
80
70 75
78 79
76
p30

78
74
72 77
78 79
76
tgt

78
74
72 77
4 3
2
NIS

2
1
0 0
13:27:50 13:28:34 13:29:17 13:30:00 13:30:43 13:31:26 13:45:07 13:45:24 13:45:42 13:45:59 13:46:16 13:46:34 13:46:51 13:47:08
time time

Figure 12. (a–d ) The change in NIS for the last four training sequences (same days as the event but
well before its occurrence) using the five-dimensional model (speeds plus tgt and p30).

the eight sequences. NIS values are higher than for the three-speed model as a
result of the increased dimensionality of this model. Based on these results, it
would be possible to suggest a novelty detection threshold set at 37.0 for the five-
dimensional observation model (approximately 2.5 times the maximum value of
NIS on the training data).
Figure 13 shows the three speeds, p30, tgt and NIS for the two test sequences.
With the first test sequence (figure 13a), NIS only rises to 4.0. When the model is
applied to the event sequence which includes the diverging speeds (figure 13b),
the NIS value rises rapidly to 60.0 before subsequently decreasing. It is clear
that the five-dimensional observation model, which includes more information
about the engine than just the three shaft speeds, would have allowed an earlier
detection of the novel event (at tZ17 : 30 : 01, if a novelty threshold of 37.0 had
been used).
The limits for normal NIS values could be set in a training phase by working out
time-averaged values over data windows of several tens of seconds. These
statistically significant values could then be used for comparison in the testing
phase. However, in testing, point-by-point values of NIS (i.e. every second) would
still be used in preference to averages owing to the need to detect sudden changes.

Phil. Trans. R. Soc. A (2007)

 Downloaded from http://rsta.royalsocietypublishing.org/ on September 13, 2016

512 P. Hayton et al.

data and NIS data and NIS

number of observations = 5 state size = 5 number of observations = 5 state size = 5
(a) observations: n1v n2v n3v p30 tgt (b) observations: n1v n2v n3v p30 tgt
90 100
n1v n2v n3v

85 90
80 80
75 70
79 80
78
p30

75
77
76 70
79 80
78
tgt

75
77
76 70
6 60
4 40
NIS

2 20
0 0
17:01:06 17:01:15 17:01:24 17:01:32 17:01:41 17:01:49 17:01:58 17:02:07 17:02:15 17:02:24 17:29:28 17:29:37 17:29:46 17:29:54 17:30:03 17:30:12 17:30:20 17:30:29 17:30:37 17:30:46
time time

Figure 13. The change in NIS for the two test sequences (five-dimensional model), (a) prior to the
event and (b) at the time of the event. Note the two different scales used for the NIS axis.

6. Conclusions

Novelty can be used to detect anomalous behaviour in a jet engine. Section 2

showed how a static model of normality can be trained to learn the normal
distribution of energy among main vibration tracked orders. A five-dimensional
vector was constructed for the vibration amplitudes of the main harmonics (0.5,
1.5, 2 and 3 times the fundamental) at each speed, normalized with respect to the
amplitude of the fundamental. Using a training dataset of normal five-
dimensional vectors, the SVM algorithm found the hyperplane that separates
the normal data from the origin in feature space with the largest margin.
Training the SVM model consists of minimizing a quadratic subject to a set of
constraints using standard quadratic programming tools, as previously described
in the SVM literature. The number of support vectors was determined
empirically using a validation set and the SVM model was then used to highlight
two events which gave rise to abnormal vibration patterns. Although the SVM
model can be used to rank novelty, with these two examples it was simply used
with a ‘novelty threshold’ determined empirically from analysis of past data.
For the dynamic model of §4, based on three or five performance parameters,
the method of Ghahramani & Hinton (1996) and Roweis & Ghahramani (1999)
was used to learn the state transition matrix, the process noise covariance matrix
and the initial state uncertainty. Linear behaviour within a restricted engine
speed range was assumed to be a sufficiently accurate approximation to the
behaviour of the engine during smooth acceleration and deceleration
manoeuvres. In order to make it possible to learn the model parameters from
training data, a number of further assumptions were introduced. The main one of
these was that the observation vector is simply the state corrupted by noise, i.e.
the measurement matrix C is the identity matrix.
A three-speed model of normality was able to identify as novel an event which
gave rise to diverging speeds prior to automatic shutdown of the engine on the
test bed. When two further parameters (HPC exit pressure and turbine gas

Phil. Trans. R. Soc. A (2007)

 Downloaded from http://rsta.royalsocietypublishing.org/ on September 13, 2016

Static and dynamic novelty detection methods 513

temperature) were added and a new five-dimensional model learnt from the same
training sequences, the level of discrimination between the novelty index (NIS)
on the anomalous data and the normal training data increased. This is not
surprising as more information is being provided, but increased dimensionality
can make the learning of the model from data more difficult.
A complete description of the engine’s behaviour for novelty detection
purposes can be provided through the use of multiple models. It is then possible
to switch between different dynamic models according to speed range and the
rate of change of speed, depending on the type of acceleration or deceleration
manoeuvre (smooth, rapid or ‘slam’).
In the continuing application of this work on the Rolls-Royce Trent family of
jet engines (Bailey et al. 2004), it is planned that novel events will be identified
during flight (on the rare occasions on which they occur) and notified to the
maintenance engineering crew on the ground once the aircraft has landed.
The authors wish to thank Dr Visakan Kadirkamanathan for his helpful discussions on learning
dynamic models from data.

Appendix A

P1 P2 P3 P4 P5 P6 P7 P8
T1 T2 T3 T4 T5 T6 T7 T8

P0
T0

P0 T0 ambient P4 T4 turbine entry

P1 T1 inlet P5 T5 high-pressure turbine exit
P2 T2 low-pressure compressor delivery P6 T6 low-pressure turbine exit
P3 T3 high-pressure compressor delivery P7 T7 exhaust
P8 T8 propelling nozzle

Figure 14. Temperature and pressure notation of a typical turbo-jet engine. Picture courtesy of
Rolls-Royce plc (1986). Picture reproduced with the permission of Rolls-Royce plc.

Phil. Trans. R. Soc. A (2007)

 Downloaded from http://rsta.royalsocietypublishing.org/ on September 13, 2016

514 P. Hayton et al.

References
Bailey, V., Utete, S., Bannister, P., Tarassenko, L., Honoré, G., Ong, M. & Nadeem, S. 2004 The
engine data store for DAME. In Proc. of the All-Hands UK e-science Meeting, Nottingham,
September 2004.
Bar-Shalom, Y. & Li, X.-R. 1993 Estimation and tracking. Artech House.
Boser, B. E., Guyon, I. M. & Vapnik, V. N. 1992 A training algorithm for optimal margin
classifiers. In Proc. of the 5th Annual ACM Workshop on Computational Learning Theory,
Pittsburgh, PA, pp. 144–152.
Chen, J. & Patton, R. J. 1998 Robust model-based fault diagnosis for dynamic systems. Boston,
MA: Kluwer Academic.
Gelb, A. (ed.) 1974 Applied optimal estimation. Cambridge, MA: MIT Press.
Ghahramani, Z. & Hinton G. E. 1996 Parameter estimation for linear dynamical systems. Report
no. CRG-TR-96-2, University of Toronto, Canada.
Hayton, P., Schölkopf, B., Tarassenko, L. & Anuzis, P. 2001 Support vector novelty detection
applied to jet engine vibration spectra. In Advances in neural information processing systems 13
(ed. T. K. Leen, T. G. Dietterich & V. Tresp), pp. 946–952. Cambridge, MA: MIT Press.
Kadirkamanathan, V., Li, P., Jaward, M. H. & Fabri, S. G. 2002 Particle filtering-based fault
detection in non-linear stochastic systems. Int. J. Syst. Sci. 33, 259–265. (doi:10.1080/
00207720110102566)
King, S. P., King, D. M., Anuzis, K., Astley, K., Tarassenko, L., Hayton, P. & Utete. S. 2002 The
use of novelty detection techniques for monitoring high-integrity plant. In Proc. of IEEE Int.
Conf. on Control Appl. 1. Glasgow. September 2002, pp. 221–226.
Korbicz, J. & Janczak, A. 2002 Artificial neural network models for fault detection and isolation of
industrial processes. Comput. Assist. Mech. Eng. Sci. 9, 55–69.
Murphy, K. 1998–2002 Bayes network Kalman filtering toolbox. http://www.ai.mit.edu/
wmurphyk, now at http://www.cs.ubc.ca/wmurphyk.
Markou, M. & Singh, S. 2003 Novelty detection: a review—part 1: statistical approaches. Signal
Process. 83, 2481–2497. (doi:10.1016/j.sigpro.2003.07.018)
Nairac, A., Townsend, N., Carr, R., King, S., Cowley, P. & Tarassenko, L. 1999 A system for the
analysis of jet engine vibration data. Integr. Comput. Aided Eng. 24, 53–65.
Rätsch, G., Mika, S., Schölkopf, B. & Müller, K. R. 2002 Constructing boosting algorithms from
SVMs: an application to one-class classification. IEEE Tran. Pattern Anal. Mach. Intell. 24,
1184–1199. (doi:10.1109/TPAMI.2002.1033211)
Rolls-Royce plc 1986 The jet engine. 4th edn. Derby, UK: Rolls-Royce plc.
Roweis, S. & Ghahramani, Z. 1999 A unifying review of linear Gaussian models. Neural Comput.
11, 305–345. (doi:10.1162/089976699300016674)
Sammon, J. W. 1969 A nonlinear mapping for data structure analysis. IEEE Trans. Comput. 18,
401–409.
Schölkopf, B., Platt, J., Shawe-Taylor, J., Smola, A. J. & Williamson, R. C. 1999 Estimating the
support of a high-dimensional distribution. Report no. TR MSR 99–87, Microsoft Research,
Redmond, WA.
Schölkopf, B., Platt J. & Smola, A. J. 2000a Kernel method for percentile feature extraction.
Report no. TR MSR 2000–22, Microsoft Research, Redmond, WA.
Schölkopf, B., Williamson, R. C., Smola, A. J., Shawe-Taylor, J. & Platt, J. C. 2000b Support
vector method for novelty detection. In Advances in neural information processing systems 12
(ed. S. A. Solla, T. K. Leen & K.-R. Müller), pp. 582–588. Cambridge, MA: MIT Press.
Tipping, M. E. & Lowe, D. 1998 Shadow targets: a novel algorithm for topographic projections by
radial basis functions. Neurocomputing 19, 211–222. (doi:10.1016/S0925-2312(97)00066-0)
Vapnik, V. 1995 The nature of statistical learning theory. Berlin, Germany: Springer.
Venkatasubramanian, V., Rengaswamy, R., Yin, K. & Kawuri, S. N. 2003 A review of process fault
detection and diagnosis—part I: quantitative model-based methods. Comput. Chem. Eng. 27,
293–311. (doi:10.1016/S0098-1354(02)00160-6)

Phil. Trans. R. Soc. A (2007)