161 research outputs found
Investigating Engineering Data by Probabilistic Measures
No full textA critical issue for data-based engineering is a lack of descriptive labels for the measured data. For many engineering systems, these labels are costly/impractical to obtain, and as a result, conventional supervised learning is not feasible. This article outlines a probabilistic framework for the investigation and labelling of engineering datasets. Two alternative probabilistic measures are suggested to define the most informative observations to investigate and annotate, in order to maximise the classification performance of a statistical model
A machine learning approach to Structural Health Monitoring with a view towards wind turbines
No full textThe work of this thesis is centred around Structural Health Monitoring (SHM) and
is divided into three main parts.
The thesis starts by exploring di�erent architectures of auto-association. These are
evaluated in order to demonstrate the ability of nonlinear auto-association of neural
networks with one nonlinear hidden layer as it is of great interest in terms of reduced
computational complexity. It is shown that linear PCA lacks performance for novelty
detection. The novel key study which is revealed ampli�es that single hidden layer
auto-associators are not performing in a similar fashion to PCA.
The second part of this study concerns formulating pattern recognition algorithms for
SHM purposes which could be used in the wind energy sector as SHM regarding this
research �eld is still in an embryonic level compared to civil and aerospace engineering.
The purpose of this part is to investigate the e�ectiveness and performance of such
methods in structural damage detection. Experimental measurements such as high
frequency responses functions (FRFs) were extracted from a 9m WT blade throughout
a full-scale continuous fatigue test. A preliminary analysis of a model regression of
virtual SCADA data from an o�shore wind farm is also proposed using Gaussian
processes and neural network regression techniques.
The third part of this work introduces robust multivariate statistical methods into
SHM by inclusively revealing how the in
uence of environmental and operational
variation a�ects features that are sensitive to damage. The algorithms that are
described are the Minimum Covariance Determinant Estimator (MCD) and the Minimum Volume Enclosing Ellipsoid (MVEE). These robust outlier methods are
inclusive and in turn there is no need to pre-determine an undamaged condition
data set, o�ering an important advantage over other multivariate methodologies.
Two real life experimental applications to the Z24 bridge and to an aircraft wing
are analysed. Furthermore, with the usage of the robust measures, the data variable
correlation reveals linear or nonlinear connections
Active Learning Approaches to Structural Health Monitoring
No full textA critical issue for structural health monitoring (SHM) strategies based on pattern recognition models is a lack of diagnostic labels for system data. In an engineering context these labels are costly to obtain, and as a result, conventional supervised learning is not feasible. Active learning tools look to solve this issue by selecting a limited number of the most informative data to query for labels. This article demonstrates the relevance of active learning, using the algorithm proposed by Dasgupta and Hsu (the DH active learner). Results are provided for applications of this technique to engineering data from aircraft experiments
On a meta-learning population-based approach to damage prognosis
No full textThe current work studies the application of population -based structural health monitoring (PBSHM) to the problem of damage prognosis. Two methods are proposed for populationinformed damage prognosis and they are evaluated according to their performance using an experimental dataset. The first method is an attempt to define a functional subspace, which includes the potential behaviour of members of the population subjected to the phenomenon of damage evolution. The second approach is a meta -learning method, the deep kernel transfer (DKT) method, which seeks to exploit information from a population in order to enhance the predictive performance of a Gaussian process. The predictive capabilities of the two methods are tested in an experimental crack -growth problem. The results reveal that the two methods are properly informed by the population to make predictions about new structures and show potential in dealing with the problem of damage evolution, which is a problem of imbalanced and difficult -to -acquire data
Machine learning at the interface of structural health monitoring and non-destructive evaluation
No full textWhile both non-destructive evaluation (NDE) and structural health monitoring (SHM) share the objective of damage detection and identification in structures, they are distinct in many respects. This paper will discuss the differences and commonalities and consider ultrasonic/guided-wave inspection as a technology at the interface of the two methodologies. It will discuss how data-based/machine learning analysis provides a powerful approach to ultrasonic NDE/SHM in terms of the available algorithms, and more generally, how different techniques can accommodate the very substantial quantities of data that are provided by modern monitoring campaigns. Several machine learning methods will be illustrated using case studies of composite structure monitoring and will consider the challenges of high-dimensional feature data available from sensing technologies like autonomous robotic ultrasonic inspection
Autonomous ultrasonic inspection using Bayesian optimisation and robust outlier analysis
No full textThe use of robotics is beginning to play a key role in automating the data collection process in Non Destructive Testing (NDT). Increasing the use of automation quickly leads to the gathering of large quantities of data, which makes it inefficient, perhaps even infeasible, for a human to parse the information contained in them. This paper presents a solution to this problem by making the process of NDT data acquisition an autonomous one as opposed to an automatic one. In order to achieve this, the robotic data acquisition task is treated as an optimisation problem, where one seeks to find locations with the highest indication of damage. The resulting algorithm combines damage detection technology from the field of data-driven Structural Health Monitoring (SHM) with novel ideas in uncertainty quantification which enable the optimisation routine to be probabilistic. The algorithm is sequential; a decision is made at every iteration regarding the next optimal physical location for making an observation. This is achieved by modelling a two-dimensional field of novelty indices across a part/structure which is derived from a robust outlier analysis procedure. The value of this autonomous approach is that the output is not only measured data, but the most desirable information from an NDT inspection – the probability that a component contains damage. Furthermore, the algorithm also minimises the number of observations required, thus minimising the time and cost of data gathering
Tool wear inspection of polycrystalline cubic boron nitride inserts
No full textIn industry, highly frequent inspection of tooling used to machine safety critical components is common place. Worn or damaged tools produce undesirable surface finishes leading often to early failure of the part due to fatigue crack growth. In the development stages of polycrystalline boron nitride tools, the tool wear inspection technique is an off-line run-to-failure method. This approach interrupts the cutting process intermittently, to measure the tool wear using optical and scanning microscopy. This method is time consuming and expensive, causing bottlenecks in production. The overall aim in industry is to develop an on-line, automated system capable of informing the operator of the tool’s imminent failure. This paper focuses on treating this process as a preventative maintenance problem by studying whether acoustic emission can be used as an indirect measurement of tool wear at any given time. Acoustic emission measurements taken from the machining process of face turning are investigated here. Basic analysis in the frequency domain using principle component analysis reveals a number of interesting insights into the process. Relationships between the sharpness of the tool and the magnitude of the frequencies suggests promising link between acoustic emission and tool wear
Development of a Mathematical Model to Design the Control Strategy of a Full Scale Roller-Rig
No full textRoller rigs have been built world-wide to investigate the dynamics of railway vehicles and they have particularly
been applied to the development of high-speed trains.
Specifically, the present paper focuses on a roller-rig for testing full scale locomotives/coaches, which is located at the Osmannoro Centre of Experimental Dynamics (Italy). The considered roller rig allows to test vehicles having up to 6 axes and 3 bogies. Aim of the roller-rig is to verify traction systems of locomotives, to test anti-slip/anti-skid control systems, to identify braking performance and to perform electromagnetic tests (radiated emissions and immunity) on the locomotive.
At present, the design of the control strategy allowing to perform the desired tests is under development. To account for roller-rig and vehicle dynamics at a design stage, a numerical model of the complete test bench (including the locomotive/coach, the roller rig, the actuating devices and the corresponding control systems) was developed and interfaced with the control strategy of the test bench. The present paper describes the simulation tool developed on purpose and preliminary experimental results are presented
On Affine Symbolic Regression Trees for the Solution of Functional Problems
No full textSymbolic regression has emerged from the more general method of Genetic Programming (GP) as a means of solving functional problems in physics and engineering, where a functional problem is interpreted here as a search problem in a function space. A good example of a functional problem in structural dynamics would be to find an exact solution of a nonlinear equation of motion. Symbolic regression is usually implemented in terms of a tree representation of the functions of interest; however, this is known to produce search spaces of high dimension and complexity. The aim of this chapter is to introduce a new representation—the affine symbolic regression tree. The search space size for the new representation is derived, and the results are compared to those for a standard regression tree. The results are illustrated by the search for an exact solution to several benchmark problems
- …
