Histechnica heeft dit boek samengesteld ter gelegenheid van het 175-jarig bestaan van de opleiding van de Technische Universiteit Delft (1842 – 2017). Histechnica wil met dit lustrumboek uw aandacht en interesse wekken voor de TUD erfgoedcollectie. De hier gepresenteerde selectie van erfgoedobjecten heeft een belangrijke rol gespeeld bij het onderwijs en/of het onderzoek aan de TUD en heeft een functie gehad die dienstbaar was aan het maatschappelijk belang. En elk van de hier voor u liggende artikelen draagt duidelijk een persoonlijke signatuur – hetgeen naar wij hopen de interesse van de lezer nog meer ten goede zal komen.@en

Research into multi-axial fatigue requires knowledge on the through time and thickness development of the propagating crack front. In order to provide this knowledge methods must be developed to monitor crack initiation and growth during a fatigue test. The acoustic emission method has been recognised as a possible means to this end.
Fatigue is a failure mode that is characterised by the accumulation of damage during cyclic loading. Fatigue initiates on the microscale as dislocation movement, and subsequently develops into crack growth and eventual failure. In the assessment of fatigue damage it is critical to estimate the remaining lifetime. Prediction models have been developed for uni-axial loaded joints. However in the field of multi-axial fatigue these models are inadequate. To improve upon these models knowledge is required on the development of the fracture during cyclic multi-axial loading.
The acoustic emission method is a procedure for non-destructive testing that is based on detecting elastic stress waves originating from changes in the microstructure of an object. In the industry two common applications of acoustic emission exist. One of these aims to extract insights into the severity of damage from waveform features, the other aims to localise and correlate acoustic events. Novel applications focus mainly on the improvement of these two fields. A state of the art method is proposed which aims to localise acoustic sources in small scale volumes as a means of imaging the crack front.
Localisation of the sources is performed using a time difference of arrive scheme, which is also known as multilateration. The accuracy of this scheme is highly dependent on the accuracy of the input parameters. Therefore the times of arrival, receiver positions and speed of sound must be established meticulously.
Extensive processing is required to transform the recorded acoustic emission data into a representation of the crack propagation. For this purpose a post-processing system has been developed which automates time of arrival picking, event building and event localisation. The developed system is an extension to the AMSY-6 acoustic emission measurement system developed by Vallen Systeme.
In order to validate the accuracy of the extended measurement system an experiment has been designed. In a volumetric geometry sources are simulated in order to assess the bias and the variance, additionally noise is added to the simulation to evaluate the sensitivity of the procedures.
The study has shown that, under ideal noise conditions, the precision can amount to a cluster of about a millimetre in size. Additionally the bias of this cluster can be reduced to a minimum by means of careful calibration of the receiver positions and the speed of sound. Both of these observations indicate that an accurate representation of the propagating crack front may be obtained through a regression model, if the conditions are known and within acceptable limits. Regarding non-ideal conditions the analysis of the noise sensitivity has shown that the accuracy is relatively stable as long as the signal to noise ratio is kept above SNR ≥ 30 dB.
In conclusion this has shown that monitoring the propagation of the crack front using volumetric acoustic emission source localisation may be possible if the conditions are right. This means that the signal to noise ratio must be kept above SNR ≥ 30 dB. If this is the case, and if enough events are recorded to perform regression both the bias and the variance should pose a problem to monitoring crack growth.

During fatigue crack growth of metallic structures, acoustic stress waves are introduced into the material. These acoustic emission waves are the result of the sudden stress release during the nucleation of the crack surface. Using dedicated acoustic emission monitoring equipment, these waves can be recorded. When multiple sensors are able to pick up the waves of a certain fatigue crack growth step, the location of this crack can be determined by using the differences in time of arrival of the signals at these sensor locations. This localization is performed by the method of multilateration, which is the same method that is used to calculate the epicentre of earthquakes.

TU Delft and the 4D-Fatigue joint industry project will use acoustic emission to predict fatigue crack depth, in order to determine the crack growth rate in a multi-axial fatigue loaded test specimen. The location of the fatigue crack will be known, so sensors can be applied very close to it. This use of acoustic emission is unconventional due to the fact that most appliances focus on a much larger scale, since this is one of the advantages of this method. The accuracy goal which is set during this project is also far exceeding traditional use, aiming at a crack depth prediction with an error less than 1mm. Little literature on acoustic emission focuses on this close range, inducing the need for additional research.

The goal of this thesis is to perform finite element simulations, in order to determine the behaviour of acoustic emission waves in close proximity of the crack. This knowledge should be used to determine what accuracy values are achievable, and whether there are ways to improve on this accuracy.

The first step in order to achieve this goal was the simulation of a Hsu-Nielsen source, more commonly referred to as a pencil lead break. This is a very simple to perform acoustic emission source, and will serve to establish appropriate finite element simulation parameters. This simple first step is also convenient to establish the effect the acoustic emission monitoring equipment has on the obtained signals. The multiple filter steps which are performed by the hardware, and the frequency response of the sensor have a great influence on the signal. Pencil lead breaks were performed on a steel plate of 5mm thickness, at several distances of a sensor. Signals produced by the simulations were verified by the experimental signals in a range of 5mm-70mm, verifying the used simulation methods.

The next step was to simulate actual fatigue growth in the same steel plate of 5mm thickness. Experimental signals were attempted to be acquired during fatigue tests of the plate. The fatigue machine that was used however, turned out to produce high amplitude noise, making it not possible to obtain the desired signals. Since verification of the simulated signals would not be possible, several source models were investigated in order to simulate fatigue crack growth acoustic emission, comparing the localization results with each other. Localization using conventional methods were able to achieve an accuracy of 0.6mm. Using a new method of wave speed determination however, which makes use of the simulation results, the localization accuracy was improved to 0.4mm.


The final step was to perform a simulation of a specimen which is used by the 4D-Fatigue JIP. This tubular specimen contains flanges, which reflect inbound waves back to the sensors, potentially influencing the localization accuracy. The simulations however, concluded that these reflections arrive after the initial wave has passed. The effect on localization was found to be negligible. Localization using conventional methods was not found result in reliable data. This decrease in accuracy compared to the accuracy found in the 5mm plate, is attributed to the larger differences in travel distance the signals have, due to the 10mm thick body of the tubular specimen. Localization using the new speed determination method, resulted in a localization accuracy of approximately 1mm.

Although localization accuracy was found to be sufficient to satisfy the initial goal of 1mm, it should be noted that the localization was performed using an estimated noise amplitude value. Also, during these numerical simulations the sensor is exactly known, which is not the case during experiments, so additional location error should be taken into account. The work in this thesis does however describe multiple phenomena which influence the localization accuracy, and forms basis for further research to improve on it.