Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

In this master thesis project the applicability of using only smartphone standalone data to monitor the condition of rolling element bearings is investigated. Using a smartphone instead of expensive measurement equipment and data acquisition systems can result in a fast and cost-effective methodology to monitor the health status of rolling element bearings. The shaft speed extraction is based on video images acquired by the smartphone camera. This methodology exploits the deformation of the video image due to the rolling shutter effect of the smartphone camera. The bearing damage detection is based on the audio of the smartphone video. The audio is captured in stereo by a dual plug-in microphone. The acoustic data is processed by three different signal processing methodologies and subsequently the resulting envelope spectra are analysed to check for bearing fault signatures. The methodologies used are: Firstly, a demodulation of the raw acoustic data by applying a Hilbert transform. Secondly, a demodulation of the acoustic data after filtering the raw acoustic data with an ideal bandpass filter based on the characteristics of the Kurtogram. Thirdly, transforming the acoustic data to the bi-frequency domain by calculating the spectral coherence. The different methodologies are validated by performing experiments with a Mild Inner Race, Small Outer Race and Large Outer Race damaged bearing running at shaft speeds in the range from 5 to 40 revolutions per second. The acoustic data resulting from a reference microphone resulted in promising results in order to detect bearing damages when using the kurtogram-based bandpass filter or the spectral coherence. Unfortunately, the acoustic data resulting from the smartphone audio has a cut-off frequency at 14.5 kHz, due to data compression. This lead to the non-detection of small bearing damages, because the signal-to-noise ratio was too low in the available frequency bands.

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

The manufacturing accuracy of gears is directly related to their reliability, operational noise and vibrations. Ever increasing quality demands have forced gear manufacturers to adopt advanced manufacturing processes to assure that the required quality is met. Current theoretical models are not capable of predicting gear quality in realistic industrial settings. As a result, manufacturers often rely on costly geometric measurements on a limited amount of samples per batch. Recently, the fields of machine learning and data analysis have been rapidly expanding. Their application in manufacturing is a common topic of research, supported by the frameworks of Industry 4.0. The use of these technologies facilitates new approaches for the modelling and monitoring of complex manufacturing procedures. At present, only a limited amount of research has been published related to monitoring and modelling of gear grinding processes. This thesis explores novel methodologies for real-time prediction of gear quality during the gear grinding process. Firstly, the important aspects of gear engineering are discussed. The current state of research is explored by reviewing the available literature. Insights gathered from this literature study are then used to develop novel methodologies for the prediction of gear quality in gears finished by grinding. Based on a production experiment, vibrations at the tool spindle have been recorded with a triaxial accelerometer. The information captured during grinding forms the basis for quality prediction. In total, three distinct methodologies are developed for the prediction of gear quality. The performance of each methodology is verified on test data and discussed in detail. This research serves as a validation of machine learning based methodologies for gear quality prediction. It serves as a basis for further work towards a final industrial application.

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Gedwongen trillingen van turbineschoepen in vliegtuigmotoren en industriële gasturbines, blijven een van de meest voorkomende oorzaken van hoge-cyclus-vermoeiingsbreuk. Deze constructies moeten zorgvuldig worden ontworpen om weerstand te bieden tegen de dynamische spanningen veroorzaakt door resonantietrillingen, die kunnen optreden binnen een breed werkingsgebied en bij verschillende motor orden. Verder wordt het ontwerp bemoeilijkt door de afwezigheid van significante materiaaldemping en de moeilijke omstandigheden waarin de turbineschoepen moeten werken. Wrijvingsdemping die plaatsvind bij de mechanische verbindingen van turbineschoepen, is een van de meest efficiënte en praktische oplossingen om resonantietrillingen te verminderen. De verbindingen waar wrijving plaatsvind kunnen eigen aan het ontwerp van turbineschoepen zijn, zoals hun bevestiging aan de voet, of kunnen worden geïntroduceerd met het doel extra wrijvingsdemping te verwezenlijken, zoals underplatform dempers. Het nauwkeurig voorspellen van de invloed van wrijvingsdemping op de dynamica van turbineschoepen blijft echter een uitdaging vanwege de niet-lineaire aard van wrijving. Deze thesis onderzoekt het dynamisch gedrag van turbineschoepen met aandacht voor underplatform dempers. Verschillende modellen van de structuur met een toenemende graad van complexiteit en veralgemening worden geconstrueerd. Een uitgebreide numerieke methodiek op basis van de harmonische balans methode wordt gepresenteerd en geïmplementeerd in Matlab. Verschillende contact -en wrijvingsmodellen worden gebruikt en hun prestaties vergeleken. Invloed van belangrijke parameters voor wrijvingsdemping wordt onderzocht, met als doel de identificatie van de optimale ontwerpwaarden van de wrijvingsdemper.

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Increasing the safety and efficiency of turbomachinery is an ongoing quest for engineers. By reducing the clearance gap between rotor and stator, the efficiency is increased. However, small perturbances in running conditions can cause the blades of the rotor to hit the stator. This phenomenon, called rotor stator interaction, can cause severe damage to the machine. Abradable liners are commonly used to diminish the effects of the contact. Additionally, new ways of monitoring blade damage while in operation are of great importance. One of these methods is blade tip timing, where probes positioned in the casing measure blade tip deflection, leading to possible damage detection. Therefore, a test rig is devised for experiments on rotor stator interaction and testing of blade tip timing methods. To achieve both at the same time, a casing segment rather than a cylindrical casing is used. The casing segment is compliantly mounted on three support bars, allowing the operator to adapt the casing segment stiffness. For the blade tip timing, an aluminium plate with circular cut-outs allows for nearly continuous probe positioning. As such, the operator is able to test multiple probe configurations, mimicking real life turbine geometric constraints. Finally, the model is analysed using the Simcenter 3D rotor dynamics solver. A complex modal analysis shows no natural frequencies occur within the operating range (0-3000rpm) of the test rig. However, as the rotor has eight blades, the eighth engine order is of interest and is excited for each rotor mode. Since this is a relatively high harmonic, the amplitude is significantly smaller than that of the first harmonic, and, thus, no cause for concern. Basement excitation analysis proves that, for excitation frequencies below 30Hz (first stator natural frequency), the blade tip timing measurements would not be excessively polluted. However, a shock response analysis, with a force of 1.5kN and a period of 1ms applied at both the blade tip and the casing segment, resulted in extremely high stress in the blade dovetail. These stress concentrations are possibly due to gluing being used instead of contact parameters, which are not supported by the rotor dynamics solver.

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

This thesis presents a numerical and experimental validation of a novel output-only modal analysis technique, which is an extension of a recently introduced wavelet based algorithm, called IFESIS (Instantaneous Frequency Estimation via Subspace Invariance properties of wavelet Structures). In the proposed methodology, a set of specific wavelet structures - Complex Shifted Morlet Wavelets (CSMW), firstly acts upon a time-domain output signal. A transformed set of complex signals, called wavelet transforms, is obtained as a result. Subsequently, the system modal parameters (eigenfrequencies and damping ratios) are estimated by applying an appropriate subspace algorithm to the entire array of resulting wavelet transforms. In this way, the procedure combines the advantages of the CSMWs with the advantages of the subspace methods. The investigated algorithm is implemented for different time-invariant and time-varying scenarios, in order to evaluate and illustrate its performance. Firstly, linear time-invariant (LTI) and time-varying (LTV) dynamic systems are simulated and the obtained time-domain responses are afterwards analyzed using IFESIS. The results demonstrate that the algorithm is able to estimate very accurately the modal parameters of the simulated systems. The numerical study further suggests that with the proposed methodology one can identify the system's vibration characteristics not only based on the standard Dirac impulse as an external force, but also using repetitive aperiodic impacts or burst noise excitation. An experimental validation of the method is demonstrated with the help of relevant time-invariant and time-varying test set-ups. The modal parameters of the test structures, estimated with IFESIS, are compared with the corresponding results obtained from a dedicated experimental modal analysis software. The two sets of results are found to be in agreement with each other. In addition, the modal shapes of the static test structure are extracted and qualitatively compared with the relevant software estimates. The implementation of IFESIS requires appropriate selection of certain internal parameters of the algorithm. Their influence on the performance is also studied, in order to arrive at optimal implementation strategies.