No description available.

No description available.

Machine condition monitoring is gaining importance in industry due to the need to increase machine reliability and decrease the possible loss of production due to machine breakdown. Often the data available to build a condition monitoring system does not fully represent the system. It is also often common that the data becomes available in small batches over a period of time. Hence, it is important to build a system that is able to accommodate new data as it becomes available without compromising the performance of the previously learned data. In real-world applications, more than one condition monitoring technology is used to monitor the condition of a machine. This leads to large amounts of data, which require a highly skilled diagnostic specialist to analyze. In this thesis, artificial intelligence (AI) techniques are used to build a condition monitoring system that has incremental learning capabilities. Two incremental learning algorithms are implemented, the first method uses Fuzzy ARTMAP (FAM) algorithm and the second uses Learn++ algorithm. In addition, intelligent agents and multi-agent systems are used to build a condition monitoring system that is able to accommodate various analysis techniques. Experimentation was performed on two sets of condition monitoring data; the dissolved gas analysis (DGA) data obtained from high voltage bushings and the vibration data obtained from motor bearing. Results show that both Learn++ and FAM are able to accommodate new data without compromising the performance of classifiers on previously learned information. Results also show that intelligent agent and multi-agent system are able to achieve modularity and flexibility.

Flywheel energy storage units have become a viable alternative to electrochemical batteries in applications such as satellites, uninterrupted power supplies, and hybrid vehicles. However, this performance is contingent upon safe operation since these flywheels can release their stored energy almost instantaneously upon failure. The research presented here investigates a health monitoring technology that may give an early indication of degraded material properties in a concentric ring composite flywheel. The existence of degraded material properties is manifested as a change in mass eccentricity due to asymmetric growth of the outermost flywheel ring. A test rig concept to investigate the technology is developed in detail using a systems engineering design process. Successful detection of the change in mass eccentricity was verified analytically through dynamic modeling of the flywheel rotor and magnetic suspension system. During steady state operation detection was determined to be feasible via measurements of the magnetic bearing currents and shaft position provided by the magnetic suspension feedback sensors. A rotordynamic analysis was also conducted and predicts successful operation to the maximum operating speed of 50,000 Rpm.

Rolling-element bearings are widely used in various mechanical and electrical applications. Accordingly, a reliable bearing health condition monitoring system is very useful in industries to detect incipient defects in bearings, so as to prevent machinery performance degradation and malfunction. Although several techniques have been reported in the literature for bearing fault detection and diagnosis, it is still challenging to implement a bearing condition monitoring system for real-world industrial applications because of the complexity of bearing structures and noisy operating conditions. The objective of this thesis is to develop a novel intelligent system for more reliable bearing fault diagnostics. This system involves two sequential processes: feature extraction and decision-making. The proposed strategy is to develop advanced and robust techniques at each processing stage so as to improve the reliability of bearing condition monitoring. First, a novel wavelet spectrum analysis technique is proposed for the representative feature extraction. This technique applies the wavelet transform to demodulate the resonance signatures that are related to bearing health conditions. A weighted Shannon function is proposed to synthesize the wavelet coefficient functions to enhance feature characteristics. The viability of this technique is verified by experimental tests corresponding to various bearing health conditions. Secondly, an enhanced diagnostic scheme is developed for automatic decision-making. This scheme consists of modules of classification and prediction: a novel neuro-fuzzy classifier is developed to effectively integrate the strengths of the selected fault detection techniques (i.e., the resulting representative features) for a more accurate assessment of bearing health conditions; a novel multi-step predictor is proposed to forecast the future states of bearing conditions, which will be used to further enhance the diagnostic reliability. The investigation results have demonstrated that the developed intelligent diagnostic system outperforms other related bearing fault diagnostic schemes.

Rolling-element bearings are widely used in various mechanical and electrical applications. Accordingly, a reliable bearing health condition monitoring system is very useful in industries to detect incipient defects in bearings, so as to prevent machinery performance degradation and malfunction. Although several techniques have been reported in the literature for bearing fault detection and diagnosis, it is still challenging to implement a bearing condition monitoring system for real-world industrial applications because of the complexity of bearing structures and noisy operating conditions. The objective of this thesis is to develop a novel intelligent system for more reliable bearing fault diagnostics. This system involves two sequential processes: feature extraction and decision-making. The proposed strategy is to develop advanced and robust techniques at each processing stage so as to improve the reliability of bearing condition monitoring. First, a novel wavelet spectrum analysis technique is proposed for the representative feature extraction. This technique applies the wavelet transform to demodulate the resonance signatures that are related to bearing health conditions. A weighted Shannon function is proposed to synthesize the wavelet coefficient functions to enhance feature characteristics. The viability of this technique is verified by experimental tests corresponding to various bearing health conditions. Secondly, an enhanced diagnostic scheme is developed for automatic decision-making. This scheme consists of modules of classification and prediction: a novel neuro-fuzzy classifier is developed to effectively integrate the strengths of the selected fault detection techniques (i.e., the resulting representative features) for a more accurate assessment of bearing health conditions; a novel multi-step predictor is proposed to forecast the future states of bearing conditions, which will be used to further enhance the diagnostic reliability. The investigation results have demonstrated that the developed intelligent diagnostic system outperforms other related bearing fault diagnostic schemes.

Engineering industries usually require nondestructive evaluation (NDE) methods to ensure quality control, safety, and optimized use of resources. Among potential NDE techniques, ultrasonic wave methods are widely used because of their versatility and affordability. For applications to layered structures, ultrasonic guided waves are naturally excited and detected, so these guided waves are the preferred choice when compared to conventional bulk waves. The main advantage of guided waves over bulk waves for layered structures is that these guided waves can propagate a much farther distance, and thus they enable long range inspection. It is important to note that guided waves are multi-mode, so a preferred mode can be selectively used, although it is sometimes more efficient to use multiple wave modes. The characteristics of guided waves, namely dispersive propagation and attenuation, are directly related to the properties of the system in which they are propagating, so the measurement of these wave characteristics can be used for material characterization and condition monitoring. Despite a number of successful techniques to experimentally measure propagation characteristics of guided waves, there is a lack of a standard procedure to obtain attenuation characteristics. This research develops such a quantitative and systematic procedure to extract attenuation characteristics from real guided wave time-domain signals. This research considers multiple wave-modes, and focuses on broadband attenuation measurements with laser ultrasonic techniques. The analytical model of guided waves with attenuation is studied in general cases, and a numerical simulation is developed to model the point source/receiver laser measurement system. The attenuation extraction technique is developed using synthetic signals generated by the simulation. Finally, this research demonstrates the use of experimentally-measured attenuation data for material characterization and condition monitoring by developing an inversion scheme to back-calculate material properties for a number of practical cases.

Flywheel energy storage units have become a viable alternative to electrochemical batteries in applications such as satellites, uninterrupted power supplies, and hybrid vehicles. However, this performance is contingent upon safe operation since these flywheels can release their stored energy almost instantaneously upon failure. The research presented here investigates a health monitoring technology that may give an early indication of degraded material properties in a concentric ring composite flywheel. The existence of degraded material properties is manifested as a change in mass eccentricity due to asymmetric growth of the outermost flywheel ring. A test rig concept to investigate the technology is developed in detail using a systems engineering design process. Successful detection of the change in mass eccentricity was verified analytically through dynamic modeling of the flywheel rotor and magnetic suspension system. During steady state operation detection was determined to be feasible via measurements of the magnetic bearing currents and shaft position provided by the magnetic suspension feedback sensors. A rotordynamic analysis was also conducted and predicts successful operation to the maximum operating speed of 50,000 Rpm.

The modern industry today is highly dependent on electric motors of differenttypes and sizes. Synchronous motors are used in applications where a fixedspeed is desired. These machines are often found in high power applicationswhere they are preferred over induction motors due to their higher efficiency.Synchronous motors represent large investments and typically drive processeswhere downtime results in significant capital losses. Thus, detecting faults atan early stage can help avoid catastrophic failures and be useful in thescheduling of maintenance. In order to detect faulty conditions before theyterminate in a failure, machine operators must perform some kind ofmonitoring on the machines. Typically, the more critical the machine is for aprocess, the more effort is put on monitoring it. Before building a monitoringsystem for a machine, one must first decide what parameters that should bemonitored. The obvious desire is to find a parameter that is easy and cheap tomeasure and at the same time can give detailed information about the workingstate of the machine.The aim of this thesis is to evaluate whether the exciter stator current is anadequate parameter to use within a monitoring system for synchronous motors.The evaluation has been made through simulations of two different setups;One using a synchronous motor in the 20 MW range fed by a synchronousmachine type exciter. And the other using the same motor but instead fed byan induction machine type exciter. It has been found that the exciter statorcurrent can be used for detection of faults associated to the rectifier and statorshort circuit of the main machine stator winding. It has not been possible todetect turn-to-turn faults in the main machine rotor.The work has been performed at ABB Corporate Research in Västerås fromJune until December 2013.

No description available.