No description available.

Arti cial neural networks (ANNs) are a powerful processing units inspired by the human brain. They can be used in many applications due to their pattern classi cation abilities, ability to model complex nonlinear input-output mappings, and their ability to adapt and learn. The relatively new Smooth Variable Structure Filter (SVSF) has recently been applied to the training of feedforward multilayered neural networks. It has shown to have good accuracy and a fast speed of convergence. In this thesis, an engine fault detection system using an ANN will be implemented. ANNs are used in engine fault detection due to the high-noise environment that engine operate in. Additionally the fault detection system must work while the engine is mounted in a vehicle, which provide additional sources of noise. The SVSF training method is evaluated and compared to other traditional training methods. Also di erent accelerometer types are compared to evaluate whether lower cost accelerometers can be used to keep the system cost down. The system is tested by inducing a missing spark fault, a fault that has a complex fault signature and is di cult to detect, especially in an engine with a high number of cylinders. / Thesis / Master of Applied Science (MASc)

Condition monitoring systems are critical for autonomous detection of damage when operating remote wind turbines. These systems continually monitor the turbine’s operating parameters and detect damage before the turbine fails. Although common in utility-scale turbines, these systems are mostly undeveloped in distributed, small-scale turbines due to their high cost and need for specialized equipment. The Cal Poly Wind Power Research Center is developing a low-cost, modular solution known as the LifeLine system. The previous version contained monitoring equipment, but lacked decision-making capabilities. The present work builds on the LifeLine by developing software-based detection of blade damage. Detection is done by monitoring of tower vibrations, rotor speed, and generator power output. First, testing is completed to inform algorithm design: the tower vibrational response is recorded, and blade damage is simulated by adding a mass imbalance to one blade. From these results, several algorithms are developed, and their performance is analyzed in a cross-validation study. The time-series method known as the Nonlinear State Estimation Technique and Sequential Probability Ratio Test (NSET+SPRT) is implemented first. This algorithm is highly successful, with a 93.3% rate of correct damage detection; however, it occasionally raises false alarms during normal operation. A custom-built algorithm known as the Adaptive Fast Fourier Transform (AFFT) is also built; its strength lies in its elimination of false alarms. The final system utilizes a joint monitoring approach, combining the benefits of the NSET+SPRT and AFFT. The final algorithm is successful, correctly categorizing 95.5% of data when operating above 120RPM, and raising no false alarms in normal operation. This version is then implemented for live monitoring on the Cal Poly Wind Turbine, allowing for robust and autonomous detection of blade damage.

No / In recent years, several accidents in pioneer gas processing industries led industries to put emphasis on real-time fault detection. Neural Network (NN) based fault (abnormal situation) detection technique played an important role in monitoring industrial safety. In this work, an attempt has been made to study the fault detection of Brahmanbaria gas processing plant using multi layered feed forward NN based system. NN based fault detection system is trained, validated and tested using data generated using the dynamic model. Preliminary results show that NN based method is able to detect the faults of Brahmanbaria Gas processing plant for fewer no of faults.

Quality assurance of electrical components of cars such as alternators and starters is an important consideration due to both commercial and safety reasons. The focus of this research is to develop a complete Fault Detection and Diagnosis (FDD) solution for alternators and starters for their implementation in test cells. The FDD would enable more reliable testing of production line parts without compromising high production throughput. Our proposed solution includes three elements: (1) background noise elimination; (2) fault detection and analysis; and (3) fault classi cation for fault type identi cation. Noise gating, Extended Multi-Scale Principle Component Analysis (EMSPCA), and Logistic Discriminant classi er were used to perform these three elements. The FDD strategy detects and extracts fault signatures from multiple sensors (which are vibration and sound measurements in this research). Included in this strategy is ltering of the background noise in sound measurements to enable operation and maintain FDD performance in noisy conditions. The EMSPCA is the core of the FDD strategy. EMSPCA breaks the fault into time-frequency scales using wavelets and applies Principle Component Analysis (PCA) on each scale. This reveals the signature of the fault. The fault signature is then examined by a classi er to match it with the correct type of faults. The total FDD strategy is automated and no operator intervention is required. The advantages of the proposed FDD strategy are: (1) high accuracy in detection and diagnosis; (2) robustness in noisy industrial conditions; and (3) no need for operators' intervention. These advantages make the proposed FDD strategy a promising candidate for mass industrial applications. / Thesis / Master of Applied Science (MASc)

As wastewater treatment plants transition to water resource recovery facilities, the need for improved control and consequently supervision increases. Despite the large volume of research that has been performed on this topic, the use in industry is scarce. Practical implementation is challenging due to the nature of the process, and a lack of standardisation in the research results in uncertainty as to the state of the art. This is one of the main challenges identified.  Experimental work is performed using the Benchmark Simulation Model No. 1 to identify monitoring requirements and evaluate the performance of univariate fault detection methods. For the former, residual based process fault signatures are used to determine minimal sensor requirements based on detectability and isolability goals. Sensor faults are the focus of the latter issue, using the Shewhart, cumulative sum, and exponentially weighted moving average control charts to detect bias and drift faults in a controlled variable sensor.  The use of a standard model and known fault detection methods is useful to establish a baseline for future work. Given the lack of standardised use in industry this is considered critical. Both proposed methods emphasise ease of visualisation which is beneficial for industrial implementation.

No description available.

No description available.

No description available.

No description available.