Middle size gas/diesel aero-derivative power generation engines are widely used on various industrial plants in the oil and gas industry. Bleed of Valve (BOV) system failure is one of the failure mechanisms of these engines. The BOV is part of the critical anti-surge system and this kind of failure is almost impossible to identify while the engine is in operation. If the engine operates with BOV system impaired, this leads to the high maintenance cost during overhaul, increased emission rate, fuel consumption and loss in the efficiency. This paper proposes the use of readily available sensor data in a Supervisory Control and Data Acquisition (SCADA) system in combination with a machine learning algorithm for early identification of BOV system failure. Different machine learning algorithms and dimensionality reduction techniques are evaluated on real world engine data. The experimental results show that Bleed of Valve systems failures could be effectively predicted from readily available sensor data.

Controlling drum level is a major and crucial control objective in thermal power plant steam boilers. The drum level as a controlled variable is highly characterized by complex non-linear process dynamics as well as measurement noise and long-time delays. Developing a data-driven process model is particularly advantageous as it could be built from ongoing operational data. Such a model could be used to assist existing controllers by providing predictions regarding the drum level. The aim of this paper is to develop such a model and to propose a control architecture that can be easily integrated into existing control hardware. For that purpose, different neural networks are used, Multilayer Perceptron (MLP), Nonlinear Autoregressive Exogenous (NARX), and Long Short Term (LSTM) neural networks. LSTM and MLP were able to capture the dynamics of the process, but LSTM showed superior performance. The results demonstrate that the use of traditional machine learning criteria to evaluate a process model is not necessarily adequate. Using the model in an open-loop and a closed-loop simulation is more suitable to test its ability to capture the dynamics of the process. A novel architecture that integrates the process model within an existing closed-loop controller is proposed. The architecture uses adaptive weights to ensure that a good model is given more influence than a bad model on the controller’s output.

This research aims to test the feasibility of Programmable Logic Controller implementation of an Artificial Neural Network based bearing fault diagnosis using vibration datasets. The main drawback of using a Programmable Logic Controller along with an Artificial Neural Network is that it does not support the parallel nature of neural networks. This drawback is not significant for relatively small applications like bearing diagnosis that involve very short execution time. In this paper, a three layer multilayer perceptron backpropagation neural network is trained using Levenberg-Marquardt training algorithm with vibration dataset consisting of four bearing status classes: normal, outer race way fault, inner race way fault and rolling element (ball) fault. Time-frequency domain and time domain input features were considered in this research. Both approaches have performed well during simulation phase. But the time-frequency feature extraction approach was observed to take too long scan cycle time to be implemented in real-time. This is due to the computationally intensive nature of Fast Fourier Transform algorithm involved during feature extraction. The time domain approach is proved to be feasible for Programmable Logic Controller implementation. The time domain input features used for neural network training were root mean square, variance, kurtosis and negative log likelihood values. The average performance obtained during simulation with 10-fold cross validation performance estimator was an error of 7.9 x10-4. The performance tests of Programmable Logic Controller implementation resulted in 100% bearing fault detection rate.