N. Barrie Jones

Improving the performance of radial basis function classifiers in condition monitoring and fault diagnosis applications where ‘unknown’ faults may occur

This paper presents a novel technique which may be used to determine an appropriate threshold for interpreting the outputs of a trained radial basis function (RBF) classifier. Results from two experiments demonstrate that this method can be used to improve the performance of RBF classifiers in practical applications.

Fault diagnostics using sliding mode techniques

Abstract This paper describes the application of sliding mode observation techniques to the problem of fault diagnostics. A specific diesel engine coolant system is considered. A non-linear sliding mode observer is used to monitor the system states. Pertinent system parameters are also monitored using the concept of the equivalent injection signal required to maintain a sliding mode. The parameter estimates from the sliding mode scheme are compared with those generated by a non-linear simulation model and are found to provide good correlation. Results from a laboratory-based power generator set (Genset) are presented to demonstrate the effectiveness of the approach.

Analytical estimates of the characteristics of surface plasmon resonance fibre-optic sensors

Abstract In surface plasmon resonance (SPR) fibre-optic sensors, the optical power transmittance is important in determining behaviour. In this paper numerical calculations are used to estimate the optical power transmittance of SPR fibre-optic sensors. These analytical estimations can be applied in SPR fibre-optic sensors both with a transducing layer and without a transducing layer. The results of these numerical calculations agree well with previously published experimental results. This work will aid in the design of SPR fibre-optic sensors in terms of geometrical structure, materials and dynamic range as well as allowing the prediction of performance and of limitations of the design.

Higher-order sliding mode control of a diesel generator set:

Abstract Diesel engines are used as prime movers and independent power sources in many applications because of their advantages in terms of fuel efficiency, robustness and reliability. This paper presents the results of isochronous speed controller design and experimental analysis of robustness and efficiency of a turbocharged diesel generator set (genset). The steady state performance and transient response of the genset speed utilizing second-order sliding mode techniques with a super twisting sliding mode control algorithm are assessed. The algorithm does not require the time derivative of the sliding variable. It only uses the measured genset speed and does not require the use of an observer. A modification made to the algorithm shows improvement of the genset performance over a wide operating envelope. The influence of this methodology and modification of the algorithm on overall generator performance, in particular in the presence of large load changes and in terms of fuel efficiency, exhaust emissions, starting speed transient response and steady speed variation, are assessed. An algorithm for tuning the higher-order sliding mode (HOSM) controller for the genset is established and presented. The robustness of the controller is investigated and the performance is compared with that obtained by a commercial genset controller and a classical proportional-integral controller.

Improving the performance of CMFD applications using multiple classifiers and a fusion framework

In this paper, we demonstrate that the performance of a condition monitoring and fault diagnosis (CMFD) system may be improved by combining the outputs from three ‘primary’ classifiers using a novel, hybrid, data-fusion approach. The resulting classifier system involves four key processing stages. In Stage One, three diverse primary classifiers are employed. In Stage Two, the outputs from the primary classifiers are combined using majority voting. In Stage Three, any unclassified patterns from Stage Two are reassessed using Dempster-Shafer theory. Finally, in Stage Four, a simple rule base (‘expert system’) is used to integrate the results from the earlier stages. We demonstrate the effectiveness of this framework on a data set intended to allow the detection of static thermostatic valve faults in a diesel engine cooling system. Overall performance of the classifier system is shown to improve from approximately 89% (using the best of the primary classifiers) to approximately 99% (using the framework). In addition, the misclassification level of the original primary classifiers is shown to be approximately 10%, while the equivalent rate for the fusion classifier is approximately 1%. We go on to describe the application of the same classifier framework to a very different problem domain (medical diagnostics) and obtain a similar improvement in system performance. Here, overall performance of the classifier system is again shown to improve from approximately 87% (using the best of the primary classifiers) to approximately 99% (using the framework). In this case, the misclassification level of the original primary classifiers is approximately 12%, while the equivalent rate for the fusion classifier is approximately 1%. The results suggest that this framework may be appropriate for use in a range of application areas.

Applying MLP and RBF classifiers in embedded condition monitoring and fault diagnosis systems

In this paper, results are presented from a comprehensive series of studies aimed at assessing the suitability of multilayered perceptron (MLP) and radial basis function (RBF) networks for use in embedded, microcontroller-based, condition monitoring and fault diagnosis (CMFD) applications. Our assessment criteria include the performance of each classifier on a range of CMFD-related problems, such as situations where there may be multiple faults present simultaneously, or where ‘unknown’ faults may occur. In addition, the processor and memory requirements of each classifier are compared and discussed. On the basis of the results obtained in these studies, it is argued that each form of classifier has both strengths and weaknesses, and that neither is suitable for use in all CMFD applications. The paper concludes by demonstrating that, where memory and processor limits allow, the best performance may be obtained through use of a fusion classifier containing both MLP and RBF components.

Error analysis of two-wavelength absorption-based fibre-optic sensors

An error analysis of two-wavelength absorption-based fibre-optic sensors is presented in this paper. A general formulation to express the measurement error and the error induced by the instability of the light power input into the absorbing sample to be measured is derived. Using this formulation, two common and important errors in this type of sensor, temperature fluctuation of light sources and bending optical fibres, are numerically discussed as illustrative examples. The analyses can be applied to both transmission-type and reflection-type absorption-based fibre-optic sensors.

Theoretical analysis of the evanescent wave absorption coefficient for multimode fibre-optic evanescent wave absorption sensors

In an optical fibre of circular cross section, leaky skew rays transmit power in a different way from trapped and meridional rays. In this paper, a theoretical analysis of the evanescent wave absorption coefficient for leaky skew rays in multimode optical fibre evanescent wave absorption sensors is presented. Further, a comprehensive expression for the effective evanescent wave absorption coefficient is obtained. Some numerical results are given to illustrate this theoretical analysis. This work could be applied to optimize the design of fibre-optic evanescent wave absorption sensors.