Recently, the use of smell in clinical diagnosis has been rediscovered due to major advances in odour sensing technology and artificial intelligence. It was well known in the past that a number of infectious or metabo...
详细信息
ISBN:
(纸本)0780376013
Recently, the use of smell in clinical diagnosis has been rediscovered due to major advances in odour sensing technology and artificial intelligence. It was well known in the past that a number of infectious or metabolic diseases could liberate specific odours characteristic of the disease stage and among others, urine volatile compounds have been identified as possible diagnostic markers. A newly developed electronic nose based on chemoresistive sensors has been employed to identify in vitro 13 bacterial clinical isolates, collected from patients diagnosed with urinary tract infections, gastrointestinal and respiratory infections, and in vivo urine samples from patients with suspected uncomplicated UTI who were scheduled for microbiological analysis in a UK Health Laboratory environment. An intelligent model consisting of an odour generation mechanism, rapid volatile delivery and recovery system, and a classifier system based on a neuralnetworks, genetic algorithms, and multivariate techniques such as principal components analysis and discriminant function analysis-cross validation. The experimental results confirm the validity of the presented methods.
Recent developments in the instrumentation of plants has led to multivariate statistical process control (MSPC) techniques becoming increasingly popular for process monitoring in the chemical industry over the last fe...
Recent developments in the instrumentation of plants has led to multivariate statistical process control (MSPC) techniques becoming increasingly popular for process monitoring in the chemical industry over the last few years. This paper examines one such algorithm, the partial least squares (PLS), and shows how the basic principles of this linear technique can be extended into the nonlinear domain via the application of radial basis function (RBF) neuralnetworks. Results showing the successful application of these methods to fault detection in a validated model of an industrial overheads condenser and reflux drum plant are also given.
Artificial neuralnetworks offer an alternative strategy for the nonlinear control of unmanned underwater vehicles (UUVs). This paper investigates the use of a multi-layered perceptron (MLP) network in controlling an ...
详细信息
Artificial neuralnetworks offer an alternative strategy for the nonlinear control of unmanned underwater vehicles (UUVs). This paper investigates the use of a multi-layered perceptron (MLP) network in controlling an UUV over a sea-bed profile and compares the use of applying chemotaxis learning over that of the more commonly employed backpropagation algorithm. The results show for differing sized MLPs the chemotaxis algorithm produces a successful controller over the sea bed profile in an improved training time. To further vindicate the chemotaxis network, it was then presented with the problem of meeting a new profile to travel over. The results show from several simulation runs that the chemotaxis network provides a robust controller over numerous sea bed profiles of which it had no prior knowledge.< >
Current implementations of aircraft flight controlsystems are an amalgam of numerous, discrete algorithms and various switching mechanisms. The switching mechanisms bring the algorithms into play either independently...
详细信息
Current implementations of aircraft flight controlsystems are an amalgam of numerous, discrete algorithms and various switching mechanisms. The switching mechanisms bring the algorithms into play either independently or in some form of combination. Advanced computing techniques, such as expert systems, can play a role by separating out the algorithm scheduling and making it more visible. By lessening the complexity both the integrity and capability should be improved. An alternative and more controversial approach is to use adaptive, learning and optimisation techniques such as neuralnetworks. These offer advantages by removing the need for algorithm creation through learning, provide graceful degradation when elements fail and can adapt to subsystem failures. However at present this approach is not welcomed as conventional verification and validation techniques cannot be employed. The author discusses the role of non-algorithmic software.< >
neuralnetworks have been employed in a range of automotive modelling and control tasks. This paper discusses the recently introduced local model network structure. The structure can be viewed as a hybrid model which ...
详细信息
neuralnetworks have been employed in a range of automotive modelling and control tasks. This paper discusses the recently introduced local model network structure. The structure can be viewed as a hybrid model which smoothly combines local models (of linear or nonlinear form) by interpolation. Interpolation of local models is implemented in the form of a basis-function network. The benefits of this structure include: existing knowledge can be directly integrated, and the model can be fully analysed and interpreted in a local manner. The author describes the local model network and discusses its relevance for automotive modelling and control tasks.< >
The article describes the concept and performance of neuro fuzzy modelling and neuro fuzzy model based control. The technique has been applied to a highly nonlinear neutralisation process. It is shown that the neuro f...
详细信息
The article describes the concept and performance of neuro fuzzy modelling and neuro fuzzy model based control. The technique has been applied to a highly nonlinear neutralisation process. It is shown that the neuro fuzzy model is easy to interpret and provides accurate predictions. The proposed neuro fuzzy model based control strategy utilises local incremental models which can naturally eliminate steady state offsets. It is shown that the neuro fuzzy model based controller gives much improved performance compared to that of a linear model based controller.
The process under investigation in this work is the integrated dry route (IDR) process of British Nuclear Fuels plc. (BNFL), which is a nuclear fuel processing plant, where non-catastrophic faults are known to occur a...
详细信息
The process under investigation in this work is the integrated dry route (IDR) process of British Nuclear Fuels plc. (BNFL), which is a nuclear fuel processing plant, where non-catastrophic faults are known to occur and a reliable early fault diagnosis scheme was required for operator advice. This paper describes the application of artificial neural network techniques to the diagnosis of non-catastrophic faults in the IDR process which operates at a few different operating points. The techniques involved developing methods to preprocess the data by statistical scaling, reducing the neural network input space using principal component analysis and training and testing the neuralnetworks. Results are presented to illustrate the performance of the developed scheme on application to the IDR process data.
Describes the results of a simulation study of two fault detection approaches applied to a hydraulic test rig: an eigenstructure assignment approach; and a neural network based approach. The rig has a predominant natu...
详细信息
Describes the results of a simulation study of two fault detection approaches applied to a hydraulic test rig: an eigenstructure assignment approach; and a neural network based approach. The rig has a predominant natural frequency of the order of 20 Hz and is more representative of real industrial systems due to the presence of operating point nonlinearity, measurement noise and load disturbances.< >
Learning methods are currently of great interest to (sections) of the controlsystems community. There is an ever increasing volume literature on neuralnetworks. At best, this work has shown that, appropriately chose...
详细信息
Learning methods are currently of great interest to (sections) of the controlsystems community. There is an ever increasing volume literature on neuralnetworks. At best, this work has shown that, appropriately chosen intelligent control type schemes, or architectures are capable of synthesising control laws for partially known systems. A large class of these approaches employ iterative techniques to adjust the variable parameters of the architecture and hence develop discrete time models of the inverse dynamics of the process to be controlled. The resulting trained network is then used as the basis of a nonlinear control law. It is important to note, however, that this general approach has several major drawbacks. These drawbacks are some of the fundamental reasons why the adaptive control community moved away from gradient based methods in favour of a formal stability theory. Currently, a number of research groups, including The Advanced systems Research Group at Southampton, are investigating the feasibility of analogue network designs for the control of continuous time nonlinear dynamic systems. One idea which has already received considerable attention is to combine an adaptation law with a sliding mode, or variable structure controller to give a globally stable closed-loop system with good tracking properties. The presentation reviews progress to-date with particular emphasis on stability theory. Some ongoing work and areas for short to medium term further development are also briefly outlined.< >
Supervised learning of classifications under l/sub 2/ costing of error trains learning classifiers to recall Bayesian a posteriori probabilities of the possible classes, given observed measurements. This result leads ...
Supervised learning of classifications under l/sub 2/ costing of error trains learning classifiers to recall Bayesian a posteriori probabilities of the possible classes, given observed measurements. This result leads to a number of insights concerning the validation of training, access to the likelihood function, creating networks of networks, incorporation of prior probabilities (which may vary in real-time), and how to choose the training set. The author focuses on the latter two points. Contextual information in the form of priors is used to generalise training data to economise on both training and computation. Structural generalization is the process whereby data is generalised architecturally rather than parametrically. A training procedure and postprocessing technique are given which enable learning under one set of prior classification probabilities to be generalized to give (asymptotically) Bayes optimal classifications under all others.< >
暂无评论