Neural network approach to sensitivity and uncertainty analysis

Computer simulation of the dynamic evolution of complex systems has become a fundamental tool for many modern engineering activities. In particular, risk-informed design projects and safety analyses require that the system behavior be analyzed under several diverse conditions in the presence of substantial model and parameter uncertainty which must be accounted for. In this paper we investigate the capabilities of artificial neural networks of providing both a first-order sensitivity measure of the importance of the various parameters of a model and a fast, efficient tool for dynamic simulation, to be used in uncertainty analyses. The dynamic simulation of a steam generator is considered as a test-bed to show the potentialities of these tools and to point out the difficulties and crucial issues which typically arise when attempting to establish an efficient neural network structure for sensitivity and uncertainty analyses.     

Prediction Intervals for Neural Networks via Nonlinear Regression

Standard methods for computing prediction intervals in nonlinear regression can be effectively applied to neural networks when the number of training points is large. Simulations show, however, that these methods can generate unreliable prediction intervals on smaller datasets when the network is trained to convergence. Stopping the training algorithm prior to convergence, to avoid overfitting, reduces the effective number of parameters but can lead to prediction intervals that are too wide. We present an alternative approach to estimating prediction intervals using weight decay to fit the network and show via a simulation study that this method may be effective in overcoming some of the shortcomings of the other approaches.     

Adaptation of Learning Rule Parameters Using a Meta Neural Network

This paper proposes an application-independent method of automating learning rule parameter selection using a form of supervisor neural network (NN), known as a meta neural network (MNN), to alter the value of a learning rule parameter during training. The MNN is trained using data generated by observing the training of a NN and recording the effects of the selection of various parameter values. The MNN is then combined with a normal learning rule to augment its performance. Experiments are undertaken to see how this method performs by using it to adapt a global parameter of the resilient backpropagation and quickpropagation learning rules.     

Dealing with missing values in neural network-based diagnostic systems

Backpropagation neural networks have been applied to prediction and classification problems in many real world situations. However, a drawback of this type of neural network is that it requires a full set of input data, and real world data is seldom complete. We have investigated two ways of dealing with incomplete data — network reduction using multiple neural network classifiers, and value substitution using estimated values from predictor networks — and compared their performance with an induction method. On a thyroid disease database collected in a clinical situation, we found that the network reduction method was superior. We conclude that network reduction can be a useful method for dealing with missing values in diagnostic systems based on backpropagation neural networks.     

A decision support tool for allocating hospital bed resources and determining required acuity of care

Limitations in health care funding require physicians and hospitals to find effective ways to utilize resources. Neural networks provide a method for predicting resource utilization of costly resources used for prolonged periods of time. Injury severity knowledge is used to determine the acuity of care required for each patient and length of stay is used to determine duration of inpatient hospitalization. Neural networks perform well on these medical domain problems, predicting total length of stay within 3 days for pediatric trauma (population mean and S.D. 4.37±45.12) and within 4 days for acute pancreatitis patients (7.75±79.19).     

A comparison of backpropagation and general regression neural networks in quantative trace metal analysis using anodic stripping voltammetry and cybernetic instrumentation

A comparison is made between backpropagation and general regression neural networks for the prediction of parts per billion lead concentration when used to process data obtained from digested curry powder by the electrochemical analysis method of differential pulse, anodic stripping at a thin film mercury electrode (TFME). Two data sets are used, one requiring the net to classify an unknown analytical data vector into one of a number of previously learnt concentrations, and one requiring the net to predict the probable concentration of an unknown sample by interpolation of the already learnt concentrations. For both of these data sets the general regression neural network is shown to train faster and to provide results superior to those obtained by backpropagation.     

Evaluation of the use of sublevel open stoping in the mining of moderately dipping medium-thick orebodies

The flow of blasted ore during mining of moderately dipping medium-thick orebodies is a challenge. Selecting a suitable mining system for such ore bodies is difficult. This paper proposes a diamond layout sublevel open stoping system using fan blastholes with backfilling to mine such orebodies. To evaluate the performance of system the relationships between ore recovery and stope footwall dip angle, footwall surface roughness, drawpoint spacing and production blast ring burden were investigated. An ore recovery data set from 81 laboratory physical model experiments was established from combinations of the listed factors. Various modules in a back propagation neural network structure were compared, and an optimal network structure identified. An ore recovery backpropagation neural network (BPNN) forecast model was developed. Using the model and sensitivity analysis of the factors affecting the proposed open stope mining system, the significance of each factor on ore recovery was studied. The study results were applied to a case study at the Shandong Gold Group Jiaojia Gold Mine. The results showed that the application of a BPNN and sensitivity analysis models for ore recovery prediction in the proposed mining system and field experimental results confirm that the suggested mining method is feasible.     

Using neural networks for the diagnosis of localized defects in ball bearings

Two neural network based approaches, a multilayered feed forward neural network trained with supervised Error Back Propagation technique and an unsupervised Adaptive Resonance Theory-2 (ART2) based neural network were used for automatic detection/diagnosis of localized defects in ball bearings. Vibration acceleration signals were collected from a normal bearing and two different defective bearings under various load and speed conditions. The signals were processed to obtain various statistical parameters, which are good indicators of bearing condition, and these inputs were used to train the neural network and the output represented the ball bearing states. The trained neural networks were used for the recognition of ball bearing states. The results showed that the trained neural networks were able to distinguish a normal bearing from defective bearings with 100% reliability. Moreover, the networks were able to classify the ball bearings into different states with success rates better than those achieved with the best among the state-of-the-art techniques.     

Predicting product quality with backpropagation: A thermoplastic injection moulding case study

This application uses live data from a thermoplastic injection moulding manufacturer to examine the feasibility and effectiveness of using backpropagation artificial neural networks for predictive quality control. Preprocessing and post processing of live data, formulating neural predictive strategies, selecting architecture and parameters, and handling of temporal aspects are topics. Performance of the neural networks are compared to other quality control methods, including control charts and statistical techniques. This case study demonstrates that even manufacturers who have modest expertise in computing and limited hardware and software availability can successfully use neural networks for data analysis and modelling.     

Evolutionary training of hardware realizable multilayer perceptrons

The use of multilayer perceptrons (MLP) with threshold functions (binary step function activations) greatly reduces the complexity of the hardware implementation of neural networks, provides tolerance to noise and improves the interpretation of the internal representations. In certain case, such as in learning stationary tasks, it may be sufficient to find appropriate weights for an MLP with threshold activation functions by software simulation and, then, transfer the weight values to the hardware implementation. Efficient training of these networks is a subject of considerable ongoing research. Methods available in the literature mainly focus on two-state (threshold) nodes and try to train the networks by approximating the gradient of the error function and modifying appropriately the gradient descent, or by progressively altering the shape of the activation functions. In this paper, we propose an evolution-motivated approach, which is eminently suitable for networks with threshold functions and compare its performance with four other methods. The proposed evolutionary strategy does not need gradient related information, it is applicable to a situation where threshold activations are used from the beginning of the training, as in “on-chip” training, and is able to train networks with integer weights.