Exploratory Analysis of Metallurgical Process Data with Neural Networks and Related Methods

1. Introduction to neural networks. 1.1 Background. 1.2 Artificial neural networks from an engineering perspective. 1.3 Brief history of neural networks. 1.4 Structures of neural networks. 1.5 Training rules. 1.6 Neural, network models. 1.7 Neural networks and statistical models. 1.8 Applications in the process industries. 2. Training of neural networks. 2.1 Gradient descent methods. 2.2 Conjugate gradients. 2.3 Newton's method and quasi-Newton method. 2.4 Levenberg-Marquardt algorithm. 2.5 Stochastic methods. 2.6 Regularization and pruning of neural network model. 2.7 Pruning algorithms for neural networks. 2.8 Constructive algorithms for neural network. 3. Latent variable methods. 3.1 Basics of latent structure analysis. 3.2 Principal component analysis. 3.3 Nonlinear approaches to latent variable extraction. 3.4 Principal component analysis with neural networks. 3.5 Example 2: feature extraction from digitised images of industrial flotation froths with auto associative neural networks. 3.6 Alternative approaches to nonlinear principal component analysis. 3.7 Example 1: low-dimensional reconstruction of data with non linear principal component methods. 3.8 Partial least squares (PLS) models. 3.9 Multivariate statistical process control. 4. Regression models. 4.1 Theoretical background to model development. 4.2 Regression and correlation. 4.3 Multicollinearity. 4.4 Outliers and influential observations. 4.5 Robust regression models. 4.6 Dummy variable regression. 4.7 Ridge regression. 4.8 Continuum regression. 4.9 Case study: calibration of an on-line diagnostic monitoring system for comminution in a laboratory-scale ball mill. 4.10 Nonlinear regression models. 4.11 Case study 1: modelling of a simple bimodal function. 4.12 Nonlinear modelling of consumption of an additive in a gold I-each plant. 5. Topographical mappings with neural networks. 5.1 Background. 5.2 Objective functions for topographic maps. 5.3 Multidimensional scaling. 5.4 Sammon projections. 5.5 Example 1: artificially generated and benchmark data sets. 5.6 Example 2: visualization of flotation data from a base metal flotation plant. 5.7 Example 3: monitoring of a froth flotation plant. 5.8 Example 4: analysis of the liberation of gold with multi-dimensionally scaled maps. 5. 9 Example 4: monitoring of metallurgical furnaces by use of topographic process maps. 6. Cluster analysis. 6.1 Similarity measures. 6.2 Grouping of data. 6.3 Hierarchical cluster analysis. 6.4 Optimal partitioning (K-means clustering). 6.5 Simple examples of hierarchical and K-means cluster analysis. 6.6 Clustering of large data sets. 6.7 Application of cluster analysis in process engineering. 6.8 Cluster analysis with neural networks. 7. Extraction of rules from data with neural networks. 7.1 Background. 7.2 Neurofuzzy modeling of chemical process systems with ellipsoidal radial basis function neural networks and genetic algorithms.7.3 Extraction of rules with the artificial neural network decision tree (ANN-DT) algorithm. 7.4 The combinatorial rule assembler (CORA) algorithm.7.5 Summary. 8. Introduction to the modelling of dynamic systems. 8.1 Background. 8.2 Delay coordinates. 8.3 Lag or delay time. 8.4 Embedding dimension. 8.5 Characterization of attractors. 8.6 Detection of nonlinearities. 8.7 Singular spectrum analysis. 8.8 Recursive prediction. 9. Case studies: dynamic systems analysis and modelling. 9.1 Effect of noise on periodic time series. 9.2 Autocatalysis in a continuous stirred tank reactor. 9.3 Effect of measurement and dynamic noise on the identification of an autocatalytic process. 9.4 Identification of an industrial platinum flotation plant by use of singular spectrum analysis and delay coordinates. 9.5 Identification of a hydrometallurgical process circuit. 10. Embedding of multivariate dynamic process systems. 10.1 Embedding of multivariate observations. 10.2 Multidimensional embedding methodology. 10.3 Application of the embedding method. 10.