Multivariate real-time monitoring using principal component analysis and projection of latent structures.

Methods such as principal component analysis, projection to latent structures and their generalisations work very well for the modelling and analysis of the large and complex data sets generated by modern industrial processes such as mining. The methods deal with multitudes of colinear, noisy and incomplete data and give easily interpretable results such as scores, deviations from the model and variable contributions. To use projection methods for process monitoring, a model for the well-functioning process is first developed from historical data and then put on line. The resulting multivariate control charts indicate good or poor functioning and the presence of faults. The charts are traditional ones, but instead of showing the values of individual variables they display scores that summarise all or large parts of the process variables in the optimum manner. Contribution plots are used for interpreting faults and finding likely causes.
Methods such as principal component analysis, projection to latent structures and their generalisations work very well for the modelling and analysis of the large and complex data sets generated by modern industrial processes such as mining. The methods deal with multitudes of colinear, noisy and incomplete data and give easily interpretable results such as scores, deviations from the model and variable contributions. To use projection methods for process monitoring, a model for the well-functioning process is first developed from historical data and then put on line. The resulting multivariate control charts indicate good or poor functioning and the presence of faults. The charts are traditional ones, but instead of showing the values of individual variables they display scores that summarise all or large parts of the process variables in the optimum manner. Contribution plots are used for interpreting faults and finding likely causes.