Process monitoring and fault diagnosis using random forests

[摘要] ENGLISH ABSTRACT: Fault diagnosis is an important component of process monitoring, relevant in the greater context of developingsafer, cleaner and more cost efficient processes. Data-driven unsupervised (or feature extractive) approachesto fault diagnosis exploit the many measurements available on modern plants. Certain current unsupervisedapproaches are hampered by their linearity assumptions, motivating the investigation of nonlinear methods.The diversity of data structures also motivates the investigation of novel feature extraction methodologies inprocess monitoring.Random forests are recently proposed statistical inference tools, deriving their predictive accuracy from thenonlinear nature of their constituent decision tree members and the power of ensembles. Random forestcommittees provide more than just predictions; model information on data proximities can be exploited toprovide random forest features. Variable importance measures show which variables are closely associatedwith a chosen response variable, while partial dependencies indicate the relation of important variables to saidresponse variable.The purpose of this study was therefore to investigate the feasibility of a new unsupervised method based onrandom forests as a potentially viable contender in the process monitoring statistical tool family. Thehypothesis investigated was that unsupervised process monitoring and fault diagnosis can be improved byusing features extracted from data with random forests, with further interpretation of fault conditions aided byrandom forest tools. The experimental results presented in this work support this hypothesis.An initial study was performed to assess the quality of random forest features. Random forest features wereshown to be generally difficult to interpret in terms of geometry present in the original variable space. Randomforest mapping and demapping models were shown to be very accurate on training data, and to extrapolateweakly to unseen data that do not fall within regions populated by training data.Random forest feature extraction was applied to unsupervised fault diagnosis for process data, and comparedto linear and nonlinear methods. Random forest results were comparable to existing techniques, with themajority of random forest detections due to variable reconstruction errors. Further investigation revealed thatthe residual detection success of random forests originates from the constrained responses and poorgeneralization artifacts of decision trees. Random forest variable importance measures and partialdependencies were incorporated in a visualization tool to allow for the interpretation of fault conditions.A dynamic change point detection application with random forests proved more successful than an existingprincipal component analysis-based approach, with the success of the random forest method again residing inreconstruction errors.The addition of random forest fault diagnosis and change point detection algorithms to a suite of abnormalevent detection techniques is recommended. The distance-to-model diagnostic based on random forestmapping and demapping proved successful in this work, and the theoretical understanding gained supports theapplication of this method to further data sets.

[发布机构] Stellenbosch University