Process monitoring with economic performance functions: Feasibility assessment for milling circuits

Siwella, Mollin (2017-12)

Thesis (MEng)--Stellenbosch University, 2017.

Thesis

ENGLISH SUMMARY: Milling circuits are operated with a throughput maximisation objective, consistent with achieving a liberation index that promotes optimal mineral recovery in downstream flotation. To meet this objective, this study investigates the feasibility of developing and using economic performance functions to monitor milling circuits. To this end, industrial and simulation cases were studied. The industrial case study investigated the development of a reliable economic performance function (EPF) that relates controlled variable behaviour to a profit function. To this end, regression models were developed between key controlled variables (mill load, mass pull and particle size) and each of two profit functions, i.e. mineral recovery and financial profit. Subsequently, the development and industrial implementation of a simple and convenient process monitoring tool that assists operators and engineers to make operational decisions based on economic performance predictions was to be assessed. Since fault conditions are unfavourable for profitable operation, the simulation case study assessed the feasibility of fault detection with EPFs. Three faults i.e. increased ore hardness, poor steel ball quality and a load cell drift were simulated and the economic impact of each fault was investigated. Results for the industrial case study showed difficulty to develop a reliable EPF with industrial data. All but four EPF models were poorly fit and showed an adjusted R-squared value below a selected threshold of 0.6. However, the four identified models were characterised with poor predictions of plant test data. Consequently, the results were not useful for developing a process monitoring tool that could be implemented at the industrial operation. A number of factors such as data quality, data pre-treatment and the model structure used in the study influenced industrial case study results. To mention a few of these factors, results suggested robust model predictive controller action that rendered data variability insufficient for EPF development purposes. Indications of poorly structured data due to measurement lags, and disturbances in data sequences from faulty data treatment were additional limitations that influenced the quality of results obtained. Moreover, a low sampling frequency for assay composite samples may have contributed to missing in-between shift events. Simulation case study results showed degraded economic performance at fault inception for two (increased ore hardness and load cell drift) of the simulated faults, to suggest opportunity for fault detection with EPFs. Significance test results pointed to at least one difference in the economic performance indices (EPIs) for the simulated faults. Although not fully explored in this study, significance test results suggested opportunity for fault prioritisation with the EPI that allows decisions about the corrective action to be made based on the severity of the impact on economic performance. Since reliable EPF development was the main limitation in this study, a comparative assessment in a different operation with well-structured data is recommended. Fault prioritisation with the EPI may also be an area of interest for future work. However, the shortcomings identified in some of the simplifying assumptions made when deriving the EPI will need to be addressed.

AFRIKAANS OPSOMMING: Maalkringe het deurvoermaksimalisering ten doel, gegee dat ʼn vrylatingsindeks wat optimale mineraalwinning in stroomafflottering verkry word. Met hierdie doelwit voor oë is hierdie studie gemik daarop om die haalbaarheid van die ontwikkeling en gebruik van ekonomiese prestasiefunksies om ʼn maalkring te monitor. Daarom is nywerheids- en simulasiegevalle bestudeer. Die nywerheidsgevallestudie het die ontwikkeling van ʼn betroubare ekonomiese prestasiefunksie (EPF) wat beheerdeveranderlike-gedrag korreleer met ‘n winsfunksie ten doel. Daarom is regressiemodelle tussen sleutel beheerdeveranderlikes (meullading, massavloei, en partikelgrootte) en elk van twee winsfunksies, naamlik mineraalherwinning en finansiële wins, ontwikkel. Vervolgens is die ontwikkeling en nywerheidsimplementering van ʼn eenvoudige en geskikte prosesmoniteringsinstrument wat operateurs en ingenieurs help om operasionele besluite op grond van voorspellings van ekonomiese prestasie te neem, geassesseer. Aangesien fouttoestande nadelig vir winsgewende werking is, het die simulasiegevallestudie die haalbaarheid van foutspeuring deur middel van EPF’s bepaal. Drie foute, naamlik verhoogde ersthardheid, swak staalbalkwaliteit en ʼn ladingselafwyking, is gesimuleer en die ekonomiese impak van elke fout is ondersoek. Die resultate van die nywerheidsgevallestudie het op ʼn paar probleme met die ontwikkeling van ʼn betroubare EPF met nywerheidsdata gedui. Alle EPF modelle, uitgesluit vier, het swak passing getoon, met ʼn aangepaste R-kwadraat waarde van minder as die gekose limiet van 0.6. Die vier geïdentifiseerde modelle is gekenmerk deur die swak voorspelling van aanlegdata. Die resultate was dus nie bruikbaar vir die ontwikkeling van ʼn prosesmoniteringsinstrument wat by die nywerheidsproses geïmplementeer kon word nie. ʼn Aantal faktore, soos datakwaliteit, datavoorverwerking en die modelstruktuur wat gebruik is, het die resultate van die nywerheidsgeval beïnvloed. Om ʼn paar van die faktore te noem, die resultate van die nywerheidsgevallestudie het gedui op robuuste model-voorspellende-beheerderaksie, wat dataveranderlikheid onvoldoende vir die doeleindes van EPF-ontwikkeling gemaak het. Aanduidings van swak gestruktureerde data as gevolg van metingsagterstande sowel as steurings in datareekse as gevolg van foutiewe datahantering was verdere beperkinge wat die gehalte van die verkrege resultate beïnvloed het. Daarby kon ʼn lae steekproeffrekwensie vir saamgestelde komposisiemonsters daartoe bygedra het dat tussenskofvoorvalle oor die hoof gesien is. Resultate van die simulasiegevallestudie het verlaagde ekonomiese prestasie met die ontstaan van die fout vir twee die gesimuleerde foute (verhoogde ersthardheid en ʼn ladingselafwyking) getoon, wat geleentheid vir foutspeuring met EPF’s voorstel. Betekenistoetsresultate het gedui op ten minste een verskil in die ekonomiese prestasie-indekse (EPI’s) vir die gesimuleerde foute. Alhoewel dit nie ten volle in hierdie studie ondersoek is nie, het betekenistoetsresultate die geleentheid vir foutprioritisering met die EPI voorgestel wat dit moontlik maak om besluite oor die nodige regstellende optrede te neem op grond van die graad van die impak op ekonomiese prestasie. Aangesien betroubare EPF-ontwikkeling die hoofbeperking in hierdie studie was, word ʼn vergelykende assessering in ʼn ander proses met goed gestruktureerde data aanbeveel. Foutprioritisering met die EPI kan ook ʼn area vir toekomstige navorsing wees. Die tekortkominge wat geïdentifiseer is in sommige van die vereenvoudigende aannames wat met die ontwikkeling van die EPI gemaak is, sal egter steeds aandag moet geniet.

Please refer to this item in SUNScholar by using the following persistent URL: http://hdl.handle.net/10019.1/102992
This item appears in the following collections: