Development of a human digital twin for human-centric dispatching for an assembly process

Date
2023-12
Journal Title
Journal ISSN
Volume Title
Publisher
Stellenbosch : Stellenbosch University
Abstract
ENGLISH ABSTRACT: Assembly workers experience higher stress levels than other positions. Reducing the stress of assembly workers is important for workers and companies as it has an impact on long-term health, employee commitment to the company and job satisfaction. The research presented in this thesis develops a new approach to strain reduction using an assembly process. Considering workers' preferences increases their autonomy and control over their work, which is expected to reduce strain. This is implemented in worker dispatching for an assembly process. The approach presented in this thesis considers the individual worker and their preferences regarding the type of task in the dispatching procedure. In addition, machine learning is used to predict the worker's strain in a particular task to evaluate the worker with the least expected strain in that task from a group of workers. By incorporating preferences, worker autonomy can be increased. Combined with strain prediction using machine learning, which incorporates individual worker strain based on various factors into the dispatching system, worker strain can be reduced. A human-centric dispatching system that includes a human digital twin as a digital representation of the individual worker is developed. The human digital twin consists of three elements: data storage, strain prediction and strain monitoring. The worker attributes included in the data storage are availability, age, assembly competence, preference regarding the type of task, and worker strain using heart rate as an objective measure. The worker's strain is predicted using task-specific training data from three scooter assembly workstations at the "Werk 150" logistics learning factory at the University of Reutlingen, Germany. The worker attributes included in the training data for strain prediction are task type, age, time of day, preference, and assembly skill. A Random Forest regressor trained on the augmented dataset 1 is used to predict the median heart rate. The mean average error is 5,64 beats per minute and the deviation between predicted and test values is 39,66 %. The developed dispatching procedure considering preferences and including strain prediction is evaluated in Werk 150. The field experiment indicated that using the developed human-centric dispatching system including the worker's human digital twin leads to a decrease in strain. Using the NASA TLX as a subjective strain measurement, the average worker strain decreases by 27 % measured with the NASA RTLX and by 33 % measured with the weighted NASA TLX across all three assembly tasks and subscales. Using heart rate as an objective strain measurement, the strain decreases by 42,27 % when measuring mean heart rate and by 41,08 % for median heart rate. Compared to random task assignment, the developed human-centric dispatching system, therefore, reduces worker strain on average by 35,84 %.In conclusion, considering the preferences of the workers in dispatching and combining this with strain prediction leads to strain reduction. This is a starting point for increased human-centricity in production and a step forward in the implementation of the Industry 5.0 concept.
AFRIKAANSE OPSOMMING: Monteerwerkers ervaar hoer stresvlakke as ander pos1s1es. Die vermindering van die stres van monteerwerkers is belangrik vir werkers en maatskappye, aangesien dit 'n impak op langtermyngesondheid, werknemer verbintenis tot die maatskappy en werksatisfaksie het. Die navorsing wat in hierdie tesis aangebied word, ontwikkel 'n nuwe benadering tot spanningvermindering vir 'n samestellingsproses. Deur werkers se voorkeure in ag te neem, word hul outonomie en beheer oor hul werk verhoog, wat na verwagting spanning sal verminder. Dit word gei:mplementeer vir werkersversending vir 'n monteerproses. Die benadering wat in hierdie tesis aangebied word, neem die individuele werker en hul voorkeure ten opsigte van die tipe taak in die versendingsproses in ag. Daarbenewens word masjienleer gebruik om die werker se spanning in 'n bepaalde taak te voorspel om die werker met die minste verwagte spanning in daardie taak van 'n groep werkers te evalueer. Deur voorkeure in te sluit, kan 'n werker se outonomie verhoog word. Gekombineer met spanningvoorspelling deur middel van masjienleer, wat individuele werkerspanning gebaseer op verskeie faktore in die versendingstelsel insluit, kan werkerspanning verminder word. 'n Mensgesentreerde versendingstelsel is ontwikkel wat 'n menslike digitale tweeling insluit as 'n digitale voorstelling van die individuele werker. Die menslike digitale tweeling bestaan uit drie elemente: Databerging, spanningvoorspelling en spanningmonitering. Die kenmerke van die werker wat by die databerging ingesluit is, is beskikbaarheid, ouderdom, samestellingsbevoegdheid, voorkeur met betrekking tot die tipe taak en werkerspanning deur hartklop as 'n objektiewe maatstaf te gebruik. Die werker se spanning word voorspel met behulp van taakspesifieke opleidingsdata van drie tweewiel skopfiets monteerstasies by die "Werk 150" logistieke leerskool by die Universiteit van Reutlingen, Duitsland. Die kenmerke wat by die opleidingsdata vir spanningvoorspelling ingesluit is, is taaktipe, ouderdom, tyd van die dag, voorkeur en monteervaardigheid. 'n Random Forest­ regressor wat op die uitgebreide datastel 1 opgelei is, word gebruik om die mediaan hartklop te voorspel. Die gemene gemiddelde fout is 5,64 slae per minuut en die afwyking tussen voorspelde en toetswaardes is 39,66 %. Die ontwikkelde versendingsprosedure met inagneming van voorkeure en insluiting van spanningvoorspelling is in Werk 150 geevalueer. Die laboratoriumeksperiment het aangedui dat die gebruik van die ontwikkelde mensgesentreerde versendingstelsel, insluitend die menslike digitale tweeling van die werker, tot 'n spanningsafname lei. Deur die NASA TLX as 'n subjektiewe spanningmeting te gebruik, verminder die gemiddelde werkerspanning met 27 % gemeet met die NASA RTLX en met 33 % gemeet met die geweegde NASA TLX oor al drie monteertake en subskale. Deur hartklop as 'n objektiewe spanningsmeting te gebruik, verminder die spanning met 42,27 % wanneer gemiddelde hartklop gemeet word en met 41,08 % vir mediaan hartklop. In vergelyking met ewekansige taakopdrag, verminder die ontwikkelde mensgesentreerde versendingstelsel dus werkerspanning met gemiddeld 35,84 %.
Description
Thesis (MEng)--Stellenbosch University, 2023.
Keywords
Citation